Students　have　produced　a　large　number　of　data　in　the　teaching　life　of　colleges　and　universities.　At　present,　the　development　trend　of　university　data　is　to　gradually　form　a　high-dimensional　data　storage　system　composed　of　student　status　information,　educational　administration　information,　behavior　information,　etc.　It　is　of　great　significance　to　make　use　of　the　existing　data　of　students　in　Colleges　and　universities　to　carry　out　deep-seated　and　personalized　data　mining　for　college　education　decision-making,　implementation　of　education　and　teaching　programs,　and　evaluation　of　education　and　teaching.　Student　portrait　is　the　extension　of　user　portrait　in　the　application　of　education　data　mining.　According　to　the　data　of　students’　behavior　in　school,　a　labeled　student　model　is　abstracted.　To　address　above　problems,　a　hybrid　neural　network　model　is　designed　and　implemented　to　mine　the　data　of　college　students　and　build　their　portraits,　so　as　to　help　students’　academic　development　and　improve　the　quality　of　college　teaching.　In　this　paper,　experiments　are　carried　out　on　real　datasets　(the　basic　data　of　a　college’s　students　in　Beijing　and　the　behavior　data　in　the　second　half　of　2018–2019　academic　year).　The　results　show　that　the　hybrid　neural　network　model　is　effective.　©　2021,　Springer　Nature　Switzerland　AG.