Policy Texts with Topic Detection and Information Entropy Evolution Analysis

EasyChair Preprint no. 6632, version 4

Abstract

Talent policy has always been an important tool for countries to seize the talent highland and seek innovative development. However, talent policy text with complex themes, uneven distribution and unclear structure have caused great trouble for scholars to perform Talent Management. We constructed a large-scale unannotated corpus related to talent policy from Sougou Engine and collected 287 talent policies from the local government in Guangdong Province, China, which has been fuelled by rapid market growth and an unprecedented population surge resulted in a labor force of record size. We proposed a novel clustering model called LDA2Vec, which merges LDA and Word2Vec, and performed topic evolution analysis regarding topic similarity and topic entropy. The talent policies mainly included five topics: (i) talent introduction; (ii) talent training; (iii) talent guarantee; (iv) talent incentive, (v) talent evaluation. Talent policy in Guangdong province has gone through an evolutionary process from monism to pluralism. The introduction of China's innovation-driven strategy in 2013 as the dividing line directly impacts topic content and the intensity of five topics particularly.

Keyphrases: Lda2vec., Semantic Similarity., Talent management., Topic Entropy., Topic Evolution.

@Booklet{EasyChair:6632,
  author = {Li-Xia Chen and Guo-He Feng},
  title = {Policy Texts with Topic Detection and Information Entropy Evolution Analysis},
  howpublished = {EasyChair Preprint no. 6632},

  year = {EasyChair, 2021}}