Academic Recommendation on Graph with Dynamic Transfer Chain

Academic contents update and learner's capability change over time. But nowadays, academic recommendation system does not take time factors into account. There are two challenges to capture learner's preferences and learning context accurately and dynamically. First modeling academic trend and user's cognitive level transferred by time is a hard problem. And designing dynamic algorithm to improve recommendation accuracy with implicit behavior data is difficult. In this paper, we propose Dynamic Transfer Chain (DTC) to model user's preferences and academic context over time on transaction data. Based on DTC model, we present a novel algorithm Dynamic Academic Recommendation on Graph (DARG). We evaluate the effectiveness of our method using an open dataset named CiteULike, including 9170 users, 11343 papers, 194596 user-paper pairs. The evaluation metric we used is Hit Ratio. The results show that our proposed approach gives 12.873% to 33.852% improvement over the previous counterpart, including User-KNN, Item-KNN, TUser-KNN, TItem-KNN.