CRANV2 Supplementary tests and results chapter Cyril Cleverdon Michael Keen Cranfield An investigation supported by a grant to Aslib by the National Science Foundation. Use, reproduction, or publication, in whole or in part, is permitted for any purpose of the United States Government. - 223 - Document Relevance In Chapter 4, Section 6, the effect of relevance was considered, and the results were presented and plotted for documents of'different grades of relevance according to the coordination level cut-off. Fig. 6.3T shows the same results as are given in Figs. 4.610T to 4.613T, but now grouped accord- ing to relevance grade for each coordination level. Coordination Relevance Recall Precision Level Grade Ratio Ratio 1 1 94.7 0.3 1 -2 94.2 0.5 1-3 93.2 0.9 1 -4 94.2 1.1 2 1 85.3 0.7 1 -2 80.6 1.0 1 -3 79.1 2.0 t -4 77.8 2.4 3 1 60.0 1.2 1 -2 56.1 1.8 1-3 54.5 3.4 1-4 48.8 3.6 4 1 42.1 2.2 1 -2 37.4 3.2 1 -3 32.7 5.3 1-4 29.6 5.8 5 1 25.3 3.1 1 -2 21.3 4.3 1-3 16.5 6.4 1-4 16.3 7.6 6 1 14.7 4.5 1-2 13.5 6.9 1-3 9.8 9.2 1-4 9.7 11.4 7 1 7.4 7.1 1-2 6.5 10.2 1 -3 5.3 t 6.2 1-4 5.3 19.2 8 1 3.2 14.3 1-2 3.9 25.0 1-3 2.0 25.0 1 -4 1.9 29.2 FIGURE 6.3T RESULTS FOR INDEX LANGUAGE I. 1.a FOR 42 QUESTIONS WITH 1400 DOCUMENTS-FOR FOUR GRADES OF RELEVANCE. Plotted as a series of short graphs in Fig. 6.3P, these illustrate yet again the inverse relationship of recall and precision.