大數據平台上基於顧客喜好和社群行為分析之顧客流失預測

Churn Prediction Based on the Analysis of Customers' Preferences and Social Behavior on a Big Data Platform

由於網際網路和電子商務的應用已發展得很完善，顧客有更多的選擇而使得顧客的忠誠度正逐漸下降。對於企業而言，開發新顧客的成本遠高於保留舊顧客的成本，而顧客的流失也將會對企業造成巨大的影響。因此，企業如何在利用有限資源有效地管理顧客以減少顧客流失對企業的影響，是企業必須面對的重要議題。本研究在一個大數據平台中提出一個應用於社群網路之顧客流失預測的新穎方法，此方法將整合顧客價值分析、顧客喜好、以及社群行為和信任關係之分析等四大因素，再利用分類方法，亦即隨機森林、K-NN和類神經網路，針對這四個因素進行顧客流失預測。實驗結果顯示本研究所提出的顧客流失預測模型在分析大量數據時，能有較好的預測準確度和計算效能。

Because of the well-developed Internet and e-commerce applications, customers have a lot of choices so that customer loyalty is decreasing. For businesses, the cost of developing new customers is higher than the cost of retaining old customers. Losing customers (users) will cause a huge impact on business. Therefore, how a business uses limited resources to manage customers effectively and reduce the impact of customer loss are critical issues. This work will build a big data platform and propose a novel method for predicting customer churn on a social network, which integrates customer value analysis, customer preferences, and the analysis of social behavior and trust relationship. Then, the classification methods, i.e. random forest, K-NN, and neural network, use these four factors to make the customer churn prediction. The experimental results show that the proposed churn prediction model has better prediction accuracy and computation efficiency in analyzing a huge amount of data.