Research on Twitter User Tag Preference Prediction Based on Thompson Sampling Algorithm

College of Computer Science, Beijing University of Technology, Beijing, China

^* Corresponding author: author@email.org

Abstract

Twitter's user behaviour data is crucial for studying user patterns and content recommendation. To achieve this goal, the paper first preprocesses a Twitter user dataset obtained from Kaggle. The dataset includes over 40,000 objects in JSON format, focusing on users who tweeted on trending topics and had at least 100 followers and were following at least 100 other accounts. This filtering helps to exclude spam and empty accounts. The study constructs a user-hashtag matrix and applies label encoding technology to convert it into a numerical matrix. The Thompson Sampling algorithm is then applied to predict user hashtag preferences. The experimental results demonstrate the remarkable effectiveness of the Thompson Sampling algorithm in predicting users' preferences for hashtags. By iteratively updating the alpha and beta parameters for each hashtag multiple times, the algorithm can accurately estimate users' preferences and successfully identify hashtags with high recommendation value. There are significant differences in the preference levels of different hashtags among user groups, providing an important basis for subsequent recommendations and push notifications. The findings validate the algorithm's effectiveness and contribute to optimizing social media content recommendation algorithms, ultimately enhancing user experience and benefiting content creators and advertisers. This research contributes to the advancement of social media platforms by improving content recommendation algorithms, enhancing user experience, and fostering a more engaging and personalized user environment.