Incorporating Embedding to Topic Modeling for More Effective Short Text Analysis

With the growing abundance of short text content on websites, analyzing and comprehending these short texts has become a crucial task. Topic modeling is a widely used technique for analyzing short text documents and uncovering the underlying topics. However, traditional topic models face difficulties in accurately extracting topics from short texts due to limited content and their sparse nature. To address these issues, we propose an Embedding-based topic modeling (EmTM) approach that incorporates word embedding and hierarchical clustering to identify significant topics. Experimental results demonstrate the effectiveness of EmTM on two datasets comprising web short texts, Snippet and News. The results indicate a superiority of EmTM over baseline topic models by its exceptional performance in both classification accuracy and topic coherence metrics.