Topic Modeling with Variable Neighborhood Search

In this study, we introduce a topic modeling approach using variable neighborhood search (VNS), it is denoted as VNStopic. VNS is global optimization meta-heuristics, the idea behind it is to find neighborhood solutions and gradually change these solutions to escape from valleys that contain local minima. VNStopic is proposed to alleviate the problems of traditional topic modeling approaches that often rely on local search techniques for parameter estimation. The proposed approach uses VNS to systematically explore areas that are further away from the previous best solutions/topics initially identified. Experimental results show that the proposed approach outperforms baselines in topic coherence.