An Event Extraction Method Based on Dialogue State Tracking Technology

Abstract

The present invention realizes an event extraction method based on dialogue state tracking technology through the method in the field of artificial intelligence. The method as a whole consists of three parts: event classification, sequence question generation model and argument extraction model; event classification detects whether the input text is an event, if the text does not contain an event, then outputs NULL, otherwise classifies the event type to which the text belongs; sequence question generation model Questions are automatically generated according to the event type and the predicted high-confidence argument results; the argument extraction model uses the questions generated by the sequence question generation model and the input text as input to predict the argument position, and then uses the label alignment mechanism to predict The arguments with high confidence in the arguments are added to the training set. Through the three-part model, this method can feedback the prediction results and high confidence results of all parameters; extract the common information and patterns between the two tasks, and use the learned syntactic and semantic knowledge tags; and make full use of Correlation between arguments.

Description

technical field

[0001] The invention relates to the field of artificial intelligence, in particular to an event extraction method based on a dialogue state tracking technology. Background technique

[0002] Event extraction is to detect the occurrence of a specific type of event from text and extract parameters associated with the event (eg, event participants or other attributes). It is a fundamental technique underpinning many natural language processing tasks, such as knowledge reasoning, text summarization, and event prediction. In the real world, labeling large amounts of data is a high cost. Therefore, it is necessary to study how to detect events and extract their parameters from a small amount of labeled data and a large amount of unlabeled data. Previous studies have studied each argument in isolation, without exploiting correlations between arguments. Event extraction requires extracting all arguments and their roles corresponding to each event. Doing so is cha...

Images