为挖掘潜藏在不同场景下的产品创新方向，从“场景-特征”关联视角出发，提出一种基于用户在线评论的产品创新方向识别框架。通过词频统计、TF-IDF等方法构建场景与特征的词库，并利用“场景-特征”词共现模式，以明晰两者的关联关系；构建DeBERTa-BiGRU-MHA模型进行情感判别，并测算不同“场景-特征”的用户关注强度和情感得分；计算用户对不同场景的需求满足得分和重视度水平，并运用KANO模型、前景理论，以捕获产品创新方向。选取汽车之家、易车、太平洋汽车平台中的汽车评论数据进行实验，研究发现，DeBERTa-BiGRU-MHA模型在多项性能指标上均优于基线模型，并通过对比产品参数迭代情况验证了前述所提框架的可行性。

o explore potential product innovation directions hidden in different scenarios, this study proposes a framework for identifying product innovation directions based on user online reviews, from the “scenario-feature” association perspective. A vocabulary of scenarios and features is constructed using methods such as word frequency statistics and TF-IDF, and the co-occurrence patterns of “scenario-feature” terms are used to clarify the relationship between the two; a DeBERTa-BiGRU-MHA model is constructed for sentiment analysis, and the user attention intensity and sentiment scores for different “scenario-feature” pairs are calculated; We calculate users’ demand satisfaction scores and priority levels for different scenarios, and apply the KANO model and prospect theory to capture product innovation directions. Using automotive review data from the Autohome, Yiche and PCauto platforms for experimentation, our findings reveal that the DeBERTa-BiGRU-MHA model outperforms baseline models across multiple performance metrics. Furthermore, by comparing product parameter iterations, we have validated the feasibility of the proposed framework.