Text summarization generates a summary that highlights key sentences while condensing all information from the original document into a few sentences. Abstractive and Extractive are the two types of summaries seen in general text summarization process. A hybrid summarization technique which generates an abstractive summary over an extractive summary is proposed in this paper. The initial phase incorporates the use of a semantic model to generate word embeddings for each sentence and these embeddings are used to improve the lack of semantic disintergrity in extractive summaries. The second phase takes in the concept of WordNet, Lesk algorithm and POS tagging for generating an abstractive summary from the extractive summary. The paper uses two different data sets: DUC 2004 and Daily mail/CNN for evaluating the performance over ROUGE and BLEU metric. The results highlight the relevance of developing hybrid approaches to summarization compared to complex abstractive techniques.
Automatic Text Summarization Using Word Embeddings