Language Model CNN-driven similarity matching and classification for HTML-embedded Product Data

Abstract

The Semantic Web Challenge Mining the Web of HTMLembedded Product Data aims to benchmark current technologies on the data integration tasks (1) product matching and (2) product classification, as recent years have seen significant use of semantic annotations in the e-commerce domain, but often with inconsistencies, no complete coverage or conflicting information. We introduce a transformer-based approach for textual product matching and extend it with an CNN for product classification. We compare the influence of different input feature combinations against prediction performance and introduce a technique to augment the classification task with additional information. We are able to outperform baseline results using text-only approaches.