US 12,455,913 B2
Method, device, and medium for consistency detection of a document and an abstract
Jiaze Chen, Beijing (CN); and Zhiyuan Zeng, Beijing (CN)
Assigned to Beijing Youzhuju Network Technology Co., Ltd., Beijing (CN)
Appl. No. 18/558,157
Filed by Beijing Youzhuju Network Technology Co., Ltd., Beijing (CN)
PCT Filed Aug. 16, 2022, PCT No. PCT/CN2022/112869
§ 371(c)(1), (2) Date Oct. 30, 2023,
PCT Pub. No. WO2023/035883, PCT Pub. Date Mar. 16, 2023.
Claims priority of application No. 202111070769.7 (CN), filed on Sep. 13, 2021.
Prior Publication US 2024/0232245 A1, Jul. 11, 2024
Int. Cl. G06F 16/00 (2019.01); G06F 16/34 (2019.01); G06F 16/353 (2025.01); G06F 16/383 (2019.01)
CPC G06F 16/353 (2019.01) [G06F 16/345 (2019.01); G06F 16/383 (2019.01)] 20 Claims
OG exemplary drawing
 
1. A computer-implemented method of consistency detection of a document and an abstract, comprising:
determining, by a computing device, a first sample and first annotation information, the first annotation information indicating that a first abstract and a first document included in the first sample are inconsistent, at least one of a plurality of text elements in the first abstract being labeled as inconsistent with the first document;
generating, by the computing device, a first adversarial sample by applying interference information to the first sample, the interference information being applied to the first sample and other text elements in the first abstract except for the at least one text element;
training, by the computing device, at least based on the first sample, the first adversarial sample, and the first annotation information, a consistency detection model according to a training objective, the consistency detection model being configured to detect whether an abstract is consistent with a document by processing textual data through a neural network, the training objective being configured to cause both a difference between the first annotation information and a detection result of the first sample from the consistency detection model, and a difference between the first annotation information and a detection result of the first adversarial sample from the consistency detection model to be within a predetermined threshold; and
applying, by the computing device, a source document and a target abstract to the trained consistency detection model to obtain a target detection result output from the trained consistency detection model, the target detection result indicating whether the target abstract is consistent with the source document.