Online ngram-enhanced topic model for academic retrieval

Abstract

Applying topic model to text mining has achieved a great success. However, state-of-art topic modeling methods still have potential to improve in academic retrieval field. In this paper, we propose an online unified topic model, which is ngram-enhanced. Our model discovers topics with unigrams as well as topical bigrams and is updated by an online inference algorithm with the new incoming data streams. On this basis, we combine our model into the query likelihood model and develop an integrated academic searching system. Experiment results on ACM collection show that our proposed methods outperform the existing approaches on document modeling and searching accuracy. Especially, we prove the efficiency of our system on academic retrieval problem.