To　ensure　the　sustainable　development　of　the　Internet　community,　satisfy　the　demand　for　the　network　user＇s　access,　this　paper　introduces　how　to　use　the　K-means　algorithm　to　divide　the　network　community　user　group　and　improve　the　FP-tree　algorithm　(CDFP-tree)　to　analyse　the　characteristics　of　the　network　user　group.　The　purpose　is　to　explore　the　relationship　between　the　user　group＇s　interest,　purpose　and　the　natural　attribute.　The　result　will　be　the　technical　support　for　the　personalized　network　services.　Another　focus　of　this　paper　is　placed　on　optimizing　and　improving　the　FP-tree　algorithm　which　is　based　on　the　constraint　item,　mainly　through　the　adjustment　of　the　method　of　construction　tree　and　traverse　of　the　trees　to　realize.　The　experiments　show　that　comparing　to　the　original　algorithm,　the　CDFP-tree　has　the　same　accuracy　while　the　efficiency　is　highly　improved.　And　the　results　of　the　characteristic　analysis　have　a　referenced　value　to　the　personalized　recommendation　service.　©　2012　IEEE.