差分进化神经网络集成的用户偏好模型构建

Customer preference model based on differential evolution neural network ensemble

用户偏好模型的构建是推荐成功与否的基础。通过产品特征属性与用户特征属性的映射,建立用户偏好模型,引入神经网络集成的机器学习方法来模拟偏好模型。为了提高用户偏好模型的泛化能力,提出用负相关学习算法并行训练成员神经网络,采用差分进化算法对成员网络进行优化,从而有效降低网络集成的泛化误差,提高模型精度。通过Movielens数据仿真,并与单个BP神经网络、GASEN、核密度神经网络集成等模型实验结果进行对比分析,其均方差明显减少,验证了差分进化神经网络集成的用户偏好模型具有较好的泛化能力,能客观反映用户偏好,从而取得更好的推荐效果。

The construction of customer preference model is the basis of the success of the recommendation. A customer preference model is built by mapping the feature of the product and the customer＇s feature,and the neural network ensemble of machine learning method is introduced to simulate the preference model. In order to improve the customer preference model generalization ability,a negative correlation learning algorithm is proposed to train component neural network parallel,using differential evolution algorithm to optimize the component neural networks,thus effectively reducing generalization error of network ensemble and improve model accuracy. Through Movielens data simulation,comparing with the results of the models such as single BP neural network,GASEN,nuclear density of neural networks,etc.,it shows that it significantly reduces its average variance,verifying the customer preference model of differential evolution neural network ensemble has better generalization ability to objectively reflect customer preferences,to obtain a better recommendation results.