GOW-LDA: Applying Term Co-occurrence Graph Representation in LDA Topic Models Improvement

摘要

In this paper, we demonstrate a novel approach in topic model exploration by applying word co-occurrence graph or graph-of-words (GOW) in order to produce more informative extracted latent topics from a large document corpus. According to the Latent Dirichlet Allocation (LDA) algorithm, it only considers the words occurrence independently via probabilistic distributions. It leads to the failure in term's relationship recognition. Hence in order to overcome this disadvantage of traditional LDA, we propose a novel approach, called GOW-LDA. The GOW-LDA is proposed that combines the GOW graph used in document representation, the frequent subgraph extracting and distribution model of LDA. For evaluation, we compare our proposed model with the traditional one in different classification algorithms. The comparative evaluation is performed in this study by using the standardized datasets. The results generated by the experiments show that the proposed algorithm yields performance respectably.