Knowing Your Customer from Online Chat Log Analysis

This research took online chat log of 1000 customers who were interested in cosmetics. It uses the online chat conversation topics that customers asked during a decision-making process when they contacted sellers in online shopping. Data mining techniques are used to discover knowledge for further marketing planning. The analysis is divided into two parts; the first part creates a model to predict whether the customer will buy the product or not, using the classification of the topics of the chat log. With experiments using Naive Bayes, Decision Tree and Support Vector Machines (SVM) algorithms, the result shows that SVM outperformed other algorithms with the accuracy of 86.33%. The second part is a study of the proportion of customer groups using behavior cycle data with K-means clustering method. The result found that, there are four groups of customer in this data, i.e., group 1: need information, wait and buy 40%, group 2: need information and buy 31%, group 3: ask information but do not buy 18%, and group 4: ask information, buy and continue to buy other products 11%.