Personalized Friends Recommendation Technology Research Based On User Preference In LBSN

Location-based social network has received a lot of attentions,represented by Facebook.This new social network combined locations with traditional social network,which extended the virtual network to the real world.We can learn people's location from location-based social network,and further learn life patterns and preferences.In recent years,how to find people's preference in life and provide personalized service has become one of the hot researches in academia and industry.The traditional mining user's preference methods can not comprehensively learn user's interest,and the recommended effect is not ideal through collaborative filtering recommendation method in location-based social network.To solve these problems,we proposed a new method that can find user's long-term preferences according to user's check-in location information and check-in time characteristics.We then came up with a new personalized recommendation model which considered user's social relations,check-in locations,and interests.Firstly,in order to mining user's preference through their check-in locations,we do not consider the number of times of their interesting activities,but consider the regularity.That's because the short-term hobby also can be found only consider the number of interesting activities times.In order to learn the regularity features,we present the concept of fuzzy period and fuzzy degree period.We proposed the fuzzy degree period calculation method and it's improved algorithm.Fuzzy period refers to a period not strictly,aimed at showing people regularity behavior in real life,rather than theoretical studies of strictly period;The value of fuzzy degree period represent the strength of the regularity.We first built the boolean sequence based on check-in locations according to the time series.And then we calculate the fuzzy degree period through boolean algebra method.For "pseudo-regularity" issue that may arise,we further proposed the improved algorithm based on fuzzy period.Secondly,to realize personalized recommendation,we added user's interests to the location-based social network,and built a multiple heterogeneous social network MHSN,which includes three kinds of heterogeneous vertex and three types of associated side.MHSN combines the social relations,check-in locations and user's preference to complete the following personalized friend recommendation research.Thirdly,we proposed the method of mining user's preference based on check-in information.User's hobby includes long-term hobby and short-term hobby,we mainly found long-term hobby in this thesis.For long-term hobby,user interesting activities usually present regularity.So we need not only consider the number of times of their interesting activities,but also consider its regularity.Furthermore,for the cold start problem that new user has no check-in records,we use the community discovery method to predict target user's hobby according to his friends' hobbies in the same community.Then,we established a multi-factors personalized friend recommendation model.Recommended friends had the similar hobby,closely social relations and space positions.We established the calculating hobby similar model to compute the hobby similarity between recommended friends and target user;We proposed a method of calculating social distance that based on common friends and max weight path to compute the social distance between recommended friends and target user;Because that closer space locations with target user is more likely to be accepted,and the check-in records are usually sparse,we came up with amethod of calculating spatial distance based on clustering.Lastly,after tremendous experiments and analysis on Foursquare real dataset,the result shows:compared to the check-in number based mining hobby method,our method that based on regularity has a better performance at accuracy.The improved regularity discovery method which based on compress has a better performance than original method.As for the problem of personalized friend recommendation,compared to the traditional methods:location-based collaborative filtering method and preference-based collaborative method,our recommendation model has a better accuracy.In conclusion,this thesis studies on the problems of hobby mining and personalized friend recommendation,and proposes new solutions.Theoretical analysis and experimental results show that,compared to the traditional methods our approach has obvious advantages in accuracy.