Application of Logical Analysis of Data to the TREC-6 Routing Task.
Endre Boros, Paul B. Kantor, Jung Jin Lee, Kwong Bor Ng, Di Zhao · 1997
mples is given a weight determined by the number of examples that it matches. 2 Detailed Procedures of the Official Runs 2.1 Training For each topic, we used MG to index all the judged relevant documents to build a index structure, and to compute the term frequencies and document frequencies of all word-stems. We S Permanent address: Department of Statistics, Soong Sil University, Seoul, Korea. selected 25 word-stems according to the Church criterion (Church & Gale, 1995) on distributions of term frequencies and document frequencies. We did the same for the judged non-relevant documents. For topics with more than 50 Mbytes of judged non relevant documents we randomly selected 50% of the judged documents for MG to index. (Topics: 77, 78, 82, 94, 95, 100, 108, 118, 119, 123, 125, 126, 128, 142, 161, 173, 187, 194, 228, 240, 282 ). This yields 25 word-stems from relevant documents and 25 word-stems from non relevant docu