Paper Reading AI Learner

Application of Multiple Chain-of-Thought in Contrastive Reasoning for Implicit Sentiment Analysis

2025-03-10 10:10:50
Liwei Yang, Xinying Wang, Xiaotang Zhou, Zhengchao Wu, Ningning Tan

Abstract

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.

Abstract (translated)

隐式情感分析旨在揭示那些微妙表达的情感,这些情感往往被歧义和比喻语言所掩盖。为了完成这一任务,需要大型语言模型和多步推理来识别那些未明确表述的情感。在这项研究中,我们提出了一种新颖的双反向链推理(DRCR)框架,以增强隐式情感分析的表现。受演绎推理启发,该框架包含三个关键步骤:1) 假设一种情感极性并推导出一个推理过程;2) 否定最初的假设,并推导出一个新的推理过程;3) 对比这两种推理路径,得出最终的情感极性。在此基础上,我们还引入了三重反向链推理(TRCR)框架来解决随机假设的局限性。两种方法均结合对比机制和多步推理,显著提高了隐式情感分类的准确性。实验结果表明,两种方法在各种模型规模上都优于现有方法,并达到了最先进的性能水平。这验证了将对比推理与多步骤推理相结合对于隐式情感分析的有效性。

URL

https://arxiv.org/abs/2503.07140

PDF

https://arxiv.org/pdf/2503.07140.pdf


Tags
3D Action Action_Localization Action_Recognition Activity Adversarial Agent Attention Autonomous Bert Boundary_Detection Caption Chat Classification CNN Compressive_Sensing Contour Contrastive_Learning Deep_Learning Denoising Detection Dialog Diffusion Drone Dynamic_Memory_Network Edge_Detection Embedding Embodied Emotion Enhancement Face Face_Detection Face_Recognition Facial_Landmark Few-Shot Gait_Recognition GAN Gaze_Estimation Gesture Gradient_Descent Handwriting Human_Parsing Image_Caption Image_Classification Image_Compression Image_Enhancement Image_Generation Image_Matting Image_Retrieval Inference Inpainting Intelligent_Chip Knowledge Knowledge_Graph Language_Model LLM Matching Medical Memory_Networks Multi_Modal Multi_Task NAS NMT Object_Detection Object_Tracking OCR Ontology Optical_Character Optical_Flow Optimization Person_Re-identification Point_Cloud Portrait_Generation Pose Pose_Estimation Prediction QA Quantitative Quantitative_Finance Quantization Re-identification Recognition Recommendation Reconstruction Regularization Reinforcement_Learning Relation Relation_Extraction Represenation Represenation_Learning Restoration Review RNN Robot Salient Scene_Classification Scene_Generation Scene_Parsing Scene_Text Segmentation Self-Supervised Semantic_Instance_Segmentation Semantic_Segmentation Semi_Global Semi_Supervised Sence_graph Sentiment Sentiment_Classification Sketch SLAM Sparse Speech Speech_Recognition Style_Transfer Summarization Super_Resolution Surveillance Survey Text_Classification Text_Generation Time_Series Tracking Transfer_Learning Transformer Unsupervised Video_Caption Video_Classification Video_Indexing Video_Prediction Video_Retrieval Visual_Relation VQA Weakly_Supervised Zero-Shot