Identifying Potential Customers On Microblog Platforms Using Text Classification Techniques

The interactive ways between companies and customers, and the ways customerscommunicate with each other have been changing dramatically with the rapid rise of avariety of social media, e.g., Microblog, Facebook and YouTube. Customers havebeen provided with many options to actively participate in the market. Compared totraditional media, the social media brings itself with many advantageous features suchas rapid dissemination, great interactive interface and real-time message sharing.These features, if being made full use of, can help companies improve their brandimages and awareness, and further enlarge their market shares. Microblog ischaracterized by a huge number of users, fast message propagation and a broad rangeof influence, so that microblog marketing has been made as one of the most importantparts of social media marketing in many companies. Identifying potential customers isessentially the foundation of microblog marketing.The most vital and fundamental problem in potential customer mining is how todenote the characteristics of customers effectively, which exerts critical effects on theperformance of potential customer mining. Recently there has been very little researchon this issue. Related work mainly focused on utilizing the demographic informationand microblog usage behavior of customers to describe the customers’ characteristics.This kind of methods requires complicated operations. Moreover, the accuracy ofclassifying potential customers is somewhat low due to insufficiently accuratecharacteristics description of customers (the best accuracy is about76%).In this study, we assume the preferences of the customers’ friends are of greatimportance in the characteristics description of customers, and aim to exploit theeffects of these preferences on identifying potential customers. Under this assumption,we propose a method to generate the textual descriptions of customers’ characteristicsfrom the personal and social relationship perspectives, via taking advantage of theself-defined tags of customers and their friends on microblog platforms. We furtherpropose a framework of mining potential customers by using text classificationtechniques.Extensive experiments have been conducted to evaluate our proposed methods.The results show that our proposed textual descriptions of customers’ characteristicsenable text classifiers to obtain the accuracy of classifying potential customers atabout86%on average. Various text classifiers, i.e., K Nearest Neighbors (KNN), Naive Bayes (NB), Rocchio, centroid-based Classification (Centroid) and SupportVector Machines (SVM), achieve high potential customer classification accuracy,which validate the effectiveness of the proposed framework. Among these fiveclassifiers, SVM obtains the best classification accuracy yet requires a large amountof time in modeling and classification. An excellent trade-off between effectivenessand efficiency goes to NB, and then coming with Rocchio and Centroid.With the aid of the fruitful social relationships existed on microblog platforms,the emerging of the preferences of customers’ friends into customer descriptions cannot only provide potential customer mining on microblog platforms with a newperspective and method, but also present important references to many classic CRM(customer relationship management) problems such as customer segmentation andcustomer churn.