Starting from terminanodes assocated with set of words and followinglinks to parent nodes 2302 useevidence tree to estimate likelihood thateach parent clusterisactive ingenerating set of words

(54) METHOD AND APPARATUS FOR 6,137,911 A 10/2000 Zhilyaev ..................... 382,225 CHARACTERIZING DOCUMENTS BASED 6,651,054 B1 1 1/2003 Judicibus ....................... 707/3 ON CLUSTERS OF RELATED WORDS 6,684,205 B1 1/2004 Modha et al. .... ... TO7.3 6,751,611 B2 6/2004 Krupin .......................... 707/3 75 6.820,093 B2 11/2004 de la Huerga ........... 707/104.1 (75) Inventors: Georges Harik, Mountain View, CA 7,013.298 B1 3/2006 De La Huerga ............... 707/3 (US); Noam M. Shazeer, Mountain View, CA (US)