recently,　some　researchers　have　found　that　the　abounding　search　engines　cannot　support　exploratory　search　effectively.　In　such　case,　it　requires　the　search　engines　know　better　about　the　imprecise　queries　provided　by　the　end　users.　Actually,　it＇s　hard　for　the　users　to　formulate　the　queries,　not　alone　understand　by　the　engines.　However,　in　our　study,　we　find　that　the　search　logs　in　the　web　community　are　significantly　beneficial　for　the　exploratory　search,　because　various　interests＇　communities　exist　in　the　large-scale　web　community.　In　this　paper,　we　treat　the　search　logs　in　the　distributed　search　servers　as　footprints.　And　with　these　footprints,　this　study　proposes　an　adaptive　method　to　support　effective　searching　over　large-scale　web　documents.　In　the　adaptive　method,　logistic　regression　with　trust　region　applied　in　Map-reduce　environment　is　devised,　and　it　processes　the　large-scale　web　documents　in　parallel.　Thus　it　can　effectively　classify　the　web　documents　queried　by　the　dynamic　imprecise　searching.　With　these　results,　the　method　organizes　them　with　frequency　tree　which　shares　redundant　contents　and　records　the　counting　for　ranking.　Extensive　experiments　demonstrate　the　merits　of　our　adaptive　method　to　support　the　exploratory　search.　©2010　IEEE.