基于增量矩阵分解的推荐系统研究
Research on Recommendation System Based on Incremental Matrix Decomposition
DOI: 10.12677/CSA.2021.113061, PDF,   
作者: 郭思宇, 方 睿：成都信息工程大学，计算机学院，四川 成都
关键词: 矩阵分解；推荐系统；评分预测；增量更新；Matrix Decomposition； Recommendation System； Score Prediction； Incremental Update
摘要: 矩阵分解(Matrix Factorization, MF)属于推荐系统中的一种应用广泛且较为经典的算法。它是基于用户行为的推荐算法，被广泛知道的是矩阵分解算法是一种很好的推荐算法。在Netflix Prize算法竞赛中取得了巨大成就，有很多应用矩阵分解拿到冠军的团队，矩阵分解可以有效地解决电影评分预测问题。传统的推荐模型通常采用离线训练的方法，计算出所有训练数据的预测系数用户。在实践中场景中总有一些新用户在训练集中找不到。但我们有他的历史行为记录。那我们对该用户如何进行预测评分呢？最简单的方法是将新用户的数据与旧数据集相结合，再一次进行矩阵分解操作。但是这样的操作计算成本太高了，对于时间成本的要求过高，是行不通的。对于新用户的电影评分预测如何解决这一问题，经过对比分析实验，发现增量矩阵分解算法就可以很好的帮助我们解决新用户电影评分预测的问题，本文将具体阐述如何通过增量更新的方式，大幅度的减少针对新用户电影评分预测的时间，有效的解决新用户进入系统后数据快速更新的问题。每当有新的数据产生，新的用户产品的时候，系统可以快速的对其进行训练，使得模型能够实时应用，优化用户体验。
Abstract: Matrix Factorization (MF) is a widely used and classic algorithm in recommender system. It is a recommendation algorithm based on user behavior. It is widely known that matrix factorization algorithm is a good recommendation algorithm. In the Netflix prize algorithm competition, it has made great achievements, there are many applications of matrix factorization to win the championship team; matrix factorization can effectively solve the problem of film score prediction. Traditional recommendation models usually use off-line training method to calculate the prediction coef-ficients of all training data. In practice, there are always some new users in the scene that can’t be found in the training set. But we have a record of his historical behavior. How do we rate the user? The simplest method is to combine the new user’s data with the old data set and perform matrix decomposition again. But the cost of calculation is too high, and the time requirement is too high. For the new user’s movie score prediction how to solve this problem, after comparative analysis and experiments, it is found that the incremental matrix decomposition algorithm can help us solve the problem of new user’s movie score prediction. This paper will specifically elaborate how to reduce the time of new user’s movie score prediction by means of incremental update, and effectively solve the problem of new user’s entering the system, the problem of data fast updating after the unification. Whenever there are new data and new user products, the system can train them quickly, so that the model can be applied in real time and the user experience can be optimized.
文章引用：郭思宇, 方睿. 基于增量矩阵分解的推荐系统研究[J]. 计算机科学与应用, 2021, 11(3): 596-603. https://doi.org/10.12677/CSA.2021.113061

参考文献

[1] Anyosa, S.C., Vinagre, J. and Jorge, A.M. (2018) Incremental Matrix Co-Factorization for Recommender Systems with Implicit Feedback. Companion Proceedings of the Web Conference, Lyon, 23-27 April 2018, 1413-1418. [Google Scholar] [CrossRef] 
[2] Ke, G., Meng, Q., Finley, T., et al. (2017) LightGBM: A Highly Efficient Gradient Boosting Decision Tree. NIPS 2017, Long Beach, 4-9 December 2017, 3146-3154.
[3] Covington, P., Adams, J. and Sargin, E. (2016) Deep Neural Networks for YouTube Recommendations. Proceedings of the 10th ACM Conference on Recommender Systems, Boston, 15-19 September 2016, 191-198. [Google Scholar] [CrossRef] 
[4] Pennington, J., Socher, R. and Manning, C. (2014) Glove: Global Vectors for Word Representation. Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), Doha, October 2014, 1532-1543. [Google Scholar] [CrossRef] 
[5] Mairal, J., Bach, F., Ponce, J., et al. (2010) Online Learning for Matrix Factorization and Sparse Coding. Journal of Machine Learning Research, 11, 19-60.
[6] Pilászy, I., Zibriczky, D. and Tikk, D. (2010) Fast Als-Based Matrix Factorization for Explicit and Implicit Feedback Datasets. Proceedings of the Fourth ACM Conference on Recommender Systems, Barcelona, 26-30 September 2010, 71-78. [Google Scholar] [CrossRef] 
[7] Johnson, C.C. (2014) Logistic Matrix Factorization for Implicit Feedback Data. NIPS 2014, Montreal, 13 December 2014, 27.
[8] He, X., Zhang, H., Kan, M.Y., et al. (2016) Fast Matrix Factorization for Online Recommendation with Implicit Feedback. SIGIR’16, Pisa, 17-21 July 2016, 549-558. [Google Scholar] [CrossRef] 
[9] Zhang, S., Yao, L., Sun, A., et al. (2019) Deep Learning Based Recommender System: A Survey and New Perspectives. ACM Computing Surveys (CSUR), 52, 5. [Google Scholar] [CrossRef] 
[10] Mikolov, T., Chen, K., Corrado, G.S. and Dean, J. (2013) Efficient Estima-tion of Word Representations in Vector Space. Proceedings of Workshop at ICLR, Scottsdale, 2-4 May 2013.
[11] Mikolov, T., Sutskever, I., Chen, K., et al. (2013) Distributed Representations of Words and Phrases and Their Compositionality. NIPS 2013, Lake Tahoe, 5-8 December 2013, 3111-3119.
[12] Bell, R. and Koren, Y. (2007) Scalable Collaborative Filtering with Jointly Derived Neighborhood Interpolation Weights. Seventh IEEE International Conference on Data Mining (ICDM 2007), Omaha, 28-31 October 2007,43-52. [Google Scholar] [CrossRef] 
[13] Zhou, Y., et al. (2008) Large-Scale Parallel Collaborative Filtering for the Netflix Prize. In: Fleischer R. and Xu J., Eds, Algorithmic Aspects in Information and Management, Springer, Berlin, 337-348. [Google Scholar] [CrossRef] 
[14] Chen, T. and Guestrin, C. (2016) Xgboost: A Scalable Tree Boosting System. SIGKDD 2016, San Francisco, 13-17 August 2016, 785-794. [Google Scholar] [CrossRef] 
[15] Hu, Y.F., Koren, Y. and Volinsky, C. (2008) Collaborative Filter-ing for Implicit Feedback Datasets. 2008 Eighth IEEE International Conference on Data Mining, Pisa, 15-19 December 2008, 263-272.
[16] Goldberg, D., et al. (1992) Using Collaborative Filtering to Weave an Information Tapestry. Com-munications of the ACM, 35, 61-70. [Google Scholar] [CrossRef] 
[17] Sarwar, B.M., et al. (2000) Ap-plication of Dimensionality Reduction in Recommender System—A Case Study. In: Proceedings KDD Workshop on Web Mining for e-Commerce: Challenges and Opportunities (WebKDD), ACM Press, New York. [Google Scholar] [CrossRef] 
[18] Funk, S. (2006) Netflix Update: Try This at Home. http://sifter.org/~simon/journal/20061211.html
[19] Koren, Y. (2008) Factorization Meets the Neighborhood: A Multi-faceted Collaborative Filtering Model. In: Proceedings 14th ACM SIGKDD Knowledge Discovery and Data Mining, ACM Press, New York, 426-434. [Google Scholar] [CrossRef] 
[20] Paterek, A. (2007) Improving Regularized Singular Value Decom-position for Collaborative Filtering. In: Proceedings KDD Cup and Workshop, ACM Press, New York, 39-42.
[21] Takács, G., et al. (2007) Major Components of the Gravity Recommendation System. SIGKDD Explora-tions, 9, 80-84. [Google Scholar] [CrossRef] 
[22] Salakhutdinov, R. and Mnih, A. (2008) Probabilistic Matrix Fac-torization. In: Proceedings Advances in Neural Information Processing Systems 20 (NIPS 07), ACM Press, New York, 1257-1264.