基于用户信息融合的图卷积网络报表推荐算法

Research on Report Recommendation Algorithm Based on Graph Convolution Network and User Information

传统报表工具无法自主为企业用户推荐相关业务信息,这一缺陷给电力企业经营管理带来了诸多挑战,针对该问题,提出了基于用户信息融合的图卷积网络报表推荐算法。首先,介绍了以数据中台为底座的报表工具整体架构。在此基础上,以营销业务为例,分析了用户与指标间的关联关系,并提出了基于图卷积网络的关联特征提取模型,同时在模型中融合了用户和指标的一般偏好特征,进一步提升了图聚合信息的深度,进而准确预测指标得分,并给出推荐结果。最后,在公开数据集和营销业务数据集上分别进行模型对比,并选取平均准确率、召回率和归一化折损累计增益作为评价指标,验证所提算法的准确性。结果表明,所提算法与现有算法相比,推荐效果有较大提升,可赋能电力企业经营管理,助力企业数字化转型。

Traditional reporting tools cannot recommend relevant business information to users independently.This has created many challenges in the operation and management of electric power enterprises.To address the above problems,this study proposes a graph convolutional network(GCN)report recommendation algorithm based on user information.First,the overall structure of the reporting tool based on the data center platform is introduced.Considering the marketing business as an example,the correlation between users and indicators is analyzed and a correlation feature extraction model based on a graph convolutional network is proposed.Simultaneously,the general preference characteristics of users and indicators are integrated into the model,and the depth of the graph aggregation information is further improved.Subsequently,the index score is accurately predicted,and a recommendation result is provided.Finally,the models are compared for a public dataset and marketing business dataset,and the average precision rate,recall rate,and normalized discounted cumulative gain are selected as evaluation indicators to verify the accuracy of the proposed algorithm.The results show that compared with existing algorithms,the proposed algorithm significantly improves the recommendation effect,which can empower the operation and management of electric power enterprises and help realize digital transformation.