Research On Joint Tensor And Matrix Factorization Based Algorithms For Point-of-Interest Recommendation

With the development of Internet technology,location-based social networks are gradually accepted by people.They provide the functions of issuing instant messages and sharing positioning with others,thus forming a more live online social circle.The huge amount of users and interactive information is the foundation for the extraction and mining of massive data,which attracted the attention of many scholars.But this not only provides a good opportunity for the research of scholars in the industry,but also brings a realistic challenge in the context of the big data era.Especially,the point-of-interest recommendation algorithm,which is widely adopted and conveniently used,maintains a certain degree of enthusiasm in the relevant research of location-based social networks.It aims to predict point-of-interest in the future through the user's history check-in records.This thesis explores the limitations of point-of-interest recommendation in the simplicity of user preference observation and the neglect of the impact of population on user behavior in terms of time statistics.On this basis,an algorithm based on tensor and matrix joint factorization is proposed to simulate the cognitive process of users,and long-term preferences and crowd influence are involved based on basic tensor model to predict the location categories that users may like.And then by exploiting weighted kernel density estimation function,the specific point-of-interests are filtered as recommendation results.The experimental results on real-world datasets show that compared with the state-of-the-arts,including MF-URT,CTF-ARA,and TAD-FPMC-HitsGroupDist,the proposed algorithm in term of precision is about 114%-151%,596%-679%,and 303%-600% higher than them on New York City dataset,respectively,and is about 455%-479%,484%-498%,and 112%--155% higher on Tokyo dataset,respectively.