Enabling Product Comparisons on Unstructured Information Using Ontology Matching

Information extraction approaches are heavily used to gather product information on the Web, especially focusing on technical product specifications. If requesting different sources for retrieving such specifications, the outcome is of varying formats (different languages, units, etc.). The problem of how to bring such information sets into a unique, interchangeable format is not considered in many extraction systems. We develop a generic process for semantically integrating heterogeneous product specifications with the help of a product information ontology. The approach is based on a number of measures for detecting the right product attributes in the ontology to be matched with the extracted specifications and finally normalizing the specifications’ values (e.g., concerning units). The feasibility of our approach is proven in a federated product search prototype called Fedseeko.