Learning Personalized Risk Preferences for Recommendation

SIGIR '20: The 43rd International ACM SIGIR conference on research and development in Information Retrieval Virtual Event China July, 2020, pp. 409-418, 2020.

Cited by: 0|Bibtex|Views103|DOI:https://doi.org/10.1145/3397271.3401056
EI
Other Links: arxiv.org|dl.acm.org|dblp.uni-trier.de|academic.microsoft.com
Weibo:
Among all the baseline models, we can see that all pair-wise learning methods are much better than the simple point-wise method, which demonstrates the superiority of pair-wise learning to rank methods on top-K ranking tasks

Abstract:

The rapid growth of e-commerce has made people accustomed to shopping online. Before making purchases on e-commerce websites, most consumers tend to rely on rating scores and review information to make purchase decisions. With this information, they can infer the quality of products to reduce the risk of purchase. Specifically, items with...More

Code:

Data:

0
Introduction
  • Designing personalized recommender systems is able to help users find relevant items efficiently in the context of web information overload.
  • A common observation in practical recommendation systems is that customers may use rating scores and review information to support their buying decisions [26, 28, 33].
  • According to this information, buyers can infer the quality of products to avoid wasting time and reduce the risk of purchase.
  • Integrating machine learning and the established economic principles can help to model the risk attitude of the user decision process based on large-scale user transaction logs
Highlights
  • Designing personalized recommender systems is able to help users find relevant items efficiently in the context of web information overload
  • We propose an optimization framework based on discrete choice model for our risk-aware recommendation learning framework — Risk-aware Recommendation (RARE)
  • Among all the baseline models, we can see that all pair-wise learning methods (BPR, NCF, MPUM, and RARE) are much better than the simple point-wise (SVD) method, which demonstrates the superiority of pair-wise learning to rank methods on top-K ranking tasks
  • We introduce risk attitudes into personalized recommendation systems and propose risk-aware recommendation
  • We believe that online shopping could be risky with uncertainty, and the risk attitudes of different consumers may affect their decision making processes under uncertainty/risk
  • We advance prospect theory into a personalized version based on machine learning over large-scale consumer transaction logs
Methods
  • 5.1 Dataset Description

    The authors use the consumer transaction data from different sources — Amazon1 [17, 31], Movielens 2 [15], Ciao and Epinions 3 [42, 43] — in the experiments to verify the recommendation performance of RARE4.
  • Considering that three Amazon datasets and Movielens1M only contain one certain category of items, the authors do the experiments on another two widely used datasets — Ciao and Epinions, which contain items from different categories.
  • Since the authors need prices to calculate the value function, which are not recorded by Movielens1M, Ciao, and Epinions, the authors set the price of each item in these three datasets equal to the same number, for example, $1.
Results
  • The major experimental results are shown in Table 4, besides, the authors plot the N DCG in Figure 2 under different length of recommendation list K.
  • In order to prove the validity of the improvement, the authors plot the NDCG for different lengths of recommendation lists on all datasets, shown in Fig.2.
  • The black curve represented as RARE, is always above the best baselines in all six datasets
  • These observations imply that by modeling user behaviors under uncertainty based on established risk-aware principles, the model has the ability to capture better user preferences resulting in better recommendation results
Conclusion
  • The authors introduce risk attitudes into personalized recommendation systems and propose risk-aware recommendation.
  • The authors bridge prospect theory and machine learning algorithms together to predict individual risk attitudes.
  • The authors advance prospect theory into a personalized version based on machine learning over large-scale consumer transaction logs.
  • The authors will consider user risk attitudes in other online systems beyond e-commerce recommendation, and consider other economic principles and/or learning methods to benefit recommendation systems both effectively and economically
Summary
  • Introduction:

    Designing personalized recommender systems is able to help users find relevant items efficiently in the context of web information overload.
  • A common observation in practical recommendation systems is that customers may use rating scores and review information to support their buying decisions [26, 28, 33].
  • According to this information, buyers can infer the quality of products to avoid wasting time and reduce the risk of purchase.
  • Integrating machine learning and the established economic principles can help to model the risk attitude of the user decision process based on large-scale user transaction logs
  • Objectives:

    The authors' objective is to solve the classical problem of decision-making under uncertainty, and the authors define the uncertainty in e-commerce.
  • Methods:

    5.1 Dataset Description

    The authors use the consumer transaction data from different sources — Amazon1 [17, 31], Movielens 2 [15], Ciao and Epinions 3 [42, 43] — in the experiments to verify the recommendation performance of RARE4.
  • Considering that three Amazon datasets and Movielens1M only contain one certain category of items, the authors do the experiments on another two widely used datasets — Ciao and Epinions, which contain items from different categories.
  • Since the authors need prices to calculate the value function, which are not recorded by Movielens1M, Ciao, and Epinions, the authors set the price of each item in these three datasets equal to the same number, for example, $1.
  • Results:

    The major experimental results are shown in Table 4, besides, the authors plot the N DCG in Figure 2 under different length of recommendation list K.
  • In order to prove the validity of the improvement, the authors plot the NDCG for different lengths of recommendation lists on all datasets, shown in Fig.2.
  • The black curve represented as RARE, is always above the best baselines in all six datasets
  • These observations imply that by modeling user behaviors under uncertainty based on established risk-aware principles, the model has the ability to capture better user preferences resulting in better recommendation results
  • Conclusion:

    The authors introduce risk attitudes into personalized recommendation systems and propose risk-aware recommendation.
  • The authors bridge prospect theory and machine learning algorithms together to predict individual risk attitudes.
  • The authors advance prospect theory into a personalized version based on machine learning over large-scale consumer transaction logs.
  • The authors will consider user risk attitudes in other online systems beyond e-commerce recommendation, and consider other economic principles and/or learning methods to benefit recommendation systems both effectively and economically
Tables
  • Table1: Examples showing how to calculate the prospect value for a certain user-item pair
  • Table2: Basic statistics of the experimental datasets
  • Table3: Hyperparameter settings for each dataset
  • Table4: Summary of the performance on six datasets. We evaluate for ranking (F1 and N DCG, in percentage (%) values), and K is the length of recommendation list. When RARE
  • Table5: Summary of the performance against ablations on six datasets. When RARE is the best, its improvements against the best ablation are significant at p < 0.01
Download tables as Excel
Related work
  • In this section, we will briefly introduce some background knowledge to help the readers get a better understanding of the areas that are related to our work.

    2.1 Collaborative Filtering

    Collaborative Filtering (CF) has been one of the most dominant approaches to recommender systems. Early methods of CF consider the user-item rating matrix and conduct rating prediction task with user-based [23, 38] or item-based [29, 40] collaborative filtering methods. In these methods, the user and item rating vector are considered as the representation vector for each user and item.

    With the development of dimension reduction methods, latent factor models, such as singular value decomposition (SVD) [25], non-negative matrix factorization [27], and probabilistic matrix factorization [32], are later widely adopted in recommender systems. In these aforementioned matrix factorization approaches, each user and item is learned as a latent factor representation to calculate the matching score of the user-item pair.

    Deep models have recently been further extended to collaborative filtering methods for the recommendation tasks. The relevant methods can be roughly divided into two subcategories: similarity learning methods and representational learning methods. The similarity learning approaches adopt simple user/item representations (such as one-hot) and learn a complex prediction network as the similarity function to calculate user-item matching scores [19]. Meanwhile, the representation learning approaches learn rich user/item representations and adopt a simple similarity function (e.g., inner product) for matching score calculation [49].
Reference
  • Binay Kumar Adhikari and Anup Agrawal. 2016. Does local religiosity matter for bank risk-taking? Journal of Corporate Finance 38 (2016), 272–293.
    Google ScholarLocate open access versionFindings
  • Maurice Allais. 1953. L’extension des théories de l’équilibre économique général et du rendement social au cas du risque. Econometrica, Journal of the Econometric Society (1953), 269–290.
    Google ScholarLocate open access versionFindings
  • Leif Azzopardi. 2014. Modelling interaction with economic models of search. In Proceedings of the 37th international ACM SIGIR conference on Research & development in information retrieval. 3–12.
    Google ScholarLocate open access versionFindings
  • Daniel Bernoulli. 1968. Specimen theoriae novae de mensura sortis. Gregg.
    Google ScholarFindings
  • Djallel Bouneffouf. 2014. R-UCB: a Contextual Bandit Algorithm for Risk-Aware Recommender Systems. CoRR abs/1408.2195 (2014). arXiv:1408.2195 http://arxiv.org/abs/1408.2195
    Findings
  • Djallel Bouneffouf, Amel Bouzeghoub, and Alda Lopes Ganarski. 2013. RiskAware Recommender Systems. In Neural Information Processing, Minho Lee, Akira Hirose, Zeng-Guang Hou, and Rhee Man Kil (Eds.). Springer Berlin Heidelberg, Berlin, Heidelberg, 57–65.
    Google ScholarLocate open access versionFindings
  • Heng-Tze Cheng, Levent Koc, Jeremiah Harmsen, Tal Shaked, Tushar Chandra, Hrishi Aradhye, Glen Anderson, Greg Corrado, Wei Chai, Mustafa Ispir, et al. 2016. Wide & deep learning for recommender systems. In Proceedings of the 1st workshop on deep learning for recommender systems. ACM, 7–10.
    Google ScholarLocate open access versionFindings
  • Michele Cohen, Jean-Yves Jaffray, and Tanios Said. 1987. Experimental comparison of individual behavior under risk and under uncertainty for gains and for losses. Organizational behavior and human decision processes 39, 1 (1987), 1–22.
    Google ScholarFindings
  • T.A. Domencich and D. McFadden. 1975. Urban travel demand: a behavioral analysis. NORTH-HOLLAND. https://books.google.com/books?id=Nx4pAQAAMAAJ [10] Ali Mamdouh Elkahky, Yang Song, and Xiaodong He.2015. A multi-view deep learning approach for cross domain user modeling in recommendation systems. In Proceedings of the 24th International Conference on World Wide Web. International World Wide Web Conferences Steering Committee, 278–288.
    Locate open access versionFindings
  • [11] Daniel Ellsberg. 1961. Risk, ambiguity, and the Savage axioms. The quarterly journal of economics (1961), 643–669.
    Google ScholarLocate open access versionFindings
  • [12] Andrew J. Flanagin, Miriam J. Metzger, Rebekah Pure, Alex Markov, and Ethan Hartsell. 2014. Mitigating risk in ecommerce transactions: perceptions of information credibility and the role of user-generated ratings in product quality and purchase intention. Electronic Commerce Research 14, 1 (01 Mar 2014), 1–23. https://doi.org/10.1007/s10660-014-9139-2
    Locate open access versionFindings
  • [13] Zuohui Fu, Yikun Xian, Ruoyuan Gao, Jieyu Zhao, Qiaoying Huang, Yingqiang Ge, Shuyuan Xu, Shijie Geng, Chirag Shah, Yongfeng Zhang, and Gerard de Melo. 2020. Fairness-Aware Explainable Recommendation over Knowledge Graphs. In Proceedings of SIGIR 2020. ACM, New York, NY, USA.
    Google ScholarLocate open access versionFindings
  • [14] Yingqiang Ge, Shuyuan Xu, Shuchang Liu, Shijie Geng, Zuohui Fu, and Yongfeng Zhang. 2019. Maximizing marginal utility per dollar for economic recommendation. In The World Wide Web Conference. 2757–2763.
    Google ScholarFindings
  • [15] F. Maxwell Harper and Joseph A. Konstan. 2015. The MovieLens Datasets: History and Context. ACM Trans. Interact. Intell. Syst. 5, 4, Article 19 (Dec. 2015), 19 pages. https://doi.org/10.1145/2827872
    Locate open access versionFindings
  • [16] Jerry Hausman and Daniel McFadden. 1984. Specification tests for the multinomial logit model. Econometrica: Journal of the Econometric Society (1984), 1219–1240.
    Google ScholarLocate open access versionFindings
  • [17] Ruining He and Julian McAuley. 20Ups and downs: Modeling the visual evolution of fashion trends with one-class collaborative filtering. In proceedings of the 25th international conference on World Wide Web. 507–517.
    Google ScholarLocate open access versionFindings
  • [18] Ruining He and Julian McAuley. 2016. VBPR: Visual Bayesian Personalized Ranking from Implicit Feedback. In AAAI.
    Google ScholarFindings
  • [19] Xiangnan He, Lizi Liao, Hanwang Zhang, Liqiang Nie, Xia Hu, and Tat-Seng Chua. 2017. Neural Collaborative Filtering. In WWW. 173–182.
    Google ScholarLocate open access versionFindings
  • [20] David Hillson and Ruth Murray-Webster. 2017. Understanding and managing risk attitude. Routledge.
    Google ScholarFindings
  • [21] Daniel Kahneman. 1979. Prospect theory: An analysis of decisions under risk. Econometrica 47 (1979), 278.
    Google ScholarLocate open access versionFindings
  • [22] Wang-Cheng Kang and Julian McAuley. 2018. Self-attentive sequential recommendation. In 2018 IEEE International Conference on Data Mining (ICDM). IEEE, 197–206.
    Google ScholarLocate open access versionFindings
  • [23] Joseph A Konstan, Bradley N Miller, David Maltz, Jonathan L Herlocker, Lee R Gordon, and John Riedl. 1997. GroupLens: applying collaborative filtering to Usenet news. Commun. ACM 40, 3 (1997), 77–87.
    Google ScholarLocate open access versionFindings
  • [24] Yehuda Koren. 2008. Factorization meets the neighborhood: a multifaceted collaborative filtering model. In Proceedings of the 14th ACM SIGKDD international conference on Knowledge discovery and data mining. ACM, 426–434.
    Google ScholarLocate open access versionFindings
  • [25] Yehuda Koren, Robert Bell, and Chris Volinsky. 2009. Matrix factorization techniques for recommender systems. Computer 8 (2009), 30–37.
    Google ScholarLocate open access versionFindings
  • [26] Georg Lackermair, Daniel Kailer, and Kenan Kanmaz. 2013. Importance of online product reviews from a consumer’s perspective. Advances in economics and business 1, 1 (2013), 1–5.
    Google ScholarLocate open access versionFindings
  • [27] Daniel D Lee and H Sebastian Seung. 2001. Algorithms for non-negative matrix factorization. In Advances in neural information processing systems. 556–562.
    Google ScholarLocate open access versionFindings
  • [28] Juha Leino and Kari-Jouko Räihä. 2007. Case Amazon: Ratings and Reviews As Part of Recommendations. In Proceedings of the 2007 ACM Conference on Recommender Systems (RecSys ’07). ACM, New York, NY, USA, 137–140. https://doi.org/10.1145/1297231.1297255
    Locate open access versionFindings
  • [29] Greg Linden, Brent Smith, and Jeremy York. 2003. Amazon.Com Recommendations: Item-to-Item Collaborative Filtering. IEEE Internet Computing 7, 1 (Jan. 2003), 76–80. https://doi.org/10.1109/MIC.2003.1167344
    Locate open access versionFindings
  • [30] Robert E. Marks. 2015. Searching for Agents’ Best Risk Profiles. In Proceedings of the 18th Asia Pacific Symposium on Intelligent and Evolutionary Systems, Volume 1, Hisashi Handa, Hisao Ishibuchi, Yew-Soon Ong, and Kay Chen Tan (Eds.). Springer International Publishing, Cham, 297–309.
    Google ScholarLocate open access versionFindings
  • [31] Julian McAuley, Christopher Targett, Qinfeng Shi, and Anton van den Hengel. 2015. Image-based recommendations on styles and substitutes. In SIGIR. ACM.
    Google ScholarLocate open access versionFindings
  • [32] Andriy Mnih and Ruslan R Salakhutdinov. 2008. Probabilistic matrix factorization. In Advances in neural information processing systems. 1257–1264.
    Google ScholarFindings
  • [33] Zan Mo, Yan-Fei Li, and Peng Fan. 2015. Effect of online reviews on consumer purchase behavior. Journal of Service Science and Management 8, 03 (2015), 419.
    Google ScholarLocate open access versionFindings
  • [34] Shameem Ahamed Puthiya Parambath, Nishant Vijayakumar, and Sanjay Chawla. 2019. Risk Aware Ranking for Top-k Recommendations. CoRR abs/1904.05325 (2019). arXiv:1904.05325 http://arxiv.org/abs/1904.05325
    Findings
  • [35] Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, et al. 2019. PyTorch: An imperative style, high-performance deep learning library. In Advances in Neural Information Processing Systems. 8024–8035.
    Google ScholarLocate open access versionFindings
  • [36] Bibek Paudel, Sandro Luck, and Abraham Bernstein. 2018. Loss Aversion in Recommender Systems: Utilizing Negative User Preference to Improve Recommendation Quality. CoRR abs/1812.11422 (2018). arXiv:1812.11422 http://arxiv.org/abs/1812.11422
    Findings
  • [37] Steffen Rendle, Christoph Freudenthaler, Zeno Gantner, and Lars Schmidt-Thieme. 2009. BPR: Bayesian personalized ranking from implicit feedback. In Proceedings of the twenty-fifth conference on uncertainty in artificial intelligence. AUAI Press, 452–461.
    Google ScholarLocate open access versionFindings
  • [38] Paul Resnick, Neophytos Iacovou, Mitesh Suchak, Peter Bergstrom, and John Riedl. 1994. GroupLens: an open architecture for collaborative filtering of netnews. In Proceedings of the 1994 ACM conference on Computer supported cooperative work. ACM, 175–186.
    Google ScholarLocate open access versionFindings
  • [39] Francesco Ricci, Lior Rokach, and Bracha Shapira. 2011. Introduction to recommender systems handbook. In Recommender systems handbook. Springer, 1–35.
    Google ScholarLocate open access versionFindings
  • [40] Badrul Sarwar, George Karypis, Joseph Konstan, and John Riedl. 2001. Item-based collaborative filtering recommendation algorithms. In Proceedings of the 10th international conference on World Wide Web. ACM, 285–295.
    Google ScholarLocate open access versionFindings
  • [41] Leonard J Savage. 1972. The foundations of statistics. Courier Corporation.
    Google ScholarFindings
  • [42] Jiliang Tang, Huiji Gao, Xia Hu, and Huan Liu. 2013. Exploiting Homophily Effect for Trust Prediction. In WSDM.
    Google ScholarFindings
  • [43] Jiliang Tang, Huiji Gao, Huan Liu, and Atish Das Sarma. 2012. eTrust: Understanding Trust Evolution in an Online World. In KDD.
    Google ScholarFindings
  • [44] Amos Tversky and Daniel Kahneman. 1992. Advances in prospect theory: Cumulative representation of uncertainty. Journal of Risk and uncertainty 5, 4 (1992), 297–323.
    Google ScholarLocate open access versionFindings
  • [45] Lev Virine and Michael Trumper. 2016. ProjectThink: why good managers make poor project choices. Routledge.
    Google ScholarFindings
  • [46] John Von Neumann and Oskar Morgenstern. 2007. Theory of games and economic behavior (commemorative edition). Princeton university press.
    Google ScholarFindings
  • [47] Jian Wang and Yi Zhang. 2011. Utilizing marginal net utility for recommendation in e-commerce. In Proceedings of the 34th international ACM SIGIR conference on Research and development in Information Retrieval. ACM, 1003–1012.
    Google ScholarLocate open access versionFindings
  • [48] Menahem E. Yaari. 1965. Convexity in the Theory of Choice Under Risk. The Quarterly Journal of Economics 79, 2 (1965), 278–290. http://www.jstor.org/stable/1880632
    Locate open access versionFindings
  • [49] Yongfeng Zhang, Qingyao Ai, Xu Chen, and W Bruce Croft. 2017. Joint representation learning for top-n recommendation with heterogeneous information sources. In Proceedings of the 2017 ACM on Conference on Information and Knowledge Management. ACM, 1449–1458.
    Google ScholarLocate open access versionFindings
  • [50] Yongfeng Zhang, Min Zhang, Yi Zhang, Guokun Lai, Yiqun Liu, Honghui Zhang, and Shaoping Ma. 2015. Daily-aware personalized recommendation based on feature-level time series analysis. In Proceedings of the 24th international conference on world wide web. 1373–1383.
    Google ScholarLocate open access versionFindings
  • [51] Yongfeng Zhang, Qi Zhao, Yi Zhang, Daniel Friedman, Min Zhang, Yiqun Liu, and Shaoping Ma. 2016. Economic recommendation with surplus maximization. In Proceedings of the 25th International Conference on World Wide Web. 73–83.
    Google ScholarLocate open access versionFindings
  • [52] Qi Zhao, Yi Zhang, Daniel Friedman, and Fangfang Tan. 2015. E-commerce recommendation with personalized promotion. In Proceedings of the 9th ACM Conference on Recommender Systems. ACM, 219–226.
    Google ScholarLocate open access versionFindings
  • [53] Qi Zhao, Yongfeng Zhang, Yi Zhang, and Daniel Friedman. 2017. Multi-product utility maximization for economic recommendation. In Proceedings of the Tenth ACM International Conference on Web Search and Data Mining. ACM, 435–443.
    Google ScholarLocate open access versionFindings
Full Text
Your rating :
0

 

Tags
Comments