基于评论个性化多层注意力的商品推荐算法

ITEM RECOMMENDATION ALGORITHM BASED ON REVIEW PERSONALIZATION AND MULTILAYER ATTENTION

目前基于评论的推荐算法大多都忽略了用户(商品)的个性化信息来对评论进行编码。对此提出一种单词级别、评论级别的个性化注意力机制,分别对单词和评论进行个性化编码。设计一种基于门控机制的融合方式,来更好地融合用户和商品的隐向量来提高评分预测性能。在3组公开数据集上进行对比实验,以预测评分的均方误差(MSE)作为评估指标,验证了该方法的有效性。由单词和评论的注意力得分,能够观察用户和商品具体哪些特性相匹配,进一步提高推荐结果的可解释性。

At present,most of the review-based recommendation algorithms ignore the personalized information of users(items)to code the comments.Therefore,this paper proposes a personalized attention mechanism of word-level and review-level to encode words and reviews,and designs a fusion method based on gating mechanism to better fuse the hidden vectors of users and items to improve the performance of score prediction.The effectiveness of our method is verified by the comparison experiments on three sets of open datasets,and the MSE of the prediction score is used as the evaluation index.According to the attention score of words and reviews,we can observe which features of the user and the item match,and further improve the interpretability of the recommendation results.