Automatic Topic Detection Based on Document Concept Similarity

Traditional topic detection methods basically use vector space model to calculate document similarity, which always calculate similarity by single word and neglect conception of words and conception similarity. Aimed at this problem, this paper first extracts event arguments from Web news corpus. Second, it decomposes word to concept set, and calculates concept similarity, word similarity and document similarity. Finally, it resolves automatic topic detection based on concept similarity calculation. Experiment results show that, compared to the traditional topic detection methods, the new method can improve precision and recall of topic detection efficiently.