ZeroDL: Zero-shot Distribution Learning for Text Clustering via Large Language Models

2024-06-19

arXiv.org

AI Summary

Current large language models (LLMs) struggle to accurately model text distributions via prompting alone, limiting unsupervised text clustering performance. To address this, we propose ZeroDL, a zero-shot distribution learning framework that requires no labeled data, human-crafted demonstrations, or model fine-tuning. ZeroDL introduces a novel three-stage paradigm: (1) zero-shot open-ended generation from LLMs, (2) semantic aggregation in an induced latent space, and (3) implicit distribution modeling via meta-information distillation and geometric clustering. By leveraging in-context learning to elicit generative capabilities and jointly optimizing semantic coherence and structural fidelity, ZeroDL automatically uncovers the intrinsic distributional geometry of unlabeled texts. Extensive experiments on standard text clustering benchmarks demonstrate that ZeroDL consistently outperforms state-of-the-art unsupervised and self-supervised methods, validating the effectiveness, generalizability, and practicality of zero-shot distribution learning.

Technology Category

• Natural Language Processing: (Large) Language Models
• Machine Learning: Large Multimodal Models (LMMs)
• Search and Optimization: Distributed Search

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

The recent advancements in large language models (LLMs) have brought significant progress in solving NLP tasks. Notably, in-context learning (ICL) is the key enabling mechanism for LLMs to understand specific tasks and grasping nuances. In this paper, we propose a simple yet effective method to contextualize a task toward a specific LLM, by (1) observing how a given LLM describes (all or a part of) target datasets, i.e., open-ended zero-shot inference, and (2) aggregating the open-ended inference results by the LLM, and (3) finally incorporate the aggregated meta-information for the actual task. We show the effectiveness of this approach in text clustering tasks, and also highlight the importance of the contextualization through examples of the above procedure.

Problem

• Enabling LLMs to perform tasks not fully describable in prompts
• Improving text clustering via zero-shot distribution learning
• Generating class labels to enhance LLM task understanding

Innovation

• Utilizes open-ended zero-shot inference
• Aggregates inference results effectively
• Incorporates meta-information for tasks