In view of the insufficient mining of implicit associative relationships and the problem of neglecting the temporal evolution factor, a domain-specific social event detection method via temporal evolution feature mining was proposed. The data was sliced by time and an entity interaction graph was constructed by considering the impact of duplicate event records from different sources on detection, in order to reduce the influence of database errors. Multi-relational graph convolutional network was improved, and the graph structure information of historical evolution sequence was updated by interaction relationships. Attention mechanism was used to learn core features to obtain global embedding of sequence units. Implicit association was mined sufficiently. Based on recurrent neural network, temporal evolution features were extracted to obtain the global embedding and the temporal evolution factor was mined effectively. Experiment results show that the proposed method can be applied to domain-specific social event detection task, which is better than existing methods.