Research On Point Of Interest Recommendation Based On Deep Learning

The popularization of portable mobile Internet devices such as smart phones affects people's work and life in all aspects.For tourism,people use the point of interest data on the Internet to select preferred locations for travel visits in advance.As information is easily obtained,people's work efficiency and quality of life are greatly improved.More and more devices have been connected to the Internet in recent years,and more and more Internet applications such as reviews have also been signed in.There has been a tremendous increase in the number of user check-in reviews on the Internet.How can users obtain massive amounts of data? How do users select the data that meets their needs and dig out users' potential hobbies from the data? This is a question worthy of exploration and research by researchers in related fields.The birth of the recommendation system can solve the above problems very well.The ultimate goal of point-of-interest recommendation is to provide personalized point-of-interest recommendation services for different users.Traditional recommendation methods have achieved some results in many fields,and classic recommendation models have been widely used in actual production.Use traditional recommendation models for point-of-interest recommendation.Most models rely on manual feature screening,and can only obtain feature information from a shallow layer.The extracted feature information is limited,which will cause the omission of important feature information and affecting the recommendation system's results.In response to such problems,this article uses deep learning technology to solve some of the problems of traditional point of interest recommendation,to improve the point of interest recommendation.Specifically,it can be classified into the following aspects:1.The traditional collaborative filtering recommendation method extracts the characteristics of users and items of interest from a single perspective.This will greatly limit the expressiveness of the feature vector.This paper proposes a method of point-of-interest recommendation based on multi-space interactive collaborative filtering to solve this problem.The model represents the interaction between users and points of interest from different angles,and then maps the user vector and the point of interest item vector to different spaces.Do user-project interactions from different angles in different spaces,which will get richer characteristic information.Then use the attention mechanism to fuse user and item feature vectors in different spaces.Finally,it is sent to the multi-layer perceptron,and the deep learning technology is used to further dig the interactive information between the user and the point of interest from the deep level to obtain the recommendation result.The experiment was compared with some similar methods and models with adjusted parameters to prove the effectiveness of this method.2.Compared with other recommendation domains,point-of-interest recommendation is susceptible to geographic location,and user-item interaction is usually implicit feedback.This paper proposes a point-of-interest recommendation method that combines historical preferences and neighborhood information.This method uses the characteristics of points of interest recommendation to establish a recommendation model.The main frame of the model is an automatic coding machine.The model integrates neighborhood information such as historical preference information,geographic location information,and "friend" information,so that points of interest that match the user's preferences and geographic location are set with a higher weight value in the final prediction scoring stage,so that users are more satisfied Recommended list.Through theoretical analysis and experimental analysis,it is proved that this method can improve the recommendation effect to a certain extent.