摘 要：在基于评论的推荐算法中，文本特征通常会在训练中发生损失，导致最后的特征交互不足，影响推荐效果。为了获取包含更多信息的文本特征，得到更准确的预测值，文章提出一种基于评论的多特征融合深度协同推荐算法。该算法首先对评论文本进行预处理，然后通过由卷积文本网络和双向 GRU 网络构成的 C&G 模块进行多特征提取，同时引入注意力机制，最后在融合层进行融合预测。在 Amazon Digital Music 数据集上的实验结果表明，该算法的准确度较高，推荐效果较好。
关键词：协同过滤；深度学习；特征融合；注意力机制
DOI:10.19850/j.cnki.2096-4706.2022.011.023
基金项目：安徽理工大学研究生创新基金项目（2020CX2071）
中图分类号：TP391 文献标识码：A 文章编号：2096-4706（2022）11-0091-04
A Comment-based Multi Feature Fusion Deep Collaborative Recommendation Algorithm
HU Shengli, ZHANG Hongbin
(School of Computer Science and Engineering, Anhui University of Science and Technology, Huainan 232001, China)
Abstract: In the comment-based recommendation algorithm, text features are usually lost in training, resulting in insufficient interaction of the final features and affecting the recommendation effect. In order to obtain text features that contain more information and get more accurate prediction values, this paper proposes a comment-based multi feature fusion deep collaborative recommendation algorithm. Firstly, the algorithm preprocesses the comment text, then extracts multiple features through the C&G module composed of convolutional text network and bidirectional GRU network, and introduces the attention mechanism at the same time. Finally, the fusion prediction is carried out at the fusion layer. The experimental results on Amazon Digital Music dataset show that the algorithm has high accuracy and good recommendation effect.
Keywords: collaborative filtering; deep learning; feature fusion; attention mechanism
参考文献：
[1] WU L，HE X，WANG X，et al. A Survey on Accuracyoriented Neural Recommendation: From Collaborative Filtering to Information-rich Recommendation [J/OL].arXiv:2104.13030v3 [cs.IR].[2022-04-25].https://arxiv.org/abs/2104.13030.
[2] SARWAR B，KARYPIS J，KONSTAN J，et al. Item-Based Collaborative Filtering Recommendation Algorithms [EB/OL].[2022-04-15].https://doi.org/10.1145/371920.372071.
[3] KIM D，PARK C，OH J，et al. Convolutional Matrix Factorization for Document Context-Aware Recommendation [EB/OL].[2022-04-23].https://dl.acm.org/doi/pdf/10.1145/2959100.2959165.
[4] ZHENG L，NOROOZI V，YU P S. Joint deep modeling of users and items using reviews for recommendation [EB/OL].[2022-04-20].https://dl.acm.org/doi/10.1145/3018661.3018665.
[5] CATHERINE R，COHEN W．TransNets: Learning to transform for recommendation [EB/OL].[2022-04-16].https://dl.acm.org/doi/pdf/10.1145/3109859.3109878.
[6] SEO S，HUANG J，YANG H，et al. Interpretable Convolutional Neural Networks with Dual Local and Global Attention for Review Rating Prediction [EB/OL].[2022-04-13].https://dl.acm.org/doi/abs/10.1145/3109859.3109890.
[7] CHUNG J，GULCEHRE C，CHO K H，et al. Empirical Evaluation of Gated Recurrent Neural Networks on Sequence Modeling [J/OL].arXiv:1412.3555 [cs.NE].[2022-04-13].https://arxiv.org/abs/1412.3555.
作者简介：胡胜利（1978—），男，回族，安徽淮南人，研究生导师，副教授，博士，研究方向：推荐系统；张鸿斌（1997—），男，汉族，河南驻马店人，硕士研究生在读，研究方向：推荐系统。