E-commerce Market Analysis from a Graph-Based Product Classifier

This work describes a system for analyzing e-commerce markets, by tracking the consumers' habits in a set of webstores. In order to retrieve the types of products that are mostly purchased or viewed by a sample of consumers, we extract product titles displayed at the pages browsed by the monitored sample. Then, the collected titles are used as input to an automatic classification process that aims to assign each product to a suitable category and brand. We introduce a graph-based classifier that explores the relation between the categories and brands to be determined and is built from a supervised training set. We will show that the classifier was able to obtain good results, improving a Bayesian technique and allowing a further analysis of the considered market.