Research Of Topic Tracking Based On HowNet And Topic Renewal

Topic tracking,as one task of Topic Detection and Topic Tracking(TDT),is an information processing technology,which is tracking known topic from the information flow of news media.General algorithm of topic tracking includes three basic modules: topic/news representation,similarity calculation and threshold comparison.Additionally, topic/news representation has two important parts,which are feature extraction and weight calculation.HowNet is a database of common-sense knowledge,which describes concepts in lexicons of Chinese and English equivalents,and unveils relations between concepts and between concept attributes.This thesis implements an algorithm of topic tracking based on topic renewal(TR). TR algorithm continuously adds new contents related to the topic,by using the theory of adaptive topic tracking,in order to update the topic vector and enhance the adaptability of it.This thesis proposes and implements an algorithm of topic tracking based on the normalization of characteristic terms(NCT) by using HowNet.NCT algorithm is implemented under the framework of ordinary topic tracking algorithm.And during the procedure of topic/news representation,the algorithm calculates the similarity of two words by using HowNet.Furthermore,this thesis also proposes and implements an algorithm of topic tracking based on topic renewal and the normalization of characteristic terms by using HowNet(TR&NCT).TR&NCT algorithm combines the advantages of the two algorithms mentioned above.The experiment,which is performed on TDT5 corpus,shows that:The performances of TR,NCT and TR&NCT algorithm are better than the general topic tracking algorithm;In addition,TR&NCT algorithm performs the best among all the algorithms.