En komparativ litteraturstudie av olika termkällor för query expansion
The purpose of this thesis is to make a comparative literary study of ten laboratory experiments with different kind of sources for query expansion terms. The experiments are grouped according to categories of term sources, which are: search results, collection dependent knowledge structures, collection independent knowledge structures and one that combines the two last mentioned sources into one. To enable a comparison all other variables were held as constant as possible. There is the improvement measured in mean average precision, which is used to measure the various sources potential. The result from the study shows a strong connection between the kind of source for the expansion term and improvement of the result. The experiments were structured based on results from A to B, the best result being A followed by B and so forth. The idea with these units is also to show potential common characteristics according to query expansion strategies. Unit A consists of the combined knowledge structure and has shown considerably better result compared to the others. The hypothesis of this experiment was, that different knowledge structures have various characteristics, that together reinforce each other. The experiments in unit B all use the collection as term source, including search result as such, and are also all statistically based. The only experiment using NLP technique and linguistically based measurement between terms, constitutes unit C. Unit D consists of all experiments in which collection independent sources were used for query expansion.