Research on Explainable Rating Prediction by Fusing Attribute Preference and Multi-order Interaction Information
-
摘要: 已有推荐系统主要基于用户−项目交互矩阵来学习用户和项目的向量表示, 而当交互矩阵稀疏时, 推荐系统的精度较低, 推荐的结果缺乏可解释性. 考虑到用户−项目交互行为中的评分标签信息, 提出了一种融合属性偏好和多阶交互信息的可解释评分预测方法, 并根据属性偏好对推荐结果进行解释. 首先, 基于注意力机制分析了用户和项目属性信息与评分标签的关系, 建模了节点的属性偏好特征表示; 然后, 聚合了用户−项目交互矩阵中节点自身、交互邻居和评分标签信息, 通过图神经网络学习了节点的多阶交互行为特征表示; 最后, 融合了节点的属性偏好特征和交互行为特征, 在异质类型信息空间下学习了用户和项目的语义特征表示, 利用多层感知机实现了评分预测, 并在MovieLens和Douban数据集上验证了方法的有效性. 实验结果表明, 所提方法在平均绝对误差(Mean absolute error, MAE)和均方根误差(Root mean square error, RMSE)指标上有效提高了推荐系统的精度, 缓解了数据稀疏场景下推荐模型性能较低的问题, 提升了推荐结果的可解释性.Abstract: Existing recommender systems mainly learn the vector representation of users and items based on their interaction matrix. However, when the interaction matrix is sparse, the accuracy of recommender systems is low and the recommendation results lack explanation. Considering the rating tag information from user-item interaction behaviors, this paper proposes an explainable rating prediction method by fusing attribute preference and multi-order interaction information, and explains the recommendation results through the attribute preference. First, based on the attention mechanism, we analyze the relationship between attribute information and rating tags for users and items, and model the attribute preference embedding of nodes. Second, by aggregating information about nodes, interactive neighbors and rating tags from user-item interaction matrix, we learn the multi-order interaction behavior embedding of nodes with graph neural networks. Finally, after fusing attribute preference embedding and interaction behavior embedding of nodes, the semantic embeddings of users and items are learned in the heterogeneous type-specific spaces. We make the rating prediction through the multi-layer perceptron and verify the validity of the method on the MovieLens and Douban datasets. The experimental results show that the proposed method can effectively improve the accuracy of recommender systems in mean absolute error (MAE) and root mean square error (RMSE) indexes, which alleviates the problem of poor performance of the model in the scenario of sparse data, and improves the interpretation of recommendation results.1)
1 1https://grouplens.org/datasets/movielens/ 2https://movie.douban.com/ 2)2 2https://movie.douban.com/ -
图 1 融合属性偏好和多阶交互信息的评分预测
Fig. 1 Rating prediction by fusing attribute preference and multi-order interaction information
图 3 几种方法在ML-L-S数据集上不同稀疏性的MAE结果
Fig. 3 MAE results of different methods on ML-L-S dataset with different sparsities
图 4 几种方法在ML-L-S数据集上不同稀疏性的RMSE结果
Fig. 4 RMSE results of different methods on ML-L-S dataset with different sparsities
图 5 几种方法在ML-1M数据集上不同稀疏性的MAE结果
Fig. 5 MAE results of different methods on ML-1M dataset with different sparsities
图 6 几种方法在ML-1M数据集上不同稀疏性的RMSE结果
Fig. 6 RMSE results of different methods on ML-1M dataset with different sparsities
图 7 几种方法在Douban数据集上不同稀疏性的MAE结果
Fig. 7 MAE results of different methods on Douban dataset with different sparsities
图 8 几种方法在Douban数据集上不同稀疏性的RMSE结果
Fig. 8 RMSE results of different methods on Douban dataset with different sparsities
图 9 用户和电影的评分预测可解释案例
Fig. 9 Explainable example of rating prediction for users and movies
图 10 ML-1M数据集上的用户和电影节点嵌入表示(转换前)
Fig. 10 The embedding representation of user and movie nodes on ML-1M dataset (before transformation)
图 11 ML-1M数据集上的用户和电影节点嵌入表示(转换后)
Fig. 11 The embedding representation of user and movie nodes on ML-1M dataset (after transformation)
图 12 ML-L-S数据集上的用户和电影节点嵌入表示(转换前)
Fig. 12 The embedding representation of user and movie nodes on ML-L-S dataset (before transformation)
图 13 ML-L-S数据集上的用户和电影节点嵌入表示(转换后)
Fig. 13 The embedding representation of user and movie nodes on ML-L-S dataset (after transformation)
图 14 Douban数据集上的用户和电影节点嵌入表示(转换前)
Fig. 14 The embedding representation of user and movie nodes on Douban dataset (before transformation)
图 15 Douban数据集上的用户和电影节点嵌入表示(转换后)
Fig. 15 The embedding representation of user and movie nodes on Douban dataset (after transformation)
表 1 实验数据集统计信息
Table 1 Statistical information of experimental datasets
数据库 用户数 项目数 交互数 评分等级 稀疏度(%) ML-L-S 610 9724 100836 0.5 ~ 5.0 98.30 ML-1M 6040 3883 1000209 1.0 ~ 5.0 95.74 Douban 3022 6971 195493 1.0 ~ 5.0 99.07 
下载: 导出CSV
表 2 不同方法在三组数据集上的MAE和RMSE结果
Table 2 MAE and RMSE results of different methods on three datasets
方法 ML-L-S ML-1M Douban MAE RMSE MAE RMSE MAE RMSE UserKNN 0.875 2 1.278 4 0.771 0 0.969 3 0.649 4 0.825 6 ItemKNN 0.680 8 0.886 9 0.739 4 0.925 7 0.697 4 0.872 8 BiasedMF 0.676 9 0.882 4 0.684 5 0.872 4 0.577 5 0.728 4 SVD++ 0.672 4 0.877 0 0.672 9 0.863 3 0.569 0 0.720 0 NCF 0.668 5 0.868 0 0.695 6 0.886 6 0.578 1 0.730 4 AFM 0.665 1 0.867 3 0.688 0 0.873 9 0.564 3 0.713 6 Wide&Deep 0.674 2 0.875 4 0.686 3 0.873 5 0.565 4 0.714 1 ACCM 0.662 8 0.865 7 0.673 4 0.856 6 0.578 9 0.730 1 NGCF 0.664 7 0.866 4 0.682 1 0.869 0 0.576 8 0.727 1 LightGCN 0.662 6 0.861 1 0.675 9 0.857 8 0.570 9 0.721 3 AFN 0.657 9 0.852 5 0.678 0 0.860 4 0.565 5 0.715 2 IncorAtt-
MOIntRec0.645 1* 0.837 2* 0.659 4** 0.843 3** 0.558 3 0.708 0 注: 加粗字体表示各列最优结果; 下划线字体表示各列次优结果. “*”表示p值小于0.05; “**”表示p值小于0.01.
下载: 导出CSV
表 3 IncorAttMOIntRec方法在不同嵌入维度下的MAE和RMSE结果
Table 3 MAE and RMSE results for IncorAttMOIntRec method with different embedding dimension sizes
嵌入维度 ML-L-S ML-1M Douban MAE RMSE MAE RMSE MAE RMSE 64 0.650 3 0.847 9 0.662 2 0.849 7 0.558 3 0.708 0 128 0.645 1 0.837 2 0.659 5 0.844 6 0.563 7 0.711 7 256 0.648 8 0.844 0 0.659 4 0.843 3 0.568 5 0.717 2 512 0.651 6 0.849 3 0.662 6 0.845 7 0.576 6 0.723 1 注: 加粗字体表示各列最优结果.
下载: 导出CSV
表 4 IncorAttMOIntRec方法在不同注意力维度下的MAE和RMSE结果
Table 4 MAE and RMSE results for IncorAttMOIntRec method with different attention dimension sizes
注意力维度 ML-L-S ML-1M Douban MAE RMSE MAE RMSE MAE RMSE 32 0.653 2 0.848 7 0.666 2 0.847 5 0.566 2 0.714 7 64 0.645 1 0.837 2 0.657 1 0.846 3 0.558 3 0.708 0 128 0.648 6 0.842 4 0.659 4 0.843 3 0.566 9 0.718 6 256 0.650 2 0.846 1 0.659 2 0.845 9 0.573 1 0.722 6 注: 加粗字体表示各列最优结果.
下载: 导出CSV
表 5 三组数据集上的IncorAttMOIntRec方法消融研究
Table 5 Ablation study of IncorAttMOIntRec method on three datasets
变体模型 ML-L-S ML-1M Douban MAE RMSE MAE RMSE MAE RMSE 去掉Rating-tag 0.653 8 0.854 7 0.667 9 0.847 7 0.568 3 0.713 4 去掉Multi-order
interaction0.688 4 0.890 1 0.680 2 0.866 7 0.574 6 0.722 8 去掉Att-preference 0.656 2 0.854 9 0.668 9 0.848 6 0.569 5 0.717 6 去掉Interaction 0.700 7 0.908 7 0.738 1 0.924 5 0.580 3 0.731 9 去掉MLP-
outputlayer0.667 5 0.873 6 0.710 5 0.896 2 0.568 4 0.713 7 IncorAttMOIntRec 0.645 1 0.837 2 0.659 4 0.843 3 0.558 3 0.708 0 注: 加粗字体表示各列最优结果.
下载: 导出CSV
-
[1] 汤文兵, 任正云, 韩芳. 基于注意力机制的协同卷积动态推荐网络. 自动化学报, 2021, 47(10): 2438−2448Tang Wen-Bing, Ren Zheng-Yun, Han Fang. Attention-based collaborative convolutional dynamic network for recommendation. Acta Automatica Sinica, 2021, 47(10): 2438−2448 [2] Zhang Y F, Chen X. Explainable recommendation: A survey and new perspectives. Foundations and Trends in Information Retrieval, 2020, 14(1): 1−101 [3] Lops P, De Gemmis M, Semeraro G. Content-based recommender systems: State of the art and trends. Recommender Systems Handbook. New York: Springer, 2011. 73−105 [4] 潘涛涛, 文峰, 刘勤让. 基于矩阵填充和物品可预测性的协同过滤算法. 自动化学报, 2017, 43(9): 1597−1606Pan Tao-Tao, Wen Feng, Liu Qin-Rang. Collaborative filtering recommendation algorithm based on rating matrix filling and item predictability. Acta Automatica Sinica, 2017, 43(9): 1597−1606 [5] Shi S Y, Zhang M, Liu Y Q, Ma S P. Attention-based adaptive model to unify warm and cold starts recommendation. In: Proceedings of the 27th ACM International Conference on Information and Knowledge Management. Torino, Italy: ACM, 2018. 127−136 [6] Wu L, Yang Y H, Zhang K, Hong R C, Fu Y J, Wang M. Joint item recommendation and attribute inference: An adaptive graph convolutional network approach. In: Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval. Xi＇an, China: ACM, 2020. 679−688 [7] Deshpande M, Karypis, G. Item-based top-N recommendation algorithms. ACM Transactions on Information Systems, 2004, 22(1): 143−177 doi: 10.1145/963770.963776 [8] Konstan J A, Miller B N, Maltz D, Herlocker J L, Riedl J. GroupLens: Applying collaborative filtering to Usenet news. Communications of the ACM, 1997, 40(3): 77−87 doi: 10.1145/245108.245126 [9] Adomavicius G, Tuzhilin A. Toward the next generation of recommender systems: A survey of the state-of-the-art and possible extensions. IEEE Transactions on Knowledge and Data Engineering, 2005, 17(6): 734−749 doi: 10.1109/TKDE.2005.99 [10] Herlocker J L, Konstan J A, Terveen L G, Riedl J T. Evaluating collaborative filtering recommender systems. ACM Transactions on Information Systems, 2004, 22(1): 5−53 doi: 10.1145/963770.963772 [11] Koren Y. Factorization meets the neighborhood: A multifaceted collaborative filtering model. In: Proceedings of the 14th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. Las Vegas, USA: ACM, 2008. 426−434 [12] Koren Y, Bell R, Volinsky C. Matrix factorization techniques for recommender systems. Computer, 2009, 42(8): 30−37 [13] Salakhutdinov R, Mnih A. Probabilistic matrix factorization. In: Proceedings of the 20th International Conference on Neural Information Processing Systems. Vancouver, Canada: ACM, 2007. 1257−1264 [14] Salakhutdinov R, Mnih A. Bayesian probabilistic matrix factorization using Markov chain Monte Carlo. In: Proceedings of the 25th International Conference on Machine Learning. Helsinki, Finland: ACM, 2008. 880−887 [15] Lee J, Kim S, Lebanon G, Singer Y, Bengio S. LLORMA: Local low-rank matrix approximation. Journal of Machine Learning Research, 2016, 17: 1−24 [16] Zhang Y F, Lai G K, Zhang M, Zhang Y, Liu Y Q, Ma S P. Explicit factor models for explainable recommendation based on phrase-level sentiment analysis. In: Proceedings of the 37th International ACM SIGIR Conference on Research & Development in Information Retrieval. Gold Coast, Australia: ACM, 2014. 83−92 [17] Ren Z C, Liang S S, Li P J, Wang S Q, De Rijke M. Social collaborative viewpoint regression with explainable recommendations. In: Proceedings of the 10th ACM International Conference on Web Search and Data Mining. Cambridge, UK: ACM, 2017. 485−494 [18] Chen X, Qin Z, Zhang Y F, Xu T. Learning to rank features for recommendation over multiple categories. In: Proceedings of the 39th International ACM SIGIR Conference on Research and Development in Information Retrieval. Pisa, Italy: ACM, 2016. 305−314 [19] Tan Y Z, Zhang M, Liu Y Q, Ma S P. Rating-boosted latent topics: Understanding users and items with ratings and reviews. In: Proceedings of the 25th International Joint Conference on Artificial Intelligence. New York, USA: ACM, 2016. 2640−2646 [20] Zhao K Q, Cong G, Yuan Q, Zhu K Q. SAR: A sentiment-aspect-region model for user preference analysis in geo-tagged reviews. In: Proceedings of the 31st International Conference on Data Engineering. Seoul, South Korea: IEEE, 2015. 675−686 [21] Zhang S, Yao L N, Sun A X, Tay Y. Deep learning based recommender system: A survey and new perspectives. ACM Computing Surveys, 2020, 52(1): Article No. 5 [22] Xiao J, Ye H, He X N, Zhang H W, Wu F, Chua T S. Attentional factorization machines: Learning the weight of feature interactions via attention networks. In: Proceedings of the 26th International Joint Conference on Artificial Intelligence. Melbourne, Australia: ACM, 2017. 3119−3125 [23] Cheng H T, Koc L, Harmsen J, Shaked T, Chandra T, Aradhye H, et al. Wide & deep learning for recommender systems. In: Proceedings of the 1st Workshop on Deep Learning for Recommender Systems. Boston, USA: ACM, 2016. 7−10 [24] Wang X, He X N, Wang M, Feng F L, Chua T S. Neural graph collaborative filtering. In: Proceedings of the 42nd International ACM SIGIR Conference on Research and Development in Information Retrieval. Paris, France: ACM, 2019. 165−174 [25] Yang Z X, Dong S B. HAGERec: Hierarchical attention graph convolutional network incorporating knowledge graph for explainable recommendation. Knowledge-Based Systems, 2020, 204: Article No. 106194 doi: 10.1016/j.knosys.2020.106194 [26] He X N, Liao L Z, Zhang H W, Nie L Q, Hu X, Chua T S. Neural collaborative filtering. In: Proceedings of the 26th International Conference on World Wide Web. Perth, Australia: 2017. 173−182 [27] Lian J X, Zhou X H, Zhang F Z, Chen Z X, Xie X, Sun G Z. xDeepFM: Combining explicit and implicit feature interactions for recommender systems. In: Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. London, UK: ACM, 2018. 1754−1763 [28] Chang S, Harper F M, Terveen L G. Crowd-based personalized natural language explanations for recommendations. In: Proceedings of the 10th ACM Conference on Recommender Systems. Boston, USA: ACM, 2016. 175−182 [29] Seo S, Huang J, Yang H, Liu Y. Interpretable convolutional neural networks with dual local and global attention for review rating prediction. In: Proceedings of the 11th ACM Conference on Recommender Systems. Como, Italy: ACM, 2017. 297−305 [30] Chen C, Zhang M, Liu Y Q, Ma S P. Neural attentional rating regression with review-level explanations. In: Proceedings of the World Wide Web Conference. Lyon, France: ACM, 2018. 1583−1592 [31] 冯永, 陈以刚, 强保华. 融合社交因素和评论文本卷积网络模型的汽车推荐研究. 自动化学报, 2019, 45(3): 518−529Feng Yong, Chen Yi-Gang, Qiang Bao-Hua. Social and comment text CNN model based automobile recommendation. Acta Automatica Sinica, 2019, 45(3): 518−529 [32] Li P J, Wang Z H, Ren Z C, Bing L D, Lam W. Neural rating regression with abstractive tips generation for recommendation. In: Proceedings of the 40th International ACM SIGIR Conference on Research and Development in Information Retrieval. Shinjuku, Japan: ACM, 2017. 345−354 [33] Chen X, Zhang Y F, Xu H T, Qin Z, Zha H Y. Adversarial distillation for efficient recommendation with external knowledge. ACM Transactions on Information Systems, 2019, 37(1): Article No. 12 [34] 饶子昀, 张毅, 刘俊涛, 曹万华. 应用知识图谱的推荐方法与系统. 自动化学报, 2021, 47(9): 2061−2077Rao Zi-Yun, Zhang Yi, Liu Jun-Tao, Cao Wan-Hua. Recommendation methods and systems using knowledge graph. Acta Automatica Sinica, 2021, 47(9): 2061−2077 [35] Wang H W, Zhao M, Xie X, Li W J, Guo M Y. Knowledge graph convolutional networks for recommender systems. In: Proceedings of the World Wide Web Conference. San Francisco, USA: ACM, 2019. 3307−3313 [36] Wang X, He X N, Cao Y X, Liu M, Chua T S. KGAT: Knowledge graph attention network for recommendation. In: Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. Anchorage, USA: ACM, 2019. 950−958 [37] Ying R, He R N, Chen K F, Eksombatchai P, Hamilton W L, Leskovec J. Graph convolutional neural networks for web-scale recommender systems. In: Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. London, UK: ACM, 2018. 974−983 [38] He X N, Deng K, Wang X, Li Y, Zhang Y D, Wang M. LightGCN: Simplifying and powering graph convolution network for recommendation. In: Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval. Xi＇an, China: ACM, 2020. 639−648 [39] Jin B W, Gao C, He X N, Jin D P, Li Y. Multi-behavior recommendation with graph convolutional networks. In: Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval. Xi＇an, China: ACM, 2020. 659−668 [40] Yao Q M, Chen X N, Kwok J T, Li Y, Hsieh C J. Efficient neural interaction function search for collaborative filtering. In: Proceedings of the World Wide Web Conference. Taipei, China: ACM, 2020. 1660−1670 [41] Zheng J, Liu J, Shi C, Zhuang F Z, Li J Z, Wu B. Dual similarity regularization for recommendation. In: Proceedings of the 20th Pacific-Asia Conference on Knowledge Discovery and Data Mining. Auckland, New Zealand: Springer, 2016. 542−554 [42] Cheng W Y, Shen Y Y, Huang L P. Adaptive factorization network: Learning adaptive-order feature interactions. In: Proceedings of the 34th AAAI Conference on Artificial Intelligence. Palo Alto, USA: AAAI, 2020. 3609−3616 -
计量
- 文章访问数: 935
- HTML全文浏览量: 410
- PDF下载量: 85
- 被引次数: 0
