SP500207 Query Improvement in INformation Retrieval Using Genetic Algorithms - A Report on the Experiments of the TREC Project chapter J. Yang R. Korfhage E. Rasmussen National Institute of Standards and Technology Donna K. Harman The parallel search could be continued if the individuals retrieving different relevant documents have the same power, that is, the differences in the number of relevant documents brought by each of them are small. The results in Table 15 and Table 16 come from the same topic in Table 13 and Table 14, but from the second generation (generation 1) after feedback and genetic modification to the first generation. It can be seen that due to the genetic operations, parts of several query individuals have begun converging. At this point the queries group into three clusters ofsimilarqueries, (0,3,7,8,9), (1,5,6) and (2,4)-- Table 15. Within each group several term weights are identical. However, parallel search still continues. Differences in the other weights result in query individuals that retrieve completely different sets of documents. For example, contrast query 7 with the others in its cluster. Other examples from this study show that this effect may be observed with differences in only one or two term weights. Tables 17 and 18 show that the parallel search from this topic went on to the final generation, 3, where for each term almost all of the weights had converged to a single value. But query 7 has two different values at terms 4 and 21 which cause it to retrieve totally different relevant documents than the other query individuals. However, this is not always the case. In most situations, the genetic algorithm will force the query variants to converge to a single query; that is, all of the query individuals in the final generation retrieve the same relevant documents. The example on topic 24 also shows, as in previous cases, that new relevant documents are brought to the user in each generation, as indicated by the `*` following the document numbers. Table 13. Term weights of query individuals of topic 24, the first generation (generation 0) ti t2 t3 t4 t5 t6 t7 t8 t9 tio tlI t12 t13 t14 tls t16 t17 t18 t19t20 t21 0 18 .31 .53 .95 .17 .70.23.49 1 .63 .31 .35 .92.52 .40.61 .79 2.13 .21 .96.74.41 .78.76.96 3 .57.25 .50.24.48 .41 .87.43 4.30.89 .04.65.40 .68.73.94 5.14.29.80.22.56 .72.20.99 6.60.24.45.79.08 .48 .15 .25 7.78.71 .45.70.10 .96.55.74 8 .49.61 .42.13 .26 .04.98.11 9.84.84.90.59.54 .17.65.69 .12.08 .39.28 .37.98 .54.77.65 .77.78 .82.15 .93 .87.87.67 .76.58 .39.36.20.83 .42.46.98 .03.32.76.24.59.04.96 .32.06.44.92.57.12 .36.38 .04.16.52.70.10.40.77 .24.34.23 .30 .23.84.97 .78 .43 .67.81 .16.28 .14.86.75 .21 .25 .43 .76.86 .89.98 .40.43 .13 .46.24.99.65 .94.61 .99.48 .80.74.38 .48.53.10.59.35 .14 .58.64.78.19.30.28.68 .29.57.42.31 .44.57 .38 .65 .35 .55 .36 .57 .48 .16.62.17 .55 .29.87 .26.11 .81 .19.42.35 .84.14.26.18 .48.38 .50 48