Event extraction method and device

Abstract

An embodiment of the invention provides an event extraction method and device. The method comprises the following steps: extracting words in the current sentence and mapping the words into word vectors, extracting entity types in the current sentence and mapping the entity types into entity type vectors, extracting entity relations in the current sentence and mapping the entity relations into entity relation vectors, and connecting the word vectors, the entity type vectors and the entity relation vectors into sentence vectors; inputting the sentence vector into an RNN (Recurrent Neural Network), and receiving an implicit strata vector expressing semantics of the current sentence from the RNN; in the feedforward neural network, predicting trigger words and argument roles in the current sentence on the basis of a hidden layer vector and a memory vector, wherein the memory vector comprises a relation between the trigger words determined on the basis of the previous sentence, a relation between the trigger word and the argument role determined based on the previous sentence, and a relation between argument roles determined based on the previous sentences; performing probability planning calculation on the prediction result to obtain probability distribution of trigger words and argument roles in the current sentence.

Description

technical field [0001] The invention belongs to the technical field of natural language processing (Natural Language Processing, NLP), in particular to an event extraction method and device. Background technique [0002] As a form of information, an event is defined as an objective fact that a specific person or thing interacts at a specific time and a specific place, generally speaking at the sentence level. In Topic Detection and Tracking (TDT), an event is a set of related descriptions about a topic, which can be formed by classification or clustering [0003] The elements that make up an event include: trigger word, event type, argument and argument role. Event trigger words: the core words that represent the occurrence of events, mostly verbs or nouns; event types: ACE2005 defines 8 event types and 33 subtypes. Among them, most of the event extraction uses 33 event types. Event recognition is a 34-category (33 event types + None) multiple classification task based on...

AI Technical Summary

Problems solved by technology

[0007] However, the main disadvantage of the traditional method is that the process is very long, so the error transmission is very serious, that is, the error transmission leads to performance attenuation; the prediction tasks of each link are independent, and the lack of relationship network will ignore the event trigger words and events. Interaction between elements; cannot handle global dependencies

[0008] Moreover, the traditional method relies heavily on external NLP tools, which in turn increases the error of event extraction

Images