Point of interest recommendation method based on graph neural network and context information perception

Abstract

The application discloses a kind of point of interest recommendation method based on graph neural network and context information perception, first by embedding layer generation point of interest ID, point of interest category and point of interest score embedding representation, merge into initial point of interest embedding vector;4 pictures are constructed;A batch of users and points of interest are randomly sampled as the input of model;Convolution is carried out to the constructed graph and outputs five vectors;The user and point of interest representation is weighted to obtain the final user representation and point of interest representation;Inner product operation generates the final prediction value;Each is a positive sample, and negative sample is sampled from data set, and is trained and parameter updated;Repeat several times, obtain the last recommendation model and carry out preference prediction to point of interest, and return N The highest score point of interest is recommended as a list to user.The application extracts high-order nonlinear interaction between nodes through graph structure, and models jointly temporal and spatial influence and social influence, can effectively alleviate data sparsity, and improve model recommendation effect.

Images