A Non-parametric Multi-view Model for Short Text Clustering

Short text clustering has become increasingly important with the popularity of social media. Existing methods fall into three paradigms: those based on topic models, deep representation learning, and large language models (LLMs). The first relies on the bag-of-words assumption, which ignores word order and semantic information. The second achieves strong performance but lacks interpretability. The third captures the most comprehensive semantic information but requires high-frequency, real-time API calls and incurs higher computational costs during model inference. In addition, the requirement to predefine the number of clusters remains a key issue. In this paper, we propose a Non-parametric Multi-View Model (NMVM) for short text clustering, which represents texts using two complementary views: bag-of-words and text embeddings. NMVM incorporates semantic information into a generative framework by integrating the Dirichlet process multinomial mixture model and the Dirichlet process gaussian mixture model. Notably, we propose and derive a novel and efficient clustering algorithm based on collapsed Gibbs sampling, which automatically infers the number of clusters from the data and identifies representative words for each cluster, thereby enabling fully non-parametric and interpretable clustering. Extensive experiments on six real-world datasets demonstrate the superiority of the proposed model over several strong baselines. Notably, our algorithm converges rapidly and demonstrates high efficiency. The source code is publicly available at https://github.com/chehaoa/NMVM.