Correlation between Private Message Content and Success of Transaction in Indonesian Online C2C Marketplace

Abstract

Company X, a growing C2C marketplace e-commerce in Indonesia, expects to
reduce their transaction refund. This applied research analyzed the message contents of sellers
and found a pattern in the contents with correlation with the sellers' transaction refund. Naive
Bayes text classifiers are built to classify the sample sellers' messages into several categories,
and based on the classification results a data set of sellers and their messages are built. This data
set is analyzed to acquire a pattern. This research concludes that sellers with low refund rate and
high refund rate differs in the composition of their messages, mainly when talking about item
details and stock. Those differences might open a path towards a better recommendation system
in C2C e-commerce system, by identifying bad sellers before they made any transactions.
Keywords: , ,