融合元数据及隐式反馈信息的多层次联合学习推荐方法

基于隐式反馈信息的推荐是目前推荐系统领域的重要方法，能在一定程度上解决显式信息难以获得的问题。但由于隐式数据本身的特点单纯利用隐式反馈信息往往难以获取较好的推荐性能。针对此问题，本文提出一种融合元数据及隐式反馈信息的多层次深度联合学习（Multi-level Deep Joint Learning，简称MDJL）的推荐方法。它利用双深度神经网络共同学习，其中一个网络利用隐式反馈学习用户及项目个体个性化关系，另一个网络利用元数据学习高层次群体共性化关系，从而有效地表达用户偏好，使MDJL框架在个体及群体因素间达到平衡。实验结果表明，MDJL推荐算法在MovieLens 100K和MovieLens 1M两个公开数据集上均表现出更优越的推荐性能。

Combing metadata and implicit feedback to recommendation by multi-level deep joint learning

Recommendation based on implicit feedback information is an important method in the field of recommender systems, which can solve the problem that explicit information is difficult to obtain to some extent. However, due to the implicit data itself, it is difficult to obtain better recommendation performance by using implicit feedback information. To solve this problem, this paper proposes a Multi-level Deep Joint Learning (MDJL) recommendation method integrating metadata and implicit feedback information. It uses double deep neural network learning, one network using implicit feedback learning the relationship between individual and individual user, another network using metadata for learning high level group common relationship, so as to effectively express the user preferences, the MDJL framework to achieve the balance in the individual and group factors. Experimental results show that the MDJL recommendation algorithm shows better recommendation performance on two public datasets of MovieLens 100K and MovieLens 1M.