Using Indirect Communications to Improve Relationship Strength Estimation

Abstract

The estimation of relationship strength based on phone communication data has drawn great interests from amounts of researches. This paper focuses on combining indirect communications with direct communications to improve relationship strength estimation. Firstly, algorithms for finding one-jump and two-jump indirect communications are proposed to make preparations for feature extraction. Then features from direct and indirect communications are extracted and aggregated to be used for classification. Experiment results on MIT Social Relationship dataset indicate that our method combining direct and indirect communications performs better than existing method. Especially, our method improves the performance of estimating relationship strength on pairs with no direct communications.