A movie recommendation method that fuses tags and knowledge graphs

Abstract

The invention discloses a movie recommendation method integrating tags and knowledge graphs, belonging to the technical field of recommendation systems. The method includes step 1: mapping a user's tag for a movie to a low-dimensional vector space, and constructing a user-movie tag embedding matrix T; After the first fully connected layer, the remaining features of the user are sent to the first multi-layer perceptron; the user feature matrix U is obtained; step 2: the entities in the movie knowledge map are mapped to the low-dimensional vector space, and the movie-entity embedding matrix E is obtained r Then send it to the KGCNN model to get the movie-entity feature matrix. Step 3: Input and T into the mixed attention model to calculate the mixed attention weight; send the KGCNN result together with the rest of the feature matrix of the movie into the second multi-layer perceptron; combine the mixed attention The movie feature matrix I is obtained by the force weight; Step 4: Send U and I to the second fully connected layer, and calculate the user's rating y' for the movie. The recommendation method improves the recommendation personalization and accuracy, and can be used in various fields.

Description

technical field [0001] The invention belongs to the technical field of recommendation systems, and in particular relates to a movie recommendation method integrating tags and knowledge maps. Background technique [0002] Traditional recommendation methods mainly include content-based recommendation methods and collaborative filtering-based recommendation methods. The former uses the user's historical information to recommend similar items for users; the latter is subdivided into user-based collaborative filtering and item-based collaborative filtering. filter. They all require a large number of user and item interactions for recommendation, which will cause cold-start problems for sparse data and new users, and it is difficult to give reasonable recommendation explanations. In addition, only using the simple information contained in the dataset cannot reflect the user's preference well, and it is difficult to construct accurate user features and item features, so the recomm...

AI Technical Summary

Problems solved by technology

They all require a large number of user and item interaction behaviors to make recommendations, which will cause cold start problems for sparse data and new users, and it is difficult to give a reasonable recommendation explanation

In addition, only using the simple information contained in the data set cannot reflect the user's preference well, and it is difficult to construct accurate user characteristics and item characteristics, so the recommendation performance is poor

Images