Abstract: Aiming at the problems of low efficiency of semantic feature extraction and low intelligence of data collection in the information collection system of coal mine safety hazard information, this paper proposes a key semantic intelligence extraction based on improved Recurrent Neural Network (RNN) for coal mine safety hazard information system. The characteristics of the past cognition for the recurrent neural network is applied to construct the key semantic intelligent collection model of coal mine safety hazard information based on RNN. The sentence is segmented with comma as the boundary, the key semantic features are extracted step by step, and the memory of the past feature extraction is accumulated and get security risk feature keywords. The experimental results show that the system has high-precision feature extraction and semantic mapping hit rate, which realizes intelligent collection of key information of coal mine safety hazards, improves the efficiency of daily safety production hazard investigation, and reduces the occurrence of coal mine safety accidents.