🤖 AI Summary
This work addresses key limitations of existing topic models—such as BERTopic—including stochastic instability, ambiguous embeddings, and reliance on a single data modality. To overcome these issues, we propose a trimodal graph-based topic modeling approach that integrates semantic embeddings, TF-IDF features, and metadata. Our method constructs a denoised graph using Mutual kNN and shared nearest neighbors, followed by consensus Leiden clustering and iterative centroid refinement. Additionally, we introduce a boundary-case-driven prototypical representation for topics. Evaluated across multiple benchmark datasets, the proposed method achieves an average Normalized Mutual Information (NMI) score of 0.575, substantially outperforming state-of-the-art alternatives. It ensures 100% corpus coverage with no outliers and is publicly available as an open-source PyPI library.
📝 Abstract
Topic modeling extracts latent themes from large text collections, but leading approaches like BERTopic face critical limitations: stochastic instability, loss of lexical precision ("Embedding Blur"), and reliance on a single data perspective.
We present TriTopic, a framework that addresses these weaknesses through a tri-modal graph fusing semantic embeddings, TF-IDF, and metadata. Three core innovations drive its performance: hybrid graph construction via Mutual kNN and Shared Nearest Neighbors to eliminate noise and combat the curse of dimensionality; Consensus Leiden Clustering for reproducible, stable partitions; and Iterative Refinement that sharpens embeddings through dynamic centroid-pulling. TriTopic also replaces the "average document" concept with archetype-based topic representations defined by boundary cases rather than centers alone.
In benchmarks across 20 Newsgroups, BBC News, AG News, and Arxiv, TriTopic achieves the highest NMI on every dataset (mean NMI 0.575 vs. 0.513 for BERTopic, 0.416 for NMF, 0.299 for LDA), guarantees 100% corpus coverage with 0% outliers, and is available as an open-source PyPI library.