Research On Point Of Interest Recommendation Algorithm In Check-in Data And Similarity Integration

In recent years,with the rapid development of location-based information services,such as tourist attractions,catering and other POI numbers based on location-based services continue to accumulate,showing an exponential growth,POI recommendation emerges as the times' requirement.POI recommendation help users find POIs that meet users' preferences from a large amount of POI data.POI recommendation research focuses on using user historical check-in data to effectively model user preferences and predict the next visit location.Compared with traditional recommendation,POI recommendation faces more severe problems.First,the user-POI check-in matrix has high sparseness,which only uses check-in matrix to model user preferences and reduces recommendation performance.Second,the check-in data of POI contains many different types of context information,such as comments,tags and other text information.Although the comment information can effectively alleviate the sparse check-in matrix,most of the existing POI recommendation algorithms do not distinguish the comment information,resulting in the lack of accuracy of extracted feature information.Aiming at the above problems,this paper proposes an improved algorithm,and the specific work is as follows.(1)In allusion of the high sparseness of user-interest check-in matrix,this paper proposes a method that combines user's relevant text information with check-in matrix to calculate user's similarity matrix.In recommendation period,a dynamic prediction method is proposed to dynamically fill the missing data and further alleviate sparse data and improve the recommended quality.For text information,this paper uses the latent LDA topic model to mine the user's interest topics,and calculates the similarity of user's topic feature vectors,and then integrates the check-in matrix to measure similarity.The experimental results on the real data set show that the proposed similarity fusion and dynamic prediction based recommend algorithm can effectively solves the problem of data sparseness and cold-start.The recommend performance is superior to traditional recommendation algorithms.(2)Aiming at the existing POI recommendation algorithms lack of dividing the importance of comment information,this paper proposes a convolutional neural network POI recommendation model based on check-in data.Firstly,an expert model is established from user's professionalism and trust by using check-in data;secondly,the expert model to separate the comments of different users;thirdly,two parallel convolution neural networks are used to mine the deep features of the separated comments documents,and a shared layer is established to merge the two convolution neural networks;Finally,The CNN model integrates into probabilistic matrix factorization to predict score.We implement experiments on Yelp dataset,and the experimental results show that the proposed algorithm achieve superior recommendation effect compared other advanced algorithms.