递归神经网络下混合属性信息推荐仿真

Recommended simulation of mixed attribute information under a recursive neural network

信息量的大幅增加,导致用户无法从推荐的海量数据中提取到所需的信息。为了解决上述问题,提出一种基于递归神经网络的混合属性信息推荐算法。通过数据预处理方法,删除没有任何信息评分的混合属性信息,并挖掘用户和混合属性信息之间的关系。采用已评分混合属性信息,融合极度梯度提升树(eXtreme Gradient Boosting, XGBoost)算法对混合属性信息分类。构建递归神经网络模型,采用梯度下降法对模型训练,获取用户对各个混合属性信息的概率值,并将其按照从大到小的顺序排列,形成推荐列表直接推送给用户完成推荐。实验结果表明,所提方法的HR值得到了提高,且NDCG取值的平均值为0.805,全面提升推荐结果的准确性。

The substantial increase in the amount of information makes it impossible for users to extract useful in⁃formation from the recommended data.As a result,a recommendation algorithm for mixed attribute information based on a recursive neural network was proposed.At first,the data preprocessing method was adopted to delete the mixed attribute information without any information score,and thus to mine the relationship between users and mixed attribute information.Then,the graded mixed attribute information was combined with the eXtreme Gradient Boosting(XGBoost)algorithm to classify the mixed attribute information.Moreover,a recurrent neural network model was con⁃structed,and then the gradient descent method was adopted to train the model,thus obtaining the probability value of each mixed attribute information.Finally,these values were arranged in order,thus forming a recommendation list that was directly pushed to users.Experimental results show that the HR value is improved,and the mean value of NDCG is 0.805,so the proposed method comprehensively improves the accuracy of the recommendation results.