融合情感偏离度特征的评论有用性预测模型——基于深度学习框架

Integrating Sentiment Deviation Features in Review Helpfulness Prediction Model—Based on a Deep Learning Framework

摘要: 随着电子商务平台的迅速发展和用户生成内容的爆发，在线评论成为消费者线上消费决策的重要依据。但海量评论中信息纷繁复杂，普通用户难以甄别有用信息，平台和商家亦难以高效筛选优质评论进行数据分析。因此，开发评论感知有用性预测模型的重要性日益凸显。现有研究主要依托评论文本长度、评分、评论者特征等传统变量，结合传统的回归分析方法构建预测模型，但在模型可解释性和深度学习应用方面存在不足。为此，本研究基于认知失调理论，首次引入情感偏离度——衡量评论文本情感与用户评分间不一致性的指标，并构建了融合attention机制的LSTM预测模型。本研究使用亚马逊图书评论数据进行严格的算法实验，结果表明：所提模型在预测准确性和稳定性上均显著优于基准方法，且情感偏离度显著降低了预测误差。本研究从理论上拓展了认知失调理论在评论有用性预测中的应用，为理解消费者决策过程提供了新视角；在实践上，模型可助力电商平台优化评论筛选和内容推荐，提升消费者体验及平台竞争力。

Abstract: With the rapid development of e-commerce platforms and the explosion of user-generated content (UGC), online reviews have become critical information sources for consumer decision-making. However, the complexity of massive comment data poses challenges for users to identify valuable information and for platforms to efficiently filter high-quality reviews for analysis, highlighting the growing importance of developing review helpfulness prediction (RHP) models. Existing studies primarily rely on traditional variables (e.g., review length, ratings, and reviewer characteristics) and regression-based methods, yet face limitations in model interpretability and the application of deep learning techniques. To address these gaps, this study introduces sentiment deviation—a novel metric quantifying the inconsistency between textual sentiment polarity and user ratings—grounded in cognitive dissonance theory, and constructs an LSTM prediction model integrated with an attention mechanism. Using Amazon food review data for empirical validation, our results demonstrate that the proposed model significantly outperforms baseline methods in prediction accuracy and stability, while the inclusion of sentiment deviation significantly reduces prediction errors. Theoretically, this research expands the application of cognitive dissonance theory in RHP and provides new insights into consumer decision-making mechanisms. Practically, the model empowers e-commerce platforms to optimize review filtering and content recommendation, thereby enhancing consumer experience and platform competitiveness.