Chinese lower resource cross-language abstract method integrated with source language word level information

• Main Authors: LI XIAOBO, YU ZHENGTAO, GAO SHENGXIANG
• Format: Patent
• Language: Chinese; English
• Published: 09.09.2022
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

The invention provides a Chinese lower resource cross-language abstract method integrated with source language word level information, belongs to the field of natural language processing, and aims at solving a cross-language abstract task, the Chinese lower resource cross-language abstract method comprises the steps that firstly, an extractor is used for scoring each word of a source language document, and the score of each word is more accurate; the extractor firstly uses a convolutional neural network to represent each word, and then uses a recurrent neural network to score the represented words. And then, fusing the score of each word into a cross-language abstract generation process. Due to the single-language abstract, a machine translation task and a cross-language abstract task have close correlation. Therefore, the two tasks and the cross-language abstract task are jointly learned. According to the method provided by the invention, the abstract generation quality can be effectively improved based on a