Principle Features for Tie Strength Estimation in Micro-blog Social Network

Relationship degree should be various between two entities, while static friend list in social network is incompetent to express the relationship divergence, thus tie strength is presented to quantitatively describe real social relation based on lots of features derived from historical data in online activities. We propose a model to measure tie strength within a user's social circle in Micro-blog by multivariate stepwise linear regression (MSLR). The model also determines the principle features that are affecting tie strength significantly from abundant features in Micro-blog. Features are mapped into four dimensions to excavate implications hiding behind them from abstract level by taking advantage of previous sociological study achievement. Thus, key dimensions for tie strength are also obtained. Then Native Bayesian Classifier is employed to test the performance of key features and key dimensions in distinguishing ties by strength values. We apply this model with the data derived from Sina Micro-blog and get approximately 80% accuracy of tie strength estimation. Seven principle features are obtained from nineteen ones. We conclude that principle features are consistent with key dimensions, which proves that the mapping processing is reasonable.