Event detection method and device based on hybrid attention network

Abstract

The invention discloses an event detection method and a device based on a hybrid attention network, and the method comprises the steps: constructing a hybrid attention network model which comprises amulti-language representation layer, a hybrid attention layer and a classification layer; performing translation of a source text and acquisition of a target text of multiple languages on the multi-language representation layer, performing text alignment, and converting the target text of multiple languages into vector representation of a sentence sequence; using the hybrid attention layer for carrying out context attention learning on texts of multiple languages in parallel and carrying out information fusion of cross-source languages and multiple target languages through a multi-language attention mechanism; and performing prediction classification of event types in the classification layer.

Description

technical field [0001] The invention relates to the technical field of event detection in natural language processing, in particular to an event detection method and device based on a hybrid attention network. Background technique [0002] The task of event detection is to identify event instances of a specific type from plain text. Specifically, given an input text, the event detection task needs to determine the trigger words contained in the text and the type of event described by the trigger words, which includes two subtasks of event trigger word identification and event trigger word classification. For example, given a plain text: Three elephants were shot dead. Event detection can automatically identify the trigger word "shot" and its triggered event subtype Attack (type Confict) and the trigger word "dead" and its triggered event subtype Die (type Life) from the text. [0003] Although the current research on event detection problems has made relatively great progre...

AI Technical Summary

Problems solved by technology

[0003] Although the current research on event detection problems has made relatively great progress, there are still two problems that will seriously limit the performance of current methods

One is the low recall rate problem due to data sparsity

In the case of limited training data, there are especially few training examples for some event types. The model learned from these few training examples needs to identify the correct event type from different expressions of a certain event type. is very challenging

The second is the low accuracy rate due to the ambiguity of natural language

[0005] However, in the existing event detection work, there is no model that can fully utilize the supplementary information from more languages

Moreover, due to the limitation of model design, the GMLATT model can only integrate the source language and a translated target language information.

Images