基于隐式反馈的深度强化学习MOOC课程推荐方法  

MOOC Course Recommendation Method Based on Implicit Feedback and Deep Reinforcement Learning

作　　者：张力 张怡文[2] 单春宇 曹小兰 王兴 刘书影[2] ZHANG Li;ZHANG Yiwen;SHAN Chunyu;CAO Xiaolan;WANG Xing;LIU Shuying(School of Electronics and Information Engineering,Anhui Jianzhu University,Hefei 230009,China;School of Big Data and Artificial Intelligence,Anhui Xinhua University,Hefei 230088,China)

机构地区：[1]安徽建筑大学电子与信息工程学院,安徽合肥230009 [2]安徽新华学院大数据与人工智能学院,安徽合肥230088

出　　处：《安庆师范大学学报（自然科学版）》2024年第3期63-70,共8页Journal of Anqing Normal University(Natural Science Edition)

基　　金：安徽省教育厅自然科学重点项目(KJ2020A0784,KJ2021A1159)。

摘　　要：传统课程推荐系统多利用历史交互数据来评估用户对物品的偏好,并以此进行推荐。但是细粒度和动态用户偏好获取受到限制,从而导致推荐结果不精确且缺乏可解释性。针对此问题,本文提出了基于隐式反馈的深度强化学习MOOC课程推荐方法(MCIFRL),旨在提高推荐结果的准确度和可解释性。该方法利用多层感知机模型来构建深度强化学习的决策网络,提取用户对话历史和候选物品数量信息,并生成询问属性或推荐物品的动作。随后,通过问答交互,多次询问物品本身的属性,实现精准推荐。实验基于系统预测用户对物品分数产生推荐列表,并在公开的MoocData数据集上验证。结果表明,MCIFRL推荐结果相较于其他方法,在不同推荐列表长度上的准确度均有所提升,且在HR和NDCG指标上均得到改善。Traditional course recommendation systems use historical interaction data to evaluate user preferences for items and make recommendations accordingly.However,the acquisition of fine-grained and dynamic user preferences is limited,resulting in inaccurate and uninterpretable recommendation results.To address this issue,this paper proposes a MOOC course recommendation method based on deep reinforcement learning with implicit feedback(MCIFRL),which aims to improve the accuracy and interpretability of recommendation results.The method employs a multi-layer perceptron model to build a deep reinforcement learning decision network,extracting user conversation history and candidate item quantity information,to generate actions to inquire about attributes or recommend items.Subsequently,through question-and-answer interaction,the attributes of the item itself are inquired about multiple times to achieve accurate recommendations.The experiment generates a recommendation list based on the system’s prediction of the user’s score for the items and validates it on the public MoocData dataset.The results show that compared to other methods,MCIFRL has improved the accuracy of recommendation results at different recommendation list lengths,and has improved both HR and NDCG indicators.

关 键 词：MOOC推荐 会话推荐 深度强化学习 多层感知机 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]