Topic Modeling for Expertise Finding using Fuzzy LDA-sBERT Clusters
Dipendra Sharma Kafle, Esma Talhi, Mickaël Coustaty, Antoine Doucet · 2024
Topic modeling identifies and organizes topics in document collections, often used in expert finding by treating expertise as a topic or set of topics. Most models rely on variations of LDA to acquire coherent topics, but LDA struggles to uncover latent relationships between those topics. In this work, we present our model that combines LDA with S-BERT, further refining topics with clustering. More specifically, we propose fuzzy clustering using Fuzzy C-means in order to discover latent relationships among the topics. We compare our model with baselines using CORD-19 dataset. Through experiments, we demonstrate that our fuzzy LDA-sBERT model outperforms all baseline models.