MTI for Full Text

To provide a stable base for the experiments with full text, the MTI indexing paths were run separately on each of the sections of the full text test collection. The output from each indexing path was saved and subsequently used by MTI for all of the experimental processing. The evaluation in the phase 1 experiments reported in the AMIA paper was based on the human indexing extracted from MEDLINE in December of 2003. Since new work planned for phase 2 would change the way the text from the articles was separated into sections new processing by the indexing paths in MTI was necesary. Since this processing uses the current PubMed database, the processing and evaluation must be done in the current environment. Consequently, moving our experiments to the new environment required establishing new baselines. In addition, the baseline is based on the current production version of MTI, so we prepared an updated version of the experimental, section-handling version of MTI to allow a valid comparison of their results. Finally, to support a current evaluation we extracted from Medline the indexing for the test collection articles to serve at the gold standard. Using that environment new performance baselines were established using the current production version of MTI and the updated, experimental version of MTI.