SP500207 Classification Trees for Document Routing, A Report on the TREC Experiment chapter R. Tong A. Winkler P. Gage National Institute of Standards and Technology Donna K. Harman Table 1: Performance on Baseline Experiment Rel-Ret @ 200 To pic# #Rel adsbal Max Median Min 5 55 4 45 29 4 6 68 4 46 20 4 7 92 1 46 30 1 8 133 4 37 16 2 9 157 13 41 29 13 10 149 94 109 88 46 11 61 15 55 26 9 12 82 14 56 15 3 13 93 7 93 26 7 14 156 38 73 s2 23 15 515 29 74 49 23 16 58 2 44 17 2 17 69 23 53 23 9 18 95 38 49 38. 14 19 664 74 147 99 56 20 274 111 179 121 56 21 16 12 16 14 0 22 106 28 79 40 8 23 30 2 27 7 2 24 253 37 96 41 29 25 13 1 12 9 1 Overall performance of the baseline case is mixed and our analysis any obvious correlation between performance and factors such as: does not reveal * the number of features extracted from the information need statements (the maximum number was 161, the minimum was 17, with the median being 39), * the complexity of the topics (some are straightforward-such as Topic 13 "Mitsubishi Heavy Industries Ltd.", whereas others involve complex con- difionals-such as Topic 1 [OCRerr]`Anh trust Cases Pending"), 5. We return to this point in the final section of the paper. 216