Online social networking technologies enable users to share information with their friends. Social networks such as Facebook, Google+, and Twitter have made a significant impact on how users share and exchange data. Users join these networks, share opinions, make comments, and interact with their friends. The interactions involve behavior associated with particular roles. Role analysis helps to characterize users’ interactions on a social network.
The concept of a social role is viewed as the behavioral expectations that are associated with identifiable positions in a social structure. What role does an individual play in a group? A user may play her role as a leader, a participant, a commentator, a lurker, or etc. Fournier and Lee [12] pointed out that communities are strongest when everyone plays a role. Members stay involved and add value by playing a wide variety of roles. Akaka et al. [2] mentioned that social roles act as resources for change in value networks because they can lead to social norms and establish social positions. Since social network dynamically changes, roles user playing may change as time goes on. The role change may imply position shifting of users, status switching of participants, behavior transition of people, and etc. Gleave et al. [14] showed that by identifying roles, it would be more effective to monitor some proportions of roles. Also, a better search result can be obtained if the authors of content are considered. Thus, role analysis has long been central to sociology [28][24][6][9], and becomes more and more valuable in many aspects. For example, Many researches [19][20][1][16][31] focused on finding the influential leaders in a social group. Obviously, a more general and flexible framework for discovering significant roles including influential leaders will be necessary and helpful for many applications.
Since the rise of Web 2.0 from early 2000s, content generated by users become increasingly popular and important. Velardi et al. [33] mentioned that communicative content attracts more and more interest from business, social and research community
analysts. Naaman et al. [26] pointed out that interesting and informative content is the key to increase amount of readers. McCallum et al. [23] presented the Author-Recipient-Topic (ART) model for social network analysis. ART combines the connectivity structure and probabilistic language model to capture topics, find social roles and summarize a large amount of data. Through analyzing the content created by users, we can get better understand the characteristics of users and behavior between users.
Previously proposed methods are mainly based on structural analysis of social networks rather than content-based behavior analysis. Most of them classify the users in social networks into pre-defined roles and do not consider the dynamic characteristics in social networks. To the best of our knowledge, there is no method using content-based behavioral features extracted from user-generated content and behavior patterns to identify roles without using any pre-defined roles and to explore role change patterns in social networks.
Therefore, in this thesis, we propose a content-based behavioral method to analyze the roles and role changes in social networks. Unlike previous studies, we combine content-based and behavioral features to discover roles in social networks.
The features provide us a better view to analyze users’ roles. Our proposed method first extracts the content-based behavioral features for each user, and then utilizes fuzzy c-means clustering method (FCM) [5] to identify different roles in a social group. Next, it transforms the results obtained from FCM into role change sequences, and applies PrefixSpan [29] to discovering role change patterns. The proposed method can find various roles in social networks without using any pre-defined roles and may discover additional roles that haven’t been previously aware of. Using the concept of fuzzy set to record user’ role memberships provides us a flexible way to identify multiple roles played by the user in a social network. Moreover, content-based behavioral features provide valuable information of users and better understand users’
behavior.
The contributions of this thesis can be summarized as follows. First, we propose a framework to identify various roles in social networks. Next, we present an approach to mine role change patterns. Finally, the experimental results show that the proposed method can find various roles and role change patterns in different kinds of social groups. There are six roles in a technology group: expert, kicker, leader, viewer, participant and follower; three roles of opposite political positions in a political group:
supporter, dissenter and neutralist; and three roles in entertainment groups: creator, commentator and participant. We also discover some interesting role change patterns in different groups. In the technology group, users may shift their roles from viewer to leader since they learn more and more expertise from the fan group so that their recognition increases with time. In the political group, users shift their roles from positive supporter to negative dissenter or neutralist with higher negative affectivity. It may be a warning sign for the politician.
The rest of this thesis is organized as follows. Chapter 2 surveys the related work.
Chapter 3 describes the problem definitions and our proposed framework. Chapter 4 evaluates the proposed framework. Finally, the conclusions and future work are described in Chapter 5.