1.5 Main Contributions

The following summarizes the main contributions of this thesis, which adds both theoretical insights and practical contributions to the body of existing work in the field.

Novel relevance feedback methods — We develop two query modeling methods for relevance feedback that are based on leveraging the similarity between feedback documents and the set thereof.
Comparison of relevance feedback methods — We provide a comprehensive analysis, evaluation, comparison, and discussion (in both theoretical and practical terms) of our novel and various other core models for query modeling using relevance feedback.
Concept-based query modeling — We develop a way of using document-level annotations to improve end-to-end retrieval performance. Our model naturally generates concept models, which may serve to support, for example, interaction tools for users or which can be used to determine semantic similarity between concepts using the language observed in the documents associated with the concepts.
Novel method for linking queries to concept languages — We develop and evaluate a novel way of associating concepts with queries that effectively handles arbitrary features. For example, features pertaining to the query, concepts, search history, etc.
Understanding of relevant features for concept identification in queries — We provide insights why some (groups of features) perform better than others in the context of linking queries to concepts.
Wikipedia-based query modeling — We show that using the linked concepts can be effectively used to improve diversity and ad hoc retrieval effectiveness on two large test collections.
State of the art retrieval effectiveness — Through extensive experimental evaluations on various test collections (including those from the biomedical, web, social science, and news domains) we validate and analyze our proposed models. In most cases we show consistent and significant improvements over established and state-of-the-art methods on ad hoc retrieval.