Event detection model construction method and device, electronic equipment and storage medium

Abstract

The embodiment of the invention provides a construction method and device of an event detection deep learning model based on open domain knowledge enhancement, electronic equipment and a storage medium. The method comprises the steps of obtaining annotated data and unannotated data; inputting the annotation data into a first event classification model for training; processing the first data subsetin the unlabeled data by adopting a synonym mapping algorithm according to an external semantic library to obtain an open domain trigger word recognition result; training a second event classification model in a knowledge distillation mode according to the open domain trigger word recognition result and the second data subset; and performing joint training on the trained first event classification model and the trained second event classification model to obtain an event detection deep learning model based on open domain knowledge enhancement. The event detection deep learning model based onopen domain knowledge enhancement obtained by the embodiment of the invention can effectively solve the long-tail problem of uneven distribution of various labels.

Description

technical field [0001] The present invention relates to the field of computer technology, in particular to a construction method, device, electronic equipment and storage medium of an event detection deep learning model based on open domain knowledge enhancement. Background technique [0002] Event detection aims to discover events from unstructured news reports. At present, event detection, as a basic core technology in the field of artificial intelligence, has been widely introduced into reading comprehension and text summarization tasks. [0003] The event detection task is divided into two steps, the first step detects trigger words in sentences, and the second step classifies trigger words into predefined event types. Most of the existing work focuses on the second step of event type classification, such as proposing models such as dynamic convolutional networks and hierarchical attention mechanisms. However, trigger word recognition is also very critical. There is a ...

AI Technical Summary

Problems solved by technology

For the long-tail problem, if a supervised method is used to rely only on labeled corpus for training, it is easy to overfit and perform poorly on triggers that do not appear / label sparsely; if a self-iterative method is used to expand training examples based on pseudo-labels, then Due to the uneven distribution of the seed set data itself, the expanded data set is also concentrated on multi-labeled trigger words, which cannot alleviate the long-tail problem; if the remote supervision method is used to expand more data by relying on an external knowledge base, it is limited by the knowledge base The problem of its own field limitations and low coverage cannot alleviate the long tail problem

Images