Joint Item Recommendation and Attribute Inference: An Adaptive Graph Convolutional Network Approach

SIGIR '20: The 43rd International ACM SIGIR conference on research and development in Information Retrieval Virtual Event China July, 2020, pp. 679-688, 2020.

Cited by: 2|Bibtex|Views128|DOI:https://doi.org/10.1145/3397271.3401144
EI
Other Links: arxiv.org|dl.acm.org|dblp.uni-trier.de|academic.microsoft.com
Weibo:
We proposed a Adaptive Graph Convolutional Network model for joint item recommendation and attribute inference in an attributed user-item bipartite graph with missing attribute values

Abstract:

In many recommender systems, users and items are associated with attributes, and users show preferences to items. The attribute information describes users'(items') characteristics and has a wide range of applications, such as user profiling, item annotation, and feature-enhanced recommendation. As annotating user (item) attributes is a l...More

Code:

Data:

0
Introduction
  • Collaborative Filtering (CF) is one of the most popular approaches for recommender systems, which suggests personalized item recommendation by collaboratively learning user and item embeddings from user-item behavior [31, 37].
  • As user and item attributes describe user and item content information, given the complete attribute feature vector of each user, attribute enhanced collaborative filtering models have been proposed to tackle the cold-start problem [5, 36].
  • These models extended CF with additional bias terms or modeled feature interactions for preference prediction
Highlights
  • Collaborative Filtering (CF) is one of the most popular approaches for recommender systems, which suggests personalized item recommendation by collaboratively learning user and item embeddings from user-item behavior [31, 37]
  • We propose an Adaptive Graph Convolutional Network (AGCN) for both item recommendation and attribute inference
  • When comparing our proposed AGCN model with the baselines, we empirically find AGCN improves over all baselines on three datasets with different evaluation metrics
  • When comparing the baselines of LP, Graph Regularization (GR) and Semi-Graph Convolutional Networks (GCN) that are designed for attribute inference, we find GR and Semi-GCN show
  • We proposed a AGCN model for joint item recommendation and attribute inference in an attributed user-item bipartite graph with missing attribute values
  • To tackle the missing attribute problem, AGCN was designed to iteratively performing two steps: graph embedding learning with previous learned attribute values, and attribute update procedure to update the input of graph embedding learning
Methods
  • To evaluate the effectiveness of the proposed model, the authors conduct experiments on three public datasets: Amazon-Video Games, Movielens-1M, Movielens-20M.
  • Amazon datasets contain users’ implicit feedbacks and rich product attributes [15, 29].
  • As the original dataset of product attributes are noisy with many attributes have appeared rarely, we.
  • Input: User-item bipartite Graph G; graph propagation depth K; Output: Parameter Θr in graph learning module, Θa in attribute update module; 1: Random initialize model parameters; 2: l=0; 3: Calculate initial user attribute Xl and Yl (Eq(3)); 4: while not converged do.
  • 11: Predict attribute values (Eq(9)).
Results
  • E.g., the proposed model shows about 7% improvement for item recommendation and more than 10% improvement for attribute inference compared to the best baselines.
Conclusion
  • The authors proposed a AGCN model for joint item recommendation and attribute inference in an attributed user-item bipartite graph with missing attribute values.
  • To tackle the missing attribute problem, AGCN was designed to iteratively performing two steps: graph embedding learning with previous learned attribute values, and attribute update procedure to update the input of graph embedding learning.
  • AGCN could adaptively adjusted the graph learning process by incorporating the given attributes and the estimated attributes, in order to provide weak supervised signals to facilitate both tasks.
  • Experimental results on three real-world datasets clearly showed the effectiveness of the proposed model
Summary
  • Introduction:

    Collaborative Filtering (CF) is one of the most popular approaches for recommender systems, which suggests personalized item recommendation by collaboratively learning user and item embeddings from user-item behavior [31, 37].
  • As user and item attributes describe user and item content information, given the complete attribute feature vector of each user, attribute enhanced collaborative filtering models have been proposed to tackle the cold-start problem [5, 36].
  • These models extended CF with additional bias terms or modeled feature interactions for preference prediction
  • Objectives:

    Given the user set U , item set V , user-item preference matrix R ∈ RM×N , user attribute matrix X ∈ Rdx ×M and attribute value indication matrix AX ∈ Rdx ×M , item attribute matrix Y ∈ Rdy ×N along with item attribute value indication matrix AY ∈ Rdy ×N , the goals are to recommend items to users and predict the missing attribute values of either users or items.
  • Methods:

    To evaluate the effectiveness of the proposed model, the authors conduct experiments on three public datasets: Amazon-Video Games, Movielens-1M, Movielens-20M.
  • Amazon datasets contain users’ implicit feedbacks and rich product attributes [15, 29].
  • As the original dataset of product attributes are noisy with many attributes have appeared rarely, we.
  • Input: User-item bipartite Graph G; graph propagation depth K; Output: Parameter Θr in graph learning module, Θa in attribute update module; 1: Random initialize model parameters; 2: l=0; 3: Calculate initial user attribute Xl and Yl (Eq(3)); 4: while not converged do.
  • 11: Predict attribute values (Eq(9)).
  • Results:

    E.g., the proposed model shows about 7% improvement for item recommendation and more than 10% improvement for attribute inference compared to the best baselines.
  • Conclusion:

    The authors proposed a AGCN model for joint item recommendation and attribute inference in an attributed user-item bipartite graph with missing attribute values.
  • To tackle the missing attribute problem, AGCN was designed to iteratively performing two steps: graph embedding learning with previous learned attribute values, and attribute update procedure to update the input of graph embedding learning.
  • AGCN could adaptively adjusted the graph learning process by incorporating the given attributes and the estimated attributes, in order to provide weak supervised signals to facilitate both tasks.
  • Experimental results on three real-world datasets clearly showed the effectiveness of the proposed model
Tables
  • Table1: The statistics of the three datasets (“s" means single-label attribute and “m” means multi-label attribute)
  • Table2: HR@N comparisons for item recommendation with different top-N values
  • Table3: NDCG@N comparison for item recommendation with different top-N values
  • Table4: Performance comparisons for attribute inference
  • Table5: Performance comparisons of different propagation depth K on three datasets
Download tables as Excel
Related work
  • CF has been widely used in most recommender systems due to its relatively high performance with easy to collect data [19, 28, 44]. Classical latent factor based models relied on matrix factorization for user and item embedding learning [6, 7, 23, 31]. As most users implicitly express their item preferences, Bayesian Personalized Ranking (BPR) was proposed with a ranking based loss function to deal with the implicit feedback [37]. In practice, CF based models suffer from the cold-start problem and could not perform well when users have limited rating records [33]. To tackle the data sparsity issue, many efforts have been devoted to incorporate auxiliary information in CF based models, such as user (item) attributes [35], item content [16], social network [39? ], and so on. Among them, attribute enhanced CF are widely studied as the attribute information are easy to collect in most platforms. Researchers proposed to mimic the latent factor distribution with the associated features of users and items [2]. SVDFeature extended over classical latent factor based models with additional bias terms, which are learned from the associated attributes [5]. Factorization machines modeled pairwise interactions between all features and was a generalized model since they can mimic most factorization models with feature engineering [35, 36]. All these feature enhanced CF models assume that the attribute information is complete. However, in the realworld, user and item attributes are incomplete with many missing values. As these models could not tackle the missing feature value issue, a preprocessing step is usually adopted to fill the missing values, such as each missing attribute value is filled by the average value, or a computational model to predict the missing values at first [18, 38]. Instead of using preprocessing step to tackle the missing attribute problem, we design a model that learns attribute inference and item recommendation at the same time.
Funding
  • This work was supported in part by National Key Research and Development Program of China(Grant No.2017YFB0803301), the National Natural Science Foundation of China(Grant No.61725203, 61972125, U19A2079, 61722204, 61932009 and 61732008)
Reference
  • Sami Abu-El-Haija, Amol Kapoor, Bryan Perozzi, and Joonseok Lee. 2018. N-gcn: Multi-scale graph convolution for semi-supervised node classification. In UAI. 310.
    Google ScholarFindings
  • Deepak Agarwal and Bee-Chung Chen. 2009. Regression-based latent factor models. In SIGKDD. 19–28.
    Google ScholarLocate open access versionFindings
  • Mikhail Belkin, Partha Niyogi, and Vikas Sindhwani. 2006. Manifold regularization: A geometric framework for learning from labeled and unlabeled examples. IMLR 7, Nov (2006), 2399–2434.
    Google ScholarLocate open access versionFindings
  • Lei Chen, Le Wu, Richang Hong, Kun Zhang, and Meng Wang. 2020. Revisiting Graph based Collaborative Filtering: A Linear Residual Graph Convolutional Network Approach. In AAAI. In Press.
    Google ScholarLocate open access versionFindings
  • Tianqi Chen, Weinan Zhang, Qiuxia Lu, Kailong Chen, Zhao Zheng, and Yong Yu. 2012. SVDFeature:a toolkit for feature-based collaborative filtering. JMLR 13, Dec (2012), 3619–3622.
    Google ScholarLocate open access versionFindings
  • Zhiyong Cheng, Ying Ding, Xiangnan He, Lei Zhu, Xuemeng Song, and Mohan S Kankanhalli. 2018. A 3NCF: An Adaptive Aspect Attention Model for Rating Prediction.. In IJCAI. 3748–3754.
    Google ScholarFindings
  • Zhiyong Cheng, Ying Ding, Lei Zhu, and Mohan Kankanhalli. 2018. Aspectaware latent factor model: Rating prediction with ratings and reviews. In WWW. 639–648.
    Google ScholarLocate open access versionFindings
  • Wei-Lin Chiang, Xuanqing Liu, Si Si, Yang Li, Samy Bengio, and Cho-Jui Hsieh. 2019. Cluster-gcn: An efficient algorithm for training deep and large graph convolutional networks. In SIGKDD. 257–266.
    Google ScholarLocate open access versionFindings
  • Jong-bum Choi, Sung-Bum Park, Woo-sung Shim, Young-Ho Moon, Dai-woong Choi, and Jae-won Yoon. 2013. Method and apparatus for encoding/decoding image by using adaptive binarization. US Patent 8,526,750.
    Google ScholarFindings
  • Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. 2009. Imagenet: A large-scale hierarchical image database. In CVPR. 248–255.
    Google ScholarLocate open access versionFindings
  • Mark Everingham, Luc Van Gool, Christopher KI Williams, John Winn, and Andrew Zisserman. 2010. The pascal visual object classes (voc) challenge. IJCV 88, 2 (2010), 303–338.
    Google ScholarLocate open access versionFindings
  • Neil Zhenqiang Gong, Ameet Talwalkar, Lester Mackey, Ling Huang, Eui Chul Richard Shin, Emil Stefanov, Elaine Runting Shi, and Dawn Song. 2014. Joint link prediction and attribute inference using a social-attribute network. TIST 5, 2 (2014), 27.
    Google ScholarLocate open access versionFindings
  • Will Hamilton, Zhitao Ying, and Jure Leskovec. 2017. Inductive representation learning on large graphs. In NIPS. 1024–1034.
    Google ScholarFindings
  • F Maxwell Harper and Joseph A Konstan. 2015. The movielens datasets: History and context. TIIS 5, 4 (2015), 1–19.
    Google ScholarLocate open access versionFindings
  • Ruining He and Julian McAuley. 2016. Ups and downs: Modeling the visual evolution of fashion trends with one-class collaborative filtering. In WWW. 507– 517.
    Google ScholarLocate open access versionFindings
  • Ruining He and Julian McAuley. 20VBPR: visual bayesian personalized ranking from implicit feedback. In AAAI. 144–150.
    Google ScholarLocate open access versionFindings
  • Xiangnan He, Lizi Liao, Hanwang Zhang, Liqiang Nie, Xia Hu, and Tat-Seng Chua. 20Neural collaborative filtering. In WWW. 173–182.
    Google ScholarLocate open access versionFindings
  • José Miguel Hernández-Lobato, Neil Houlsby, and Zoubin Ghahramani. 2014. Probabilistic matrix factorization with non-random missing data. In ICML. 1512– 1520.
    Google ScholarFindings
  • Yifan Hu, Yehuda Koren, and Chris Volinsky. 2008. Collaborative filtering for implicit feedback datasets. In ICDM. 263–272.
    Google ScholarLocate open access versionFindings
  • Di Jin, Ziyang Liu, Weihao Li, Dongxiao He, and Weixiong Zhang. 2019. Graph convolutional networks meet Markov random fields: Semi-supervised community detection in attribute networks. In AAAI, Vol. 33. 152–159.
    Google ScholarLocate open access versionFindings
  • Thomas N Kipf and Max Welling. 2017. Semi-supervised classification with graph convolutional networks. In ICLR.
    Google ScholarFindings
  • Ajith Kodakateri Pudhiyaveetil, Susan Gauch, Hiep Luong, and Josh Eno. 2009. Conceptual recommender system for CiteSeerX. In RecSys. 241–244.
    Google ScholarFindings
  • Yehuda Koren, Robert Bell, and Chris Volinsky. 2009. Matrix factorization techniques for recommender systems. Computer 42, 8 (2009), 30–37.
    Google ScholarLocate open access versionFindings
  • Michal Kosinski, David Stillwell, and Thore Graepel. 2013. Private traits and attributes are predictable from digital records of human behavior. PNAS 110, 15 (2013), 5802–5805.
    Google ScholarLocate open access versionFindings
  • Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. 2012. Imagenet classification with deep convolutional neural networks. In NIPS. 1097–1105.
    Google ScholarFindings
  • Qimai Li, Zhichao Han, and Xiao-Ming Wu. 2018. Deeper insights into graph convolutional networks for semi-supervised learning. In AAAI. 3538–3545.
    Google ScholarFindings
  • Shie Mannor, Dori Peleg, and Reuven Rubinstein. 2005. The cross entropy method for classification. In ICML. 561–568.
    Google ScholarLocate open access versionFindings
  • Benjamin M Marlin and Richard S Zemel. 2009. Collaborative prediction and ranking with non-random missing data. In RecSys. 5–12.
    Google ScholarFindings
  • Julian McAuley, Christopher Targett, Qinfeng Shi, and Anton Van Den Hengel. 2015. Image-based recommendations on styles and substitutes. In SIGIR. 43–52.
    Google ScholarLocate open access versionFindings
  • Miller McPherson, Lynn Smith-Lovin, and James M Cook. 2001. Birds of a feather: Homophily in social networks. Annual review of sociology 27, 1 (2001), 415–444.
    Google ScholarLocate open access versionFindings
  • Andriy Mnih and Ruslan R Salakhutdinov. 2008. Probabilistic matrix factorization. In NIPS. 1257–1264.
    Google ScholarFindings
  • Baback Moghaddam and Alex Pentland. 1995. Probabilistic visual learning for object detection. In ICCV. IEEE, 786–793.
    Google ScholarLocate open access versionFindings
  • Seung-Taek Park and Wei Chu. 2009. Pairwise preference regression for cold-start recommendation. In RecSys. 21–28.
    Google ScholarFindings
  • Damien Poirier, Isabelle Tellier, Françoise Fessant, and Julien Schluth. 2010. Towards text-based recommendations. In RIAO. 136–137.
    Google ScholarLocate open access versionFindings
  • Steffen Rendle. 2010. Factorization machines. In ICDM. 995–1000.
    Google ScholarLocate open access versionFindings
  • Steffen Rendle. 2012. Factorization machines with libfm. TIST 3, 3 (2012), 57.
    Google ScholarLocate open access versionFindings
  • Steffen Rendle, Christoph Freudenthaler, Zeno Gantner, and Lars Schmidt-Thieme.
    Google ScholarFindings
  • 2009. BPR: Bayesian personalized ranking from implicit feedback. In UAI. 452– 461.
    Google ScholarLocate open access versionFindings
  • [38] Shaoyun Shi, Min Zhang, Xinxing Yu, Yongfeng Zhang, Bin Hao, Yiqun Liu, and Shaoping Ma. 2019. Adaptive Feature Sampling for Recommendation with Missing Content Feature Values. In CIKM. 1451–1460.
    Google ScholarFindings
  • [39] Peijie Sun, Le Wu, and Meng Wang. 2018. Attentive Recurrent Social Recommendation. In SIGIR. 185–194.
    Google ScholarLocate open access versionFindings
  • [40] Peijie Sun, Le Wu, Kun Zhang, Yanjie Fu, Richang Hong, and Meng Wang. 2020. Dual Learning for Explainable Recommendation: Towards Unifying User Preference Prediction and Review Generation. In WWW. 837–847.
    Google ScholarLocate open access versionFindings
  • [41] Rianne van den Berg, Thomas N Kipf, and Max Welling. 2017. Graph Convolutional Matrix Completion. STAT 1050 (2017), 7.
    Google ScholarLocate open access versionFindings
  • [42] Xiang Wang, Xiangnan He, Meng Wang, Fuli Feng, and Tat-Seng Chua. 2019. Neural Graph Collaborative Filtering. In SIGIR. 165–174.
    Google ScholarLocate open access versionFindings
  • [43] Le Wu, Yong Ge, Qi Liu, Enhong Chen, Richang Hong, Junping Du, and Meng Wang. 2017. Modeling the evolution of usersâĂŹ preferences and social links in social networking services. TKDE 29, 6 (2017), 1240–1253.
    Google ScholarLocate open access versionFindings
  • [44] Le Wu, Qi Liu, Enhong Chen, Nicholas Jing Yuan, Guangming Guo, and Xing Xie. 2016. Relevance meets coverage: A unified framework to generate diversified recommendations. TIST 7, 3 (2016), 1–30.
    Google ScholarLocate open access versionFindings
  • [45] Le Wu, Peijie Sun, Yanjie Fu, Richang Hong, Xiting Wang, and Meng Wang. 2019. A Neural Influence Diffusion Model for Social Recommendation. In SIGIR. 235–244.
    Google ScholarLocate open access versionFindings
  • [46] Carl Yang, Lin Zhong, Li-Jia Li, and Luo Jie. 2017. Bi-directional joint inference for user links and attributes on large social graphs. In WWW. 564–573.
    Google ScholarLocate open access versionFindings
  • [47] Rex Ying, Ruining He, Kaifeng Chen, Pong Eksombatchai, William L Hamilton, and Jure Leskovec. 2018. Graph Convolutional Neural Networks for Web-Scale Recommender Systems. In SIGKDD. 974–983.
    Google ScholarLocate open access versionFindings
  • [48] Xiaojin Zhu and Zoubin Ghahramani. 2002. Learning from labeled and unlabeled data with label propagation. (2002).
    Google ScholarFindings
  • [49] Xiaojin Zhu, Zoubin Ghahramani, and John D Lafferty. 2003. Semi-supervised learning using gaussian fields and harmonic functions. In ICML. 912–919.
    Google ScholarLocate open access versionFindings
Full Text
Your rating :
0

 

Tags
Comments