ISSN 1000-1239 CN 11-1777/TP

计算机研究与发展 ›› 2020, Vol. 57 ›› Issue (6): 1227-1238.doi: 10.7544/issn1000-1239.2020.20190403

• 人工智能 • 上一篇    下一篇

基于强化学习DQN的智能体信任增强

亓法欣1,童向荣1,于雷1,2   

  1. 1(烟台大学计算机与控制工程学院 山东烟台 264005);2(纽约州立大学宾汉姆顿分校计算机科学系 纽约州宾汉姆顿市 13902) (qifaxin123@163.com)
  • 出版日期: 2020-06-01
  • 基金资助: 
    国家自然科学基金项目(61572418)

Agent Trust Boost via Reinforcement Learning DQN

Qi Faxin1, Tong Xiangrong1, Yu Lei1,2   

  1. 1(School of Computer and Control Engineering, Yantai University, Yantai, Shandong 264005);2(Department of Computer Science, State University of New York at Binghamton, Binghamton, NY 13902)
  • Online: 2020-06-01
  • Supported by: 
    This work was supported by the National Natural Science Foundation of China (61572418).

摘要: 信任推荐系统是以社交网络为基础的一种重要推荐系统应用,其结合用户之间的信任关系对用户进行项目推荐.但之前的研究一般假定用户之间的信任值固定,无法对用户信任及偏好的动态变化做出及时响应,进而影响推荐效果.实际上,用户接受推荐后,当实际评价高于心理预期时,体验用户对推荐者的信任将增加,反之则下降.针对此问题,并且重点考虑用户间信任变化过程及信任的动态性,提出了一种结合强化学习的用户信任增强方法.因此,使用最小均方误差算法研究评价差值对用户信任的动态影响,利用强化学习方法deep q-learning(DQN)模拟推荐者在推荐过程中学习用户偏好进而提升信任值的过程,并且提出了一个多项式级别的算法来计算信任值和推荐,可激励推荐者学习用户的偏好,并使用户对推荐者的信任始终保持在较高程度.实验表明,方法可快速响应用户偏好的动态变化,当其应用于推荐系统时,相较于其他方法,可为用户提供更及时、更准确的推荐结果.

关键词: 多智能体系统, 强化学习, 信任, 深度q学习, 最小均方误差方法

Abstract: Trust recommendation system is an important application of recommendation system based on social network. It combines the trust relationship between users to recommend items to users. However, previous studies generally assume that the trust value between users is fixed, so it is unable to respond to the dynamic changes of user trust and preferences in a timely manner, thus affecting the recommendation effect. In fact, after receiving the recommendation, there is a difference between actual evaluation and expected evaluation which is correlated with trust value. The user’s trust in the recommender will increase when the actual evaluation is higher than expected evaluation, and vice versa. Based on the dynamics of trust and the changing process of trust between users, this paper proposes a trust boost method through reinforcement learning. Least mean square algorithm is used to learn the dynamic impact of evaluation difference on user’s trust. In addition, a reinforcement learning method deep q-learning (DQN) is studied to simulate the process of learning user’s preferences and boosting trust value. Finally, a polynomial level algorithm is proposed to calculate the trust value and recommendation, which can motivate the recommender to learn the user’s preference and keep the user’s trust in the recommender at a high level. Experiments indicate that our method applied to recommendation systems could respond to the changes quickly on user’s preferences. Compared with other methods, our method has better accuracy on recommendation.

Key words: multi-agent systems, reinforcement learning, trust, deep q-learning (DQN), least mean square (LMS)

中图分类号: