Abstract: Due to the vast amounts of users, it is difficult for a user to make effective connections with others for common interests. Friend recommendation on online social networks, therefore, becomes a challenging research issue, which may have significant effects on sustainable developments of social networks.Most of the existing friend recommendation methods are conducted based on users' explicit information such as background, demography, interests and posts, while ignoring users' implicit information such as their social relationships. Notably, explicit information is often incomplete and not trustworthy, and cannot be appropriately used to measure user similarities.In order to effectively recommend friends, a recommendation algorithm is proposed based on user relationship information in online social networks. In the described algorithm, user relationships are characterized by using the association analysis method, and then a weighted, directed graph between network users is constructed. Based on this graph, this algorithm builds a transition matrix and uses the PageRank algorithm to calculateusers' scores that indicate the acceptance probabilities, and then recommend the users with high scores to the target user on social networks. In addition, with the consideration of the user authority in a specific social network, an enhanced friend recommendation algorithm is further developed.In order to validate the proposed approaches, friend recommendation experiments on Twitter are conducted and the users' information and their relationship data are extracted. For this purpose, two traditional methods, i.e., social filtering algorithm and the PageRank algorithm, are used to compare with the two proposed approaches based on two measures, i.e., accuracy and recall rate. Experiments results show that the proposed recommendation algorithms yield clearly better results in accuracy and recall rate than the traditional recommendation algorithms.

Key words: social networks, friend recommendation, social relationship, user authority, PageRank algorithm