College of Computer Science and Technology, Jilin University, Changchun 130012, China.
Key Laboratory of Symbol Computation and Knowledge Engineering (Jilin University), Ministry of Education, Changchun 130012, China.
Comput Intell Neurosci. 2022 Oct 18;2022:4940401. doi: 10.1155/2022/4940401. eCollection 2022.
With the rapid development of e-commerce, recommendation system has become one of the main tools that assists users in decision-making, enhances user's experience, and creates economic value. Since it is difficult to explain the implicit features generated by matrix factorization, explainable recommendation system has attracted more and more attention recently. In this paper, we propose an explainable fast recommendation model by combining fine-grained sentiment in review data (FSER, (Fast) Fine-grained Sentiment for Explainable Recommendation). We innovatively construct user-rating matrix, user-aspect sentiment matrix, and item aspect-descriptive word frequency matrix from the review-based data. And the three matrices are reconstructed by matrix factorization method. The reconstructed results of user-aspect sentiment matrix and item aspect-descriptive word frequency matrix can provide explanation for the final recommendation results. Experiments in the Yelp and Public Comment datasets demonstrate that, compared with several classical models, the proposed FSER model is in the optimal recommendation accuracy range and has lower sparseness and higher training efficiency than tensor models or neural network models; furthermore, it can generate explanatory texts and diagrams that have high interpretation quality.
随着电子商务的快速发展，推荐系统已成为辅助用户决策、提升用户体验和创造经济价值的主要工具之一。由于难以解释矩阵分解生成的隐含特征，可解释性推荐系统最近受到了越来越多的关注。在本文中，我们提出了一种可解释的快速推荐模型，该模型结合了评论数据中的细粒度情感（FSER，（Fast）细粒度情感用于可解释推荐）。我们创新性地从基于评论的数据中构建了用户评分矩阵、用户方面情感矩阵和项目方面描述性词汇频率矩阵，并通过矩阵分解方法对这三个矩阵进行了重构。用户方面情感矩阵和项目方面描述性词汇频率矩阵的重构结果可以为最终的推荐结果提供解释。在 Yelp 和 Public Comment 数据集上的实验表明，与几个经典模型相比，所提出的 FSER 模型在最优推荐准确性范围内，与张量模型或神经网络模型相比，稀疏度更低、训练效率更高；此外，它还可以生成具有高解释质量的解释性文本和图表。
