·360· 智能系统学报 第16卷 0.5 Tmall Rey,USA,2018:108-116 0.4 Le Gowalla [4]KOREN Y.Collaborative filtering with temporal dynam- ics[C]//Proceedings of the 15th ACM SIGKDD Interna- 邓0.2 tional Conference on Knowledge Discovery and Data Min- 0.1 ing.Paris,France,2009:447-456. 0 10 20 30 40 50 维度 [5]GARCIN F,DIMITRAKAKIS C,FALTINGS B.Person- alized news recommendation with context trees[C]//Pro- 图7维度参数对模型的影响 ceedings of the 7th ACM Conference on Recommender Fig.7 Impact of dimension parameter Systems.Hong Kong,China,2013:105-112. 4结束语 [6]RENDLE S.FREUDENTHALER C.SCHMIDT-THIEME L.Factorizing personalized Markov chains for next-basket LSSSAN相比AttRec方法，利用长期反馈数 recommendation[Cl//Proceedings of the 19th International 据对长期/一般偏好进行准确表达，并从结构上赋 Conference on World Wide Web.Raleigh,USA,2010: 予了相对重要的短期反馈数据更高的权重；相比 811-820. SHAN方法，LSSSAN考虑了序列性偏好和长期 [7]HIDASI B.TIKK D.General factorization framework for 数据中的相互依赖关系。 context-aware recommendations[J].Data mining and 本文在Tmall和Gowalla上对LSSSAN进行 knowledge discovery,2016,30(2):342-371 训练和测试，其效果整体优于其他先进的方案。 [8]HE RUINING,MCAULEY J.Fusing similarity models with Markov chains for sparse sequential recommenda- 且由于Gowalla数据集的反馈数据相互依赖性和 tion[C]//Proceedings of the 2016 IEEE 16th International 顺序相关性严格于Tmall数据集，模型在Gowalla Conference on Data Mining.Barcelona,Spain,2016: 上表现优于在Tmal上的表现，表明模型擅长于 191-200 处理相对严格的相互依赖关系和顺序相关性的数 [9]HIDASI B.KARATZOGLOU A,BALTRUNAS L,et al 据，也表明模型可能会因为数据集数据的弱相互 Session-based recommendations with recurrent neural net- 依赖性和弱顺序相关性而出现不稳定的情况。同 works[Cl//Proceedings of the 4th International Conference 时本文通过消融实验验证了模型结构的合理性， on Learning Representations.San Juan,Puerto Rico,2016: 并给出了当数据出现明显的弱相互依赖性和弱顺 1-10. 序相关性时的候选方案。 [10]WU Chaoyuan,AHMED A,BEUTEL A,et al.Recurrent LSSSAN在实际应用上可为众多互联网应用 recommender networks[C]//Proceedings of the 10th ACM 提供推荐模型，尤其在数据具有强相互依赖性和 International Conference on Web Search and Data Min- 顺序相关性的互联网应用上将会保证可靠的性 ing.Cambridge,UK,2017:495-503. [11]TANG Jiaxi,BELLETTI F,JAIN S,et al.Towards neur- 能；未来的工作会考虑在LSSSAN的基础上尝试 采用内存机制以提高性能，并在更多的数据集上 al mixture recommender for long range dependent user sequences[C]//Proceedings of World Wide Web Confer- 测试模型性能。 ence.San Francisco,USA,2019:1782-1793 参考文献： [12]ZHOU Guorui,ZHU Xiaoqiang,SONG Chenru,et al. Deep interest network for click-through rate prediction[Cl/ [1]孙宏超.阿里巴巴发布2020财年第三季度财报：收入增 Proceedings of the 24th ACM SIGKDD International 长38%，年活跃用户达7亿[EB/OL].[2020-02-13] Conference on Knowledge Discovery Data Mining. kuaibao.qq.com/s/20200213A0PEAW00 London,UK,2018:1059-1068. [2]WANG Shoujin,HU Liang,WANG Yan,et al.Sequential [13]ZHANG Shuai,TAY Y,YAO Lina,et al.2019.Next recommender systems:challenges,progress and item recommendation with self-attentive metric prospects[C]//Proceedings of the 28th International Joint learning[C]//Proceedings of the 33rd AAAI Conference Conference on Artificial Intelligence.Macao,China,2019: on Artificial Intelligence.Hawaii,USA,2019:9. 6332-6338 [14]YING Haochao,ZHUANG Fuzhen,ZHANG Fuzheng, [3]XU Chen,XU Hongteng,ZHANG Yongfeng,et al.Se- et al.Sequential recommender system based on hierarch- quential recommendation with user memory ical attention networks[Cl//Proceedings of the 27th Inter- networks[C]//Proceedings of the 11th ACM International national Joint Conference on Artificial Intelligence Conference on Web Search and Data Mining.Marina Del Stockholm.Sweden.2018:3926-3932Tmall Gowalla 0.5 0.4 0.3 0.2 0.1 召回率 0 20 30 40 50 10 维度 图 7 维度参数对模型的影响 Fig. 7 Impact of dimension parameter 4 结束语 LSSSAN 相比 AttRec 方法，利用长期反馈数 据对长期/一般偏好进行准确表达，并从结构上赋 予了相对重要的短期反馈数据更高的权重；相比 SHAN 方法，LSSSAN 考虑了序列性偏好和长期 数据中的相互依赖关系。 本文在 Tmall 和 Gowalla 上对 LSSSAN 进行 训练和测试，其效果整体优于其他先进的方案。 且由于 Gowalla 数据集的反馈数据相互依赖性和 顺序相关性严格于 Tmall 数据集，模型在 Gowalla 上表现优于在 Tmall 上的表现，表明模型擅长于 处理相对严格的相互依赖关系和顺序相关性的数 据，也表明模型可能会因为数据集数据的弱相互 依赖性和弱顺序相关性而出现不稳定的情况。同 时本文通过消融实验验证了模型结构的合理性， 并给出了当数据出现明显的弱相互依赖性和弱顺 序相关性时的候选方案。 LSSSAN 在实际应用上可为众多互联网应用 提供推荐模型，尤其在数据具有强相互依赖性和 顺序相关性的互联网应用上将会保证可靠的性 能；未来的工作会考虑在 LSSSAN 的基础上尝试 采用内存机制以提高性能，并在更多的数据集上 测试模型性能。 参考文献： 孙宏超. 阿里巴巴发布 2020 财年第三季度财报: 收入增 长 38%, 年活跃用户达 7 亿 [EB/OL]. [2020-02-13]. kuaibao.qq.com/s/20200213A0PEAW00 [1] WANG Shoujin, HU Liang, WANG Yan, et al. Sequential recommender systems: challenges, progress and prospects[C]//Proceedings of the 28th International Joint Conference on Artificial Intelligence. Macao, China, 2019: 6332−6338. [2] XU Chen, XU Hongteng, ZHANG Yongfeng, et al. Se￾quential recommendation with user memory networks[C]//Proceedings of the 11th ACM International Conference on Web Search and Data Mining. Marina Del [3] Rey, USA, 2018: 108−116. KOREN Y. Collaborative filtering with temporal dynam￾ics[C]//Proceedings of the 15th ACM SIGKDD Interna￾tional Conference on Knowledge Discovery and Data Min￾ing. Paris, France, 2009: 447−456. [4] GARCIN F, DIMITRAKAKIS C, FALTINGS B. Person￾alized news recommendation with context trees[C]//Pro￾ceedings of the 7th ACM Conference on Recommender Systems. Hong Kong, China, 2013: 105−112. [5] RENDLE S, FREUDENTHALER C, SCHMIDT-THIEME L. Factorizing personalized Markov chains for next-basket recommendation[C]//Proceedings of the 19th International Conference on World Wide Web. Raleigh, USA, 2010: 811−820. [6] HIDASI B, TIKK D. General factorization framework for context-aware recommendations[J]. Data mining and knowledge discovery, 2016, 30(2): 342–371. [7] HE RUINING, MCAULEY J. Fusing similarity models with Markov chains for sparse sequential recommenda￾tion[C]//Proceedings of the 2016 IEEE 16th International Conference on Data Mining. Barcelona, Spain, 2016: 191−200. [8] HIDASI B, KARATZOGLOU A, BALTRUNAS L, et al. Session-based recommendations with recurrent neural net￾works[C]//Proceedings of the 4th International Conference on Learning Representations. San Juan, Puerto Rico, 2016: 1−10. [9] WU Chaoyuan, AHMED A, BEUTEL A, et al. Recurrent recommender networks[C]//Proceedings of the 10th ACM International Conference on Web Search and Data Min￾ing. Cambridge, UK, 2017: 495−503. [10] TANG Jiaxi, BELLETTI F, JAIN S, et al. Towards neur￾al mixture recommender for long range dependent user sequences[C]//Proceedings of World Wide Web Confer￾ence. San Francisco, USA, 2019: 1782−1793. [11] ZHOU Guorui, ZHU Xiaoqiang, SONG Chenru, et al. Deep interest network for click-through rate prediction[C]// Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. London, UK, 2018: 1059−1068. [12] ZHANG Shuai, TAY Y, YAO Lina, et al. 2019. Next item recommendation with self-attentive metric learning[C]//Proceedings of the 33rd AAAI Conference on Artificial Intelligence. Hawaii, USA, 2019: 9. [13] YING Haochao, ZHUANG Fuzhen, ZHANG Fuzheng, et al. Sequential recommender system based on hierarch￾ical attention networks[C]//Proceedings of the 27th Inter￾national Joint Conference on Artificial Intelligence. Stockholm, Sweden, 2018: 3926−3932. [14] ·360· 智 能 系 统 学 报 第 16 卷