Research And Application Of Point Of Interest Recommendation Based On Union Neural Network And Mobile Context

In the past few years,with the wake of developments in artificial intelligence,mobile application has been innovated by leaps and bounds,which leads the prevalence of location-based mobile services.These mobile services help users navigate routes and explore nearby locations.Therefore,they produce large-scale location-based data that embeds abundant hints of user preferences on different locations and provides us with opportunities to predict the locations on future.This facilitates outdoor activities of users by making a list of personal recommendations and bring more commercial benefits to the third-parties.Point of interest(POI)recommendation,as one of the basic and popular mobile services,also plays an important role in the field of recommendation system.POI recommendation aims to recommend new satisfactory locations to users according to their historical records.Existing classical POI recommendation models no matter based on matrix factorization or collaborative filtering face a huge challenge where they cannot capture the preference of user deeply and effectively.Meanwhile,they have the problem of data sparsity.Neural network that is able to mine the potential relevance between user and location can solve the challenges mentioned above.Meanwhile,it is necessary to make full use of mobile context in POI recommendation.The context reveals the tendency of user on visiting location explicitly,which makes recommendation more reasonable and explainable.In conclusion,we propose two POI recommendation algorithms based on union neural network and mobile context respectively and then present a recommendation model that combines the two algorithms.The main work of this thesis is as follows:(1)This thesis introduces the research background,research status,relevant theories of POI recommendation and analyzes the shortcoming of existing algorithms.(2)This thesis presents a deep POI algorithm based on neural network.First,we construct two different multi-layer neural networks to reduce the dimensions of user vector and location vector respectively.Then,we catenate the two new vectors to obtain the potential relevance between user and location.Next,we build a deep matrix factorization to change the traditional way.Finally,the results of relevance computing and matrix factorization are sent to the union neural network for the final recommendation prediction.The effectiveness of the algorithm is verified on real data sets.(3)This thesis presents a POI algorithm based on mobile context.First,we cluster locations into a few regions and figure out the preference of user on visiting regions,which also considers the popularity of locations.Then,since the distance and category of a location has obvious influence on decision of visiting it,we model the preferences of distance and category respectively.Finally,we integrate the distance and category,which is under the prepositive probability of region.The effectiveness of the algorithm is verified on real data sets.(4)This thesis proposes three different strategies to combine the two algorithms based on union neural network and mobile context.The first and second strategies follow the recall-sort rule.The third strategy mixes the two algorithms with weights.