| CPC G06F 40/284 (2020.01) | 10 Claims |

|
1. The deep semantic feature based few-shot intent recognition method for air traffic control instructions, comprising the steps of:
first, obtaining an overall language model and an intent recognition model as follows:
constructing a fine-tuned BERT language model, a masked language model and a language model based on contrastive learning, and performing a joint pre-training to obtain an overall language model; constructing a classifier, pre-labeling a part of the to-be-predicted area air traffic control instructions, thereby obtaining small samples therefrom; extracting a deep semantic feature representation through the overall language model, inputting the deep semantic feature representation into the classifier, and performing training on the classifier to obtain an intent recognition model;
second, using the overall language model and the intent recognition model to recognize the intent of the to-be-predicted area air traffic control instructions that are not pre-labeled:
inputting the to-be-predicted area air traffic control instructions that are not pre-labeled into the overall language model, and extracting a deep semantic feature representation of the corresponding instructions; subsequently, performing intent recognition processing on the deep semantic feature representation of the corresponding instructions by using the intent recognition model to obtain a corresponding prediction result, thereby completing the intent recognition of the to-be-predicted area air traffic control instructions that are not pre-labeled.
|