Directed retrieval and extraction of high-quality product specifications

Abstract

In recent years, a large quantity of algorithms has been presented for extracting information from semi-structured sources like HTML pages. Some of them already focus on product information and are adopted, e.g., in online platforms. However, most of those algorithms do not specifically target technical product specifications and never take the localization of such specifications into account. This work focuses on automating the whole process of retrieving and extracting product specifications. It achieves a high data quality by directing the source retrieval to producer pages where product specifications are extracted in an unsupervised manner. The resulting specifications are of high relevance to consumers since they enable effective product comparisons. The success of the developed algorithms is proven by a federated information system called Fedseeko. © 2011 Springer-Verlag Berlin Heidelberg.