User Interest Mining via Multivariate Probit Regression

doi:10.3969/j.issn.1006-2475.2017.06.005

Abstract

Abstract: Mining user interest is a fundamental technique in many fields such as recommender system, personalized retrieval and online advertising. The historical actions of a user through the Web or in real word reflect his interests. However, if the user uses the Web at his first time, it is difficult to learn his interests because only few historical actions are known. To deal with this issue, we propose a variant of multivariate Probit model to learn the prior of the user’s interests based on user’s attributes. The attributes may include sign up location, sign up time and some other registration information. The posterior distribution of the model is simulated by a Markov chain Monte Carlo (MCMC) method to estimate the expectation of user’s interest. To evaluate our algorithm, we collect the information of movie stars and their movies as the evaluation dataset. The experiment on this dataset demonstrates that the prior information can effectively improve the performance on cold start users.

Key words: user interest, multivariate Probit, MCMC method

CLC Number:

TP391

SHEN Qiang-hua. User Interest Mining via Multivariate Probit Regression[J]. Computer and Modernization, doi: 10.3969/j.issn.1006-2475.2017.06.005.

References

[1] Ahmed A, Low Yucheng, Aly M, et al. Scalable distributed inference of dynamic user interests for behavioral targeting[C]// Proceedings of the 17th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. 2011:114-122.

[2] Aly M, Hatch A, Josifovski V, et al. Web-scale user modeling for targeting[C]// Proceedings of the 21st International Conference on World Wide Web. 2012:3-12.

[3] Vu T, Perez V. Interest mining from user tweets[C]// Proceedings of the 22nd ACM International Conference on Information & Knowledge Management. 2013:1869-1872.

[4] Adomavicius G, Tuzhilin A. Toward the next generation of recommender systems: A survey of the state-of-the-art and possible extensions[J]. IEEE Transactions on Knowledge and Data Engineering, 2005,17(6):734-749.

[5] 于洪,李俊华. 一种解决新项目冷启动问题的推荐算法[J]. 软件学报, 2015,26(6):1395-1408.

[6] 孙冬婷,何涛,张福海. 推荐系统中的冷启动问题研究综述[J]. 计算机与现代化, 2012(5):59-63.

[7] 何旭. 基于粒关联规则的冷启动推荐算法[D]. 漳州:闽南师范大学, 2015.

[8] 巫文佳,何旭. 基于粒关联规则的冷启动推荐方法[J]. 计算机科学, 2014,41(3):71-75.

[9] Peng Furong, Lu Jianfeng, Wang Yongli, et al. N-dimensional Markov random field prior for cold-start recommendation[J]. Neurocomputing, 2016,191:187-199.

[10] Feng He, Qian Xueming. Recommendation via user’s personality and social contextual[C]// Proceedings of the 22nd ACM International Conference on Information & Knowledge Management. 2013:1521-1524.

[11] Adomavicius G, Tuzhilin A. Context-aware recommender systems[M]// Ricci F, Rokach L, Shapira B, et al. Recommender Systems Handbook. Springer, 2011:217-253.

[12] Chib S, Greenberg E. Analysis of multivariate Probit models[J]. Biometrika, 1998,85(2):347-361.

[13] Geman S, Geman D. Stochastic relaxation, Gibbs distributions, and the Bayesian restoration of images[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1984,6(6):721-741.

[14] Gilks W R. Markov Chain Monte Carlo[M]. John Wiley & Sons, 2005.

[15] Cowles M K, Carlin B P. Markov chain Monte Carlo convergence diagnostics: A comparative review[J]. Journal of the American Statistical Association, 1996,91(434):883-904.

[16] Casella G, George E I. Explaining the Gibbs sampler[J]. The American Statistician, 1992,46(3):167-174.

[17] Hastings W K. Monte Carlo sampling methods using Markov chains and their applications[J]. Biometrika, 1970,57(1):97-109.