融合序列模式评分的策略梯度推荐算法

推荐算法在一定程度上解决了信息过载问题,但传统推荐模型在挖掘数据特性方面有待改进。为此,结合强化学习方法提出一种融合序列模式评分的策略梯度推荐算法。将推荐过程建模为马尔可夫决策过程;分析推荐基础数据特性模式,设计以序列模式评分为奖励的反馈函数,在算法的每一次迭代过程中学习;通过对累积奖励设计标准化操作来降低策略梯度的方差。将该方法应用到电影推荐中进行验证,结果表明所提方法具有较好的推荐准确性。

POLICY GRADIENT RECOMMENDATION ALGORITHM COMBINING SEQUENCE PATTERN RATING

The recommendation algorithm solves the problem of information overload to a certain extent,but the traditional recommendation model needs to be improved in mining the characteristics of the data.For these problems,we propose a policy gradient recommendation algorithm combining sequence pattern rating based on the reinforcement learning method.The recommendation process was modeled as Markov decision process;the characteristic pattern of recommended basic data was analyzed and a feedback function was designed with sequential pattern rating as reward to learn in each iteration of the algorithm;the variance of the policy gradient was reduced by designing a standardized operation on the cumulative reward;The method was applied to movie recommendation for verification,the experimental results showed that the proposed method has a good recommendation accuracy.