基于随机森林模型的电信运营商外呼推荐系统

在电信运营商领域,外呼推荐是一种重要的推荐产品和服务的途径。实现了一种基于运营商大数据的自动外呼推荐系统,该系统能够挖掘用户的行为特征并且使用机器学习的方法预测用户对于被推荐产品的接受可能性。传统推荐系统使用的模型算法为矩阵分解、大规模稀疏特征分类、神经网络等。采用随机森林算法的主要原因是随机森林具有并行化程度高、训练速度快、生成的决策树可解释等诸多优点,适合于基于电信业数据的推荐系统。该外呼推荐系统基于Hadoop、Impala和Spark等大数据处理平台及工具,使用随机森林分类器作为核心算法,将用户最近的行为特征回归为接受外呼推荐产品的可能性。在线测试表明使用该系统与当前部署的人工随机外呼相比,能够提升约41%的用户接受率；同时,根据模型算法输出特征的重要性,进一步给出了两类用户的特征分析。

Random Forest Based Telco Out-calling Recommendation System

Out-calling recommendation is widely used in recommending products and services to customers by telecommunication (telco) operators.In this paper,we developed an automatic out-calling recommendation system which relies on telco big data.This system uses data-mining methods to extract customer behaviors and machine-learning algorithm to predict the acceptance probabilities when customers are recommended to certain products.Different from most recommendation systems which use matrix factorization (MF),sparse features classification,neural network and etc,this paper used random forest,not only because the algorithm is easy to be parallel implemented and has fast training speed,but also the rules from the resulting decision trees are easy to explain.These characteristics make the random forest suitable for telco recommendation system.Our system is implemented on the top of Hadoop,Impala and Spark.Random forest is used as the core algorithm to calculate the acceptance probability when a user is recommended to a product based on user behavior features.Online testing shows that the proposed system can achieve 41% improvement compared with the current deployed random out-calling recommendation method.We also gave the customer behavior analysis according to the feature importance from the outputs of random forest.