Triples Extraction in Epidemiological Investigation Field Based on Improved Pre-training Model

Public health emergencies often pose a threat to people's lives and property. Most of the current epidemiological analyzes of infectious diseases focus on statistical analysis methods, ignoring the spatiotemporal communication relationship and semantic communication relationship centered on the infected. This article is based on real epidemiological survey data and develops a triplet in the field of epidemiological survey. Extraction work, a method based on a specific labeling strategy that improves the BERT pre-training model is proposed to solve the entity overlap problem in the traditional method. Experiments conducted on real epidemiological survey trajectory data shows that the $\mathrm{F}_{1}$ value of this method is excellent. Compared with other methods, it can be used to mine spatiotemporal semantic features in the field and study the activities of infected people and the transmission process.