Abstract:

 To solve the topic excursion problem in hot topic tracking process, an adaptive hot topic tracking model based on relevance feedback was proposed. To obtain the topic dynamic evolution procedure accurately, firstly, a correlation factor was introduced into TF-IDF (term frequency—inverse document frequency) for extracting feature words. Secondly, a formula for computing relevance degree between story and topic was constructed, and a discriminant function for determining whether the new feature word could be added into the topic lexicon was also constructed. At the same time, the methods for dynamically computing adaptive updating threshold and adaptive correlation threshold were given. Finally, in the updated topic lexicon, our approach gave the new weight to each feature word according to its contribution to the topic. The experimental results showed that the proposed method could reduce the false alarm rate 0018 and the miss alarm rate 0.063 compared with the other 3 trackers in the hot topic tracking process, which concluded that this proposed technique was more suitable for solving the problem of topic drift.

Key words: adaptive updating threshold, topic excursion, weight updating, topic tracking, relevance feedback