Predicting Friendship Strength for Privacy Preserving: A Case Study on Facebook

Effective friend classification in Online Social Networks (OSN) has many benefits in privacy. Anything posted by a user in social networks like Facebook is distributed among all their friends. Although the user can select the manual option for their post-dissemination, it is not feasible every time. Since not all friends are the same in social networks, the visibility access for the post should be different for different strengths of friendship for privacy. Previous works in finding friendship strength in social networks have used interaction and similarity based features but none of them has considered using sentiment-based features as the driving factor to determine the strength. In this paper, we develop a supervised model to estimate the friendship strength based upon 23 different features comprising of structure based, interaction based, homophily based and sentiment based features. We evaluate our model on a real-world Facebook dataset we built that has ground truth for different types of friendship: close, good, acquaintance, and casual. Our model obtains an AUROC of 0.82 in identifying acquaintances from all the other three categories, and an AUROC of 0.85 in distinguishing between close friends and acquaintances. Our experiments suggest that features like average comment length, reaction scores for likes and love, friend tag score, Jaccard similarity and closeness variable consistently perform better in predicting friendship strength across different classifiers. In addition, combining language-based features with homophilic, structural and interaction features produces more accurate and trustworthy model to evaluate friendship strength.