Studying social influence on Twitter

The study of social influence has a long history in research areas of sociology and marketing. In recent years, with the rapid growth of Online Social Networks (OSNs), online influence has received lots of attention from both academic community and industry. For the work in this dissertation, we consider the Twitter platform and aim to address two problems: 1) feature selection for measuring social influence; and 2) influence maximization for marketing campaigns. While many researchers focus on measuring social influence on Twitter, there is still lacking of a comprehensive analysis of feature selection. Most existing studies directly utilize their own pre-defined features to build the model without evaluation and judgement for the seselected features. In order to find principal features for measuring user influence on Twitter, we select manifest features based on sociology knowledge. Besides principal manifest features, we identify hidden features and map them to the attributes of influencers in the research area of social science. Furthermore, we propose a hybrid feature selection method for predicting user influence. After evaluating the quality of features by utilizing a filter method, a reduced feature subset is obtained. Following the principles of wrapper methods, we assess the feature subset at each searching step. Finally, an optimal feature set with a high degree of accuracy for predicting user influence is obtained. Influence maximization is the most fundamental and important problem when studying social influence. In this work, we identify a specific influence maximization problem as selecting a set of seed users to maximize the effectiveness of advertising campaigns on Twitter. When studying influence maximization problem, we develop our solution with new ideas focusing on : 1) the definition of influence; 2)the influence probability model; 3) the influence diffusion model; and 4) the seed nodes selection algorithm. The proposed influence maximization approach has taken into consideration of social ties, user interactions, and the characteristics of advertising information propagation on Twitter. Our work provides a solid generic solution for promoting products or services in online social networks like Twitter.