Please wait a minute...
浙江大学学报(工学版)
浙江大学学报(工学版)  2020, Vol. 54 Issue (2): 311-319    DOI: 10.3785/j.issn.1008-973X.2020.02.012
计算机技术、信息工程     
融合用户信任和影响力的top-N推荐算法
张雪峰(),陈秀莉,僧德文*()
杭州电子科技大学 复杂系统建模与仿真教育部重点实验室，浙江 杭州 310018
Top-N recommendation algorithm combining user trust and influence
Xue-feng ZHANG(),Xiu-li CHEN,De-wen SENG*()
Key Laboratory of Complex Systems Modeling and Simulation, Hangzhou Dianzi University, Hangzhou 310018, China
 全文: PDF(946 KB)   HTML
摘要：

针对现有基于信任的推荐方法通常直接利用社交网络的二值信任关系来提高推荐质量，较少考虑用户间信任强度的差异和潜在影响的问题,提出结合用户信任和影响力的混合推荐算法进行top-N项目推荐. 采用自动编码器对用户行为进行无监督的初始特征优化，将高维、稀疏的用户行为压缩成低维、稠密的用户及项目特征向量；提出融合用户交互信息、偏好度和信任的新型信任度量模型，发掘社交网络中用户间的隐含信任关系，重构社会信任网络；将社会信任网络的拓扑结构和用户的交互信息融入结构洞算法，通过改进的结构洞算法来识别网络中的影响力用户，提高top-N项目推荐性能. 实验在FilmTrust、Epinions、Ciao这3个标准数据集上进行对比验证，实验结果证明了所提算法的有效性.
关键词： 社会化推荐;  用户信任;  影响力;  矩阵分解;  自动编码器    
Abstract:

A hybrid recommendation algorithm with the incorporation of user trust and social influence was proposed for top-N item recommendation, in view of the existing trust-aware recommendation systems, which directly use the binary trust relationship of social networks to improve the quality of recommendation, and less consider the difference of trust intensity and potential impact between users. The auto-encoder is used to perform unsupervised initial feature optimization on user behavior, and the high-dimensional and sparse user behaviors are compressed into low dimensional and dense users and item feature vectors. A novel trust value measurement model that combines user interaction information, preferences, and trust is brought up to explore the implicit trust relationship between users in social networks and reconstruct the social trust network. The improved structure hole algorithm is used to identify the influential users in the network and improve the top-N item recommendation performance, which integrates the topological structure of the social trust network and the user's interactive information. Comparison verification was conducted on three standard datasets, FilmTrust, Epinions and Ciao, and experimental results demonstrated the effectiveness of the proposed algorithm.
Key words: social recommendation    user trust    influence    matrix factorization    auto-encoder
收稿日期: 2019-01-05 出版日期: 2020-03-10
CLC:  TP 391  
基金资助: 浙江省基础公益研究计划资助项目（LGF19F020015）；中国高等教育学会高等教育科学研究“十三五”规划课题资助项目（2018GCLYB12）
通讯作者: 僧德文     E-mail: zhangxf@hdu.edu.cn;sengdw@hdu.edu.cn
作者简介: 张雪峰（1980—），男，讲师，从事推荐系统和智能计算研究. orcid.org/0000-0002-7735-6342. E-mail： zhangxf@hdu.edu.cn
服务　　
把本文推荐给朋友
加入引用管理器
E-mail Alert
作者相关文章　　
张雪峰
陈秀莉
僧德文

引用本文:

张雪峰,陈秀莉,僧德文. 融合用户信任和影响力的top-N推荐算法[J]. 浙江大学学报(工学版), 2020, 54(2): 311-319.

Xue-feng ZHANG,Xiu-li CHEN,De-wen SENG. Top-N recommendation algorithm combining user trust and influence. Journal of ZheJiang University (Engineering Science), 2020, 54(2): 311-319.

链接本文:

http://www.zjujournals.com/eng/CN/10.3785/j.issn.1008-973X.2020.02.012    或    http://www.zjujournals.com/eng/CN/Y2020/V54/I2/311

图 1  特征提取流程图
数据集 用户数量 项目数量 评分记录 评分稀疏度/%
Epinions 40 163 139 738 664 824 0.01
Ciao 7 375 99 746 139 738 0.04
FilmTrust 1 508 2 071 40 163 1.14
表 1  数据集稀疏性分析
图 2  各算法得出的排名与实际影响力的相关性分析
排名 Epinions Ciao FilmTrust
P@10 $\alpha $ $\beta $ $z$ $\mu $ P@10 $\alpha $ $\beta $ $z$ $\mu $ P@10 $\alpha $ $\beta $ $z$ $\mu $
1 0.010 486 0.5 2.0 2.0 0.5 0.023 78 0.5 2.0 2.0 2.0 0.352 258 2.0 1.0 1.0 0.5
2 0.010 467 1.0 2.0 0.5 0.5 0.023 609 0.5 0.5 0.5 0.5 0.351 964 2.0 0.5 2.0 2.0
3 0.010 379 2.0 1.0 2.0 0.5 0.023 411 1.0 2.0 1.0 1.0 0.351 819 0.5 1.0 0.5 1.0
4 0.010 243 1.0 2.0 1.0 1.0 0.023 396 0.5 2.0 1.0 0.5 0.351 819 1.0 1.0 0.5 0.5
5 0.010 231 0.5 2.0 0.5 2.0 0.023 36 2.0 0.5 0.5 1.0 0.351 775 0.5 0.5 0.5 0.5
表 2  3个数据集上精确率排名Top-5的参数配置
图 3  参数$s$、$\delta $在3个数据集上的精确率结果
图 4  参数$k$在3个数据集上的精确率结果
数据集 方法 N=5 N=10
$P@N$ $F1@N$ ${\rm{NDCG}}@N$ $P@N$ $F1@N$ ${\rm{NDCG}}@N$
Epinions MostPop 0.011 690 0.012 98 0.012 334 0.009 171 0.013 05 0.016 238
GBPR 0.009 353 0.011 03 0.012 296 0.007 560 0.011 11 0.016 095
FISM 0.011 470 0.013 07 0.012 808 0.009 020 0.013 15 0.016 361
FST 0.011 790 0.013 30 0.013 988 0.009 187 0.013 28 0.016 930
FSTID- 0.012 310 0.014 02 0.014 355 0.010 240 0.014 59 0.017 588
FSTID 0.012 430 0.014 15 0.014 470 0.010 480 0.014 76 0.017 832
Ciao MostPop 0.026 770 0.024 36 0.025 906 0.021 420 0.026 62 0.033 443
GBPR 0.022 280 0.020 63 0.022 319 0.018 270 0.021 16 0.028 759
FISM 0.027 040 0.024 95 0.026 185 0.021 410 0.026 87 0.032 510
FST 0.027 410 0.025 23 0.027 240 0.021 740 0.027 20 0.034 910
FSTID- 0.028 300 0.026 44 0.027 389 0.023 290 0.029 14 0.035 503
FSTID 0.029 240 0.026 82 0.027 634 0.023 610 0.029 50 0.035 932
FilmTrust MostPop 0.417 000 0.409 50 0.409 529 0.350 300 0.451 80 0.538 924
GBPR 0.412 400 0.405 10 0.372 923 0.347 000 0.445 80 0.500 997
FISM 0.417 100 0.408 70 0.413 404 0.350 300 0.451 60 0.540 511
FST 0.419 100 0.409 90 0.419 351 0.351 400 0.452 10 0.545 109
FSTID- 0.419 800 0.411 60 0.426 273 0.353 200 0.454 10 0.547 688
FSTID 0.420 500 0.412 40 0.427 569 0.353 300 0.454 80 0.551 260
表 3  Top-N项目推荐对比实验结果
数据集 MostPop GBPR FISM FST FSTID
Epinions 8.98 97.20 106.20 47.00 63.60
Ciao 0.38 8.86 7.70 20.00 11.61
FilmTrust 0.05 0.92 1.34 5.00 3.83
表 4  各算法在3个数据集上的实际运行时间
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: ACM, 2000: 195-204.
2 RESNICK P, IACOVOU N, SUCHAK M, et al. GroupLens: an open architecture for collaborative filtering of netnews [C]// Proceedings of the 1994 ACM Conference on Computer Supported Cooperative Work. North Carolina: ACM, 1994: 175-186.
3 BALABANOVI M, SHOHAM Y Fab: content-based, collaborative recommendation[J]. Communications of the ACM, 1997, 40 (3): 66- 72
doi: 10.1145/245108.245124
4 MARSDEN P V, FRIEDKIN N E Network studies of social influence[J]. Sociological Methods and Research, 1993, 22 (1): 127- 151
doi: 10.1177/0049124193022001006
5 KAUTZ H, SELMAN B, SHAH M Referralweb: combining social networks and collaborative filtering[J]. Communications of the ACM, 1997, 40 (3): 63- 65
doi: 10.1145/245108.245123
6 MA H, YANG H, LYU M R, et al. Sorec: social recommendation using probabilistic matrix factorization [C]// Proceedings of the 17th ACM Conference on Information and Knowledge Management. California: ACM, 2008: 931-940.
7 YANG B, LEI Y, LIU J, et al Social collaborative filtering by trust[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39 (8): 1633- 1647
doi: 10.1109/TPAMI.2016.2605085
8 JAMALI M, ESTER M. A matrix factorization technique with trust propagation for recommendation in social networks [C]// Proceedings of the 4th ACM Conference on Recommender Systems. New York: ACM, 2010: 135-142.
9 MA H, ZHOU D, LIU C, et al. Recommender systems with social regularization [C]// Proceedings of the 4th ACM International Conference on Web Search and Data Mining. Hong Kong: ACM, 2011: 287-296.
10 GUO G, ZHANG J, YORKE-SMITH N A novel recommendation model regularized with user trust and item ratings[J]. IEEE Transactions on Knowledge and Data Engineering, 2016, 28 (7): 1607- 1620
doi: 10.1109/TKDE.2016.2528249
11 KOREN Y. Factorization meets the neighborhood: a multifaceted collaborative filtering model [C]// Proceedings of the 14th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. Las Vegas: ACM, 2008: 426-434.
12 BURT R S. Structural holes: the social structure of competition [M]. Cambridge: Harvard university press, 2009.
13 GUO G, ZHANG J, ZHU F, et al Factored similarity models with social trust for top-N item recommendation[J]. Knowledge-based Systems, 2017, 122: 17- 25
doi: 10.1016/j.knosys.2017.01.027
14 RENDLE S, FREUDENTHALER C, GANTNER Z, et al. BPR: Bayesian personalized ranking from implicit feedback [C]// Proceedings of the 25th Conference on Uncertainty in Artificial Intelligence. Montreal: ACM, 2009: 452-461.
15 PAN W, CHEN L. GBPR: group preference based bayesian personalized ranking for one-class collaborative filtering [C]// Twenty-Third International Joint Conference onArtificial Intelligence. Beijing: ACM, 2013: 2691-2697.
[1] 李诺,郭斌,刘琰,景瑶,於志文. 神经协同过滤智能商业选址方法[J]. 浙江大学学报(工学版), 2019, 53(9): 1788-1794.
[2] 董立岩,金佳欢,方塬程,王越群,李永丽,孙铭会. 基于非负矩阵分解的Slope One算法[J]. 浙江大学学报(工学版), 2019, 53(7): 1349-1353.
[3] 赵廷廷,王喆,卢奕南. 基于传播概率矩阵的异构信息网络表示学习[J]. 浙江大学学报(工学版), 2019, 53(3): 548-554.
[4] 厉小军,柳虹,施寒潇,朱柳青,张亚辉. 基于深度学习的课程推荐模型[J]. 浙江大学学报(工学版), 2019, 53(11): 2139-2145.
[5] 韩勇, 宁连举, 郑小林, 林炜华, 孙中原. 基于社交信息和物品曝光度的矩阵分解推荐[J]. 浙江大学学报(工学版), 2019, 53(1): 89-98.
[6] 涂鼎, 陈岭, 陈根才, 吴勇, 王敬昌. 基于在线层次化非负矩阵分解的文本流主题检测[J]. 浙江大学学报(工学版), 2016, 50(8): 1618-1626.
[7] 居斌, 钱沄涛, 叶敏超. 基于结构投影非负矩阵分解的协同过滤算法[J]. 浙江大学学报(工学版), 2015, 49(7): 1319-1325.
[8] 苏凯, 马良荔, 孙煜飞, 郭晓明. 面向Web服务QoS预测的非负矩阵分解模型[J]. 浙江大学学报(工学版), 2015, 49(7): 1358-1366.
[9] 岳克强, 孙玲玲, 游彬, 楼立恒. 基于欠定盲分离的并行识别防碰撞算法[J]. 浙江大学学报(工学版), 2014, 48(5): 865-870.
[10] 崔建涛, 厉小润, 赵辽英. 高光谱图像亚像元级地物端元提取方法[J]. J4, 2012, 46(10): 1857-1865.
[11] 厉小润, 伍小明, 赵辽英. 非监督的高光谱混合像元非线性分解方法[J]. J4, 2011, 45(4): 607-613.
[12] 赵武锋 严晓浪. 基于多尺度梯度角和SVM的正面人脸识别方法[J]. J4, 2008, 42(4): 590-592.