Detection of E-Commerce Systems with Sparse Features and Supervised Classification

Enriching web shop pages with structured data has recently become popular in e-commerce. It is mainly driven by search engines favouring those pages. While structured data in e-commerce is mainly generated automatically by shop extensions, this data covers only a small share of the market, resulting in a major hamper for applications operating on aggregated data. In this context, more than 90% of product detail pages on the web are generated by only 7 e-commerce systems. Meanwhile, little research addresses methods to automatically detect e-commerce systems. Automated detection would allow to design system-specific extractors able to grow the amount of structured data in e-commerce. Therefore, we propose a novel approach to this problem, which filters features generated from HTML tag attributes with an e-commerce specific white list. We evaluate 6 classification algorithms on the problem and discuss computational effort. We can show that this approach is capable of detecting the 6 most important e-commerce systems with a F1-score of 0.9 by analyzing only one HTML page per web shop. We evaluate our findings on an independent dataset and on reference shop sites.