ABSTRACT. Drawing on the theories of knowledge recombination, this study aims to unpack the essential tension between tradition and innovation in scientific research. Using the American Physical Society data and computational methods, we analyze the impact of knowledge spanning on both citation counts and disruptive innovations. The findings show that: First, there is an inverted U-shaped relationship between knowledge spanning and citation counts, and this nonlinear effect is moderated by team size. Second, knowledge spanning has a positive linear impact on disruptive innovation, and its effect is also moderated by team size. Third, there is a U-shaped relationship between disruption and citation counts. This study contributes to the theories of knowledge recombination by suggesting that when evaluating the quality of scientific research with disruptive innovations, the essential tension disappears.

ABSTRACT. With the continuous maturity and development of image classification, the anti-noise ability of the model has become an important research topic. This study proposes a (natural and horizontal) viewable texture image classification method with robustness to noise. Firstly, by adding different intensities and different types of Gaussian and salt and pepper noise to each Brodatz texture image in the data set, each Brodatz texture image in the data set is converted into corresponding image natural visibility graph (INVG) and image horizontal visibility graph (INVG). Then, the degree distribution P (k) is extracted and input into different classifiers. Principal component analysis (PCA) is applied to the input vector to reduce the number of features and avoid final over fitting, and a 5-fold cross validation procedure is adopted. The results show that the proposed method has better anti-noise ability, and the classification accuracy is higher than that of the relevant comparative literature.

ABSTRACT. Dimension governs dynamical processes on networks. The social and technological networks which we encounter in everyday life span a wide range of dimensions, but studies of spreading on finite-dimensional networks are usually restricted to one or two dimensions. To facilitate investigation of the impact of dimension on spreading processes, we define a flexible higher-dimensional small world network model and characterize the dependence of its structural properties on dimension. Subsequently, we derive mean field, pair approximation, intertwined continuous Markov chain and probabilistic discrete Markov chain models of a COVID-19-inspired susceptible-exposed-infected-removed (SEIR) epidemic process with quarantine and isolation strategies, and for each model identify the basic reproductive number, which determines whether an introduced infinitesimal level of infection in an initially susceptible population will shrink or grow. We apply these four continuous state models, together with discrete state Monte Carlo simulations, to analyse how spreading varies with model parameters. We observe discrepancies from Monte Carlo which are consistent across competing continuous state models, which we characterise by simultaneously considering network efficiency and network size.

ABSTRACT. Heart disease is a worldwide public health problem related to morbidity and mortality. With the rapid development of computer-aided techniques, various machine learning methods have been emerged to predict and diagnose heart disease. Thus, developing an effective predictive model to accurately identify heart patients is a hotspot in medical research currently. This study proposes a hybrid method DBN-RF, which utilizes deep belief network to mine the inner patterns of heart disease and embeds random forest to perform a supervised fine-tuning on network parameters, as shown in Fig.1. The proposed method is evaluated on a combined heart disease dataset containing 11 clinical features. The results indicate that our proposed method outperforms other state-of-the-art methods in terms of accuracy, recall, precision and F1-score. Finally, we detect and describe important features with tangible influence on heart disease.

ABSTRACT. The advancement of globalization has led to increasingly close exchanges between countries, and in this environment, it is crucial for countries to grasp the development motivations of other countries in time to seize development opportunities. Motivation, as an intrinsic drive that determines behavior, is abstract in nature, it is difficult to measure and assess motivation itself directly. To address this problem, this paper improves Friedkin's social influence model that has been applied to motivation descriptions, constructs dynamic alliance networks and hostile networks and dynamic sensitivity matrices based on open-source data with China and the United States as the core, to remotely assess the process of motivation change of countries under the influence of international relations. Taking integrated circuits as the analysis case, it is found that after evolution, most countries' motivations for development shows a decreasing trend, while the United States maintains the highest motivation due to its stronger power, and China's motivation continues to rise and maintain an upward trend in recent years; finally, the evolution results are compared with the real data to find that this dynamic motivation model can reflect the motivation changes of countries to a certain extent.

ABSTRACT. Link prediction is a classic complex network analytical problem to predict the possible links according to the known network structure information. Considering similar nodes should present closer embedding vectors with network representation learning, in this paper we propose a graph attention network method based on node similarity (SiGAT) for link prediction. Firstly, we calculate similar nodes in the networks by traditional method—Jaccard index and obtain new links for original links. Then, the new links be assigned an optimal weight through the graph attention network mechanism, that is SiGAT. SiGAT employs network embedding to characterize a pair of nodes associations and then trains a classifier to predict novel potential network links. We can obtain more accurate node embeddings by aggregating features from added neighbors. Finally, the results of experiments on four datasets, such as the Yeast dataset, Cora dataset, BIO-CE-HT dataset and Human proteins (Vidal) dataset, show that SiGAT outperforms existing popular approaches.

ABSTRACT. Exploration behavior is an important research topic in behavioral neuroscience. Most analyses on trajectories of animal exploratory movements are often confined on local patterns defined by spatial and behavioral features, therefore neglect global structure underlying the whole sequence of movements. We develop a novel framework for the analysis of exploration trajectories based on complex network theory. The framework consists of three steps: (1) Quantifying the behavior changes by a vector of speed and curvature, and using a line simplification algorithm to split the animal trajectories into different segments according to speed and curvature; (2) Learning the main graph of the segments based on the method of reverse graph embedding to visualize the spatial changes of each segment; (3) Based on the temporal information and the learning results above-mentioned, building a complex network model with nodes containing both animal behavioral and spatial information, and analyzing the statistics of the whole network and communities to acquire the trajectory’s local and global characteristics. Applying our approach to analyze the trajectories of mice exploring an open field environment, we identify four categories of behavior patterns along with the transition information between them during exploration. Our results are more comprehensive and systematic as compared with those in previous studies, and thus indicates that complex network is a sensible tool for the analysis of trajectory data.

ABSTRACT. Diversity is one of the typical characteristics of scientific research. For example, at the individual level, a scientist often has diverse and constantly updated research interests. At the group level, a team with more diverse members in terms of gender, age, discipline, background, etc. tends to have stronger academic creativity and impact. In order to explore the relationship between the diversity of a scientist’s collaborative relationships and his academic output and impact, we conducted an analysis based on historical publication data from journals of the American Physical Society (APS). Firstly, the number of published papers is used to measure the output of a scientist, and normalized C10 is used to measure the cumulative citations of all papers published by a scientist within 10 years after the paper is published. Then, we use the author information of a paper to construct the scientist collaboration network, and the node degree, clustering coefficient and k-core to measure the topological structure characteristics of scientists in the network. Based on the above three indicators, the Shannon entropy is calculated to measure the diversity of scientists' collaborators. Our study found that from the perspective of individuals, if a scientist has more collaborators or is in a more core position in the collaboration network, he often has stronger academic impact. But such a positive correlation does not apply to the clustering coefficient, that is, scientists with small clustering coefficient in the network have greater academic influence. From the perspective of diversity, Shannon entropy based on degree, clustering coefficient and k-core has a bell-shaped nonlinear relationship with C10. In other words, there is an optimal range for the diversity of scientists' collaborators, that is, about 0.5-0.7 (0.5 for k-core, 0.6 for degree and 0.7 for clustering coefficient). Such results suggest that, pursuing too many or refusing to extend collaborators in scientific research is not conducive to a scientist to improve his academic influence. Our research may provide some reference for young scientists to find partners and form research teams.

ABSTRACT. In this paper, the distributed state estimation problem of continuous-time linear time-invariant (LTI) systems with input disturbance is considered. A group of agents communicating through an undirected graph are used to measure the outputs of the target system, where each one can only access a portion of the outputs. A big challenge in the distributed state estimation problem is induced by the disturbance and the partial output information measurement. First, with the observer structure decomposition (OSD), a high-gain observer is designed for each agent to robust the input disturbance to compensate the observable portion of the state. Second, a consensus protocol with time-varying nodal coupling gain is introduced for each agent to compensate the un-observable portion of the state, where the time-varying nodal coupling gain is designed with the help of the node-based adaptive strategy. It is found that a sufficient small high-gain observer parameter and a connected communication among the multiple agents can lead to a robust distributed observer of the target system, which can realize the distributed state estimation on the target system robustly.

ABSTRACT. We quantitatively uncovered the scale-free and small-world properties of the wind-turbine trade network(WTN) by analyzing the structure evolution. The WTN is largely dominated by the key countries from Europe, America, and Asia. From the time evolution analysis, the structural differentiation of WTN leads to the consistent influence weakening of the North America cluster. The unparalleled influence of European countries and raising of Asia, especially China, are the principal characters of the WTN in the recent decades. Further review the backbone network by filtering many noisy trade links revealed the sparsification of the WTN with the increase of the filtering threshold and core countries are Denmark, Germany, Spain, the USA, and China. The node removal treatment and supply shortage simulation schemes are used to study the robustness of WTN. The WTN is found to be much more stable when facing random trade conflict compare to target trade conflict, which depicts the pivotal role of the highly connected countries. Under the homogeneous sensitivity scenario, the complex interactions among countries lead to the large-scale contagion of the supply shortage and further devastate the functional integrity of the WTN. The overreacting of key countries are very likely to induce the large-scale breakdown. Core countries such as Denmark, Germany, Spain, Netherlands, China, and the USA are crucial for the risk transmission and their mutual effects through the export cut-off will be re-amplified and further undermine the structure of the WTN. In order to alleviate the aftermath of the large-scale disruption, we incorporate the heterogeneous sensitive parameter setting. The simulation result shows that the contagion risk of the supply shortage can be suppressed conspicuously. The study about the structure, evolution, vulnerability, and the risk spreading mitigate strategy for the WTN have important practical implications for the sustainable energy governance of global governments.

ABSTRACT. Forest fires are becoming more and more common disasters, we have built a Behavior fire network model which suit for people's agricultural activities and residential behaviors disturbances to the forest complex network system in China. Come from graph theoretic model, forest fire complex network define as node, link, pattern, node’s degrees distribution, etc. Forests are carbon sinks and precursors to Earth's fossil fuels. The canopy density value Cfire of the fire-prone forest is generally between the highest value Cmax and the lowest value Cmin of the forest canopy density value Cx. Take discrete combustibles (fuel) as nodes (Vertices), the lengths of the most combustibles (Lenths) as weighted edges (Edge), and take the FMC value as the weight value (Weight) to construct the fire dynamic evolution and spread of the forest fires complex network of the processing. This model can be used for microscopic, single fire spread speed, and can also be used for large-scale, macroscopic forest fire control. Fire statistics experiments show that the data of network model based on the FMC (Fuel Moisture Content) of combustibles and topographic and meteorological factors are superior to the traditional Rothemel physical parameter model. Dynamic model of fire network modeling are describe on following: If the combustible node Vo is artificially ignited, according to the principle of the most flammable (O-D shortest distance algorithm principle), the forest fire will proceed along the shortest distance path until the next combustible node Vd. If the FMC of the Vn combustible material is too high at this time, the heat energy of the fire point will be consumed, so the fire point cannot continue to promote combustion, at last, the fire speed is zero and the fire point will disappear automatically. If the FMC of Vn is too low, the fire point will continue to advance along the path of FMC min until all combustibles are consumed. If we want to control Fire network then we use the four steps: Crowd Control, Node control, Fire path control and edge connection control, and FMC control. Lastly, the mode of fire behavior Fire Network Modeling can be predict fire disaster and digital ecology model。

ABSTRACT. In making great strides to the machine learning-based era in artificial intelligence, are we really making much progress? It appears that researchers generally face the dichotomies on making a trade-off between the precision and scalability of machine learning-based frameworks and the explainability and hyperparameter-free of conventional ones. Faced with this situation, interdisciplinary studies seem to be a promising future. In terms of the field of recommender systems, a significant application of artificial intelligence, this article reveals the flaw of ProbS (a prevalent conventional recommendation framework) in self-adaptive perception towards different recommendation scenarios. To make up for the flaw, this article proposes two nodal feature generation methods for representing users and items on the advice of the essential thought from machine learning-based recommendation frameworks, which are hyperparameter-free, explainable and scenario-oriented. Based on the ProbS framework and the generated features, this article proposes the Adaptive and Interpretable ProbS (AIProbS) model, a recommendation model with self-adaptive perception, quantification and control abilities and with precise, hyperparameter-free and explainable instincts. To evaluate its performance on diverse real recommendation scenarios, this article designs control experiments, revealing that the AIProbS can achieve state-of-the-art performance on model precision, compared with baseline models of both conventional and machine learning-based frameworks.

ABSTRACT. Deep Neural Networks (DNN) are known to be vulnerable to adversarial samples, the detection of which is crucial for the wide application of these DNN models. There are various differences between clean and adversarial samples, from the input space to the model hidden layer, which has been utilized to expose adversarial samples. However, most of these detection methods are limited to a single model and adversarial features exposed on it, making it vulnerable to adaptive attacks. To address the problem, we propose Graph-Guided Testing (GGT), which is a multiple-model based detection algorithm that robust to adaptive attacks. GGT generates diverse models with the guide of graph characteristics, and it detects adversarial samples by their instability on the multi-model decision boundaries. GGT is efficient since the generated model has only about 5% floating point operations of the original model. The experiments validate that GGT performs much better than other state-of-the-art methods against adaptive attacks.

ABSTRACT. Percolation theory has been widely used to study phase transitions in network systems. It has also successfully explained various macroscopic spreading phenomena across different fields. Yet, the theoretical frameworks have been focusing on direct interactions among nodes, while recent empirical observations have shown that indirect interactions are common in many network systems like social and ecological networks, among others. By investigating the detailed mechanism of both direct and indirect influence on scientific collaboration networks, here we show that indirect influence can play the dominant role in behavioral influence. To address the lack of theoretical understanding of such indirect influence on the macroscopic behavior of the system, we propose a new percolation mechanism of indirect interactions called induced percolation. Surprisingly, our new model exhibits unique anisotropy property. Specifically, directed networks show first order abrupt transitions as opposed to the second order continuous transition in the same network structure but with undirected links. A mix of directed and undirected links leads to rich hybrid phase transitions. Furthermore, a unique feature of non-monotonic pattern is observed in network connectivities near the critical point. We also present an analytical framework to characterize the proposed induced percolation, paving way to further understand network dynamics with indirect interactions.

ABSTRACT. In order to achieve the goal of carbon neutrality, China has proposed policies such as development capacity and double control of energy consumption. However, the problem of inconsistent implementation has led to tight coal supply, soaring coal prices and frequent power restrictions. Under the goal of carbon neutrality, the paper addresses the issue of government coal price regulation. We use complex networks and aggregation ranking methods to study the influence of coal prices in China provinces. Firstly, the paper calculates the transfer entropy between coal price nodes based on the time series data of coal prices in provinces. And then, the paper uses the sliding time window method to establish a coal price transfer entropy network. Secondly, the paper analyzes the network properties between coal price nodes, and finds the ranking of coal price influence of each province under different network indicators. Finally, the aggregation ranking method is applied to aggregate the coal price influence rankings of different network indicators to obtain a comprehensive coal price influence ranking, and the validity of the aggregation ranking is verified through relevant analysis. The paper's research will have important implications for China's macro-regulation of coal prices under the carbon neutrality target.

ABSTRACT. This paper mainly studies the parameter estimation problem of wireless sensor networks when sensing and communicating information in complex environments are subject to adversarial attacks. The wireless sensor network is an important support for distributed estimation in the actual environment, but the sensing and communication environment of the network is not absolutely safe. To alleviate the impact of adversarial attacks on the network, a novel AP-DLMS algorithm with adaptive threshold attack detection and malicious punishment strategy is proposed. In this algorithm, to detect the location of malicious attacks, the adaptive threshold is constructed by the observation matrix and the network topology, and then the estimated deviation of each node is obtained by designing the standard reference estimation. Meanwhile, in order to weaken the impact of data tampering on network performance, a punishment factor is designed to reduce the fusion weight of malicious nodes. Besides, we also propose a new attack model, namely the probabilistic random attack model. Finally, the influence of the key parameters in the adaptive threshold on the performance of the AP-DLMS algorithm is explored through simulation experiments, and the mean square performance of the proposed algorithm under the FDI continuous and time-sharing attack, Gaussian attack and probabilistic random attack models are analyzed respectively. The effectiveness of the algorithm is verified. The results show that the algorithm has better robustness in the adversarial network environment.

ABSTRACT. Higher-order network (HON) modeling shows a good capacity of effectively improving the accuracy of the representation of natural complex systems by reconstructing higher-order dependencies involving three or more components from time-resolved data. However, extracting higher-order dependencies from high-dimensional and time-stamped data would increase model complexity. Furthermore, referring to Ockham’s razor, it is of particular importance for higher-order network models to balance its power to represent complex systems against complexity. While this is an essential area of research, no research has contributed to a systematic study of evaluating the necessity of higher-order dependencies from data. Our work aims to propose a framework based on a hypothesis test for detecting higher-order dependencies: recurring, significant, dependent patterns. Through a comprehensive empirical evaluation and analysis, significant higher-order dependencies are obtained. Experiments in the APS (American Physical Society) citation dataset and synthetic web clickstream dataset demonstrate the validation of our method finally. Results on the APS citation dataset show that after citing articles published in more specialized journals, the knowledge flow of an article in a journal, such as Rev Mod Phys, tends to return to articles published in the same specialized journal or field. Results on synthetic web clickstream dataset show that assuming a 5% chance of getting a type 1 error, our framework could effectively capture higher-order user preference patterns. The framework provides statistical evidence for the importance and necessity of higher-order dependencies.

ABSTRACT. Information driven by different sources on emergent epidemic can typically motivate people to adopt behavioral responses, which contributes the reduction of infection risk, thus, suppressing the epidemic spread. However, most previous studies usually assume that the multi-source information and behavioral responses are constant, ignoring the dynamical evolution of epidemic and information, especially when individuals interact with each other dynamically. Here, we aim at studying the effects of dynamical multi-source information, e.g., self-awareness and information diffusion, and individual’s behavioral responses on the co-evolution of epidemic and information in time-varying multiplex networks. We propose the UAU-SIS (unaware-aware-unaware susceptible-infected-susceptible) model with time-varying self-awareness and behavioral responses. With a microscopic Markov chain approach, we derive the epidemic threshold for the proposed model. Experimental results show that timevarying behavioral responses can effectively suppress the epidemic spread with an increased epidemic threshold, while time-varying self-awareness can only reduce the outbreak size. In addition, the role of multi-source information in suppressing epidemic spread is limited. When the information transmission rate is beyond some critical value or the information efficiency is low, it will no longer affect the epidemic spread. Thus, our study is expected to provide helpful guidance for coping with the epidemic like COVID-19.

ABSTRACT. Flow networks describe the self-organization and the structure of evolving complex systems, representing systems out of thermodynamic equilibrium. The channels of the network forming a complex system are conducting flows driven by thermodynamic forces. Those channels are formed by doing work on the constraints to the motion of the flows, thus minimizing the time and energy for the flows along with the network, i.e. obeying the Principle of Least Action, upon which all motion in nature occurs. The flow paths are the paths of least action, with curvature representing a geodesic (shortest line in curved space) through the system. That curvature can be represented by the metric tensor from differential geometry, and the lowest possible action on the flow network (action efficiency) can be used as a numerical measure for its level of organization. We present a model of a flow network, its measure for action efficiency, and simple metric tensor descriptions. We follow the system through its reorganization (evolution) towards more action-efficient states. We study the power-law scaling relation (proportionality) between action efficiency and the size of the system. We use data for specific systems and computer simulations of Agent-Based Models of self-organization, forming the flow network and its dependency on the size of the system. As the simplest model, we use ant-based simulations on Python software. Our results confirm the size-complexity rule and the size-rate of increase of complexity rule, which means that the larger the systems are the faster they find the least action state of the flow network and the higher levels of action efficiency they achieve, which means higher levels of evolution and organization.

ABSTRACT. The distribution of the ratio of consecutive eigenvalue spacing of random matrices has emerged as an important tool to study spectral properties of many-body systems. We numerically investi- gates the eigenvalues ratio distribution of various model networks, namely, small-world, Erdős- Rényi random, and (dis)assortative random having a diagonal disorder in the corresponding adja- cency matrices. Without any diagonal disorder, the eigenvalues ratio distribution of these model networks depicts Gaussian orthogonal ensemble (GOE) statistics. Upon adding diagonal disorder, there exists a gradual transition from the GOE to Poisson statistics depending upon the strength of the disorder. The critical disorder (w c ) required to procure the Poisson statistics increases with the randomness in the network architectures. We relate w c with the time taken by maximum entropy random walker to reach the steady-state. These analyses will be helpful to understand the role of eigenvalues other than the principal one for various networks dynamics such as transient behaviour.

ABSTRACT. Team interaction networks can significantly affect team performance (Mason & Watts, 2012; Amelkin et al, 2018). To date, direct social learning (where edges help by sharing high-performing solutions) has obscured the effects of indirect social learning (where edges obfuscate individual contributions by sharing collective performance). In this work, we use models of greedy learners that only receive feedback about collective performance to isolate the performance effects of task structures and team networks. We demonstrate that tasks can favor teams with different network structures, particularly when future probabilities of finding solutions are tied to network properties. Furthermore, self-edges, high eigenvector centrality, and low degree centrality improve team performance across a wide variety of tasks. We call this performance benefit to certain network characteristics network intelligence. Our findings suggest that teams may benefit from adapting their structures on certain tasks to improve their performance, but that consistent individual feedback and distributed coordination responsibilities may provide teams with more enduring performance gains.

ABSTRACT. Hierarchies are the hidden backbones of complex systems, and their analysis allows for a deeper understanding of their structure and how they evolve. We consider languages to be also complex adaptive systems. Hence, we analyzed the hierarchical organization of historical syntactic networks from German that were created from a corpus of texts from the 11th to 17th centuries. We tracked the emergence of syntactic structures in these networks and mapped them to specific communicative needs. We named these emerging structures "communicative hierarchies", and we hypothesize that the communicative needs of speakers are the organizational force of syntax. Thus, we propose that the emergence of communicative hierarchies is what shapes syntax, and that these hierarchies are the prerequisite to the Zipf's law. The emergence of communicative hierarchies indicates that the objective of language evolution is not only to increase the efficiency of transferring information. Language is also evolving to increase our capacity to communicate more sophisticated abstractions as we advance as a species.

ABSTRACT. Complex rheology of colloidal gels arises from the evolution of microstructure that forms due to interactions between particles: colloids form bonds, which in turn form networks that govern the mechanics of colloidal gels. What is clear is that particles form clusters at mesoscale, and that elasticity in gels arises from a percolated network at the macroscale. Thus, understanding the physics of this particulate network is the key to controlling and designing gels with desirable properties. We borrow concepts from network science to characterize the particulate network in gels. We associate each colloidal particle to a vertex, and each inter-particle bond to an edge; we then proceed to the analysis of the networks of bonds, completely unaware of the spatial coordination of the particles, and to reveal structural signatures both at local and global scales. We employ network science tools to identify colloidal clusters and coarse grain the network further with distinct features for gels of different energies. The cluster networks from gels of different energy potential show distinct features, and a simple spring network made out of connecting the clusters recovered the mechanics of our colloidal gel, suggesting that the appropriate length scale for a network analysis is set by the cluster size rather than individual particles. Finally, we show that network resilience analysis, made possible using a Girvan-Newman criterion, shows similar features to those from the elasticity of the spring network when strength of attraction between colloids is varied. That is, we show network resilience correlates with the elasticity of the network when strength of attraction between colloids is varied.

ABSTRACT. Information war has brought a completely different form of operation from the past. This form of joint operation is not only limited to the changes of combat structure and combat strength but also pays more attention to the formation of overall combat strength. This paper considers the complexity of the correlation of the equipment system of system, and from the perspective of the system of system, proposes a new equipment system of system portfolio development method, which provides a new idea for the equipment development demonstration. First, based on the interrelationship of various types of equipment, constructing the equipment heterogeneous information network[1-5]; secondly, the equipment function chain is extracted on this basis; Third, designing the deep reinforcement learning optimization algorithm with the maximum ability value of the equipment function chain and giving the equipment development scheme; finally, the algorithm is solved to verify its effectiveness by taking the development of an equipment system of systems as an example.

ABSTRACT. Global warming, extreme climate events, earthquakes and their accompanying socioeconomic disasters pose significant risks to humanity. Yet, due to the nonlinear feedbacks, multiple interactions and complex structures of the Earth system, the understanding and, in particular, the prediction of such disruptive events represent formidable challenges to both scientific and policy communities. During the past years, the emergence and evolution of Earth system science has attracted much attention and produced new concepts and frameworks. Especially, novel statistical physics and complex networks based techniques have been developed and implemented to substantially advance our knowledge of the Earth system, including climate extreme events, earthquakes and geological relief features, leading to substantially improved predictive performances. We present here a comprehensive review on the recent scientific progress in the development and application of how combined statistical physics and complex systems science approaches such as critical phenomena, network theory, percolation, tipping points analysis, and entropy can be applied to complex Earth systems. Notably, these integrating tools and approaches provide new insights and perspectives for understanding the dynamics of the Earth systems. The overall aim of this review is to offer readers the knowledge on how statistical physics concepts and theories can be useful in the field of Earth system science.

ABSTRACT. The El Niño Southern Oscillation (ENSO) is one of the most prominent interannual climate phenomena. Early and reliable ENSO forecasting remains a crucial goal, due to its serious implications for economy, society, and ecosystem. Despite the development of various dynamical and statistical prediction models in the recent decades, the “spring predictability barrier” remains a great challenge for long-lead-time (over 6 mo) forecasting. To overcome this barrier, here we develop an analysis tool, System Sample Entropy (SysSampEn), to measure the complexity (disorder) of the system composed of temperature anomaly time series in the Niño 3.4 region. When applying this tool to several near-surface air temperature and sea surface temperature datasets, we find that in all datasets a strong positive correlation exists between the magnitude of El Niño and the previous calendar year’s SysSampEn (complexity). We show that this correlation allows us to forecast the magnitude of an El Niño with a prediction horizon of 1 y and high accuracy (i.e., root-mean-square error = 0.23° C for the average of the individual datasets forecasts). For the 2018 El Niño event, our method forecasted a weak El Niño with a magnitude of 1.11±0.23° C. Our framework presented here not only facilitates long-term forecasting of the El Niño magnitude but can potentially also be used as a measure for the complexity of other natural or engineering complex systems.

ABSTRACT. To help inform university-reopening policies, we collected survey data on social contact patterns and developed an agent-based model to simulate the spread of SARS-CoV-2 in university settings. In the model, each person-agent is assigned to one role category of a) faculty and staff, b) student living on campus, c) student living off campus, or d) student living at a sorority or fraternity. Each location-agent plays one role of: a) recreation center or any gym or other shared exercise spaces; b) union, dining centers, and coffee shops on campus; c) bars, restaurants, and coffee shops off campus; d) stores and other types of services off campus; and e) other types of social gathering such as sport, religious, and social events. In Figure 1, each person-agent has contact with people it regularly meets in the three contact lists, and meets other people at the five types of locations. More specifically, the people within each person’s contact lists are randomly selected from the whole population during model initialization, and remain unchanged throughout each simulation run. By contrast, the people each person meets at specific locations are dynamic. Considering different immunization effectiveness, we simulate SARS-CoV-2 spreading in a university population under two scenarios: 1) relaxation of non-pharmaceutical interventions (NPIs) and 2) adoption of NPIs, such as wearing masks. In addition, we perform a thorough model calibration and accurate sensitivity analyses on the immunization effectiveness. Considering reproduction number R_0=3 and 70% immunization effectiveness, we estimated that at least 80% of the university population immunized through natural infection or vaccination is needed for safe university reopening with relaxed non-pharmaceutical interventions. In contrast, at least 60% of the university population immunized through natural infection or vaccination is needed for safe university reopening, when non-pharmaceutical interventions are adopted. Still, attention needs to be paid to large-gathering events that could lead to infection size spikes. At an immunization coverage of 70%, continuing non-pharmaceutical interventions, such as wearing masks, could lead to a 78.39% reduction in the maximum cumulative infections and a 67.59% reduction in the median cumulative infections. However, even though this reduction is very beneficial, still there is a possibility of non-negligible size outbreaks, because the maximum cumulative infection size is equal to 1.61% of the population, which is substantial.

ABSTRACT. One year after COVID-19 was declared a pandemic, a dozen vaccines have already been approved for use. Fast and targeted vaccination has become the main goal to reduce morbidity and mortality, often with prioritization criteria based on the risk of developing severe infection – i.e. the elderly, individuals with comorbidities. Yet, vaccination coverage remains heterogeneous across the globe, with very ill-protected regions. Reactive vaccination could be efficient, in combination with mass vaccination, to accelerate the decline of epidemic or to curb local case resurgence. Reactive vaccination consists in targeting the individuals most at risk of infection, e.g. those in workplaces where virus circulation is detected. Quantifying its effect requires modeling complex interplays between different time scales (time for vaccines being effective, incubation and propagation timing, detection system effectiveness) as well as natural history of infection, pace of vaccination, vaccines hesitancy, and overall changes in human behavior due to non-pharmaceutical intervention. To do so, we used an agent-based model integrating real-time demographic and contact data to build a multilayer temporal network of contacts, where layers represent different settings (household, workplaces, schools, transport, and community). The model takes as input a synthetic population reproducing a typical medium-sized town, with workplace sizes and school types. We simulated on this network COVID-19 epidemics and vaccination distribution, exploring different epidemic situations, initial coverage, and vaccination pace. We found that in most scenarios, reactive vaccination leads to a higher reduction in cases than non-reactive strategies, even with lower vaccine effectiveness. However, we found that in the case of high vaccine coverage, reactive strategy can be less effective than high pace non-reactive strategy. We also found that in case of flare-ups, reactive vaccination, reactive vaccination could help to prevent flare-ups, if reactive vaccination is implemented quickly and supported by enhanced test-trace-isolate and triggers an increased vaccine uptakes. These results provide valuable information on the inclusion of reactive strategy in public health policy to protect the at-risk populations from COVID-19.

ABSTRACT. After one year of stop-and-go COVID-19 mitigation, in the spring of 2021 European countries still experienced sustained viral circulation due to the Alpha variant. As the prospect of entering a new pandemic phase through vaccination was drawing closer, a key challenge remained on how to balance the efficacy of long-lasting interventions and their impact on the quality of life. Focusing on the third wave in France during spring 2021, we simulate intervention scenarios of varying intensity and duration, with potential waning of adherence over time, based on past mobility data and modeling estimates. We identify optimal strategies by balancing efficacy of interventions with a data-driven “distress” index (Fig.1), integrating intensity and duration of social distancing. We show that moderate interventions would require a much longer time to achieve the same result as high intensity lockdowns, with the additional risk of deteriorating control as adherence wanes. Shorter strict lockdowns are largely more effective than longer mod- erate lockdowns, for similar intermediate distress and infringement on individual freedom. Our study shows that favoring milder interventions over more stringent short approaches on the basis of perceived acceptability could be detrimental in the long term, especially with waning adherence.

ABSTRACT. Mathematical models are one of the key tools to combat epidemics of emerging diseases. As the recent COVID-19 pandemic has shown, it is essential to properly model human contact patterns to truly understand the epidemic dynamics. The classical approach is based on mixing matrices that encode the number of interactions between certain groups of the population, being age-mixing matrices the most common of them. More recently, the use of networks has sparked a plethora of discoveries in the field of mathematical epidemiology. Despite their differences, it is generally possible to reproduce the same results with both approaches as long as one properly calibrates them. However, once this had been done, if one introduces non-pharmaceutical interventions aimed at reducing the number of contacts in the population one might get very different outcomes.

In this work, we first build multiplex contact networks using highly detailed sociodemographic data representing the interactions of individuals in schools, workplaces, households, and the general community. Then, we derive the age-mixing matrices that would encode exactly the same set of human interactions from these networks. We implement a classical SIR model, as well as a model for influenza and for COVID-19, using each of these approaches, and calibrate them to yield the same results under a baseline scenario. We aim to understand what would each model predict when a certain intervention is put in place.

In particular, we mimic the closure of schools, workplaces, and the reduction of contacts in the community that has been observed during the COVID-19 pandemic, and study how stringent they have to be to control an outbreak. Our results show that when interactions are encoded in networks, the models predict that a milder intervention is enough to stop an outbreak in comparison to the approach with age-mixing matrices (see figure 1). This has very important policy implications given that the large majority of models that have been used during the COVID-19 pandemic are based on approaches akin to the age-mixing matrix one. However, we acknowledge that building such networks is a very complex process that requires a large amount of data, and the models build on networks are not free from caveats. As such, we also explore the strengths and weaknesses of each approach and give some recommendations regarding the strength of the claims that can be done with one approach or the other.

ABSTRACT. To mitigate the scope of an acute infectious outbreak, a rapid and effective control of the level of contacts within the population, a key factor for the spread of infectious disease, is required. We devised a framework based on GPS crowd-sourcing data collected from mobile phones capable of predicting the short-term fate of an epidemic outbreak and enabling us to evaluate the epidemic situation under various scenarios. In short, mobile phones send pings with location and time stamps. From the co-location of devices, we can identify contacts and reconstruct (parts of) the actual contact network in the population. We correct for the effect of the unseen part of the population using Horvitz-Thompson network sampling theory. We find that the "contact index", a combination of the first two moments of the contact number distribution, predicts actual epidemic dynamics with a lead of about 17 days in the case of COVID-19, thus also showing the relevance of the "friendship paradox" in the context of epidemic spread.

ABSTRACT. Until now, many studies on the Ising model have been actively conducted in complex networks. Because the Ising spin model defined on a complex network can describe opinion dynamics that describes forming a huge social public opinion through the exchanging individuals' opinions.

The complex networks, unlike lattices or random networks, have a unique structure called a hub, and thus complex networks show new properties in phase transitions that do not appear in lattices or random networks. In other words, the topological structure of the network further affects the physical properties and the phase transition of the system. From a sociological point of view, it means that the social structure affects the formation of public opinion.

Inspired by this result, this research group also tried to systematically study how the social structure affects the formation of public opinion. However, we have introduced a multiplex network structure to more accurately describe the process of actual public opinion formation. Because people's opinions are not limited to one topic, but they have their own opinions on various topics, and it is common to exchange opinions with neighborhood on all of them.

Therefore, we considered the Ashkin-Teller model defined on the multiplex network and systematically studied how the topological structure of complex networks affects the phase transition of the Ashkin-Teller model as in previous studies. Through this study, it was found that the structure of the complex network had a profound effect on the phase transition of the Ashkin-Teller model, and it was confirmed that the formation of social public opinion can vary depending on how people are connected socially.

However, this study was limited to the case where inter-layer interaction was attractive, and no study has been conducted yet on the case of repulsive interaction. From a sociological point of view, the inter-layer interaction corresponds to the process of exchanging individual thinking patterns rather than opinions of a specific topic. The case, where people oppose the thinking patterns of neighbors around them, is common such as academia and art circles, therefore, it is essential to study the case where inter-layer interaction is repulsive, not attractive.

In this study group, considering the case where inter-layer interaction is repulsive, we investigate how the complex network structure affects the phase transition of the system.

As a result, we find various types of phase transitions that do not appear in lattice structures or random networks, and it was also confirmed that this was due to the hubs of complex networks.

ABSTRACT. The state of many physical, biological and socio-technical systems evolves by combining smooth local transitions and abrupt resetting events, be them endogenous or exogenous, to a set of reference values. The inclusion of the resetting mechanism not only provides a more realistic description of the modeled system but also leads to novel phenomenology not present in reset-free cases. However, most models where stochastic resetting is studied address the case of a finite, often a single one, number of uncorrelated variables, such as the position of non-interacting random walkers.

Network growth with node deletion can be framed as a stochastic resetting problem where an arbitrarily large number of degrees of freedom ---the nodes--- are coupled, both in the resetting and non-resetting (growth) events. Indeed, when a link is added in the network there are two nodes that simultaneously increase their degree by one unit. When a node, say with degree $k$, is removed from the network, that node looses all its links and there are $k$ other nodes that decrease their degree by one unit. It is known that the introduction of resetting induces interesting properties that qualitatively differ with respect the reset-free case, such as the emergence of a non-equilibrium steady state (NESS) in the degree distribution. However, the dynamical properties regarding how this NESS is approached and the implications that resetting has on system-wide properties like the emergence of a giant component have been unexplored. Similarly, little is known about the properties of a node to attain a certain degree value for first time.

In this work, we analytically tackle these questions. Our system is made of $N$ nodes, connected via undirected edges. Two processes compete in the formation of the network. On the one hand, links between two nodes picked uniformly at random are added at rate $\alpha N/2$. On the other hand, with rate $rN$, nodes selected uniformly at random are removed, along with all their edges. The master equation for the degree distribution $p_k(t)$ reads \begin{equation} \label{eq:MastEq} \diff{p_k} = \alpha p_{k-1} - \alpha p_{k} - r p_{k} + r(k+1)p_{k+1} - r k p_{k} + \delta_{k0} r. \notag \end{equation} The first two terms are associated to the uniformly random addition of a link, while the other ones correspond to resetting process. The third term encodes the direct resetting of a node of degree $k$, while the fourth and the fifth ones correspond to the loss of one link due to the removal of a neighbor. The final term stands for the incoming flux of nodes that are reset to degree $0$.

We are able to exactly solve these master equations, thus providing information not only on the steady state but also on the relaxation towards it. The access to the full-time dependence of the degree distribution allows us to develop a time-dependent percolation theory, from which we discuss the emergence, or lack thereof, of a percolating structure as a function of time and the growth and resetting parameters. We find that the criticality condition for its appearance satisfies \begin{equation} \label{eq:crit_point} 1 = \frac{2\alpha}{3r} \frac{1-2e^{-rt} + e^{rt}}{1 + e^{rt}}. \end{equation} Fig.~1$(a)$ shows the size giant component $S$ as a function of time and of the resetting parameter, for fixed $\alpha = 1$. We see that as resetting becomes more and more probable, the giant component needs more time to emerge and its stationary value tends to be smaller. There is a value of resetting rate, $2 \alpha / 3$, above which the giant component never emerges, no matter how long we wait. Fig.~1$(b)$ further evinces the non-trivial impact of the growth and resetting rate on the percolation transition, where it is presented the time evolution of the size of the giant component for several pairs $(\alpha, r)$. We see that if their ratio is the same, the curves tend to the same stationary value, as expected from the long-time limit of the degree distribution and the mean degree. However, the absolute values of $(\alpha, r)$ do have a strong impact on the critical point and the time scale to reach the stationary value, as it can be appreciated in the ordering of the curves. We see, moreover, that theory and simulations match very well in all cases.

We finally study the first-passage properties to reach a certain degree $k^*$ for a node that starts with no edges. We find an analytical expression for the mean-first passage time (MFPT), and report the existence of a crossover value $k_c=k_c(\alpha, r)$ such that if $k^* > k_c$ then the MFPT grows faster than an exponential, while it is exponential for $k^* \leq k_c$.

Our work unveils new results in the classical problem of network growth and, in turn, serves as an illustrative, natural and solvable example of a coupled multiparticle system subjected to resetting, hence advancing the state of the art in the theory of stochastic resetting.

ABSTRACT. Network theory has been successfully applied to explore a great variety of complex systems. Particularly interesting is the study of network local patterns that can shed light on the emergence of global properties. In this perspective, network motifs represent a fruitful example. While binary motifs have received great attention, only few works have proposed an extension of the motif concept to the weighted case. This work aims to fill this gap with a method of few ingredients: (i) unbalanced weighted networks; (ii) a sink node and (iii) a random walker with a limited number of steps. The sink node is introduced to compensate the excess of ingoing flows and balance the network. In this way, the random walker moving for a limited number of steps along network links, can describe all subgraphs of dimension smaller than and equal to the maximum number of steps as mutually exclusive events such that their total occurrence probability sum up to 1.We applied our approach to different real networks and test the significance of weighted motifs occurrences using a randomization technique based on the entropy maximization of the system under local constraints: degree and strength. We consider a maximum of 3 steps offering both analytical results and simultations outcome. In principle, we can include any number of steps, but this exponentially increases the computational complexity of the problem. Figure 1(a) reports the frequency of the eight weighted motifs (fig.1 top) for all networks in our sample. It offers a first idea about the different nature of the systems according to their local organization in subgraphs. In order to wash away possible statistical effects, in fig. 2(b)-(c) we look at z-scores, obtained by the comparison between occurrences in real networks and related benchmark models. Z-score profiles allow to (i) investigate more on each system looking at its organization in significant weighted motifs; (ii) make comparisons between the over(under)- representation of weighted motifs in different systems and relate them with the functioning mechanism or the evolutionary process that brought to network formation. Firstly, we found that networks belonging to the same field show very similar motifs significance profiles. Furthermore, focusing on specific weighted motifs we can shed light on underlying functioning mechanism of different systems. Investigating more on the systems features, we can link these patterns with functioning mechanisms peculiar of the network. Our approach also allows us to identify specific subpatterns characterizing almost all the different systems,revealing critical functional features (f.e.,motif VI). Furthermore, it is possible to focus on selected nodes, study the over time variation of the weighted motifs composition, and relate them with exogenous shocks as economic, political, and social events. In the paper we show all these possible applications testing the temporal evolution of weighted motifs for three countries in the world trade web dataset.

ABSTRACT. Biological systems need to react to stimuli over a broad spectrum of timescales. If and how this ability can emerge without external fine-tuning is a puzzle. This problem has been considered in discrete Markovian systems where results from random matrix theory could be leveraged. Indeed, generic large transition matrices are governed by universal results, which predict the absence of long timescales unless fine-tuned. Previous work has considered an ensemble of transition matrices and motivated a temperature-like variable that controls the dynamic range of matrix elements. Findings were applied to fMRI data from 820 human subjects scanned at wakeful rest. The data was quantitatively understood in terms of the random model, and brain activity was shown to lie close to a phase transition when engaged in unconstrained, task-free cognition – supporting the brain criticality hypothesis in this context. In this work, the model is advanced in order to discuss the effect of matrix asymmetry, which controls entropy production, on the previous results. We introduce a new parameter that controls the asymmetry of these discrete Markovian systems and show that when varied over an appropriate scale this factor is able to collapse Shannon entropy measures. This collapse indicates that structure emerges over a dynamic range of both temperatures and asymmetry.

ABSTRACT. The dynamics of networks on Heider’s balance theory moves toward reducing the tension by constantly reevaluating the interactions to achieve a state of balance. Conflict of interest, however, is inherent in most complex systems; frequently, there are multiple ideals or states of balance, and moving towards one could work against another. In this paper, by introducing the competitive balance theory, we study the evolution of balance in the presence of conflicts of interest. In our model, the assumption is that different states of balance compete in the evolution process to dominate the system. We ask whether, through these interactions, different states of balance compete to prevail their own ideals or a set of coexisting ideals in a balanced condition is a possible outcome. The results show that although there is a symmetry in the type of balance the system either evolves towards a symmetry breaking, where one of the states of balance dominates the system, or, less frequently, the competing states of balance coexist in a jammed state.

ABSTRACT. Graph matching consists of aligning the vertices of two unlabeled graphs in order to maximize the shared structure across networks; when the graphs are unipartite, this is commonly formulated as minimizing their edge disagreements. In this talk, we address the common setting in which one of the graphs to match is a bipartite network and one is unipartite. Commonly, the bipartite networks are collapsed or projected into a unipartite graph, and graph matching proceeds as in the classical setting. This potentially leads to noisy edge estimates and loss of information. We formulate the graph matching problem between a bipartite and a unipartite graph using an undirected graphical model, and introduce methods to find the alignment with this model without collapsing. We theoretically demonstrate that our methodology is consistent, and provide non-asymptotic conditions that ensure exact recovery of the matching solution. In simulations and real data examples, we show how our methods can result in a more accurate matching than the naive approach of transforming the bipartite networks into unipartite, and we demonstrate the performance gains achieved by our method in simulated and real data networks, including a co-authorship-citation network pair, and brain structural and functional data.

ABSTRACT. To understand the emergence of hashtag popularity in online social networking complex systems, we study the largest Chinese microblogging site Sina Weibo, which has a Hot Search List (HSL) showing in real time the ranking of the 50 most popular hashtags based on search activity. We investigate the prehistory of successful hashtags from 17 July 2020 to 17 September 2020 by mapping out the related interaction network preceding the selection to HSL. We have found that the circadian activity pattern has an impact on the time needed to get to the HSL. When analyzing this time we distinguish two extreme categories: a) "Born in Rome", which means hashtags are mostly first created by super-hubs or reach super-hubs at an early stage during their propagation and thus gain immediate wide attention from the broad public, and b) "Sleeping Beauty", meaning the hashtags gain little attention at the beginning and reach system-wide popularity after a considerable time lag. The evolution of the repost networks of successful hashtags before getting to the HSL show two types of growth patterns: "smooth" and "stepwise". The former is usually dominated by a super-hub and the latter results from consecutive waves of contributions of smaller hubs. The repost networks of unsuccessful hashtags exhibit a simple evolution pattern.

ABSTRACT. The large-scale interaction data of the online live streaming platform provides experimental datasets for the quantitative analysis of human behavior, and offers a new opportunity for the mining of the online interaction mechanism with collective dynamics. Given the lack of empirical research on real-time collective interaction, this paper collected a one-year-long comprehensive dataset of real-time livestreaming statistics, involving more than 1.9 million anchors from Douyu (the largest live streaming platform in China), and designed a generalized evolution model for exploring the interaction mechanism between anchors and viewers. First, we construct a viewer-anchor bipartite interaction network representing the dynamics of the entities in the platform, and then propose an evolution model with adjustable preference strength of viewer-anchor interaction. The preference strength can be adjusted with two parameters: the fraction of random choice and the preference coefficient of viewers. Experiments on empirical datasets show that the model can accurately and robustly predict the evolution process when all viewers have linear preference on the number of existing viewers attracted by the anchor when they select an anchor to interact with. This paper reveals the dominating mechanism of preferential attachment for the viewers selecting an anchor and reflects the human tendency and preference for valuable content, confirming the cumulative effect of reputation or word-of-mouth in social systems. Our study provides a quantitative model for exploring the interactive behavior characteristics and internal mechanism of large-scale online crowds in live streaming, and is of great significance for describing and predicting the formation and development of social relationships in more general settings.

ABSTRACT. The structure of a social network has important implications for how information spreads across that network. In the case of social media, it is critical to study the ties among users in order to understand information propagation, because information exposure and amplification is, in significant part, determined by who follows whom. In this work, we study how different attributes are associated with following someone on Twitter. We anticipate for the Twitter graph to be substantially homophilous along various dimensions (McPherson et al 2001). Of special interest is political homophily: there is evidence of substantial polarization regarding conversations in this platform, measuring partisanship as following particular accounts (Barberá 2015; Garimella & Weber 2017), and some experimental evidence of polarization relative to tie formation (Mosleh et al. 2021). Twitter ties are also associated with geographical location and entangled with existing physically bounded ties, and thus impacted by national borders, language, or number of air flights (Takhteyev et al. 2012; Stephens and Poorthuis 2015). However, not much is known regarding individual attributes of Twitter users beyond location or partisanship, and previous research uses either small samples or imperfect measures of political attitudes based on activity in the platform itself. Here we use a distinct subsample of 1.5 million Twitter accounts matched to U.S. voter records (Hughes et al 2021; Grinberg et al 2019), including information on location, age, gender, race/ethnicity, partisanship (inferred based on voter records data) and party registration for each user. We examine the correlates of reciprocal ties, asymmetric ties, and degree, and find generally strong homophily across the different attributes we examine. Geographical proximity is the main predictors of a follower relationship, and, while partisanship is important, its association is not as strong as for other variables, such as age, race, or median income of the tract. In addition, political homophily is lower using the measure of partisanship inferred from voter records than when using a measure based on following behavior (Barberá et al. 2015). We identify a local information ecosystem influenced by demographic homophily and physical proximity and based in reciprocal ties. In contrast, asymmetric ties tend to reflect a national information ecosystem, driven by active, celebrity hubs. This evidence proves the importance of social cleavages beyond partisanship in the Twitter follower network and helps understanding how information flows in this platform may be biased by these cleavages.

ABSTRACT. Online media are widely held responsible for the rise of political polarization throughout the Western world. But popular narratives of 'filter bubbles' and 'echo chambers' have recently been heavily criticized, because users clearly do communicate with political opponents online [1]. So how do users of online media polarize? We approach this question using a novel dataset, derived from the discussion forums of a major German-speaking news platform. This dataset contains over 94,000 users and 46 million interactions between them. Crucially, the interactions comprise up-votes as well as down-votes, and can thus be represented as edges with both signed and temporal information. This unique combination of features allow us to investigate the formation of political alliances and the emergence of political conflict in real time. We focus on debates surrounding the highly contentious European refugee crisis (2015-16), a notoriously turbulent year regarding corruption scandals which led to the Austrian government collapsing (2019) and the months comprising the start of the COVID-19 pandemic (2020).

We theorize political polarization as an increase of structural balance between actors with opposing ideological positions (see [2]). We test this model by quantifying the trajectory of structural polarization in the network of signed interactions as a longitudinal user analysis across years (similarly done in [3]). We do so by finding an optimal bi-partition of the signed network and defining a normalized upper bound of balance based on the amount of frustrated edges in that partition (i.e. edges that violate the assumptions of our partition model) [4]. The signed interactions are defined by the total balance of interactions between two users during that time period, with the minimum requirement of 2 interactions for positive edges and only 1 interaction for negative edges, given a clear positivity bias in the user behaviour.

Our results are congruent with the political developments over the same period: Overall, the level of structural polarization is increasing, and moments of acute political crisis and conflict coincide with peaks in structural polarization. Following a start of the migration crisis where humanitarian help and unity prevailed, several controversial government decisions increased the social tensions from the start of 2016. This led to abrupt alternations of power by very different political parties in the small interval of five years (2016-2020). We examine the context of the peaks in polarization through a detailed analysis of the social and political circumstances of each time period provided by the news articles texts. We also find seasonality patterns such as a tendency of increased polarization in the first quarters of each year, probably explained by periods of political agenda planning. To give a better sense of the structural network changes, in Figure 1 we show a network visualization of the users present in four specific time periods, which contain similar percentages of (+)/(-) edges, and show different polarization structures.

Furthermore, we have performed additional analysis in the direction of dimensionality reduction methods (similarly done in [1]) to identify the latent ideology space in which users are positioned according to their ideological orientation in a left-right spectrum. Our research fills the research gap of analyzing dynamic signed networks in an online community, as previous research on online interaction could either only analyze positive link dynamics or static snapshots of social signed networks (see [5]).

[1] Pablo Barberá, John T Jost, Jonathan Nagler, Joshua A Tucker, and Richard Bonneau. Tweeting from left to right: Is online political communication more than an echo chamber? Psychological science, 26(10):1531–1542, 2015. [2] Simon Schweighofer, Frank Schweitzer, and David Garcia. A weighted balance model of opinion hyperpolarization. Journal of Artificial Societies and Social Simulation, 23(3):5, 2020. [3] Ernesto Estrada. Rethinking structural balance in signed social networks. Discrete Applied Mathematics, 268:70–90, 2019. [4] Samin Aref and Mark C Wilson. Measuring partial balance in signed networks. Journal of Complex Networks, 6(4):566–595, 2018. [5] Pedro Calais Guerra, Wagner Meira Jr, Claire Cardie, and Robert Kleinberg. A measure of polarization on social media networks based on community boundaries. In Seventh international AAAI conference on weblogs and social media, 2013.

ABSTRACT. Introduction. In a politically polarized environment, people tend to rely on news media that share similar political preferences for information on political issues. In this study we use network analysis to investigate Indonesian news media landscape on Twitter during the 2019 Indonesian presidential election. Objective. We aim (i) to elaborate on the political affiliation of Indonesian news media, both in terms of news production and content consumption; (ii) to characterize the topological properties of media networks in each political affiliation segment; and (iii) to investigate the relationship between political affiliation and media credibility, by examining several indicators namely number of media, total news share, and number of core media. Data. We archive election-related tweets based on a number of keywords related to the name of the presidential/vice-presidential candidates: Joko Widodo/KH. Maaruf Amin (JW-MA) and Prabowo Subianto/Sandiaga Uno (PS-SU). In this study, we only processed tweets containing news URLs from 560 Indonesian news media. Method. We build a bipartite network between users and news media based on the presence of a media URL in tweets. The network is then projected and validated into a media network that represents the Indonesian news-media landscape. We identify the political attitudes of news media using an audience-based approach, and extract media clusters based on their political alignments, then analyze their internal networks separately. Result. In this study, we reveal the existence of polarization in the spectrum of media political affiliation, both in terms of production and consumption of news on social media. The pro JW-MA news-media has larger information production capacity than the pro PS-SU news-media. This study also shows differences in the topological characteristics of media sub-networks in each segment of political affiliation. The pro JW-MA news media forms an information ecosystem that is more hierarchical, has structural resilience, and is able to facilitate the distribution of information better than the other two media sub-networks. Furthermore, the correlation between political affiliation and news media credibility reveals the role of low-credibility media in the dynamics of information production and consumption during elections. Low-credibility media tends to have extreme political tendencies, dominates the core structure of media networks and collectively has a high level of information exposure. We highlight these findings and link them to the spread of misinformation on social media during elections.

ABSTRACT. For the last 30 years, the web has been used as a space of debate and knowledge creation, including historical knowledge. It has fostered the emergence of several communities dedicated to discussing topics about history. Thus, the digital space might have the power to provide a more democratic history that relies on the inclusion of different voices in historical sources. But at the same time, it also raises questions about editing and authority. When attempting to understand authority relations on the web, one factor gains special prominence — moderation. Moderation involves actions of exclusion, organization, and establishment of norms; thus, moderators' decisions influence everything seen, valued, and said by web users. Here, we aim to understand the dynamics of knowledge creation considering moderation bias.

Our focus is to identify the role of moderators in the creation of knowledge on the major Reddit history forums by analyzing how moderators and users established authority relations. For that, we developed a mixed-methods approach: we analyze these forum’s rules and perform temporal social network analysis based on dialogues established on them. The analysis shows that the rules have become progressively more extensive and stricter over the years, creating appropriate ways for posting submissions and commenting. When comparing these rules with the networks of dialogues, we note that forums with more strict rules might inhibit the active participation of most users. Thus, moderators gain prominence: they determine the rules that a user must follow to become an authority, but by restricting the debate through the rigidity of the rules, they create the conditions to become the central authorities.

ABSTRACT. Discoveries are essential milestones for the progress of our society. Therefore, unveiling the hidden mechanisms behind the emergence of new ideas is not only interesting from a scientific point of view, but also has a tangible sociological and economical impact. Recently, different mathematical approaches have been proposed to investigate and model the dynamics leading to the emergence of the new. Among these, of particular interest are random processes with reinforcement, such as urn models and biased random walks. These models successfully replicate the basic signatures of real-world discovery and innovation processes. However, they neglect the effects of social interactions. In particular, by considering the exploration dynamics as the one of a single entity and thus neglecting the multi-agent nature of the process, these models (i) do not capture the heterogeneity of the pace of the individual explorers; (ii) do not include the benefits brought by collaborations, more broadly, social interactions. Indeed, empirical evidences of these mechanisms have been found in various contexts, from music-listening and language to politics and voting.

In this talk I give insights on how our peers can influence our experience of the new, using theoretical and data-driven mathematical models. In the first part of the talk, I introduce the model proposed in Physical Review Letters 125.24 (2020), p. 248301, where each explorer is associated with an urn model with triggering (UMT) that governs its exploration and discovery dynamics. Urns are coupled through the links of a complex network, so that each exploration process is also subjected to interactions with the processes of the neighboring nodes, and explorers can exploit opportunities (possible discoveries) coming from their social contacts, in a cooperative manner. We study the impact of the network topology on the exploration dynamics and we find that the pace of discovery of an explorer strongly depends on its position in the social network, as shown in the attached Fig.(a) for the Zachary Karate Club network. Notice the higher pace of discovery displayed by the notoriously central nodes, suggesting that a strategic location on the social network correlates with the discovery potential of an individual. In particular, we show that the ranking of the nodes that distinguishes the fastest explorers can be predicted analytically by using the eigenvector centrality. This highlights that the structural (not just local) properties of the nodes can strongly affect the agents' ability to discover novelties.

In the second part of the talk I investigate a data set containing the whole listening histories of a large, socially connected sample of users from the online music platform Last.fm. We demonstrate in arXiv:2202.05099 that users exhibit highly heterogeneous discovery rates of new artists and that their social neighborhood significantly influences their behavior. In particular, we find that more explorative users tend to interact with peers more prone to explore new content. We capture this phenomenology in a data-driven modeling scheme where users are represented by random walkers exploring a graph of artists, shown in the attached Fig.(b), and interacting with each other through their social links. As in the previous model, each agent is associated to an urn, but, differently from it, they interact with their peers by checking the current artist listened by one of their friends. Interestingly, even starting from a uniform population of agents (no natural differences among the individuals), our model predicts the emergence of strong heterogeneous exploration patterns, with users clustered according to their musical tastes and propensity to explore.

We hope our work can represent a significant step forward to develop a general framework to understand how social interactions shape discovery and innovation processes.

ABSTRACT. Collaboration has become particularly important as modern science and technology are increasingly developed by teams. However, we lack a quantitative understanding of the expertise disparity among coauthors before the team is formed. Here, we propose a new method to quantify the expertise distance of researchers based on their prior publication history. We find that across science and technology, teams with more diverse expertise have no significant advantage in the short-term (2 years) or mid-term (5 years) impact, but exhibit higher long-term (10 years) impact, largely because they increasingly attract more cross-disciplinary influence. When other dimensions of diversity are missing, diversifying coauthor expertise appears particularly beneficial for teams assembled within the same institution or country. Overall, our method may have policy implications for team assembly, funding, hiring, and promotion processes, and sheds new light on a wide range of measures of similarity and diversity in science, technology, and society.

ABSTRACT. We use various graph embedding and text embedding methods to learn vector representations of American Physical Society (APS) papers. We then evaluate each embedding's ability to capture the disciplinary structure of the APS's Physics and Astronomy Classification Scheme (PACS). Through our experiments, we have found that graph embedding methods perform better than text embedding methods at this task.

ABSTRACT. An increasing number of studies hint that negative emission technologies (NETs) represent a central element to reach net--zero emissions and meet the Paris Agreement. However, as of today, NETs are far from fully--developed technologies deployable at large scale. Indeed, the path from scientific advances to widespread practical solutions is typically long and uncertain. Here, we take an innovation network perspective, with the aim to quantify the multidimensional nature of knowledge spillovers generated from 20 years of NETs research. Our findings show that scientific advances in NETs generate significant, positive and large knowledge spillovers within science, and from science to technology and policy. However, stark differences applies across carbon removal solutions. Except for Direct Air Capture (DAC), the distance between NETs' scientific knowledge and practical technological developments remains considerable. Solutions such as Bio--energy with carbon Capture and Storage (BECCS) and Blue Carbon (BC) are gaining relative momentum in policy and public debate, while DAC lags behind in such respect. In addition, we identify cities and countries that can serve as hubs for supporting collaborations. Our results suggest that innovation might play an essential role in dealing with the climate change crisis, but we might need to move beyond the conventional innovation policy toolkit, closing a coordination gap between science, technology and policy.

ABSTRACT. In the innovation process from pure scientific research to applied scientific research to technology and finally to commercialization, patenting activities play a critical role. Following large investments into graphene field since the discovery in 2004 from both public and private sources, we collected 139,264 patent applications related to graphene from Derwent Innovation. Performing co-clustering on their International Patent Classification (IPC) subclasses resulted in seven technological topics: (0) composites, (1) synthesis, (2) batteries, (3) sensors, (4) water treatment, (5) catalyst, and (6) devices. Since regional and assignee information are also available, we naturally addressed the following scientific questions: (i) who are the regional leaders in technological competitions at the aggregate level, (ii) how did the technological leaders obtain their positions, and (iii) did the leaders invest across multiple technologies or only around their core competencies? From the proportions of patent applications, there were three clear trends in graphene technology: (1) synthesis was the first dominant topic, but was replaced by (6) devices since 2008, which was in turn replaced by (2) batteries from 2013 onwards. Following our previous study on graphene science where the order of scientific emergence is theory and experimental tests - synthesis and functionalization - sensors - supercapacitors and electrocatalysts, we found that the evolution sequence of (1) - (0) - (3) - (6) - (5) - (2) - (4) in graphene technology agrees roughly with what was happening in graphene science. Between 2004 and 2017, the top 4 regions in terms of number of patent applications were China, United States, Japan, and South Korea. Changes in relative rankings of these four regions across the seven technological areas indicated differences in investment preferences. When we further classify assignees into three types (universities, corporates, and others) and the entities into three groups (lower quartile, interquartile, and upper quartile), their patent application histories tell an interesting story of the difference in patenting strategies between universities and corporations, as well as between large entities and small entities. In both corporations and universities, the lower quartile groups showed a clear trend towards greater specialization as their patent portfolios grew. On the other hand, entities with large patent portfolios became highly diverse in technology management over time. In general, the patent portfolios of interquartile and upper quartile universities are higher than their corporate counterparts, leading us to ask: (iv) why are university patents more diverse than corporate patents. We found evidences at the aggregate level to support our hypothesis that companies aim to use their patents to develop products. Hence, they will buy useful patents instead of paying licensing fees, but sell off patents they cannot use. On the other hand, universities do not develop any products, so they keep all their patents and focus on licensing instead. This hypothesis were further confirmed by two case studies on SAMSUNG and RICE UNIVERSITY. Finally, we found that the differences between China and United States at the regional level can be explained by their technological landscapes being dominated by universities and large corporations respectively.

ABSTRACT. Acknowledgement shows the credit for contributed scholars [1]. However, the interpersonal acknowledgement dataset has not been established, mainly because of the difficulty distinguishing scholars only by the mentioned names without any ID or affiliation in acknowledgement statements. In this work, we extracted and detected acknowledged scholars recorded in eight open access journals (e.g., PLOS ONE and Scientific Reports) until 2021, utilizing named-entity recognition and citation and collaboration networks based on Microsoft Academic Graph(MAG), which is commonly used in bibliometric research. Consequently, 180,375 acknowledged scholars in 89,731 papers are identified. Manual evaluation of this dataset assures 98.2% precision. Moreover, detected acknowledged scholars contain unique scholar ID compatible with MAG, contributing to the acceleration of acknowledgement research and integrating well-studied scientometrics research, such as citation and collaboration network analysis. The data is available in https://commons.sk.tsukuba.ac.jp/data_en. Furthermore, as an example application of these acknowledgement data, we established the acknowledgement network at the journal level to insist on the differences between authorship and contributions via acknowledgement in terms of interdisciplinarity with Shannon entropy [2]. To measure interdisciplinarity of acknowledgement contexts, we distinguish acknowledgement type into four categories; Investigation and Analysis, Materials and Resources, Peer Communication and Writing, by keyword matching. The result revealed that contributions for Investigation and Analysis via acknowledgement tend to have more interdisciplinarity than authorship. These datasets and findings lead to significant attention toward the credit not appearing in authorship and the discipline interdisciplinarity via acknowledgement statement.

[1] Paul-Hus, A., Mongeon, P., Sainte-Marie, M., & Larivière, V. (2017). The sum of it all: Revealing collaboration patterns by combining authorship and acknowledgements. Journal of Informetrics, 11(1), 80-87. [2] Shannon, C. E. (1948). A mathematical theory of communication. The Bell System Technical Journal, 27(3), 379-423.

ABSTRACT. Our civilization has become obsessed with controlling and avoiding any possible risks, rather than realizing that risk is a fundamental ingredient to succeed in any endeavor and to achieve sustainable development. We call this risk avoidance syndrome as the “zero-risk society”. Human societies are unavoidably exposed to diverse endogenous and exogenous shocks. While attempting to remove these risks, the “zero-risk society” syndrome has been contributing to the relative technological stagnation and innovation fatigue in the past decades, and carries the danger of eventually leading to more fragile societies in the future. To explore the causes and solutions, we propose five generating mechanisms, namely increasing wealth and aging, rising inequality of opportunity, ‘illusion of control’ created by technology, herding and imitation catalysed by media, and management shaped by extremes and overreactions. In order to falsify the importance of these five factors, we embody them into social network models, stochastic processes and statistical physical models. The models allow us to explore the importance of the five factors in changing the social risk propensity during wealth evolution and opinion dynamics. From the results, we observe how each factor pushes the society out-of-equilibrium and identify the presence of critical transitions separating risk-embracing societies and zero-risk societies. We analyse the existence of early warning signals and symptoms preceding these bifurcations. In order to address the potentially devastating crisis posed by a “zero-risk society” syndrome, especially at a time where Mankind needs to foster an extraordinary challenging energy transition to address climate change and Human-Earth environment sustainability, we propose (i) to promote a culture in education and entrepreneurship where failure is seen as part of the learning process, (ii) to aggressively fund explorative and high-risk projects, and (iii) to honor and reward risk-takers, explorers and creative inventors as “Hollywood stars”, thereby bringing creative momentum towards a sustainable human society. The models are confronted with practical case studies, such as venture capitalism, and inspire recommendations to build better sustainable social systems.

ABSTRACT. We propose a new multi-objective portfolio optimization incorporating idiosyncratic and systemic risks of financial networks. The two risks are measured by the idiosyncratic variance and the network clustering coefficient derived from the asset correlation networks, respectively. We construct three types of financial networks in which nodes indicate assets and edges are based on three correlation measures. Starting from the multi-objective model, we formulate and solve the asset allocation problem. We find that the optimal portfolios obtained through the multi-objective with networked approach have a significant over-performance in an out-of-sample framework. This is further supported by the less drawdown during the periods of stock market fluctuated downward. Through analyzing different datasets, we also show that improvements made to portfolio strategies are robust.

ABSTRACT. The recent Russian invasion into Ukraine highlighted the vulnerability of the network of global food supply. In an interconnected world the local loss of a basic staple crop can lead to shortages around the globe that are not limited to the crop itself but include all food products that rely on its availability as an input for their production. Therefore, a thorough risk assessment has to include the trade networks of multiple food products as well as their interconnections by production processes. The importance of losses caused by a lack of different inputs to produce a given food product and the resulting multidimensionality of systemic risk have not been widely acknowledged yet. Here we construct a multilayer-network model of global trade and production of food, based on the supply- and use-tables of a physical multi-regional input-output model for 191 countries and 125 products. We use the propagation of supply shocks on this network to compute the relative losses (RL) of a given product that a country suffers if another product is no longer produced in a different country. This total RL can exceed the losses attributed to direct trade of the former product by more than an order of magnitude, as we exemplify by studying the effects of a lost agricultural production in Ukraine on countries around the globe. Computing the RL for all combinations of shock-source and shock-target countries and food products, enables us to quantify systemic risk as a multidimensional object. We advocate to take into account this multidimensionality in the future design of more resilient supply chains to enhance global food security

ABSTRACT. As the largest blockchain platform that supports smart contracts, Ethereum has suffered more and more serious security problems. A large number of high-risk transactions such as phishing scams are increasing on Ethereum year by year. However，the existing phishing scams detection techniques are mainly developed from machine learning models, which lack of interpretability. To increase the interpretability of the model, this paper proposes a novel risk assessment technique to assess the risk of Ethereum accounts, which includes an anonymous score to measure transaction anonymity and a network propagation mechanism to formulate the relations between accounts and transactions. To evaluate the effectiveness of the risk assessment technique, we conduct related experiments on the real-world transaction data collected from Ethereum and obtained satisfactory conclusions.

ABSTRACT. In a prediction market, investors buy and sell contracts linked to the outcome of real-world events such as “Donald Trump will be Re-Elected President in 2024”. The contract prices range between $0 and $1, and represent investors perceived likelihood of the event’s outcome (1). Prices are driven by supply and demand as the investors react to new information they believe is pertinent to an event. Thus, the price of a contract is largely driven by human behavior as information travels over the network of investors. Additionally, the price of a contract rises (falls) to $1 ($0) as a market closes and the outcome of an event is realized; providing a ground truth that can be exploited to determine properties of financial systems (Figure 1, A). Prediction market literature has largely discussed the accuracy of these markets, asking whether the contract (e.g., a presidential candidate) with the highest price is actually predictive of the winner (2). Comparatively little research attempts to explain the dynamics that brought the market to an (in)accurate prediction. Here, study these dynamics using 2859 contracts provided by a popular online prediction market – PredictIt. Our data covers a wide variety of events from 2014 to 2021 ranging from elections, legislative actions, and career milestones of politicians. Despite this heterogeneity, we find striking universal patterns in the distribution of fluctuations and the number of traded contracts over time (Figure 1, B). In addition, we quantify the long-term memory of these time series with the Hurst exponent; allowing us to classify time series as mean-reverting, random-walk like, and trending. We find that Hurst exponent distributions also exhibit universal patterns, and time series with similar Hurst exponents have similar shapes. Our findings suggest that complexity of human interactions that drive prediction market dynamics can be embedded in a low-dimensional space of variables, opening the door to mechanistic modelling of these systems.