Identification of topic evolution: network analytics with piecewise linear representation and word embedding

Understanding the evolutionary relationships among scientific topics and learning the evolutionary process of innovations is a crucial issue for strategic decision makers in governments, firms and funding agencies when they carry out forward-looking research activities. However, traditional co-word network analysis on topic identification cannot effectively excavate semantic relationship from the context, and fixed time window method cannot scientifically reflect the evolution process of topics. This study proposes a framework of identifying topic evolutionary pathways based on network analytics: Firstly, keyword networks are constructed, in which a piecewise linear representation method is used for dividing time periods and a Word2Vec mode is used for capturing semantics from the context of titles and abstracts; Secondly, a community detection algorithm is used to identify topics in networks; Finally, evolutionary relationships between topics are represented by measuring the topic similarity between adjacent time periods, and then topic evolutionary pathways are identified and visualized. An empirical study on information science demonstrates the reliability of the methodology, with subsequent empirical validations.