Multi-Dimensional Point-of-Interest Recommendation In Location-Based Social Networks

With the maturity of mobile positioning technology and the popularity of smart devices,traditional social networks have gradually shifted to locationbased social networks,attracting more and more users.At the same time,along with the process of urbanization,a large number of new points of interest(such as shopping malls,restaurants,parks,attractions,etc.)have emerged,and the entertainment life of residents has become more abundant.The interaction between users and interest points has prompted the interest point recommendation issue to become the current research hotspot.The point of interest recommendation refers to the analysis of the historical check-in data of the user in the location social network,recommending the points of interest that the user may be interested in but have not visited.In the research,people find that the check-in data is composed of multi-dimensional information.Different dimensions reflect different aspects of the user's behavioral habits,and the rational use of this information can provide recommendation quality.Therefore,the multidimensional point of interest recommendation becomes a new research direction that can be used for tourism planning and social interaction.This article is devoted to the study of the multi-dimensional interest point recommendation problem,and proposes a new recommendation method by fully tapping the dimensionality information of the constituent social network check-in data.The main research work is as follows:(1)The issue of the influence of the category dimension on the recommendation was not considered in the recommendation process.Digging deeper into the category information in the user's check-in record,the user-interest point check-in matrix is decomposed into categories,and a time decay function is introduced to reflect the change trend of the user category preference,and each time the different types of check-in are assigned different weights,and then combined.The popular dimension predicts the user's subjective assessment of interest points.(2)For the problem of not considering the influence of different domains in the social dimension in the process of recommendation,build a category social network matrix in combination with social networks and domain information,introduce a PageRank algorithm to calculate the degree of authority of social friends on different domains,and then combine social The collaborative filtering algorithm predicts the user's objective evaluation of interest points.Finally,the subjective and objective scores of the linear combination forecast are designed and a multi-dimensional joint recommendation algorithm is designed to effectively integrate spatial,temporal,social,category,and popular dimension information.Foursquare real dataset is chosen to verify the algorithm in this thesis.The algorithm proposed in this paper is compared with several currently representative algorithms.The evaluation index selects precision and recall.The result proves that compared with the current advanced recommendation algorithm,the proposed MJRA algorithm recommendation results improve precision by 15% and recall rate by 10%.