由于自编码神经网络能够提取数据从低层到高层的特征,发现样本间潜在的相关性,为了提高推荐系统的精确度提出一种基于降噪自编码的推荐算法。首先利用ZCA白化对评分数据进行预处理,对处理后的数据加入随机噪声并构建自编码神经网络模型,再对模型进行预训练和微调得出网络权重,最后根据训练的网络权重对测试样本进行重构,预测用户评分并计算评分误差。实验结果表明,基于降噪自编码神经网络能有效提高推荐精度。
As the autoencoders neural network can extract the features of data from low-level to high-level, and find the potential correlation between the samples, in order to improve the accuracy of the recommendation system, this paper proposes a recommendation algorithm based on denoising autoencoders. Firstly, this paper pretreats score data by ZCA whitening, adds the random noise in the processed data, and builds the autoencoders neural network, then obtains the network weights through the model pretraining, fine-tuning. Finally according to the trained network weights the test sample score is reconstructed to forcast user rating and calculate score error. Experimental results show that denoising autoencoders can effectively improve the recommendation accuracy.