Research of Topic Detection Integrated with Correlation and Graph Analysis Theory

How to fast and effectively detect valuable topics and their developing trends from enormous texts has become a huge technical challenge. Topic detection is an important tool for implementing this task. Currently, most methods carried out their research under the assumption that topics were independent, which ignored inherent relationships between topics. In order to solve the above problems, topic correlation and graph analytical methods are integrated to establish a theoretical framework for topic detection. Firstly, correlated topic model is introduced to measure topic relevance and generate semantic term graph. Secondly, correlation graph algorithm is proposed in which the quantified topic correlation is adopted to extract co-occurrence relations. Co-occurrence term graph is obtained to realize the mining of important and hidden topics. Topic correlation is simultaneously used in analyzing semantic and co-occurrence relationships in our approach to discover meaningful topics and their changing trends. The simulation results verify the validity of the proposed theory.