[Purposes]In response to the problem that knowledge graph based reinforcement learning rec-ommendation algorithms only consider the interpretability of recommendation results without considering the diversity of recommendation results,this paper proposes a new knowledge graph based reinforcement learning recommendation algorithm to improve the diversity and accuracy of recommendation results while solving the interpretability problem of recommendation results.[Methods]This article proposes an evaluation metric ETD for the diversity of recommendation results,and introduces the Ran-dome_Beam_Search search algorithm in the path inference module of the algorithm to improve the diver-sity of recommendation results.At the same time,an attention mechanism is introduced in the reinforce-ment learning module to improve the accuracy of recommendation results.[Findings]Compared with the interpretable PGPR recommendation algorithm,the algorithm proposed in this article improved the diver-sity of recommendation results by 28.4%and accuracy by 0.056 percentage points on the Beauty dataset;the accuracy of recommendation results on the Clothing dataset improved by 0.035 percentage points.[Conclusions]The algorithm proposed in this article not only solves the interpretability problem of rec-ommendation systems,but also improves the diversity and accuracy of recommendation results.
维普
万方数据