An automatic feedback thesaurus approach and its parallel implementations

Both search-service users and developers are facing the word-mismatch problem during Internet Web search. As solutions developed in the traditional Information Retrieval environment are not suitable, a technique called “Automatic Feedback Thesaurus” (AFT) is introduced, implemented and tested in the present study. The development of AFT gives insights in designing a new type of user-oriented thesaurus called “User Preference Model”. Both sequential and pipelined parallel implementations of this model were proposed. The possibility of using this thesaurus model on documents was also briefly discussed.; Parallel Processing was used to compensate for the high complexity of AFT. Two workstation clusters were built. One cluster consisted of out-of-date workstations reflecting the goal of combining retired equipment into valuable computing resources. Another cluster was constructed by using recent personal computers in order to achieve a high performance/cost ratio. Before these two workstation clusters were built, a parallel information retrieval system called “PAIRS” was implemented on a transputer system. Valuable experience in constructing parallel computers and implementing parallel systems was gained through development and performance analysis of PAIRS.; Methods in analyzing performance of parallel systems were also discussed and traditional technique was fine-tuned to become more accurate. A new concept called “Expansion Potential” (EP) is introduced in the present study. This concept was used on the performance analysis of the two workstation clusters built for AFT.