A Hot Topic Detection Approach on Chinese Microblogging

Abstract

This paper proposes an approach of hot topic detection on microblogging, that is, do microblogging texts clustering and heat evaluation so as to find hot topics. For the shortness and sparseness and the existence of synonyms and polysemy of the massive microblogging texts, when modeling the microblogging texts, microblogging text vectors are mapped to low-dimensional feature vector space to achieve the purpose of dimensionally reduction and denoising using Latent Semantic Indexing (LSI). For the shortcomings of traditional single-pass algorithm on topic detection, this paper proposes a two-level Hierarchical Agglomerative Clustering combined single-pass clustering method. Finally, according to text and propagation characteristics of microblogging, this paper proposes the topic heat evaluation model. Experimental studies on real world microblogging data show the method in this paper works well on massive microblogging texts and can effectively dig out hot topics.