基于深度学习的智能学习资源推荐算法

Intelligence learning resource recommendation algorithm based on deep learning

为了提高资源推荐性能,采用广义回归神经网络完成资源推荐。首先,提取推荐系统的用户和资源特征,选择两者的特征差异值之和作为推荐系统目标函数,然后构建广义回归神经网络(Generalized regression neural network,GRNN)资源推荐模型。考虑到GRNN训练效果对平滑因子和核函数中心的依赖性强的特点,引入差分进化(Differential evolution,DE)算法对GRNN的平滑因子和核函数中心偏移因子进行优化求解:选择最小特征差异值求解函数作为DE算法适应度函数,通过DE算法的多次交叉、变异和选择操作,获得最优平滑因子和偏移因子。最后采用优化后的平滑因子和偏移因子进行GRNN资源推荐,生成特征差异较小的候选资源序列作为资源推荐序列。试验证明,选择合理的DE算法交叉速率和差分缩放因子,能够获得较好的平滑因子和偏移因子,GRNN也能够获得更好的推荐效果。和常用资源推荐算法比较,对于3种不同的训练样本,该文算法能够获得更优的资源推荐准确率,且RMSE值较低。

In order to improve the performance of resource recommendation,generalized regression neural network is used to complete resource recommendation.Firstly,the features of users and resources in the recommendation system are extracted,and the sum of their feature differences is selected as the objective function of the recommendation system.Then,a generalized regression neural network(GRNN)resource recommendation model is constructed.Considering the strong dependence of GRNN training effect on smoothing factor and kernel function center,differential evolution(DE)algorithm is introduced to optimize the smoothing factor and kernel function center offset factor of GRNN.The minimum feature difference value solving function is selected as the fitness function of DE algorithm,and the optimal smoothing factor and offset factor are obtained through multiple crossover,mutation and selection operations of DE algorithm.Finally,the optimized smoothing factor and offset factor are used for GRNN resource recommendation,and the candidate resource sequence with small feature difference is generated as the resource recommendation sequence.Experiments show that a reasonable cross rate and differential scaling factor of DE algorithm can obtain better smoothing factor and offset factor,and GRNN can obtain better recommendation effect.Compared with common resource recommendation algorithms,it can obtain better resource recommendation accuracy for three different training samples,and the RMSE value is lower.