In　current　search　engines,　there　are　two　ways　to　display　the　results　of　Question　style　User　Queries　(QUQs),　the　natural　results　and　Featured　Snippet　(FS).　Actually,　a　search　engine　that　triggers　a　FS　can　better　satisfy　the　user＇s　information　seeking　need,　and　the　data　format　of　FS　is　usually　formulated　as　Question　Answer　Pairs　(QAPs)　in　the　dictionary.　In　this　setting,　an　answer　can　be　retrieved　as　a　FS　if　and　only　if　the　QUQ　and　the　question　in　QAPs　are　matched.　The　traditional　retrieval　method　is　basically　based　on　keywords,　which　failed　to　bridge　the　semantic　gap.　On　the　other　hand,　neural　matching　methods　may　not　be　deployed　online　directly　due　to　the　high　flexibility　requirements　in　complex　real-world　scenarios.　To　this　end,　this　paper　combines　retrieval　model　and　matching　model　in　a　unified　system　for　FS　triggering.　This　system　contains　two　stages:　the　recall　stage　and　the　ranking　stage.　In　the　recall　stage,　for　a　QUQ,　we　use　the　vector-based　retrieval　model　rather　than　the　BOW　(bag　of　words)-based　one　to　ensure　accurate　and　quick　recall　of　possible　candidates　from　QAPs.　In　the　ranking　stage,　we　use　the　ensemble　method　on　multiple　models,　including　pre-trained　network　BERT,　to　boost　matching　performance.　To　improve　the　flexibility　and　adaptability　of　the　system,　two　query　analysis　techniques,　i.e.　query　term　weighting　and　query　term　fuzzy　method　are　also　incorporated　in　the　matching　network.　We　conduct　extensive　experiments　on　real-word　data.　The　experimental　results　demonstrate　the　superiority　of　our　system.　©　2019　IEEE.