HOAX DETECTION BASED ON IDENTIFICATION OF OPINION LEADERS WHICH SUPPORTED BY OUTLIER USERS IN SOCIAL MEDIA

• Main Author: Ady Sanjaya, Samuel
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/34030
• Institution: Institut Teknologi Bandung

Social media is social interaction between people where they create, share, or exchange information and ideas in virtual communities and networks. Social media provides a two-way diffusion of information that allows users to share their information with other users, access information and communicate. Along with the growing population of Twitter users who can register for free and unconditionally, causing some irregularities. Social media is not rarely used as a tool to spread false news, spam, or do things that are not fair. Outlier detection has only been limited to user groupings: outliers and non outliers, even though there are characteristics of other users who play a role in social media networks. Most of the messages circulating on social media only come from some people who have high centrality that is passed on by other users. People who have high centrality are people who have a status or power called Opinion leaders. But in social media, high centrality is not only owned by people with this group, there are several anonymous accounts that have high centrality because they are supported by outlier accounts. Content in this condition can be suspected as content that is not good or even indicated as a Hoax. The importance of this identification is because Opinion Leaders have a high base of followers, so that it will speed up the process of absorption of information and will be harmful if the content distributed is indicated as Hoax. Outlier users who are controlled by certain groups or are part of a particular group will potentially spread content indicated Hoax to make it false belief in a mass manner. The use of Principal Component Analysis (PCA) in the initial stage before clustering is used for selecting attributes that have high weight. K-Means clustering was chosen as an unsupervised learning method to handle dynamic and unlabeled data. This study used 231,488 Twitter data from 25,880 users. By doing centroid analysis, making assumptions and analyzing characteristics, users have been grouped into: normal users (42%), spam (35.78%), active (20.06%) and opinion leaders (2.1%). Interaction in social media is dominated by Active users (20%) and Opinion leaders (2%) with accumulated interactions reaching 54.6%, with roles as destination nodes (passive). The number of outlier user clusters that make some opinion opinions and active users as sources of information, will be defined as "outliers in information diffusion". The information redirector published to broadcast several topics of hoaxes triggered by Opinion leaders and spread by massive outlier users. Detecting this content can be the initial presumption of content that is indicated to be a hoax to avoid spreading "false beliefs".