考虑多层次潜在特征的个性化推荐模型

doi:10.11959/j.issn.1000-0801.2022035

摘要/Abstract

摘要：

个性化推荐已成为解决信息过载的最有效手段之一，也是海量数据挖掘研究领域的热点技术。然而传统推荐算法往往只使用用户对物品的评分信息，而缺少对用户与物品潜在特征的综合考虑。基于因子分解机、宽神经网络、交叉网络和深度神经网络的融合，提出一种新的考虑多层次潜在特征的模型，可以提取用户与物品的浅层潜在特征、低阶非线性潜在特征、线性交叉潜在特征以及高阶非线性潜在特征。在 4 个常用的数据集上的实验结果表明，考虑用户与物品多层次潜在特征可以有效提高个性化推荐的预测精度。最后，研究了嵌入层维度以及神经元数量等因素对新模型预测性能的影响。

关键词: 个性化推荐, 层次化潜在特征, 深度学习

Abstract:

Personalized recommendation has become one of the most effective means to solve information overload, and it is also a hot technology in the research field of massive data mining.However, traditional recommendation algorithms often only use the user’s rating information on the item, and lack a comprehensive consideration of the potential characteristics of the user and the item.The factorization machine, wide neural network, crossover network and deep neural network were combined to extract the shallow latent features, low-order nonlinear latent features, linear cross latent features, and high-order nonlinear latent features of users and items.Thus, a new deep learning personalized recommendation model with multilevel latent features was established.The experimental results on four commonly used data sets show that considering the multi-level potential features of users and items can effectively improve the prediction accuracy of personalized recommendations.Finally, the influence of factors such as the dimensions of the embedding layer and the number of neurons on the prediction performance of the new model was studied.

Key words: personalized recommendation, hierarchical latent feature, deep learning

中图分类号:

TP391.3

申情, 郭文宾, 楼俊钢, 余强国. 考虑多层次潜在特征的个性化推荐模型[J]. 电信科学, 2022, 38(2): 71-83.

Qing SHEN, Wenbin GUO, Jungang LOU, Qiangguo YU. Personalized recommendation model with multi-level latent features[J]. Telecommunications Science, 2022, 38(2): 71-83.

图/表 16

图1

表1

表2

不同数据集按比例划分训练集与测试集的SR"

训练集占比	80%	$70 %$	50%	30%
MovieLens-100k	0.949 6	$0 . 9559$	0.968 5	0.981 1
MovieLens-1m	0.964 3	$0 . 9687$	0.977 7	0.986 6
MovieLens-latest-small	0.986 4	$0 . 9881$	0.991 5	0.994 9
FilmTrust	0.990 9	$0 . 9920$	0.994 3	0.996 5

表2

表3

表4

表5

表6

表7

表8

表9

表10

图2

图3

图4

图5

表11

参考文献 27

[1]	项亮 . 推荐系统实践[M]. 北京: 人民邮电出版社, 2012.
	XIANG L . The Development of recommendation system[M]. Beijing: Posts ＆ Telecom Press, 2012.
[2]	BPRATTOL ， CARTAS ， FENUG ,et al. Semantics-aware content-based recommender systems:Design and architecture guidelines[J]. Neurocomputing, 2017(254): 79-85.
[3]	YANG S , KORAYEM M , ALJADDA K ,et al. Combining content-based and collaborative filtering for job recommendation system:a cost-sensitive statistical relational learning approach[J]. Knowledge-Based Systems, 2017(136): 37-45.
[4]	RENDLE S , . Factorization machines[C]// Proceedings of 2010 IEEE International Conference on Data Mining. Piscataway:IEEE Press, 2010: 995-1000.
[5]	ZHANG L B , LUO T J , ZHANG F ,et al. A recommendation model based on deep neural network[J]. IEEE Access, 2018,6: 9454-9463.
[6]	BI J W , LIU Y , FAN Z P . A deep neural networks based recommendation algorithm using user and item basic data[J]. International Journal of Machine Learning and Cybernetics, 2020,11(4): 763-777.
[7]	CHENG H T , KOC L , HARMSEN J ,et al. Wide ＆ deep learning for recommender systems[C]// DLRS 2016:Proceedings of the 1st Workshop on Deep Learning for Recommender Systems.[S.l.:s.n.], 2016: 7-10.
[8]	GUO H F , TANG R M , YE Y M ,et al. DeepFM:a factorization-machine based neural network for CTR prediction[C]// Proceedings of the Twenty-Sixth International Joint Conference on Artificial Intelligence. California:International Joint Conferences on Artificial Intelligence Organization, 2017.
[9]	WANG R X , FU B , FU G ,et al. Deep ＆ cross network for ad click predictions[C]// ADKDD'17:Proceedings of the ADKDD'17.[S.l.:s.n.], 2017: 1-7.
[10]	GUILHERME BRANDAO MARTINS , JOAO PAULO PAPA , HOJJAT ADELI . Deep learning techniques for recommender systems based on collaborative filtering[J]. Expert Systems, 2020,37(6): 1-21.
[11]	MYLAVARAPU B K . Collaborative filtering and artificial neural network based recommendation system for advanced applications[J]. Journal of Computer and Communications, 2018,06(12): 1-14.
[12]	TANG H , LEI M , GONG Q ,et al. A BP neural network recommendation algorithm based on cloud model[J]. IEEE Access, 2019(7): 35898-35907.
[13]	薛峰, 刘凯, 王东 ,等. 基于深度神经网络和加权隐反馈的个性化推荐[J]. 模式识别与人工智能, 2020,33(4): 295-302.
	XUE F , LIU K , WANG D ,et al. Personalized recommendation algorithm based on deep neural network and weighted implicit feedback[J]. Pattern Recognition and Artificial Intelligence, 2020,33(4): 295-302.
[14]	杨洁, 朱咸军, 周献中 ,等. 基于混杂社会网络的个性化Web服务推荐方法[J]. 电子学报, 2020,48(2): 341-349.
	YANG J , ZHU X J , ZHOU X Z ,et al. Personalized web service recommendation based on heterogeneous social network[J]. Acta Electronica Sinica, 2020,48(2): 341-349.
[15]	HUI B , ZHANG L , ZHOU X ,et al. Personalized recommendation system based on knowledge embedding and historical behavior[J]. Applied Intelligence, 2022(52): 954-966.
[16]	ZHEN W , ALLAM M , LIANG M B . Research on e-commerce personalized recommendation system based on big data technology[C]// Proceedings of 2021 IEEE 2nd International Conference on Information Technology,Big Data and Artificial Intelligence (ICIBA)，Piscataway:IEEE Press, 2021, 909-913.
[17]	邵英玮, 张敏, 马为之 ,等. 融合商品潜在互补性发现的个性化推荐方法[J]. 软件学报, 2020,31(4): 1090-1100.
	SHAO Y W , ZHANG M , MA W Z ,et al. Integrating latent item-item complementarity with personalized recommendation systems[J]. Journal of Software, 2020,31(4): 1090-1100.
[18]	FU M S , QU H , YI Z ,et al. A novel deep learning-based collaborative filtering model for recommendation system[J]. IEEE Transactions on Cybernetics, 2019,49(3): 1084-1096.
[19]	YAN W J , WANG D , CAO M J ,et al. Deep auto encoder model with convolutional text networks for video recommendation[J]. IEEE Access, 2019(7): 40333-40346.
[20]	TANG J X , WANG K . Personalized top-N sequential recommendation via convolutional sequence embedding[C]// Proceedings of WSDM '18:Proceedings of the Eleventh ACM International Conference on Web Search and Data Mining.[S.l.:s.n.], 2018: 565-573.
[21]	邓路佳, 刘平山 . 基于GMM-FMs的广告点击率预测研究[J]. 计算机工程, 2019,45(5): 122-126.
	DENG L J , LIU P S . Research on click-through rate prediction of advertisement based on GMM-FMs[J]. Computer Engineering, 2019,45(5): 122-126.
[22]	YANG D Q , CHEN L H , LIANG J Q ,et al. Social tag embedding for the recommendation with sparse user-item interactions[C]// Proceedings of 2018 IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining (ASONAM). Piscataway:IEEE Press, 2018: 127-134.
[23]	郁豹, 李振华, 张凯 ,等. 基于DeepFM模型的广告推荐系统研究[J]. 计算机应用与软件, 2019,36(7): 307-310,316.
	YU B , LI Z H , ZHANG K ,et al. Advertisement recommendation system based on deepfm model[J]. Computer Applications and Software, 2019,36(7): 307-310,316.
[24]	LIAN J X , ZHOU X H , ZHANG F Z ,et al. xDeepFM:combining explicit and implicit feature interactions for recommender systems[C]// Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery ＆ Data Mining. New York:ACM Press, 2018.
[25]	NAUMOV M , MUDIGERE D , SHI H J M ,et al. Deep learning recommendation model for personalization and recommendation systems[EB]. 2019.
[26]	SONG W P , SHI C C , XIAO Z P ,et al. AutoInt:automatic feature interaction learning via self-attentive neural networks[C]// Proceedings of CIKM '19:Proceedings of the 28th ACM International Conference on Information and Knowledge Management.[S.l.:s.n.], 2019: 1161-1170.
[27]	HONGTAO X , QINGSHENG Z , HONGCHUN Q ,et al. User-based collaborative recommendation filtering algorithm using extremely valued ratings[J]. International Journal of Digital Content Technology and Its Applications, 2011,5(9): 47-54.

数据集	数据集大小	评分区间
MovieLens-100k	10万余条评分，943位用户，1 682部电影	1～5
MovieLens-1m	100 万余条评分，6 040位用户，3 952部电影	1～5
MovieLens-l atest-small	100 836 条评分，610 位用户，9 742部电影	0.5～5
FilmTrust	35 497条评分，1 508位用户、2 071部电影	0.5～4

训练集占比	80%	70%	50%	30%
UB	0.825 4±0.022 7	0.838 5±0.013 2	0.842 9±0.017 9	0.882 3±0.027 2
FM	0.744 9±0.002 3	0.745 8±0.005 4	0.747 7±0.007 9	0.761 3±0.009 4
DNN	0.732 5±0.009 3	0.741 4±0.019 9	0.745 8±0.013 1	0.760 5±0.017 2
DeepFM	0.719 9±0.005 3	0.729 5±0.006 9	0.742 2±0.004 5	0.759 1±0.006 4
DCN	0.723 2±0.007 3	0.730 1±0.005 7	0.744 9±0.007 8	0.758 3±0.008 9
WD	0.719 4±0.005 4	0.728 9±0.004 9	0.7411±0.009 5	0.759 3±0.010 1
本文模型	0.710 2±0.004 3	0.724 9±0.006 7	0.735 2±0.007 9	0.754 7±0.005 3

训练集占比	80%	70%	50%	30%
UB	1.028 3±0.034 1	1.043 3±0.017 3	1.054 2±0.021 5	1.094 3±0.029 5
FM	0.944 1±0.003 3	0.945 3±0.007 7	0.948 5±0.009 4	0.967 5±0.013 1
DNN	0.936 3±0.011 1	0.942 7±0.014 9	0.947 4±0.006 6	0.966 1±0.009 2
DeepFM	0.920 1±0.006 9	0.928 9±0.006 1	0.938 6±0.008 6	0.959 2±0.011 7
DCN	0.922 9±0.009 4	0.931 1±0.008 3	0.941 6±0.010 9	0.959 7±0.009 1
WD	0.919 4±0.006 8	0.927 3±0.009 1	0.937 9±0.014 2	0.958 1±0.007 4
本文模型	0.909 8±0.005 3	0.926 1±0.004 3	0.931 2±0.005 5	0.952 4±0.008 5

训练集占比	80%	70%	50%	30%
UB	0.821 3±0.008 1	0.830 3±0.012 4	0.851 4±0.015 8	0.876 9±0.019 3
FM	0.674 5±0.018 2	0.675 6±0.019 6	0.706 8±0.021 4	0.734 2±0.025 1
DNN	0.692 6±0.019 7	0.701 8±0.020 3	0.719 4±0.024 6	0.738 5±0.026 8
DeepFM	0.672 8±0.007 8	0.672 7±0.009 1	0.688 4±0.012 6	0.706 8±0.013 4
DCN	0.673 6±0.014 1	0.673 3±0.015 7	0.689 5±0.017 3	0.714 5±0.016 2
WD	0.674 7±0.009 3	0.673 9±0.010 6	0.691 7±0.011 4	0.718 1±0.013 6
本文模型	0.663 5±0.011 7	0.668 1±0.012 9	0.685 2±0.013 5	0.702 3±0.015 1

训练集占比	80%	70%	50%	30%
UB	1.019 8±0.007 2	1.028 7±0.009 1	1.047 6±0.012 7	1.073 7±0.015 3
FM	0.862 3±0.017 8	0.864 7±0.019 4	0.885 4±0.021 3	0.915 8±0.023 8
DNN	0.877 4±0.015 4	0.886 4±0.018 3	0.897 2±0.020 6	0.923 8±0.026 8
DeepFM	0.856 5±0.008 7	0.862 3±0.009 5	0.873 4±0.010 7	0.899 1±0.011 3
DCN	0.857 2±0.005 7	0.863 7±0.008 6	0.875 9±0.007 4	0.903 4±0.009 1
WD	0.858 6±0.017 5	0.863 1±0.013 2	0.879 3±0.014 4	0.908 9±0.012 8
本文模型	0.849 3±0.019 7	0.857 2±0.014 5	0.872 6±0.015 9	0.891 2±0.013 7