A friend recommendation method based on representation learning of global attention mechanism

Abstract

A friend recommendation method for social network service platform based on representation learning of global attention mechanism, establish network model, use LINE method to transform social network into structure embedding vector in Euclidean space, based on attribute vector of adjacent nodes, use CNN network to calculate The feature matrix, considering the attention weight between adjacent nodes, obtains the correlation matrix of two nodes, and then performs row and column average pooling and softmax function on the correlation matrix to obtain the information embedding vector of node pairs; embedding the network structure Add the information embedding vector in proportion to calculate the embedding vector of the node. When the objective function reaches the set target value, the embedding vector of all nodes is obtained, and the embedding vector of the node is used to calculate the Pearson coefficient. Users with high Pearson coefficient are was recommended as a friend. The present invention introduces a global attention mechanism to consider the information among social network users, so that the accuracy of the representation vector is improved, and the performance of the friend recommendation algorithm is improved.

Description

technical field [0001] The invention relates to the field of social network recommendation, in particular to a method for recommending friends on a social network service platform based on representation learning of a global attention mechanism. Background technique [0002] Due to the rapid development of computer technology, big data already exists in all aspects of people's life, work, study and so on. An important form of expression of big data - the network has shown an important role. Networks, such as social networks, word coexistence networks, and communication networks, widely exist in various real-world applications. A social network is like a forum: Friends share each other's knowledge, experiences and insights, providing users with a constant flow of diverse information. Friends can conduct related discussions around a certain topic of interest, and at the same time, they can follow people with the same interests. Users can obtain the knowledge they want from ...

AI Technical Summary

Problems solved by technology

[0005] In order to overcome the incomplete acquisition of network user information and low accuracy of embedded vectors in existing friend recommendation methods, and to improve the performance of existing recommendation algorithms, the present invention proposes a global attention mechanism-based approach with high accuracy. The friend recommendation method of social network service platform based on network representation learning considers the influence of attention weight among users in the social network, and realizes the embedding of network structure and user information, so a data based on global attention mechanism is proposed Representation method, which effectively improves the performance of friend recommendation method

Images