【摘要】 互联网信息陡增，导致信息过载，为客户更加精准地推荐商品变得越来越困难．与传统推荐算法相比，基于图神经网络的推荐算法可以更好地提取客户与商品之间的关联关系．但是，在此类算法中潜在特征的乘法内积的简单线性组合无法准确捕获客户交互数据的复杂结构．针对这类问题，提出了基于感知器的图卷积推荐算法，即在利用图神经网络提取关联关系时，使用感知器分别对客户和商品的特征进行提取．使用召回率和归一化折损累计增益作为评价指标，在3组公共数据集中进行了对比实验．实验结果表明，该方法比已有相关算法的效果有所提升.更多还原

【Abstract】 The proliferation of information on the Internet today has led to information overload,making it increasingly difficult to recommend products to customers more accurately.Compared with the traditional recommendation algorithm,the recommendation algorithm based on graph neural network can better extract the relationship between customers and products.However,in such algorithms,simple linear combinations of multiplicative inner products of latent features cannot accurately capture the complex structure of customer interaction data.To solve these problems,a perceptron based graph convolution recommendation algorithm is proposed in this paper.When the graph neural network is used to extract the correlation relationship,the perceptron is used to extract the features of customers and goods respectively.Comparative experiments were conducted in three sets of public data sets,using recall and normalized cumulative loss gain as evaluation indexes.And the experimental results show that the effectiveness of the proposed method is improved compared with the existing algorithms.更多还原