基于SDAE及极限学习机模型的协同过滤应用研究

发布时间：2018-06-14 18:35

本文选题：推荐系统 + 协同过滤　；参考：《计算机应用研究》2017年08期

【摘要】：鉴于传统的协同过滤推荐算法在处理冷启动和数据较稀疏的问题上表现不佳,提出一种将堆栈降噪自编码器(stacked denoising autoencodes,SDAE)与最近邻推荐方法相结合的混合SDAE推荐模型。使用逐层自编码的思想将极限学习机与降噪自编码器堆叠形成基于极限学习机(extreme learning machine,ELM)计算的堆栈降噪自编码器的深度学习模型,最终用模型提取的抽象特征应用于最近邻算法预测打分。通过多组数据集上各种模型的实验结果表明,在稀疏度低于8%时,与余弦相似度模型和皮尔森相似度模型相比,混合SDAE推荐模型实验效果分别提高了11.3%和21.1%;与潜在矩阵分解模型相比,混合SDAE模型收敛所需的迭代次数少近30%;而在与相似度模型和矩阵分解模型的三组比较实验中,混合SDAE模型的稳定性也表现最良好,所提出的混合SDAE模型收敛速度较快,并有效解决了数据稀疏与冷启动的问题。
[Abstract]:In view of the poor performance of the traditional collaborative filtering recommendation algorithm in dealing with cold start and sparse data, a hybrid SDAE recommendation model combining stack de-noising denoising autoencodesSDAE and nearest neighbor recommendation method is proposed. The idea of layer by layer self-coding is used to stack the extreme learning machine and the noise reduction self-encoder to form a depth learning model of the stack noise reduction self-encoder based on the extreme learning machine ELM. Finally, the abstract features extracted by the model are applied to the prediction scoring of nearest neighbor algorithm. The experimental results show that when the sparsity is lower than 8, it is compared with the cosine similarity model and Pearson similarity model. The experimental results of the hybrid SDAE recommended model were improved by 11.3% and 21.1g, respectively. Compared with the potential matrix decomposition model, the hybrid SDAE model required nearly 30 fewer iterations to converge, while in the three groups of experiments, it was compared with the similarity model and the matrix decomposition model. The stability of the hybrid SDAE model is also the best. The proposed hybrid SDAE model converges faster and effectively solves the problems of data sparsity and cold start.
【作者单位】：武汉理工大学计算机科学与技术学院;
【基金】：国家自然科学基金资助项目(21376185) 湖北省技术创新重大专项基金资助项目(2016AAA058)
【分类号】：TP391.3

【相似文献】