IRENA: information retrieval engine based on natural language analysis
Avi Arampatzis, Cornelis H. A. Koster, T. Tsoris · 1997
The experimental IRENA system was developed to study the improvement of precision and recall in document retrieval systems by means of Natural Language Processing (NLP) techniques. The NLP component deals with the morphological, lexical and syntactical part of the English language. For the purpose of syntactical analysis of both queries and documents, the power of the AGFL formalism was explored in describing and developing a syntactical analyzer for the English noun phrase with a large lexicon. The noun phrase co-occurrence hypothesis was formulated and tested as a new relevance criterion in achieving high levels of precision. Furthermore, the problem of calculating recall in non-indexed collections was partially solved by introducing a new measure, relative recall. The system was tested on a large corpus of English language documents concerned with pop music. The results of this experiment are reported, and some conclusions drawn on the viability of the techniques. In the approach taken, all linguistic knowledge is encapsulated in the grammar and the lexicon. The language is merely a parameter. Consequently, the same techniques can also be applied for other, more inflected languages.