Optimising Search Engines Using Evolutionally Adapted Language Models in Typed Dependency Parses

Abstract

In this paper, an approach to automatic optimisation of the retrieval quality of search engines using a language model paradigm is presented. The topics of information retrieval (IR) and natural language processing (NLP) have already been investigated. However, most of the approaches were focused on learning retrieval functions from existing examples and pre-set feature lists. Others used surface statistics in the form of n-grams or efficient parse tree utilisations – either performs poorly with a language open to changes. Intuitively, an IR system should present relevant documents high in its ranking, with less relevant following below. To accomplish that, semantics/ontologies, usage of grammatical information and document structure analysis were researched. An evolutionary enrichment of language model for typed dependency analysis acquired from documents and queries can adapt the system to the texts encountered. Futhermore, the results in controlled experiments verify the possibility of outperforming existing approaches in terms of retrieval quality.