Learning Style Integrated Deep Reinforcement Learning Framework for Programming Problem Recommendation in Online Judge System

Abstract Exercise recommendation is an integral part of enabling personalized learning. Giving appropriate exercises can facilitate learning for learners. The programming problem a specific application the exercise recommendation. Therefore, innovative framework problems that integrate learners’ styles proposed. In addition, there are some difficulties to be solved in this framework, such as quantifying behavior, representing problems, and strategies. For behavior strategies, algorithm based on deep reinforcement (DRLP) DRLP includes design action space, action-value Q -network, reward function. Learning style embedded into through space make recommendations more personalized. To represent DRLP, multi-dimensional integrated representation model proposed quantify difficulty feature, knowledge point text description, input output description problems. particular, Bi-GRU introduced learn texts’ contextual semantic association information from both positive negative directions. Finally, simulation experiment carried out with actual data 47,147 learners LUOGU Online Judge system. Compared optimal baseline model, effect has improved (HR, MRR, Novelty have increased by 4.35%, 1.15%, 1.1%), which proves rationality -network.