Searching for Semantic Relations between Named Entities in I-CAB.
Fabio Celli · 2009
This paper describes the experiments for the extraction from I-CAB of three different semantic realtion types (role, social, location) holding between two named entities in Italian. The method adopted exploits the information about the part-of-speech of the context before and between the two entities. This method, called Part-Of-Speech Context Count (POSCo) obtained an F1-measure of 0.812 using simply decision trees and decision lists algorithms. The rules extracted from this experiment are implemented in a semantic relation annotator (called RA) compatible with the TextPro tool suite. 1 Introduction and related work I-CAB [6] (Italian Content Annotation Bank) is a Italian corpus made of 525 news documents taken from a local newspaper named ”LAdige”, annotated with temporal expressions and 4 named entity types (person, organization, location and geo-political entity), the same entity types can be detected automatically