Data　mining　is　used　to　find　useful　information　from　massive　data.　Frequent　pattern　mining　is　one　important　task　of　data　mining.　Recently,　the　researches　on　frequent　pattern　mining　for　semi-structured　data　have　made　some　progresses,　and　it　also　have　a　lot　of　focuses　for　data　stream.　However,　only　a　few　studies　focus　on　both　semi-structured　data　and　data　stream.　This　paper　proposes　an　algorithm　named　SPrefixTreeISpan.　We　segment　the　semi-structured　data　stream　first,　and　then　uses　the　pattern-growth　method　to　mine　each　segment.　In　the　end,　we　maintain　all　the　results　on　a　structure　called　patternTree.　At　the　same　time,　the　mining　algorithm　is　optimized　by　the　inevitable　parent-child　relationship　and　the　inevitable　child-parent　relationship　extracted　from　XML　schema.　Experiment　shows　that　SPrefixTreeISpan　has　better　performance.