Twitter Buzzer Detection System Using Tweet Similarity Feature and Support Vector Machine

Abstract

Over the past few years, people have been able to get and share information through social media easily. Some of that information can be a false issue created by a buzzer account that intends to influence people into a specific opinion. Politicians often use social media to maintain a good image in society by utilizing buzzer accounts. The main characteristic of a buzzer account is that they upload the same content repeatedly within a certain period. Before analyzing data taken from social media such as Twitter, we need a buzzer detection system to filter data from buzzer users. This research attempts to build a buzzer detection system using text processing and classification method. We use the similarity of tweets as a feature for the buzzer detection system by applying Cosine Similarity to the Term Frequency - Inverse Document Frequency (TF-IDF) feature of the tweets. In addition, we will use other features such as the number of followers, number of followings, the intensity of tweets, the ratio of retweets, and the ratio of tweets that contain links as additional features in this study. This research uses these features as inputs to the Support Vector Machine model to determine whether an account is a buzzer or not. This system has promising results by having 89% accuracy, 86.67% precision, 70.91 % recall, and 78% F1-score.