融合时间上下文与特征级信息的推荐算法

doi:10.3778/j.issn.1673-9418.2105008

摘要/Abstract

摘要： 针对基于自注意力机制的序列推荐模型忽略了各类辅助信息，导致模型不能利用它们捕捉多层次序列关系模式等问题，提出了一种融合时间上下文和特征级信息的推荐算法（ITFR）。首先将物品表示与其每一个属性表示连接起来输入到一个注意力网络，经过注意力加权后得到一种基于属性的物品表示。然后ITFR应用感知时间间隔的自注意力区块和基于物品-属性的自注意力区块分别捕捉物品与交互序列时间间隔之间的关系模式和物品与属性之间的隐式关系。最后将两个自注意力区块的输出表示连接起来，并将其作为联合输出表示输入到全连接层用于下一个物品的推荐。在两个公开数据集上进行实验，采用命中率（HR）和归一化折损累计增益（NDCG）两种性能指标进行评估。在Beauty数据集中，相较于最优的基线方法，HR@10和NDCG@10分别提升了4.6%和5.1%；在MovieLens-1M数据集中，HR@10和NDCG@10分别提升了1.7%和1.5%。实验结果表明，融入辅助信息增强序列表示的方法可以提升推荐性能。

关键词: 推荐算法, 序列推荐, 自注意力机制, 时间信息

Abstract: Aiming at the problem that the sequence recommendation models based on the self-attention mechanism ignore all kinds of auxiliary information, which makes the model unable to use them to capture multi-level sequence relationship patterns, a recommendation algorithm integrating time context and feature-level information is proposed (ITFR). Firstly, the item representation is connected with each of its attribute representations and input into an attention network. After attention is weighted, an attribute-based item representation is obtained. Then, ITFR applies the self-attention block of the perception time interval and the self-attention block based on the item-attribute to capture the relationship pattern between the item and the interaction sequence time interval and the implicit relationship between the item and the attribute, respectively. Finally, the output representations of the two self-attention blocks are connected, and the joint output representation is input to the fully connected layer for the recommendation of the next item. Experiments are conducted on two public datasets, and two performance indicators of hit rate (HR) and normalized discounted cumulative gain (NDCG) are used for evaluation. In the Beauty dataset, compared with the optimal baseline method, HR@10 and NDCG@10 are increased by 4.6% and 5.1%, respectively. In the Movie-1M dataset, HR@10 and NDCG@10 are increased by 1.7% and 1.5%, respectively. Experimental results show that the method of incorporating auxiliary information to enhance sequence representation can improve recommendation performance.

Key words: recommendation algorithm, sequence recommendation, self-attention mechanism, time information

沈义峰, 金辰曦, 王瑶, 张家想, 卢先领. 融合时间上下文与特征级信息的推荐算法[J]. 计算机科学与探索, 2023, 17(2): 489-498.

SHEN Yifeng, JIN Chenxi, WANG Yao, ZHANG Jiaxiang, LU Xianling. Integrating Time Context and Feature-Level Information for Recommendation Algorithm[J]. Journal of Frontiers of Computer Science and Technology, 2023, 17(2): 489-498.

参考文献

[1] MOONEY R J, ROY L. Content-based book recommending using learning for text categorization[C]//Proceedings of the 5th ACM Conference on Digital Libraries, San Antonio, Jun 2-7, 2000. New York: ACM, 2000: 195-204.
[2] GOLDBERG D, NICHOLS D, OKI B M, et al. Using colla-borative filtering to weave an information tapestry[J]. Com-munications of the ACM, 1992, 35(12): 61-70.
[3] SHANG S, DING R G, ZHENG K, et al. Personalized trajectory matching in spatial networks[J]. The VLDB Journal—The International Journal on Very Large Data Bases, 2014, 23(3): 449-468.
[4] YAP G E, LI X L, PHILIP S Y. Effective next-items recommendation via personalized sequential pattern mining[C]//LNCS 7239: Proceedings of the 17th International Conference on Database Systems for Advanced Applications, Apr 15-19, 2012. Berlin, Heidelberg: Springer, 2012: 48-64.
[5] HE R N, MCAULEY J. Fusing similarity models with Markov chains for sparse sequential recommendation[C]//Proceedings of the 16th International Conference on Data Mining, Barcelona, Dec 12-15, 2016. Washington: IEEE Computer Society, 2016: 191-200.
[6] WU C Y, AHMED A, BEYTEL A. Recurrent recommender networks[C]//Proceedings of the 10th ACM International Conference on Web Search and Data Mining, Cambridge,Feb 6-10, 2017. New York: ACM, 2017: 495-503.
[7] HIDASI B, QUADRANA M, KARATZOGLOU A, et al. Parallel recurrent neural network architectures for feature-rich session-based recommendations[C]//Proceedings of the 10th ACM Conference on Recommender Systems, Boston, Sep 15-19, 2016. New York: ACM, 2016: 241-248.
[8] ASWANI A V, SHAZEER N, PARMAR N, et al. Attention is all you need[C]//Proceedings of the Annual Conference on Neural Information Processing Systems 2017, Long Beach, Dec 4-9, 2017. Red Hook: Curran Associates, 2017: 5998-6008.
[9] RENDLE S, FREUDEN T C, SCHMIDT T L. Factorizing personalized Markov chains for next-basket recommendation[C]//Proceedings of the 19th International Conference on World Wide Web, Raleigh, Apr 26-30, 2010. New York:ACM, 2010: 811-820.
[10] HIDASI B, KARAT Z A, BALTRUNAS L, et al. Session-based recommendations with recurrent neural networks[J]. arXiv:1511.06939, 2015.
[11] CHEN X, XU H T, ZHANG Y F, et al. Sequential recom-mendation with user memory networks[C]//Proceedings of the 11th ACM International Conference on Web Search and Data Mining, Marina Del Rey, Feb 5-9, 2018. New York:ACM, 2018: 108-116.
[12] YING H C, ZHUANG F Z, ZHANG F Z, et al. Sequential recommender system based on hierarchical attention networks[C]//Proceedings of the 27th International Joint Conference on Artificial Intelligence, Stockholm, Jul 13-19, 2018: 3926-3932.
[13] LIU Q, ZENG Y F, MOKHOSI R, et al. STAMP: short-term attention/memory priority model for session-based recommendation[C]//Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, London, Aug 19-23, 2018. New York: ACM, 2018: 1831-1839.
[14] MA C, KANG P, LIU X. Hierarchical gating networks for sequential recommendation[C]//Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, Anchorage, Aug 4-8, 2019. New York: ACM, 2019: 825-833.
[15] KANG W C, MCAULEY J. Self-attentive sequential recom-mendation[C]//Proceedings of the 2018 IEEE International Conference on Data Mining, Singapore, Nov 17-20, 2018.Washington: IEEE Computer Society, 2018: 197-206.
[16] SUN F, LIU J, WU J, et al. BERT4Rec: sequential recom-mendation with bidirectional encoder representations from Transformer[C]//Proceedings of the 28th ACM International Conference on Information and Knowledge Management,Beijing, Nov 3-7, 2019. New York: ACM, 2019: 1441-1450.
[17] LI J C, WANG Y J, MCAULEY J J. Time interval aware self-attention for sequential recommendation[C]//Proceedings of the 13th ACM International Web Search and Data Mining, Houston, Feb 3-7, 2020. New York: ACM, 2020: 322-330.
[18] ZHANG T T, ZHAO P P, LIU Y, et al. Feature-level deeper self-attention network for sequential recommendation[C]//Proceedings of the 28th International Joint Conference on Artificial Intelligence, Macao, China, Aug 10-16, 2019: 4320- 4326.
[19] WU J B, CAI R Q, WANG H N. Déjà vu: a contextualized temporal attention mechanism for sequential recommendation[C]//Proceedings of the 2020 International World Wide Web Conference, Taipei, China, Apr 20-24, 2020. New York:ACM, 2020: 2199-2209.
[20] HONG L J, DAVISON B D. Empirical study of topic modeling in Twitter[C]//Proceedings of the 3rd Workshop on Social Network Mining and Analysis, Paris, Jun 28, 2009. New York: ACM, 2010: 80-88.
[21] WANG Z, LONG M, ZHANG Y. A hybrid document feature extraction method using latent Dirichlet allocation and Word2Vec[C]//Proceedings of the IEEE 1st International Conference on Data Science in Cyberspace, Changsha, Jun 13-16, 2016. Washington: IEEE Computer Society, 2016:98-103.
[22] MIKOLOV T, CHEN K, CORRADO G, et al. Efficient estimation of word representations in vector space[J]. arXiv: 1301.3781, 2013.