Enhancing User Emotion via a Session-Based News Recommender System Using Touch Interaction Patterns

News recommender systems are utilized to mitigate information overload through identification of user preferences and filtering of irrelevant content. However, anonymous browsing limits access to historical data. Session-based approaches address this challenge by relying solely on current interactions. In parallel, positive emotional experiences are recognized as crucial factors for increasing user engagement and revisit intention. In this study, a session-based news recommender system considering user interests and emotional states is proposed. A website was developed to collect user touch interaction data and self-reported emotions. Machine learning models were trained to predict emotions from these interaction patterns. Recommendation models were designed to incorporate predicted emotions, content features, and psychological insights. The regression tree method outperformed other models and when deployed, the average emotional index was raised from 21.52 to 46.86, and session duration was extended from 46.40 s to 51.14 s. It is confirmed that integrating emotional awareness enhances engagement.