通过对比正反推理路径，提升隐含情感分析准确率
Application of Multiple Chain-of-Thought in Contrastive Reasoning for Implicit Sentiment Analysis
- 构建正反双向推理链，对比两种假设的合理性
- 在多个模型规模上超越现有方法，达当前最优性能
- 适合需要精准识别隐含情绪的场景，如舆情分析
隐含情感分析旨在挖掘被模糊或比喻语言掩盖的微妙情绪。为此，需依赖大语言模型与多步推理来识别未明确表达的情感。本文提出一种新型双逆向思维链框架（DRCR），受演绎推理启发，包含三个步骤：1）假设情感极性并推导推理过程；2）否定初始假设并生成新推理路径；3）对比两条推理链以确定最终情感极性。在此基础上，进一步提出三重逆向思维链（TRCR）框架，解决随机假设带来的局限性。两种方法均结合对比机制与多步推理，在不同模型规模下均显著提升隐含情感分类准确率，实验验证其有效性。结果表明，对比推理与多步推理的融合对隐含情感分析具有显著优势。
原文摘要 · Abstract (English)
Implicit sentiment analysis aims to uncover emotions that are subtly expressed, often obscured by ambiguity and figurative language. To accomplish this task, large language models and multi-step reasoning are needed to identify those sentiments that are not explicitly stated. In this study, we propose a novel Dual Reverse Chain Reasoning (DRCR) framework to enhance the performance of implicit sentiment analysis. Inspired by deductive reasoning, the framework consists of three key steps: 1) hypothesize an emotional polarity and derive a reasoning process, 2) negate the initial hypothesis and derive a new reasoning process, and 3) contrast the two reasoning paths to deduce the final sentiment polarity. Building on this, we also introduce a Triple Reverse Chain Reasoning (TRCR) framework to address the limitations of random hypotheses. Both methods combine contrastive mechanisms and multi-step reasoning, significantly improving the accuracy of implicit sentiment classification. Experimental results demonstrate that both approaches outperform existing methods across various model scales, achieving state-of-the-art performance. This validates the effectiveness of combining contrastive reasoning and multi-step reasoning for implicit sentiment analysis.
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。