Title:
PRE-TRAINING LANGUAGE MODEL-BASED SUMMARIZATION GENERATION METHOD
Document Type and Number:
WIPO Patent Application WO/2022/104967
Kind Code:
A1
Abstract:
Disclosed in the present invention is a pre-training language model-based summarization generation method, comprising: acquiring text information of a summarization to be generated, and performing multi-feature weighted language pre-training processing on the text information to obtain a candidate summarization; inputting the candidate summarization into a pre-training language model to obtain output data of the pre-training language model; and inputting the output data of the pre-training language model into a decoder model to obtain a target summarization. In the embodiments of the present invention, pre-training processing is performed on the text information of the summarization to be generated, and then the target summarization can be generated after processing performed by an encoder and a decoder, and therefore the method in the embodiments of the present invention can better improve the quality and readability of the summarization.
Inventors:
CAI SHUBIN (CN)
ZHOU HUAIFENG (CN)
MING ZHONG (CN)
FENG XIAOGANG (CN)
ZHOU HUAIFENG (CN)
MING ZHONG (CN)
FENG XIAOGANG (CN)
Application Number:
PCT/CN2020/136138
Publication Date:
May 27, 2022
Filing Date:
December 14, 2020
Export Citation:
Assignee:
UNIV SHENZHEN (CN)
International Classes:
G06F16/34; G06F40/126
Foreign References:
| CN111723547A | 2020-09-29 | |||
| CN109829161A | 2019-05-31 | |||
| CN110119444A | 2019-08-13 | |||
| CN111897949A | 2020-11-06 | |||
| US20170372696A1 | 2017-12-28 |
Attorney, Agent or Firm:
JOHNSON INTELLECTUAL PROPERTY AGENCY (SHENZHEN) (CN)
Download PDF: