A Film Recommendation Algorithm Based On Tensor Decomposition

EasyChair Preprint 1146

Abstract

In today's increasingly complex Internet information, the emergence of a recommendation system provides a list of information for users with unclear goals. The  recommenddation system belongs to the category of information filtering systems. When a large amount of useful information comes together, data mining and prediction algorithms are used to predict the user's possible preferences. Algorithms based on project recommenddation, content recommendation and integration are the main recom-mended methods used by people.

The recommendation algorithm is the core of the recommendation system. The commonly used general recommendation algorithms are content-based collaborative filtering algorithms and hybrid recommendation algorithms. This article uses a new algorithm based on traditional algorithms, that is, a three-dimensional tensor is used to describe the potential correlation model of three entities: user, item, and tag. Based on the use of primary metadata to construct the initial high-order tensor, the high-order singular value decomposition (HOSVD) is applied to the tensor. The dimensionality reduction is performed to correlate the latent semantics among the three types of entities and improve the accuracy of the tag recommendation system.

In order to verify the effectiveness of the algorithm used in this paper, we selected a certain amount of data in the real database to test the algorithm. The test results show that the proposed method has significantly improved the RMS error performance index than the item-based recommendation algorithm

Keyphrases: High–order singular value decomposition, Recommended System, tag recommendation, tensor decomposition

@booklet{EasyChair:1146,
  author    = {Yulong Han and Junfeng Liang and Jingyi Liu and Ruxuan Wang},
  title     = {A Film Recommendation Algorithm Based On Tensor Decomposition},
  howpublished = {EasyChair Preprint 1146},
  year      = {EasyChair, 2019}}