Scalable Parameter-Light Spectral Method for Clustering Short Text Embeddings with a Cohesion-Based Evaluation Metric

2025-11-24

AI Summary

Short-text embedding clustering typically requires a pre-specified number of clusters, limiting its practicality in fully unsupervised settings. To address this, we propose a scalable spectral clustering framework that automatically estimates the optimal number of clusters by analyzing the eigenstructure of the Laplacian matrix—constructed from cosine similarity—eliminating manual specification. We further introduce Cohesion Ratio, an information-theoretic cohesiveness metric highly correlated with external evaluation measures, enabling reliable unsupervised quality assessment. Our method is compatible with standard clustering algorithms (e.g., K-Means, Hierarchical Agglomerative Clustering) and incorporates adaptive sampling for computational efficiency. Extensive experiments across six short-text benchmarks and four embedding models demonstrate that our approach significantly outperforms mainstream parameter-free methods—including HDBSCAN and OPTICS—in clustering accuracy, while maintaining high efficiency, robustness, and broad applicability.

Technology Category

• Machine Learning: Clustering
• Natural Language Processing: Summarization
• Data Mining & Knowledge Management: Anomaly/Outlier Detection

Application Category

• Web Mining and Content Analysis: Normalization, clustering, classification, and summarization of Web text
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics

Abstract

Clustering short text embeddings is a foundational task in natural language processing, yet remains challenging due to the need to specify the number of clusters in advance. We introduce a scalable spectral method that estimates the number of clusters directly from the structure of the Laplacian eigenspectrum, constructed using cosine similarities and guided by an adaptive sampling strategy. This sampling approach enables our estimator to efficiently scale to large datasets without sacrificing reliability. To support intrinsic evaluation of cluster quality without ground-truth labels, we propose the Cohesion Ratio, a simple and interpretable evaluation metric that quantifies how much intra-cluster similarity exceeds the global similarity background. It has an information-theoretic motivation inspired by mutual information, and in our experiments it correlates closely with extrinsic measures such as normalized mutual information and homogeneity. Extensive experiments on six short-text datasets and four modern embedding models show that standard algorithms like K-Means and HAC, when guided by our estimator, significantly outperform popular parameter-light methods such as HDBSCAN, OPTICS, and Leiden. These results demonstrate the practical value of our spectral estimator and Cohesion Ratio for unsupervised organization and evaluation of short text data. Implementation of our estimator of k and Cohesion Ratio, along with code for reproducing the experiments, is available at https://anonymous.4open.science/r/towards_clustering-0C2E.

Problem

• Estimating cluster count from Laplacian eigenspectrum structure
• Scaling clustering to large datasets via adaptive sampling
• Evaluating cluster quality without labels using Cohesion Ratio

Innovation

• Spectral method estimates clusters from Laplacian eigenspectrum
• Adaptive sampling enables scalable clustering on large datasets
• Cohesion Ratio evaluates cluster quality without ground-truth labels