The invention relates to a fast symbol community discovery system for automatically determining the number of communities. The system comprises an adjacency matrix A and a network model(NModel) of a network N; the network model (NModel)=(n, K, Z, [pi], [theta], [omega]), wherein n is corresponding to the node number of the network; K is the number of models and corresponding to the number of communities in the network; Z is a n*K dimension matrix, wherein Zi is the indication vector of K dimension; [pi]={ [pi]1, [pi]-1, [pi]0} is a three dimension vector, wherein the [pi] 1 represents the connecting efficiency of positive linkage within the block, [pi]2 represents the connecting efficiency of negative linkage within the block and [pi] 0 represents the efficiency of zero linkage within theblock; [theta]={ [theta]1, [theta]-1, [theta]0} is a three dimension vector, wherein the [theta] 1 represents the connecting efficiency of positive linkage within the block, [theta]2 represents the connecting efficiency of negative linkage within the block and [theta] 0 represents the efficiency of zero linkage within the block; [omega] is a K-dimension vector which indicates the ratio relation ofnodes distributing in different blocks and the formula is satisfied: the summation of [omega]k equals to 1, wherein k is from 1 to K. Parallel computing of parameter estimation and model selection isrealized, which effectively reduces the time complexity of symbolic community mining.