POI Recommendation Model Using Multi-Head Attention in Location-Based Social Network Big Data

Abstract

A point of interest (POI) recommendation model using deep learning in location-based social network big data is proposed. Firstly, the features of POI are divided into inherent features composed of attributes such as geographical location and category, and semantic features of relevance composed of spontaneous access by users. Secondly, the inherent attribute features and semantic features of POI are extracted by constraint matrix decomposition and word vector model respectively, and the two hidden vectors are spliced into the feature vectors of POI to solve the problems of data sparsity and cold start. Finally, the multi-head attention is used to obtain the key information of user preferences, and a deep learning recommendation framework is constructed to model the nonlinear interaction between features. Experiments show that when the recommendation list is 10, the precision and recall of the proposed method are 0.118 and 0.135 respectively, which are better than the comparative recommendation method.