Nowadays　many　researches　have　focused　on　the　Affinity　Propagation　(AP)　algorithm　for　community　detection　for　its　advantages　of　near-linear　complexity　and　no　prerequisite　for　any　object　function　or　cluster　number.　In　view　of　different　influences　from　common　neighbors,　we　propose　an　improved　Affinity　Propagation　algorithm　which　is　based　on　adjacency　matrix　and　considers　self-similarity　and　similarity　among　nodes　having　common　but　disconnected　neighbors.　Two　AP　algorithms　based　on　common　neighbors　using　Local　Naive　Bayes　and　Super-mean　Random　Walk　are　proposed.　The　experiments　on　both　the　artificial　datasets　and　the　real-world　datasets　demonstrate　that　the　quality　of　communities　discovered　by　the　improved　algorithms　provide　an　effective　solution　with　a　better　stability.