Domain-specific text error correction method, storage medium and device

• Main Authors: BAO LINA, LI RONGSHENG, LIU WEI, DENG HUA, QI YIFEI, YANG YANG, WU XIA, CUI HAOYI, DENG YANGWU, MIAO DEYU, LIU KANGKANG
• Format: Patent
• Language: Chinese; English
• Published: 03.02.2023
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

The invention discloses an error correction method for a specific field text, a storage medium and a device. The method comprises the following steps: constructing a data set for specific field training; performing first-stage training on the text generation model by using the general field data set; transforming the text generation model by adding a part-of-speech supervision signal, and then performing second-stage training on the transformed text generation model by using a data set for training in a specific field; and performing sentence-level error correction on the trained text generation model, and decoding to generate a corrected text. Labels do not need to be manually labeled, and the labels can be automatically generated according to existing data; common grammar characteristics and prior knowledge in the general field can be inherited; adaptability correction can be carried out for a specific field, especially a power field, and specific nouns and common terms in the specific field can be identifi