Web　Intelligence　(WI)　is　a　new　and　active　research　field　in　current　AI　and　IT.　Intelligent　B2C　Portals　are　an　important　research　topic　in　WI.　In　this　paper,　we　first　investigate　and　analyze　the　architecture　of　a　B2C　portal　for　personalized　recommendation　from　the　viewpoint　of　conceptual　levels　of　WI.　Aiming　at　knowledge-level　data　mining　in　a　B2C　portal,　we　present　a　new　improved　learning　algorithm　of　Bayesian　Networks,　which　consists　of　two　major　contributions,　namely,　reducing　Conditional　Independence　(CI)　test　costs　by　few　lower　order　CI　tests　and　accelerating　search　process　by　means　of　sort　order　for　candidate　parent　nodes.　Experimental　results　on　benchmark　ALARM　data　sets　show　that　the　improved　algorithm　has　high　accuracy,　and　is　more　efficient　in　the　time　performance　than　other　algorithms.　Finally,　we　apply　this　algorithm　to　learning　Customer　Shopping　Model　(CSM)　in　an　intelligent　recommendation　system.　By　a　number　of　experiments　on　real　world　data,　we　find　that　the　recommendation　method　based　on　the　learned　CSM　outperforms　some　traditional　ones　in　rates　of　coverage　and　precision.　©　2007　-　IOS　Press　and　the　authors.　All　rights　reserved.