Applying proper noun detection in Spanish for information retrieval

Abstract

In this work an automatic method for proper noun detection in Spanish documents is presented. The objective is to check if it can be applied to improve the retrieval performance. A priori we assume that an indexing process that incorporates more information of the document also provides better retrieval results for classical information retrieval. But our results show that this is not true. A lot of tests were carried out to obtain the best performance for all situations: single proper nouns, compound proper nouns, compound proper nouns plus single proper nouns, different weighting schema for proper nouns, etc. The results were discouraging, as the retrieval performance was deteriorated in all the tests. The worst case is detecting compound proper nouns. The effect is less dramatic if the single nouns of the compound ones are considered.