Subject-fused PGN-GAN text abstract model

• Main Authors: LYU SHUAI, SUN WENBO, GUO JIFENG, FEI YUXIAO
• Format: Patent
• Language: Chinese; English
• Published: 07.09.2021
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

The invention relates to a subject-fused PGN-GAN abstract generation method. In order to generate an abstract closer to subject information and a source text, an LDA subject model is tried to be fused on the basis of a pointer generation network, and the subject information is fused into a sequence-to-sequence model combining a pointer network and a GAN. According to the model, priori knowledge of human beings is simulated by using a theme, so that the abstract is more combined with theme generation, and after a subject term vector is obtained, the subject term vector and a context are jointly synthesized into a new vector to influence text generation. The result of the pointer generation network model added with the generative adversarial network is smoother than that of other generated abstracts, and meanwhile, due to introduction of theme information, the result is closer to the meaning of an original text. 本发明涉及一种融合主题的PGN-GAN摘要生成方法，为了能够生成更加贴近主题信息和源文本的摘要，本发明在指针生成网络的基础上尝试融入了LDA主题模型，并将主题信息融入到结合指针网络和GAN的序列到序列