A quantitative evaluation of the enhanced topic based vector space model

En savoir plus sur le livre

This contribution presents a quantitative evaluation procedure for Information Retrieval models, specifically applied to the enhanced Topic-based Vector Space Model (eTVSM). As an ontology-based model, its effectiveness is closely tied to the quality of the underlying ontology. The evaluation involved testing the eTVSM with various ontologies to assess their impact on effectiveness. Key findings include: first, the eTVSM's effectiveness is comparable to that of the classic Vector Space Model when a trivial ontology is used, where every term acts as an independent concept. Second, the effectiveness of the eTVSM improves with an ontology that resolves synonyms, which was automatically derived from the WordNet ontology. Third, more complex ontologies derived from WordNet significantly decreased the eTVSM's effectiveness, even falling below the classic Vector Space Model. Lastly, a manually created and optimized ontology enhanced the eTVSM's effectiveness, surpassing the best levels reported in literature for the Latent Semantic Index model with similar document sets.