[Objective] In view of the the shortcomings of the existing event extraction models based on sequence labeling and text generation, this paper proposed a joint event extraction model using an automatically constructed template to elicit the knowledge of pre-trained language model.

[Methods] This paper designed an automatic template construction strategy based on the Event Prompt to generate unified prompt templates. The Event Prompt Embedding Layer was introduced for the Event Prompt in the encoding layer, and then connected to the pre-trained BART model to capture the semantic information of the sentence, generated the corresponding prediction sequence, and extracted the event trigger words and arguments of the corresponding event type contained in the original text from the prediction sequence, realized the joint extraction of event trigger words and arguments.

[Results] In the event dataset containing complex event information text, the F1 of event trigger word extraction reached 77.67%, and the F1 of event argument extraction reached 65.06%, 2.43% and 1.62% higher than the optimal baseline method respectively.

[Limitations] This paper only considered sentence level text, and only optimized the prompt in the encoding layer.

[Conclusions] The prompt based optimization can reduce the cost of template construction while maintaining the same or even better performance. And this model can recognize the complex event information contained in the text, which effectively improved the effect of multi-label classification of event elements.