递归神经网络下多属性信息模糊推荐仿真

Simulation of Multi-Attribute Information under Recurrent Neural Network

由于海量信息资源的搜索需要较高的时间成本,因此信息推荐已经成为解决当前网络资源交互环境信息过载问题的关键方法。但是多属性信息数据结构复杂,单属性数据推荐算法难以取得理想的精度和结果。考虑用户信任信息形式和传播路径,提出基于递归神经网络的多属性信息模糊推荐算法。利用递归神经网络结构,存储用户历史兴趣信息。通过用户对项目的感兴趣程度建立评分矩阵,利用余弦相似性的思想计算项目属性相似性、综合相似性以及用户相似性。引入权值因子,将综合相似性与用户相似性的预测评分线性整合,得到多属性信息推荐列表,选取top-N推荐给用户。实验测试结果显示,提出方法下列表内相似性指标值ILS始终小于0.3,推荐精准度高于90%,平均绝对误差可控制在0.2以下。以上所得数据均可证明所提方法可综合考虑目标的多属性特征,为用户推荐感兴趣程度最高的对象。

At present,information recommendation has become the key method to solve the problem of information overload in current network resource interaction.However,the data structure of multi-attribute information is complex,so the single-attribute data recommendation algorithm is difficult to achieve ideal accuracy and results.With consideration of the form and propagation path of user trust information,this paper presented a fuzzy recommendation algorithm for multi-attribute information based on a recursive neural network.Firstly,the recursive neural network was used to store the historical interest information.Then,a scoring matrix was established by the user's interest in the project.Based on the cosine similarity,the attribute similarity,comprehensive similarity and user similarity between projects were calculated.Moreover,the weight factor was introduced to linearly integrate the prediction scores of comprehensive similarity and user similarity and thus to obtain a multi-attribute information recommendation list.Finally,top-N was selected to recommend to users.Experimental results show that the ILS of similarity index is always less than 0.3,and the recommended accuracy is higher than 90%.Meanwhile,the average absolute error can be controlled within 0.2.The above data prove that the proposed method can comprehensively consider the multi-attribute features of the target and recommend the object with the highest degree of interest for users.