Topic Tracking Algorithm Based on Topic Structure Characteristics

Abstract

Topic tracking task is used for public opinion monitoring, and its key technology is text classification algorithm. However, existing text classification algorithms need large-scale train corpus during training, while topic tracking task only provides a small amount of train corpus, resulting in that it has poor performance. We analyze story description contents in train corpus, and find that the report description contents of the same topic have the topic structure characteristics of high similarity. We use topic information to represent highly similar topic structure characteristics to make up for the lack of train corpus in text classification algorithm, and fuse topic structure characteristics and text classification algorithm to make the topic tracking algorithm consider topic structure characteristics, and propose Topic Tracking Algorithm based on Topic Structure characteristics (TTATS). To verify its performance, we carry out quantitative and qualitative experiments. The experimental results of multiple dimensions show that it has preferable topic tracking performance.