田阳, 武浩. 基于双层注意力机制的对偶正则化评分预测[J]. 云南大学学报(自然科学版), 2021, 43(4): 681-689. doi: 10.7540/j.ynu.20200398
引用本文: 田阳, 武浩. 基于双层注意力机制的对偶正则化评分预测[J]. 云南大学学报(自然科学版), 2021, 43(4): 681-689. doi: 10.7540/j.ynu.20200398
TIAN Yang, WU Hao. The dual regularization score prediction based on double attention hierarchical attention mechanis[J]. Journal of Yunnan University: Natural Sciences Edition, 2021, 43(4): 681-689. DOI: 10.7540/j.ynu.20200398
Citation: TIAN Yang, WU Hao. The dual regularization score prediction based on double attention hierarchical attention mechanis[J]. Journal of Yunnan University: Natural Sciences Edition, 2021, 43(4): 681-689. DOI: 10.7540/j.ynu.20200398

基于双层注意力机制的对偶正则化评分预测

The dual regularization score prediction based on double attention hierarchical attention mechanis

  • 摘要: 基于深度学习的推荐系统已成为学界发展的趋势,针对其因数据稀疏而造成的预测精度下降的问题,提出了一个基于评论信息的深度协同过滤推荐系统. 首先,嵌入表示用户或物品的评论文本,将其送入BiGRU层以增强长文本中前后单词的关联性;然后,采用双层注意力机制来分配不同评论对于中间表征向量贡献度的权重;最后,利用概率矩阵分解法融合用户及物品表征向量,从而预测出用户对物品的评分. 实验结果表明,此模型可以显著减少评分预测的误差,有效提高推荐的精度.

     

    Abstract: The recommendation system based on deep learning is the trend of current development. In order to solve the problem about the prediction accuracy decreases due to the sparse data, in this paper, a deep collaborative filtering recommendation system based on comment information is proposed. Firstly, the user’s or item’s comment text is embedded and represented, and it is sent to BiGRU layer to enhance the relevance of the words before and after the long text. Then, a double-layer attention mechanism is used to allocate the weight of different comments’ contribution to the model. Finally, PMF is used to fuse the user model and the item model to predict the user’s rating on the item. The experimental results show that this model can significantly reduce the error of score prediction, so as to effectively improve the accuracy of recommendation.

     

/

返回文章
返回