Micro Text Classification Based on Balanced Positive-Unlabeled Learning

2025-03-17

AI Summary

To address the dual macro- and micro-level class imbalance arising in text quality control—characterized by sparse negative instances, absent fine-grained labels, and high similarity between positive and negative samples—this paper proposes the Balanced Fine-Grained Positive-Unlabeled (BFGPU) learning framework. Methodologically, it reformulates coarse-grained binary classification as a fine-grained PU learning task; designs a theoretically grounded, two-level imbalance-aware PU loss function; and incorporates pseudo-label rebalancing with dynamic threshold adjustment. Evaluated on multiple public and real-world datasets, BFGPU consistently outperforms state-of-the-art methods, demonstrating superior robustness and accuracy even under extreme class imbalance. The framework effectively mitigates both label scarcity and feature-space ambiguity, enabling reliable fine-grained discrimination without explicit negative supervision.

Technology Category

• Machine Learning: Multi-class/Multi-label Learning & Extreme Classification
• Natural Language Processing: Learning & Optimization for NLP
• Reasoning under Uncertainty: Relational Probabilistic Models

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

In real-world text classification tasks, negative texts often contain a minimal proportion of negative content, which is especially problematic in areas like text quality control, legal risk screening, and sensitive information interception. This challenge manifests at two levels: at the macro level, distinguishing negative texts is difficult due to the high similarity between coarse-grained positive and negative samples; at the micro level, the issue stems from extreme class imbalance and a lack of fine-grained labels. To address these challenges, we propose transforming the coarse-grained positive-negative (PN) classification task into an imbalanced fine-grained positive-unlabeled (PU) classification problem, supported by theoretical analysis. We introduce a novel framework, Balanced Fine-Grained Positive-Unlabeled (BFGPU) learning, which features a unique PU learning loss function that optimizes macro-level performance amidst severe imbalance at the micro level. The framework's performance is further boosted by rebalanced pseudo-labeling and threshold adjustment. Extensive experiments on both public and real-world datasets demonstrate the effectiveness of BFGPU, which outperforms other methods, even in extreme scenarios where both macro and micro levels are highly imbalanced.

Problem

• Addresses extreme class imbalance in text classification.
• Transforms coarse-grained PN classification to fine-grained PU classification.
• Proposes BFGPU framework for improved performance in imbalanced scenarios.

Innovation

• Transforms PN classification to PU learning
• Introduces BFGPU with unique loss function
• Uses rebalanced pseudo-labeling and threshold adjustment