Quantitative analysis of trust factors on social network using data mining approach

Computing trust factors quantitatively is an important task in social network analysis, especially for online social networks where face-to-face interactions are spared. Trust factors can be used to infer degree of friendship as well as investigating what contribute to customer loyalty in e-commerce on online social networks. In the past, trust factors were either obtained via direct questioning during survey or qualitative estimation; questions to be asked are `how much do you trust something/somebody in multi-level Likert scale?' Respondents replied by indicating one of the following, for example: Very much, Much; Somewhat; Little; Never. The results are then counted usually by simple frequency distribution. In this paper, an alternative method by data mining is presented that infers quantitatively the relative importance of each trust factor with respective to the predicted class. This is done by Feature Selection algorithms prior to constructing a decision tree. The advantage of using data mining method over simple statistic is that each factor (or Relative Importance Factor) acts as a predictor variable that foretells how the likelihood of an expected occurrence. In other words, a-priori knowledge can be assumed in this case; the relative importance can be assumed known without the fact that actually occurred. Such relative importance measures would be useful for estimating the weights of variables that are used in some prediction formula. A case study of estimating trust and therefore inducing recommendation in Facebook is given.