融合项目类别信息的非侵入式嵌入序列推荐

Non-intrusive embedding for item category information fusion in sequential recommendation

针对基于自注意力机制的序列推荐模型的数据稀疏问题和融合项目辅助信息导致的信息过载问题,提出融合项目类别信息的非侵入式嵌入序列推荐模型(NIESR)。通过融合项目类别信息辅助自注意力机制计算项目相似度,缓解数据稀疏问题;将不同种类的输入分别用一个自注意力机制提取特征,将每个注意力分数矩阵分别与项目序列的嵌入表征点积后拼接,缓解信息过载问题。在3个公开数据集上验证了所提模型的有效性,评估指标平均提高4.5%,最高提升13.3%。

Aiming at the problems of data sparsity and information overload caused by fusing side information of items,a non-intrusive embedding for item category information fusion in sequential recommendation model(NIESR)was proposed.The problem of data sparsity was alleviated by fusing item category information to assist self-attention mechanism to calculate item simila-rity.To alleviate the information overload problem,a variety of information input from the model was input into a self-attention network separately,and each attention score matrix was dotted with the embedding representation of item sequence and then connected.The validity of the proposed model is verified on three public datasets,and the evaluation metrics increase by 4.5%on average and 13.3%at the highest.