Traditional　information　retrieval　technologies　cannot　provide　proper　search　results　for　people　with　different　interests　and　background　in　different　time.　WebSifter,　an　assistant　system　for　personal　web　search,　extracts　users＇　information　to　integrate,　filter　and　rearrange　the　search　results.　The　system　includes　modules　of　information　collection,　behavior　analysis,　interest　set　construction,　and　result　production.　It　gains　the　users＇　interests　by　the　implicit　methods　and　explicit　methods,　takes　linear　regression　model　to　analyze　users＇　behaviors,　and　uses　a　multilevel　model　to　describe　users＇　interests.　WebSifter　can　dynamically　learn　the　relationship　between　users＇　behaviors　and　interests,　and　it　can　effectively　integrate　users＇　current　interests　and　past　interests　to　deal　with　the　search　results.