GSMRecIT: A graph-based multi-task sequential recommender for simultaneous prediction of next item, interval, and duration

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Recent studies primarily focus on predicting the next items in user interactions, with only a few simultaneously predicting time intervals. However, in scenarios involving games or live broadcasts, it is necessary not only to predict the next interactive item but also the time interval and duration of the user's next interaction. To bridge this gap, this study employs a graph neural network (GNN) to embed user sequences, enhancing overall information integration and encoding, thus improving forecast accuracy. Additionally, the model incorporates an attention network to assign weights to various sequence data, effectively capturing users' long-term and short-term preferences. Given these capabilities, the proposed model can make top-N predictions concerning items, time intervals, and interaction durations of interest to users. Experimental results indicate that the proposed model significantly enhances prediction accuracy, demonstrating its great potential for improving predictions of next-item interactions in contexts with multimedia, live-streaming, or gaming activities.