图卷积神经网络 在电信运营商潜客挖掘中的应用研究
2021-08-16唐崔巍王琼徐海勇黄岩
唐崔巍 王琼 徐海勇 黄岩



【摘要】 基于运营商5G套餐的精准营销场景,将用户数据、产品数据、访问行为数据等多源异构数据进行图结构转化,文章采用InfoMap算法做图数据节点聚类,使用图卷积神经网络技术对图数据深度计算挖掘,并在群簇节点间通过随机删边技术进行结构优化。结果显示,基于InfoMap算法的图卷积神经网络与随机删边技术相结合的潜客挖掘模型的推荐准确性具有明显提升。在数字化转型时期,该潜客挖掘算法为运营商提供了精准营销、客户管理等方面的新思路。
【关键词】 图卷积神经网络 InfoMap 随机删边技术 潜客挖掘
Research on the Application of Graph Convolution Network in the Telecom OperatorsPotential Customer Mining
TANG Cui-wei, WANG Qiong,XU Hai-yong, HUANG Yan,
(China Mobile Information Technology Co., Ltd., Beijing 100037,China)
Abstract: Based on the 5G package precision marketing scenario, this paper converted users, products and behaviors data into graph-structured data.This paper applied the InfoMap algorithm to cluster nodes of graph data, utilized graph convolution neural network technology to carry out deep computation and mining of graph data, performed random drop edge technology among cluster nodes of the graph data to achieve structural optimization. The results showed that this studys model which was based on InfoMap algorithm combining graph convolution neural network with random drop edge technology model improved the recommend accuracy significantly. In the digital transformation period, the potential customer mining algorithm provides operators with new ideas in precision marketing and customer management.
Keywords: graph convolution neural network; InfoMap; dropedge; potential customers mining
引言:
随着 “5G+工业互联网”的应用场景迅速发展,企业数字化、智能化转型正在加速形成,运营商的业务更趋于互联网化和个性化,深度把握市场需求,为客户创造价值对运营商来说至关重要。业务推荐主要是通过指标分段或以存量标签为规则筛选目标用户,再将配置好的商品名称、商品图片、商品价格、优惠方式以及商品链接进行推送,然而这种推荐模式存在定位人群不精准,推送商品内容与目标人群不匹配等问题,对用户信息数据的利用率仍待提升。因此,提升运营商潜客挖掘的匹配度和准确率愈发重要。本文基于运营商线上用户群体,将推荐引擎与5G套餐推荐场景相结合,采用GCN图挖掘算法进行5G套餐的用户订购行为和用户特征挖掘分析,深度训练优化5G套餐订购潜客挖掘模型,挖掘与现有订购5G套餐用户的特征相似用户,从而输出更精准有效的潜在客户名单,实现用户数据价值充分挖掘。……
