SP500207 Text Retrieval with the TRW Fast Data Finder chapter M. Mettler National Institute of Standards and Technology Donna K. Harman the query, we would have had 135 good documents out of 262. Turning in only 200, wed expect to have around 103 relevant, which would have been near the high score. Table III - Topic 36 Results as a Function of the Number of Technical Terms Num Tech Rel Docs Terms Ret Ret Prec Recall 10 26 31 0.84 0.17 9 34 42 0.81 0.22 8 41 59 0.69 0.26 7 50 77 0.65 0.32 6 58 100 0.58 0.37 5 77 139 0.55 0.55 4 96 176 0.55 0.62 3 111 210 0.53 0.71 2 120 222 0.54 0.77 1 131 240 0.55 0[OCRerr]. 84 0 135 262 0.52 0.87 5.0 Future Plans During 1993 we hope to continue researching and evaluating better methods for query construction. Our objectives will be: * Design and test a method of sequencing the execution of FDF queries to insure that 200 documents will be retrieved for each topic, * Develop methods and algorithms to semi-automate manual query construction, * Use the extensive relevance judgements from TREC-I to test techniques to generate FDF queries from statistical analysis of the relevant documents for each topic, and * Examine the feasibility of using the FDF's term weighting capability to allow it to act as a back-end processor for other text retrieval techniques. 6.0 Acknowledgments The FDF system is the result of extensive development by many people over the last 8 years. My role has been that of a reporter on the basic system's capabilities and the manner in which they might be applied to a TREC-like problem. 317