Research And Implementation Of Friend And Location Recommendations In Instagram Social Network

The rapid development of online social networks provides users with a platform to record information and share resources. However, among massive data information, it is difficult for users to find accurate information to meet their current needs. Fortunately, recommendation system is one of effective approaches to solving this problem, the reason lies in that it can identify the users’preferences based on their mined historical data, which is used to recommend interesting resources to users. In LBSNs(Location Based Social Networks), as the location dimension being introduced into online social networks, recommendation is more challenging than traditional recommendation systems. Information related with location dimension, for example, user attributes and position expressions, are more complex, which results in recommendation issues in LBSNs have been becoming a hot research topic both in academic and industrial area.Friend and location recommendations are typical representives of recommendations in LBSNs. The former is usually implemented based on the number of common friends or attribute similarity, such as age, career, address, and so on. The latter is basically processed depending on spatial clustering of geographic location or longest common subsequences. However, the lackness of description information about user activity preferences and user trust relationships leads to unsatisfactory recommendation quality. Aiming at this issue, we address friend and location recommendations considering some factors, including users’ activity preferences, social trust relationships, location scoring and physical distance.Aiming at false recommendation and ignorance of users’ preferences in current friend recommendation algorithms, we implement friend recommendation in LBSNs based on users’ social trust and similarity of users’ preferences. Since users’ preferences are reflected by their activities, we calculate users’ activity similarity to find their similar friends. At the same time, as trust values between users can reflet their degree of correlation, thus the recommendation results considering trust relationships are more reasonable. Aiming at sparse location ratings and cold start problem for new users in current location recommendation algorithms, we implement location recommendation in LBSNs based on a matrix factorization method depending on users’ social trust and location similarity relationship to alleviate the sparseness issue. At the same time, we implement location recommendation by utilizing location accessing history of their trusted friends.Extensive experiments are carried out based on Instagram, which is a popular social network application for sharing photos. Compared with traditional recommendation algorithms, FRBTA obtained better performance in precision, recall, F-measure and MAP, due to the preference effect resulted from users’ activities and reliable friend candidates produced by social trust relationships. Similarily, LRBTA obtained better performance in metrics mentioned above, since location scoring matrix is implemented by a matrix factorization method. This matrix is dependind on both social trust relationship and location similarity.