Abstract:

The recommendation system is becoming popular as an important way to solve the problem of information overload in the era of information explosion. Traditional recommendation systems generally have shortcomings such as low precision and unclear evaluation criteria. Introducing machine learning, especially deep learning techniques, into the recommendation system will effectively improve the above defects and bottlenecks. A hybrid recommendation system based on self-attention model is proposed. Firstly, the attention model in deep neural network is used to weight the item attributes of specific recommended products to providing user recognition degree of pre-recommended products. Then, the traditional loss ranking model is replaced by Ada-Boosting model, which makes the related evaluation indicators such as accuracy and recall rate greatly improved. Finally, based on the existing recommendation system evaluation indicators, a new evaluation criteria called ARD is introduced for the first time. By evaluating ARD, the recommendation system performance can be more accurately evaluated in the user experience dimension.

Key words: self-attention model, Ada-Boosting, collaborative filtering, hybrid recommendation

摘要：

推荐系统作为信息爆炸时代下解决信息过载问题的重要方式受到了越来越大的关注。传统的推荐系统普遍存在精度不高、评价标准不明确等缺陷，将机器学习尤其是深度学习技术引入推荐系统将有效改善上述缺陷及瓶颈。提出了一种基于注意力模型的混合推荐系统，利用深度神经网络中的注意力模型对特定推荐商品的物品属性进行加权分配，获得预推荐商品的用户认可度评分；通过自适应增强模型替换传统的损失排序模型，使得精确度、召回率等相关评价指标获得较大提升。在现有推荐系统评价指标的基础上，首次引入了用户群体评价认可度指标，通过认可度指标可以在用户体验维度对推荐系统性能给出更精确的评价。

关键词: 注意力模型, 自适应增强, 协同过滤, 混合推荐