A Network Topic Detection Approach Using the Reduction of Tolerance Semantic Blocks
Meng Zu-qian · Journal of Chinese Computer Systems · 2013
To overcome the shortcomings in the existing network topic detection approaches based on word clustering,this paper makes full use of the semantic information given by semi-structured documents and co-occurrence frequency in the word semantic context to define the semantic similarity degree between words,and then establish the word co-occurrence network for a given document set,thus modeling the semantic association between words. On these bases,a concept of tolerance semantic block is proposed,and by using the construction,abruption,and reduction of tolerance semantic blocks,a network topic detection approach is presented.The results obtained by this approach are relatively stable,and they are both brief and expressive word subsets. Therefore,the proposed approach is well applicable to network topic detection. Numerical experimentation also shows that the proposed approach really has these advantages and that it is effective and feasible for network topic detection.