| 摘 要: 推荐算法作为推荐引擎实现的核心而得到广泛研究。在各类推荐算法中,大部分对于用户行为特征属 性、用户人口属性、物品特征属性,以及用户—物品关联特征属性等参数的应用方式存在局限性。它们一般采用相似度 计算、或模型计算等方法,其特征提取及参数的调优依赖于事前定义,存在参数优化效率低的问题。本文结合机器学习 技术,提出一种混合推荐算法,即(MMLHC算法),以多层神经网络作为参数优化计算的模型,应用Mahout库实现算 法,实验结果显示算法能有效去除原始输入数据的噪声、奇异点,在模型的各层之间优化权重参数与偏差,输出数据去 噪平滑,正常拟合。相似度与精确度的计算指标良好。 |
| 关键词: 推荐算法;机器学习;多层神经网络;隐藏层;可视层 |
|
中图分类号: TP312
文献标识码: A
|
|
| Implementation of a Hybrid Recommendation Algorithm Based on Mahout |
|
TANK Ke
|
( School of Computer Science, Chengdu College of University of Electronic Science and Technology, Chengdu 611731, China)
kevindoctor@aliyun.com
|
| Abstract: Recommendation algorithms are widely studied as the core of recommendation engine implementation, but most types of the current recommendation algorithms have limitations on the application of user behavior characteristic attribute, user population attribute, item characteristic attribute and user-item relation characteristic attribute. They generally use similarity-based calculation, or model-based calculation methods, and their feature extraction and parameter tuning depends on prior de nition, with low effectiveness of parameter optimization. This paper proposes a hybrid recommendation algorithm of Multilayer Machine Learning Hybrid Compute (MMLHC). Multi-layer neural networks are considered as model of parameter optimization calculation and Apache Mahout implements the algorithms. The experimental results show that the algorithm can effectively remove the noise and singular points of the original input data, optimize the weight parameters and deviations between the layers of the model, and smooth the output data denoising and normal tting. The index of similarity and accuracy is improved. |
| Keywords: recommendation algorithm; machine learning; multi-layer neural network; hidden layer; visual layer |
