Carlos M. Lorenzetti

Formal Characterization and Empirical Analysis of Incremental Context-Based Retrieval Methods

PhD in Computer Sciences Thesis - Universidad Nacional del Sur

March 18,  2011

by Carlos M. Lorenzetti

Abstract

The Web has become a potentially infinite information resource, turning into an essential tool for many daily  activities. This resulted in an increase in the amount of information available in users' contexts that is not taken into account by current information retrieval systems. This thesis proposes a semisupervised information retrieval technique that helps users to recover context relevant information. The objective of the proposed technique is to reduce the vocabulary gap existing between the knowledge a user has about a specific topic and the relevant documents available in the Web. This thesis presents a method for learning novel terms associated with a thematic context. This is achieved by identifying those terms that are good descriptors and good discriminators of the user’s current thematic context. In order to evaluate the proposed method, a theoretical framework for the evaluation of search mechanisms was developed. This served as a guide for the implementation of an evaluation framework that allowed to compare the techniques proposed in this thesis with other techniques existing in the literature. The experimental evidence indicates that the methods proposed in this thesis present significant improvements over previously published techniques. In addition the evaluation framework was equipped with novel evaluation metrics that favor the exploration of novel material and incorporates a semantic relationship metric between documents. The algorithms developed in this thesis evolve high quality queries, which have the capability of retrieving results that are relevant to the user context. These results have a positive impact on the way users interact with available resources.

Chapter 1 (only in Spanish) -->

Full Thesis file[pdf, 2.1 MB] (spanish)