With　the　rapid　development　of　microblogs　in　recent　years,　accurate　prediction　of　microblog　user　profiles　is　valuable　for　marketing,　personalized　recommendation,　and　legal　investigation.　Microblog　users　post　rich　contents　everyday　and　build　a　complex　friendship　network　with　＂following＂　behaviors.　Both　of　user-generated　content　and　friendship　network　are　crucial　for　user　profiling.　In　this　work,　we　propose　a　neural-network　based　model　for　user　profiling.　It　takes　advantages　of　both　user-generated　content　and　friendship　network　with　attentional　multiscale　convolutional　neural　networks　and　graph　embeddings.　We　evaluate　our　model　on　SMP　CUP　2016　dataset　whose　task　is　to　infer　age,　gender　and　region　of　microblog　users.　The　experiment　results　show　that　utilizing　information　from　user　generated　content　and　friend　network,　our　method　obtains　the　state-of-the-art　performance　on　all　of　three　sub-tasks.