Selecting a Relevant Set of Examples to Learn IE-Rules

The growing availability of online text has lead to an increase in the use of automatic knowledge acquisition approaches from textual data, as in Information Extraction (IE). Some IE systems use knowledge learned by single-concept learning systems, as sets of IE rules. Most of such systems need both sets of positive and negative examples. However, the manual selection of positive examples can be a very hard task for experts, while automatic methods for selecting negative examples can generate extremely large example sets, in spite of the fact that only a small subset of them is relevant to learn. This paper briefly describes a more portable multi-concept learning system and presents a methodology to select a relevant set of training examples.