Head-word cross-sentence event argument detection method

• Main Authors: CHEN WEIQING, CUI YING, PAN LEI, GAO XIANG, LIAO HONGZHOU
• Format: Patent
• Language: Chinese; English
• Published: 11.02.2022
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

The invention provides a head-word cross-sentence event argument detection method, which can reduce the cross-sentence argument detection complexity and improve the accuracy. According to the technical scheme, the method comprises the steps that corpus preprocessing is conducted, a candidate head-word set is established, shallow semantic vectors and deep semantic vectors of candidate head-words and trigger words are calculated, a trigger word association dependency arc is detected based on a double-affine transformation neural network model, the head-words corresponding to the trigger words are detected based on secondary double-affine transformation, and argument classification is completed; the head-word is selected from a head-word-argument category set, an argument is detected from adjacent words of the head-word, the dynamic words and position vectors of the argument are spliced,the spliced dynamic words and position vectors are input into a multilayer perceptron (MLP) model to achieve realize feature mo