User Stance Detection and Prediction Considering Most Frequent Interactions

Document Type


Publication Date

Summer 10-15-2022


— Detecting and predicting users' stances on social media platforms is a challenging problem. There are many applications in detecting and predicting users' stances in many fields like politics, digital marketing and social science. Most research works in this area are using the users' textual contentslike tweets or posts as indicators for their stances in a given topic. Recently, some works started to consider users' online activities like networks of accounts that users interact with in mentions, replies or retweets as indicators of their stances. There are challenges in using users' online activities to detect or predict their stances in given topics. One of these challenges is that most of users' interactions might not be relevant to the given topics. Moreover, the large number of users' interactions might have a significant impact in detecting their stances efficiently. The main objective of this paper is to overcome these challenges by considering the most relevant users' online interactions to improve the stances prediction and detection tasks. To meet this objective, we selected only the most relevant interactions to given topics using a filtration method based on the users' interaction frequency. Our results show, in most cases, that considering the most frequent and relevant users' interactions outperforms the research works that are including all the interactions of users.