Dominating Factors Affecting Individual Retweeting Behavior

Abstract

In Chap. 3, we identify some influential factors that have an positive impact on individual retweeting behavior, such as topical relevance, information richness, soical tie strength, etc. One may wonder whether these factors only play an important role in theroy or, are these factors still important when predicting individual retweeting behavior? Furthermoer, to the best of our knowledge, virtually no scholarly effort has been undertaken to figure out the relative importance of those factors when predicting individual retweeting decision. Instead, a large number of features are indiscriminately introduced into the prediction model without examining the relevance of these features. The existence of redundant features not only increases data collection cost, but also tends to generate an overfitted model which predicts poorly on future observations not used in model training, known as the curse of dimensionality. Thus, it is necessary to rank the priority of these factors and find out the dominating ones. To tackle the above problems, we first pick out a specific user to illustrate the feature (also called factor in the monograph) selection process. The results confirm that only a small subset of predictors have an influential impact on individual retweeting behavior. And then, based on a large sample, we commit ourselves to find out factors that are not only important in theory in terms of explaining individual retweeting behavior, but also important in practice in terms of predicting individual retweeting behavior. Finally, we obtain a subset of dominating factors which not only save the cost of collecting trivial features but also improve the prediction performance to some extent, under certain classification algorithms such as support vector classification (SVC) or logistic.