BERT-BiLSTM-CRF dangerous chemical named entity identification method fusing multiple features

The invention discloses a BERT-BiLSTM-CRF dangerous chemical named entity identification method fusing multiple features, and the method comprises the steps: obtaining dangerous chemical data, and carrying out the cleaning and preprocessing: removing noise data, carrying out the word segmentation of the data, and marking a named entity; a 3D-CNN framework is constructed to extract context morphological features, and pinyin, radicals and stroke numbers are introduced as Chinese character features; constructing a BiLSTM model for extracting context semantic features, introducing an attention mechanism, and learning more complex context sequence features; and through the constructed CRF part, the named entity tag sequence is decoded, and the dangerous chemical named entity is identified. According to the method, through the constructed BERT-BiLSTM-CRF model fusing multiple features, the dangerous chemical named entities can be accurately identified, dangerous chemical information can be extracted, and the method