Interest Point Recommendation Algorithm Based On Deep Neural Network And Spatio-Temporal Perception

With the rapid progress of location-based social networks(Location Based Social Network,LBSN),personalized point-of-interest recommendations are also becoming popular,which can help users discover locations that may be of interest.However,since the point of interest recommendation is an implicit feedback,it is difficult to interact between the user and the point of interest.If the user's check-in behavior is not "differentiated",it will lead to inaccurate mining of user preferences.The number of check-ins accounts for only a small proportion of the entire social network of locations,which makes check-in data suffer from high sparseness.Therefore,how to effectively improve the performance of points of interest recommendation has become an urgent problem to be solved.In view of the above limitations and according to the characteristics of point-of-interest recommendation,a detailed and in-depth study on point-of-interest recommendation is carried out.First of all,in view of the existing point of interest recommendation system treats the user's check-in behavior equally,and cannot accurately reflect user preferences.This article analyzes the combination of deep neural network and recommendation system and uses the deep neural network structure based on autoencoder to learn users.Check-in preferences.The algorithm obtains the hidden vector representation of the user through the hidden layer of the encoder;then,when learning the hidden vector representation,the influence of geographic location and social relationship is added to further improve the performance;and the attention mechanism is added in the encoding process of the encoder.The multi-dimensional attention mechanism adaptively distinguishes user preferences from the user's historical check-in records,and finally obtains the check-in probability of users who have not visited points of interest,and sorts them in descending order,thereby determining the real points of interest recommendation list.Secondly,in view of the inadequate mining of relevant attributes in user sign-in information by existing methods,which leads to the problem of data sparseness,this paper uses the fusion of multiple sign-in features in location social networks to recommend points of interest.The algorithm uses contextual features such as check-in time,geographic location,and point of interest categories in the location social network check-in data set to find the potential relationship between them and calculate the correlation between them,and then obtain the check-in preference and geographic location in the time dimension.The check-in preference in the location dimension and the check-in preference in the category dimension integrate the check-in time,geographic location,and point of interest category,and integrate the check-in preferences of the three dimensions to obtain the end-user check-in preference to determine the point of interest recommendation list.Finally,for the two points of interest recommendation algorithms proposed in this article,we have conducted extensive experimental comparisons and analyses on Foursquare and Gowalla datasets by comparing the algorithm in this article with some existing algorithms,which proves the interest in this article.Point recommendation algorithm can improve recommendation performance.