| Resumen |
We argue for that taking into account semantic relations between words in the text can improve information retrieval performance. We implemented the process of information retrieval with simplified Conceptual Graph-like structures and compare the results with those of the vector space model. Our semantic representation, combined with a small simplification of the vector space model, gives better results. In order to build Conceptual Graph-like representation, we have developed a grammar based on the dependency formalism and the standard defined for Conceptual Graphs (CG). We used noun pre-modifiers and noun post-modifiers, as well as verb frames, extracted from VerbNet, as a source of definition of semantic roles. VerbNet was chosen since its definitions of semantic roles have much in common with the CG standard. We experimented on a subset of the ImageClef 2008 collection of titles and annotations of medical images. |
