Recommendation method based on collaborative filtering

Abstract

The invention discloses a recommendation method based on collaborative filtering. The recommendation method comprises: calculating the similarity between any two projects; constructing a project similarity matrix according to the similarity between the any two projects; obtaining a nearest neighbor set of each project according to the project similarity matrix; calculating a predicated score of each project predicated by a user according to the nearest neighbor set; calculating the similarity of any two users according to the predicated score of each project and the confidence degree of the predicated score; calculating a nearest neighbor set of each user according to the similarity of the any two users; according to the nearest neighbor set of each user, scoring projects, which are not scored, of each user, and generating a recommendation set of the user according to a scoring result. By utilizing the reliability of nearest neighbor collaborative filtering recommendation, the reliability of complementary data is improved on the basis of the complementary data when data sparsity is not enough, so that the collaborative filtering can play a better role.

Description

technical field [0001] The invention relates to the technical field of personalized recommendation, in particular to a recommendation method based on collaborative filtering. Background technique [0002] With the increasing number of users and products in the e-commerce system, user rating data is extremely sparse in the entire product space. When the data is sparse, there are few common scoring items, and the calculation results are often inaccurate or even impossible to calculate. Taking the data set commonly used in experiments as an example, the sparsity of movielens is 95.5%, and the sparsity of netfix leads to the problem that the similarity cannot be calculated or calculated inaccurately. In large-scale e-commerce systems, items rated by users generally do not exceed 1% of the total number of items, and items rated by two users are even less. For example, the user rating data received by eachmovie website in 18 months is: 72916 Each user has 2,811,983 evaluation val...

AI Technical Summary

Problems solved by technology

[0006] Cosine similarity: By treating the user's rating as a vector in n-dimensional space, the similarity between two users can be measured by the cosine angle The amount of calculation is small, but the user's scoring scale problem is ignored in the application

Images