Social Network Friendship Prediction Problem Research

Friendship prediction is always one essential problem in related field of social network.It could be obtained by related algorithm involves friendship between users,after gaining the result,we could apply it to variety of recommendation systems including the functions of friends recommendation in social APP and goods recommendation in shopping websites,even to predict the change of friendship or friendship in real life.Based on interaction data between users which include photos,private messages,comments and mutual friends,friendship strength could be calculated by algorithm.There are two research directions about predicting friendship discussed in our paper,one is subdividing the interactive modes into as many aspects as possible,it’s an effective way to make prediction results more accuracy,the other one is optimizing calculation,lowering time complexity,so that the algorithm could have higher universality,no matter how much the size of dataset is,calculation would cost less time.The author of this paper does some research about problem of friendship prediction,the main research achievements and contributions are as follows:The calculation of friendship strength is used to solving many problems,such as if new links could emerge between users,finds missing links and predicts which are true friends among all friends in social networks.The paper which bases on calculation of friendship proposes a new question,which is predicting the change of friendship strength in the future,this research is helpful for recommendation system of shopping websites and social networks to improve its accuracy.The paper optimizes the group similarity in original friendship strength equation,the original group similarity treats directly mutual friends of users equally without discrimination.On the contrary,according to the contact frequency between mutual friends and users,we divide directly mutual friends of users into two types,mutual friends who contact users more frequent are given higher significance,meanwhile,mutual friends who contact less have lower proportion in the calculation.As a result,experiment proves that the accuracy improves obviously after optimizing group similarity as mentioned before.As to another original friendship strength calculation,we add variance of interaction,recency and longevity into it.The calculation mentioned just before considers expectation of interaction between users,however,it couldn’t assure if interaction between users is durable and stable.The original equation only considers interaction between users without taking the moment when users built friendship or interacted into account,experiments verify optimization contributes to the improvement of prediction accuracy,what’s more,from the analysis of experimental results,we could infer recency is essential to accuracy improvement.