SP500207 Overview of the First Text REtrieval Conference (TREC-1) chapter D. Harman National Institute of Standards and Technology Donna K. Harman 0.9 0.8 0.7 0.6 0 0.5 [OCRerr] 0.4 0.3 0.2 0.1 0 Best Automatic Adhoc 0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 Recall flilirpi crnlpl siemsl 1.00 Figure 5. The Best Adhoc Results using Automatic Query Construction. Figure 6 shows the recall/precision curve for the three ThEC-1 runs with the highest 1 1-point averages using manual construction of queries. It should be noted that varying amounts of manual intervention were used, and this should be considered when comparing results. These curves show differences in that the "clartb" and "gecrd2" have initially a high precision, but lose this precision as recall increases, whereas the 91cnqs12" method has a lower initial precision, but higher precision aL the higher recall levels. This may be a function of the very different methods being used. The "clartb" system adds noun phrases found in likely relevant documents to improve the query terms taken from the topic (see Evans paper), whereas the "cnqse" system uses more general thesaurus entries to expand the query (see Nelson paper). The `3gecrd2" system uses a totally different approach of constructing elaborate Boolean pattern matchers (see Jacobs, Krupka & Rau paper). Best Manual Acihoc 0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00 Recall 0.9 0.8 0.7 0.6 0 [OCRerr] 0.5 0.4 0.3 0.2 0.1 0 cnqs[OCRerr] clartb gecrd2 Figure 6. The Best Adhoc Results using Manual Query Construction. 14