一种基于混合深度学习的推荐算法

为解决推荐算法中的冷启动和数据稀疏性问题，提高推荐的效率，提出一种基于混合深度学习的推荐算法。该算法将深度学习中的半自动编码器和多层感知机模型有效结合，通过半自动编码器模型解决稀疏数据，并融合相关辅助信息解决冷启动问题。它先用半自动编码器提取用户和项目的深层次特征，再将提取的潜在特征输入到多层感知机中进行非线性融合，完成评分预测。实验结果表明，相比于其他算法，该算法可以更好地处理稀疏数据和冷启动问题，使推荐准确性得到不同程度的提升。在给定数据集上，该算法比传统矩阵分解算法的均方根误差提升了约46%。

A Hybrid Deep Learning Algorithm for Recommendation System

In order to improve cold start and data sparsity in the recommendation algorithm and enhance the recommendation effect,this paper proposes a hybrid deep learning based recommendation algorithm HSAEM (hybrid semi autoencoder and multilayer perceptron).It effectively combines the semi automatic encoder in deep learning with the multilayer perceptron model.It improves the sparse data through the semi automatic encoder model,solves the cold start problem by fusing auxiliary information,extracts the deep features of users and projects,and sends the extracted potential characteristics to the multilayer perceptron for nonlinear fusion,thus implementing the score predicts.Results show that HSAEM helps raise accuracy to various degrees by more effective dealing with cold start and data sparsity,its RMSE standing 4.6% more accurate than the traditional matrix factorization algorithm on a given data set.