Mapping product descriptions to a large ontology
Henrik Oxhammar · DSpace repository (University of Tartu) · 2006
In this paper we describe an information retrieval approach for mapping online business information texts to concepts in a large ontology.We adopt the traditional vector space model by representing the texts as queries and the concept labels in the ontology as documents.Because of the size of the ontology and the fact that concept labels are very sparse and generic, we conducted additional experiments for reducing the set of concepts, as well as the enrichment and enlargement of concept labels.The documents in our collection were of too poor quality for this task, and although we show that our enrichment technique did provide us with an ontology with good overall similarity to our query collection, individual concepts did not include enough terms for our method to achieve good results.18000000 "Clothing and accessories" 18500000 "Leather clothes" 18510000 "Leather clothing accessories" 18512000 "Leather belts and bandoliers" 18512100 "Belts" 18512200 "Bandoliers"The ontology defines 8323 unique concepts of this kind.By a concept's code, it is possible to derive a number of useful facts.First, we can determine at what level the concept is defined.E.g., 18512000 "Leather belts and bandoliers" resides on level five.Leaf concepts, i.e., concepts that have the finest granularity, make up almost 68% (5644) of the total number of concepts.(Warin et al., 2005) The ontology is a strict taxonomy, i.e., concepts are related by the hyponomy/hyperonomy (sub/super) relationships.Therefore, it is pos-