Den kombinerade effekten av query-expansion och query-strukturer på återvinningseffektiviteten i ett probabilistiskt system
This thesis deals with query formulation in full text retrieval. The variables studied were query expansion and query structur[INS: e :INS , and the aim of the thesis is to study their co-effects on retrieval performance in a probabilistic system. The expansion was made with synonyms, and the structure by expressing facets and phrases. The study was performed using QPA, Query Performance Analyzer, which includes InQuery retrieval system and a sub-collection of TREC documents with its topics. The measurement used was precision at 11 DCV points. The best result was obtained by queries with no expansion and a structure where all terms had equal influence. Any other combinations of expansion and structure used in this study were contraproductive. The results point out that extensive expansion including all kinds of term relations, and a faceted structure, as used in earlier studies, are more favourable than the rather limited expansion and structures used in this study. The possible influence of the fact that all queries included phrases is discussed. The methods to structure phrases and how to create their synonyms are discussed as important variables for further investigations. The importance of how expansion terms are chosen as well as good term sources is also underlined.