Abstract
This paper presents a clustering-based approach to topic modeling via analyzing word networks based on the adaptation of a community detection algorithm. Word networks are constructed with different word representations, and two types of topic assignments are introduced. Topic coherence score and the document clustering results are reported for topic model evaluation. Experimental results showed that it achieved comparable results with the current best. It also showed that the proposed approach produced a higher performance as the number of most relevant words gets larger in$C_{cv}$coherence score.