·414· 智能系统学报 第16卷 表5兀，参数对于Foursquare和CA数据集影响 Table 5 Influence of parameters on Foursquare and CA datasets % 数据集 评价你 8h 16h 24h 32h 40h ACC@5 20.68 20.42 21.07 20.42 20.33 Foursquare MAP 14.04 14.10 14.17 13.83 14.04 ACC@5 16.97 17.06 17.00 16.87 16.64 CA MAP 12.89 13.11 13.11 12.58 12.72 5结束语 tions on systems,man,and cybernetics:systems,2015, 45(1):129-142. 本文研究了下一个兴趣点推荐的问题，提出 [6]LI Huayu,GE Yong,HONG Richang,et al.Point-of-in- 了一种新的基于会话的时空循环神经网络模型， terest recommendations:learning potential check-ins from 该模型能够同时考虑序列信息、两类时间信息、 friends[C]//Proceedings of the 22nd ACM SIGKDD Inter- 空间信息以及用户偏好进行个性化的下一个兴趣 national Conference on Knowledge Discovery and Data 点推荐。为了解决用户不同的时空偏好问题，本 Mining.San Francisco,USA,2016:975-984. 文对RNN神经网络进行改进，提出利用时空转移 [7]ZHANG Zhiyuan,LIU Yun,ZHANG Zhenjiang,et al. 矩阵对用户不同的时空偏好进行建模，使RNN模 Fused matrix factorization with multi-tag,social and geo- 型能够捕获用户签到行为的时空信息。通过在真 graphical influences for POI recommendation[J].World 实的数据集上对模型进行性能测试，结果表明， wide web,2019,22(3):1135-1150. 提出的模型显著优于当前主流相关模型。下一步 [8]孟祥福，张霄雁，唐延欢，等.基于地理-社会关系的多样 性与个性化兴趣点推荐).计算机学报，2019,42(11)： 工作将探索更丰富的数据集，进一步结合用户基 2574-2590. 本属性、地点基本属性来解决兴趣点推荐中出现 MENG Xiangfu,ZHANG Xiaoyan,TANG Yanhuan,et.al 的冷启动问题，并且整合其他辅助的信息（如图 A diversified and personalized recommendation approach 像、用户评分和文本评论等)作更精确的下一个 based on geo-social relationships[J].Chinese journal of 兴趣点推荐。 computers,.2019,42(11):2574-2590. 参考文献： [9]XINXin,CHEN Bo,HE Xiangnan,et al.CFM:convolu- tional factorization machines for context-aware recom- [1]RENDLE S,FREUDENTHALER C.SCHMIDT-THIEME mendation[C]//Proceedings of the Twenty-Eighth Interna- L.Factorizing personalized Markov chains for next-basket tional Joint Conference on Artificial Intelligence.Macao, recommendation[C//Proceedings of the 19th International China,2019:3926-3932. Conference on World Wide Web.North Carolina.USA. [10]HE Jing,LI Xin,LIAO Lejian.Category-aware next 2010:811-820 point-of-interest recommendation via listwise bayesian [2]LIU Qiang,WU Shu,WANG Diyi,et al.Context-aware personalized ranking[C]//Proceedings of the Twenty- sequential recommendation[Cl//Proceedings of the IEEE Sixth International Joint Conference on Artificial Intelli- 16th International Conference on Data Mining.Barcelona, gence.Melbourne,Australia,2017:1837-1843. Spain,2016:1053-1058 [11]ZHANG Zhiqian,LI Chenliang,WU Zhiyong,et al. [3]FENG Jie,LI Yong,ZHANG Chao,et al.DeepMove:pre- NEXT:a neural network framework for next POI recom- dicting human mobility with attentional recurrent net- mendation[J].Frontiers of computer science,2020,14(2) works[C]//Proceedings of the 2018 World Wide Web Con- 314333 ference.Lyon,France,2018:1459-1468. [12]ZHANG Lu,SUN Zhu,ZHANG Jie,et al.Modeling hier- [4]LIU Qiang,WU Shu,WANG Liang,et al.Predicting the archical category transition for next POI recommendation next location:a recurrent model with spatial and temporal with uncertain check-ins[J].Information sciences,2020, contexts[C]//Proceedings of the Thirtieth AAAl Confer- 515:169-190. ence on Artificial Intelligence.Phoenix,USA,2016: [13]CHO K.VAN MERRIENBOER B,GULCEHRE C,et al. 194-200. Learning phrase representations using RNN encoder-de- [5]YANG Dingqi,ZHANG Daqing,ZHENG V W,et al. coder for statistical machine translation[C]//Proceedings Modeling user activity preference by leveraging user spa- of the 2014 Conference on Empirical Methods in Natural tial temporal characteristics in LBSNs[J].IEEE transac- Language Processing.Doha,Qatar,2014:1724-1734.表 5 πt 参数对于 Foursquare 和 CA 数据集影响 Table 5 Influence of πt parameters on Foursquare and CA datasets % 数据集 评价 πt / 8 h 16 h 24 h 32 h 40 h Foursquare ACC@5 20.68 20.42 21.07 20.42 20.33 MAP 14.04 14.10 14.17 13.83 14.04 CA ACC@5 16.97 17.06 17.00 16.87 16.64 MAP 12.89 13.11 13.11 12.58 12.72 5 结束语 本文研究了下一个兴趣点推荐的问题，提出 了一种新的基于会话的时空循环神经网络模型， 该模型能够同时考虑序列信息、两类时间信息、 空间信息以及用户偏好进行个性化的下一个兴趣 点推荐。为了解决用户不同的时空偏好问题，本 文对 RNN 神经网络进行改进，提出利用时空转移 矩阵对用户不同的时空偏好进行建模，使 RNN 模 型能够捕获用户签到行为的时空信息。通过在真 实的数据集上对模型进行性能测试，结果表明， 提出的模型显著优于当前主流相关模型。下一步 工作将探索更丰富的数据集，进一步结合用户基 本属性、地点基本属性来解决兴趣点推荐中出现 的冷启动问题，并且整合其他辅助的信息 (如图 像、用户评分和文本评论等) 作更精确的下一个 兴趣点推荐。 参考文献： 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. North Carolina, USA, 2010: 811−820. [1] LIU Qiang, WU Shu, WANG Diyi, et al. Context-aware sequential recommendation[C]//Proceedings of the IEEE 16th International Conference on Data Mining. Barcelona, Spain, 2016: 1053−1058. [2] FENG Jie, LI Yong, ZHANG Chao, et al. DeepMove: pre￾dicting human mobility with attentional recurrent net￾works[C]//Proceedings of the 2018 World Wide Web Con￾ference. Lyon, France, 2018: 1459−1468. [3] LIU Qiang, WU Shu, WANG Liang, et al. Predicting the next location: a recurrent model with spatial and temporal contexts[C]//Proceedings of the Thirtieth AAAI Confer￾ence on Artificial Intelligence. Phoenix, USA, 2016: 194−200. [4] YANG Dingqi, ZHANG Daqing, ZHENG V W, et al. Modeling user activity preference by leveraging user spa￾tial temporal characteristics in LBSNs[J]. IEEE transac- [5] tions on systems, man, and cybernetics: systems, 2015, 45(1): 129–142. LI Huayu, GE Yong, HONG Richang, et al. Point-of-in￾terest recommendations: learning potential check-ins from friends[C]//Proceedings of the 22nd ACM SIGKDD Inter￾national Conference on Knowledge Discovery and Data Mining. San Francisco, USA, 2016: 975−984. [6] ZHANG Zhiyuan, LIU Yun, ZHANG Zhenjiang, et al. Fused matrix factorization with multi-tag, social and geo￾graphical influences for POI recommendation[J]. World wide web, 2019, 22(3): 1135–1150. [7] 孟祥福, 张霄雁, 唐延欢, 等. 基于地理-社会关系的多样 性与个性化兴趣点推荐 [J]. 计算机学报, 2019, 42(11): 2574–2590. MENG Xiangfu, ZHANG Xiaoyan, TANG Yanhuan, et. al A diversified and personalized recommendation approach based on geo-social relationships[J]. Chinese journal of computers, 2019, 42(11): 2574–2590. [8] XIN Xin, CHEN Bo, HE Xiangnan, et al. CFM: convolu￾tional factorization machines for context-aware recom￾mendation[C]//Proceedings of the Twenty-Eighth Interna￾tional Joint Conference on Artificial Intelligence. Macao, China, 2019: 3926−3932. [9] HE Jing, LI Xin, LIAO Lejian. Category-aware next point-of-interest recommendation via listwise bayesian personalized ranking[C]//Proceedings of the Twenty￾Sixth International Joint Conference on Artificial Intelli￾gence. Melbourne, Australia, 2017: 1837−1843. [10] ZHANG Zhiqian, LI Chenliang, WU Zhiyong, et al. NEXT: a neural network framework for next POI recom￾mendation[J]. Frontiers of computer science, 2020, 14(2): 314–333. [11] ZHANG Lu, SUN Zhu, ZHANG Jie, et al. Modeling hier￾archical category transition for next POI recommendation with uncertain check-ins[J]. Information sciences, 2020, 515: 169–190. [12] CHO K, VAN MERRIËNBOER B, GULCEHRE C, et al. Learning phrase representations using RNN encoder-de￾coder for statistical machine translation[C]//Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing. Doha, Qatar, 2014: 1724−1734. [13] ·414· 智 能 系 统 学 报 第 16 卷