Public emergencies often have a great impact on society. To assist the establishment of a social early warning mechanism and improve the efficiency of emergency management of public emergencies, we analyze the text of public emergencies and propose a new word discovery model based on multi-granularity to construct a lexicon of public emergencies. In terms of corpus selection, considering the suddenness of the event, the idea of constructing a lexicon based on official release documents and supplemented by specialized vocabularies is proposed. Taking the new crown epidemic text as an example, we first realize the multi-granularity neologism discovery from words to words and from words to phrases, supplemented with high-frequency terms of epidemic reports published regularly by the National Scientific and Technological Nomenclature Validation Committee, and finally combined the neologism dictionary with the basic dictionary of jieba to generate the domain lexicon. The experiment proves that the domain dictionary constructed by the model improves the F-value of participle by 20.57% compared with the base dictionary.