【摘要】 在解决新用户冷启动问题时,固定不变的Epsilon参数会使传统Epsilon-greedy算法收敛缓慢。为此,提出一种改进的Epsilon-greedy算法。利用免疫反馈模型动态调整Epsilon参数,从而使算法快速收敛。使用蒙特卡罗模拟方法对算法进行实验验证,结果表明,该算法能够在用户与推荐系统交互较少的情况下为用户进行有效推荐,且推荐效果优于传统的Epsilon-greedy、Softmax和UCB算法。更多还原

【Abstract】 When solving the cold-start problem of new users,f ixed and invariant Epsilon parameters will slow the convergence of traditional Epsilon-greedy algorithm. Therefore,an improved Epsilon-greedy algorithm is proposed.Immune feedback model is used to dynamically adjust the Epsilon parameters so that the algorithm converges quickly.Monte Carlo simulation is used to validate the proposed algorithm. Results show that this algorithm can effectively recommend to users when they have little interaction with the recommendation system, and the recommendation effect is better than the traditional Epsilon-greedy algorithm,Softmax algorithm and UCB algorithm.更多还原