Movie recommendation method based on weighted heterogeneous information networks

Abstract

The invention relates to a movie recommendation method based on weighted heterogeneous information networks. The movie recommendation method includes the steps: (1) calculating to obtain similarities among users based on different meta paths by the aid of semantic information and edge attribute information in the heterogeneous information networks; (2) respectively applying the similarities calculated based on the different meta paths to a collaborative filtering recommendation algorithm based on the users to obtain a user scoring value based on each similarity; (3) distributing different weights for each user scoring value by the aid of a supervised learning algorithm to fuse a user final scoring value completely considering information of various meta paths. Experimental results on a classical data set extending Movie Lens show that the accuracy of the method is remarkably improved as compared with a traditional algorithm.

Description

technical field [0001] The invention relates to a movie recommendation method based on a weighted heterogeneous information network, in particular to a new collaborative filtering recommendation method based on meta-path calculation between users in a weighted heterogeneous information network, which belongs to the technology of data mining and machine learning field. Background technique [0002] Collaborative Filtering (Collaborative Filtering) is the fastest growing and most widely used algorithm in the history of recommendation system development. Its basic idea is that similar users select similar products. According to the K neighbors most similar to the target user, the target Item ratings for recommendation. Among them, neighbor users are obtained based on the similarity between users, and the most commonly used similarity measurement methods are Pearson correlation coefficient and Cosine similarity. However, the traditional similarity measurement method only consi...

AI Technical Summary

Problems solved by technology

This patent has the following defects: (1) Essentially, the information used in this invention is still only user-item matrix data, or only uses the very sparse rating information of users on items, and does not introduce other attribute information or social relationship information; (2) When calculating the similarity between users or calculating the similarity between items to cluster items, the invention still uses the classic similarity measurement methods in collaborative filtering technology: cosine similarity, pearson correlation coefficient, modified cosine similarity degree, etc., and did not propose a new similarity measurement method; (3) the invention did not consider the polarization impact of users on different item ratings in a fine-grained manner

This patent has the following defects: (1) This patent introduces the social information of users, that is, the friendship between users, etc., but the information considered is not comprehensive, and it does not consider other information between users and items except ratings. Rich attribute information of the item

(3) The patent does not take fine-grained consideration of the polarization impact of users on different item ratings

Images