Prioritized Named Entity Driven LDA for Document Clustering

Topic modeling methods like LSI, pLSI, and LDA have been widely studied in text mining domain for various applications like document representation, document clustering/classification, information retrieval, etc. However, such unsupervised methods are effective over corpus with well separable topics. In real-world applications, topics might be of highly overlapping in nature. For example, a news corpus of different terror attacks has highly overlapping keywords across reporting of different terror events. In this paper, we propose a variant of LDA, named as Prioritized Named Entity driven LDA (PNE-LDA), which can address the issue of overlapping topics by prioritizing named entities related to the topics. From various experimental setups, it is observed that the proposed method outperforms its counterparts in entity driven overlapping topics.