Context-based Evolution Research Over Topic And Topic Relations

Automatic extraction of semantic information and its evolution from large-scalecorpus has appealed to many experts and scholars in recent years. Topics are regardedas the latent semantic meanings underlying document collection. And topic evolutionfocuses on how to extract topics from document collection of multiple periods andrepresent their changes over time.However, it is found that semantic relatedness also exists among topics of thesame period by analyzing the results of topic model. This research proposes a newconcept, namely topic context, to describe the semantic relatedness among topics.Those topics, which often concur with a topic T in a document, are called the contextof the topic T. On the one hand, topic context can improve topic evolution results afterit is integrated into the calculation of topic relating. On the other hand, topic contextcan be used to extract the semantic relations among topics and construct semanticmap for them. Combining topic evolution and topic relation extraction, we can get thechanges of topic relations over time or topic relations evolution.This article has conducted meaningful research over the following aspects:(a) Propose the novel idea of topic context to mine semantic relations amongtopics of the same period.(b) Improve topic evolution by integrating topic context into topic relating.(c) Combine topic evolution and topic relations (calculated by using topiccontext) to implement topic relations evolution.The experiments on NPC&CPPCC news reports from2008to2012and NIPSscientific literature from2007to2011have shown that the method we proposed hasnot only improved the results of topic evolution but also mined the evolution ofsemantic relations among topics, which characterizes the relationships changes overtime.