A novel recommendation scheme with multifactorial weighted matrix decomposition strategies via forgetting rule. (May 2021)

• Record Type: Journal Article
• Authors: Chen, Jianrui; Lu, Yanqing; Shang, Fanhua; Zhu, Tingting
• Is Part Of: Engineering applications of artificial intelligence. Volume 101(2021)
• Publication Date: 2021-05
• Subjects: Matrix decomposition -- Forgetting function -- User interest -- Multifactorial weighted matrix -- Recommendation system; Engineering -- Data processing -- Periodicals; Artificial intelligence -- Periodicals; Expert systems (Computer science) -- Periodicals; Ingénierie -- Informatique -- Périodiques; Intelligence artificielle -- Périodiques; Systèmes experts (Informatique) -- Périodiques; Artificial intelligence; Engineering -- Data processing; Expert systems (Computer science); Periodicals; 620.00285
• DOI: 10.1016/j.engappai.2021.104191
• Languages: English
• ISSNs: 0952-1976

Abstract

Abstract: Recommendation systems play more and more important roles in many fields, such as movie recommendation, book recommendation, music recommendation. Sparse data and large-scale data result in low recommendation efficiency and a cold start problem. Recently, matrix decomposition plays an active role in recommendation with its high recommendation efficiency and fast recommendation speed. But in most matrix completion mechanisms, time influence and users features are not fully utilized. In this paper, a continuous function is constructed to grasp the time-varying rule of users' interests, and the scores of different scoring time are processed by this function. Thus, the processed scores contain not only the user interests but also their memory rule. Moreover, user interests may change slowly over time, so we keep the original rating matrix in the recommendation process. Finally, a new matrix decomposition optimization model is constructed by considering the score matrix that combines time information and the original score matrix. Compared with the recent matrix decomposition recommendation algorithms, the effectiveness of our proposed algorithm is verified based on the recommended evaluation metrics and several datasets.