摘要： 传统协同过滤算法仅依靠用户评分数据的低维向量方法，存在推荐结果精确度低以及冷启动问题。为此，提出一种新的动态混合推荐算法，将栈式降噪自动编码器融入到基于用户的协同过滤中，学习用户的深层次特征，并与基于用户项目属性偏好的相似度融合。在预测生成阶段，设置时间衰减项，动态预测访问概率，及时更新用户兴趣变化，从而提高推荐质量。在MovieLens数据集上的实验结果表明，与UB-CF、AE和SDAE-IA算法相比，该算法推荐性能明显提高。

关键词: 协同过滤, 自动编码器, 项目属性, 相似度, 时间衰减

Abstract: Traditional collaborative filtering algorithms only rely on low-dimensional vector method of user rating data,which results in low accuracy of recommendation results and cold-start problems.Therefore,this paper proposes a new dynamic hybrid recommendation algorithm,which integrates the stacked denoising Autoencoder(AE) into the user-based collaborative filtering to learn users'deep features,and is integrated with the similarity based on user item attribute preference.In the prediction generation stage,the time attenuation item is set to dynamically predict the accessing probability and update the changes of user interest in time,so as to improve the quality of recommendation.Experimental results on the MovieLens dataset show that the recommendation performance of the algorithm is significantly improved compared with the UB-CF,AE,and SDAE-IA algorithms.

Key words: collaborative filtering, Autoencoder(AE), item attribute, similarity, time attenuation

中图分类号: 

• TP18