Research On Location-based Friends Recommendation

With the vigorous development of information technology,people's life has been changed greatly.Especially in the field of information technology,the rapid development of the Internet has brought tremendous changes to the whole world and has an influence on people's life style and ways of thinking.People's social needs have increased a lot.Therefore,social network as a tool for communication has become an essential part of life.Social networks such as Twitter and Sina Weibo have become very popular because they are regarded by people as the reflection and extension of their real life.But there are so many users in the social network that different people have different interests and preference.If the users make friends indiscriminately on the social network,they may have an ineffective communication with others or they can't get the right information they need,which will cause bad experience.A good system should filter the massive amounts of information and select the content that one user most interest and concern,which will improve the efficiency of the system and save time for users.Thus,how to find a proper friend in the social network is important.Recommender system is designed for solving this problem by synthesizing and analyzing available information to help users to find friends they interested and integrate into social community.How to recommend friends in social networks has received substantial interest from both academia and industry for the sake of improving users satisfaction and loyalty and maintain their stickiness.Traditional friend recommendation approaches in social network try to find friends with common interests measured with the contents of users' published posts and their following relationships.However,most of these approaches may only suggest 'similar' people to be friends,which is only suitable for virtual social space instead of the real world.For instance,if friends locate far away from each other,it is difficult for them to meet together to participate in some interesting activities.Therefore,it is easy to recommend friends in social network,but it is relatively hard to make friends not only in virtual social space but also in real world.Location-based social networks(LBSN)add location information to an existing social network so that users can share location-embedded information.A user's location history provides rich contextual information and has significant correlation to his/her real social behaviors.More specifically,a user's historical locations(e.g.,check-ins and geo-tagged photos)reflect his/her experience,living pattern,preference and interest.Further,the location interdependency between two users includes not only that they co-occur in the same physical location or share similar location histories but also their common interests,behaviors,and activities,which can be inferred from their location histories.In a word,the location dimension connects social networks with reality,bridging the gap between the virtual social space and the physical world.Therefore,considering locations into friend recommendation in social network can meet needs of the real world.In this thesis,we propose a location sensitive friend recommendation framework for social network from the perspective of both virtual social space and real world.This means we try to find friends that they can not only communicate with each other by social network,but also can participate in some real-life activities face to face.The main idea of our framework is that we assume people are more likely to be friends on both virtual social space and real world if their daily activities share more location overlaps besides the common interests.Our framework considers two kinds of information of social network to recommend friends,i.e.,user following relationships and check-in locations.We explore three metrics to capture the geographical relations between users.Then we exploit three integration methods,i.e.,weighted arithmetic method,weighted geometric method and geographic regularization method,to fuse the geographical relations and user following relationships modeled by matrix factorization to recommend friends.Experiments on the real dataset indicate that our methods outperform the baseline methods in terms of precision,recall and F1.We also find that the weighted arithmetic method achieves the best performance among these integration methods.