融合评分和社会化标签的两阶段深度推荐方法

当前融合评分和标签的推荐方法对两种数据的挖掘程度有限,且大多数局限在提取浅层的线性特征层面.深度学习技术被成功应用于推荐方法,然而数据的稀疏性导致学习的潜在特征效果不好,因此,提出一种融合评分和社会化标签的两阶段深度推荐方法.首先,利用堆叠降噪自编码器分别从评分和社会化标签中提取用户、项目的潜在特征;其次,将学习的潜在特征进行拼接作为用户、项目完整的潜在特征,并与原始评分相结合构建监督学习数据集;最后,将构建的数据集作为BP神经网络的输入以训练评分预测模型.为降低训练误差,通过联合训练的方式进行参数学习.基于MovieLens、Last.FM数据集的实验表明,该方法与几种基准方法相比有更好的推荐性能.

Two-stage deep recommendation method combining rating and social tags

The current recommendation methods that integrate ratings and social tags do not fully mine the features contained in the information, and most of them are limited to extracting shallow linear features. Deep learning technology has been successfully applied into the recommendation methods, but the learned latent factors may not be effective due to the sparsity of information. Thus, this paper proposed a two-stage deep recommendation method that combined rating and social tags. Firstly, it used the stacked denoising autoencoder to extract latent features of users and items from ratings and social tags; secondly, concatenated the learned latent features as the complete latent features of users and items, and combined with the original ratings to construct a supervised learning dataset; finally, used the constructed dataset as the input of the BP neural network to train the rating prediction model. To reduce the training error, it performed parameter learning through joint training. Experiments based on MovieLens and Last. FM datasets show that the proposed method has better recommendation performance than several benchmark methods.