Information extraction and classification from free text using a neural approach

Many approaches to Information Extraction (IE) have been proposed in literature capable of finding and extract specific facts in relatively unstructured documents. Their application in a large information space makes data ready for post-processing which is crucial to many context such as Web mining and searching tools. This paper proposes a new IE strategy, based on symbolic and neural techniques, and tests it experimentally within the price comparison service domain. In particular the strategy seeks to locate a set of atomic elements in free text which is preliminarily extracted from web documents and subsequently classify them assigning a class label representing a specific product.