Social　annotation　systems　enable　users　to　annotate　large-scale　texts　with　tags　which　provide　a　convenient　way　to　discover,　share　and　organize　rich　information.　However,　manually　annotating　massive　texts　is　in　general　costly　in　manpower.　Therefore,　automatic　annotation　by　tag　prediction　is　of　great　help　to　improve　the　efficiency　of　semantic　identification　of　social　contents.　In　this　paper,　we　propose　a　tag　prediction　model　based　on　convolutional　neural　networks　(CNN)　and　bi-directional　long　short　term　memory　(BiLSTM)　network,　through　which,　tags　of　texts　can　be　predicted　efficiently　and　accurately.　By　Experiments　on　real-world　datasets　from　a　social　Q&A　community,　the　results　show　that　the　proposed　CNN-BiLSTM　model　achieves　state-of-the-art　accuracy　for　tag　prediction.