用小模型实现情感分类新突破，比大模型更准更敏感。
CMHL: Contrastive Multi-Head Learning for Emotionally Consistent Text Classification
- 设计单模型架构，联合预测情绪、效价和强度，增强逻辑一致性。
- 在Dair-AI数据集上达到93.75%的F1分数，超越56倍大的模型。
- 适合需要高敏感度与可解释性的心理健康监测场景。
文本情感分类（TEC）是自然语言处理中最具挑战性的任务之一。现有方法依赖大语言模型（LLMs）和多模型集成。本文挑战了规模越大性能越好的假设，提出一种新型单模型架构CMHL，通过三个关键创新提升情感逻辑一致性：(1) 多任务学习，同时预测主要情绪、效价和强度；(2) 基于Russell环形模型的心理学辅助监督；(3) 新颖的对比矛盾损失，惩罚相互冲突的预测（如同时高置信度判断喜悦与愤怒）。仅用125M参数，模型在dair-ai Emotion数据集上取得93.75%的新最优F1分数，显著优于56倍大的LLMs及sLM集成（86.13%-93.2%）。在Reddit Suicide Watch和Mental Health Collection（SWMH）数据集上也表现出跨领域泛化能力，F1达72.50%（优于MentalBERT和MentalRoBERTa的68.16%-72.16%），召回率73.30%（高于67.05%-70.89%），显著提升心理危机检测敏感度。结果表明，架构智能（而非参数量）才是推动TEC进步的关键。通过嵌入心理学先验与显式一致性约束，一个精心设计的单模型可超越大规模LLMs与复杂集成，为情感计算提供高效、可解释且临床相关的范式。
原文摘要 · Abstract (English)
Textual Emotion Classification (TEC) is one of the most difficult NLP tasks. State of the art approaches rely on Large language models (LLMs) and multi-model ensembles. In this study, we challenge the assumption that larger scale or more complex models are necessary for improved performance. In order to improve logical consistency, We introduce CMHL, a novel single-model architecture that explicitly models the logical structure of emotions through three key innovations: (1) multi-task learning that jointly predicts primary emotions, valence, and intensity, (2) psychologically-grounded auxiliary supervision derived from Russell's circumplex model, and (3) a novel contrastive contradiction loss that enforces emotional consistency by penalizing mutually incompatible predictions (e.g., simultaneous high confidence in joy and anger). With just 125M parameters, our model outperforms 56x larger LLMs and sLM ensembles with a new state-of-the-art F1 score of 93.75\% compared to (86.13\%-93.2\%) on the dair-ai Emotion dataset. We further show cross domain generalization on the Reddit Suicide Watch and Mental Health Collection dataset (SWMH), outperforming domain-specific models like MentalBERT and MentalRoBERTa with an F1 score of 72.50\% compared to (68.16\%-72.16\%) + a 73.30\% recall compared to (67.05\%-70.89\%) that translates to enhanced sensitivity for detecting mental health distress. Our work establishes that architectural intelligence (not parameter count) drives progress in TEC. By embedding psychological priors and explicit consistency constraints, a well-designed single model can outperform both massive LLMs and complex ensembles, offering a efficient, interpretable, and clinically-relevant paradigm for affective computing.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。