Title:
KEY INFORMATION EXTRACTION METHOD BASED ON FINE ANNOTATION TEXT, AND APPARATUS AND STORAGE MEDIUM
Document Type and Number:
WIPO Patent Application WO/2021/203581
Kind Code:
A1
Abstract:
A key information extraction method based on fine annotation text, and an apparatus and a storage medium. The method comprises: S110, performing pre-training on text data by means of a BERT pre-training model so as to obtain word vectors, and combining the obtained word vectors to form matrix text data (S110); S120, inputting the matrix text data into a key information extraction model, training the key information extraction model by using a CMRC data set, and obtaining key information according to the matrix text data (S120); and S130, sorting the obtained key information according to a preset sorting rule, and using key information that meets a set selection rule as an output (S130). According to the present method, the problem of performing automatic annotation on text segment fragments is solved, such that the annotation cost is greatly reduced, and the technical effect of providing powerful support for downstream tasks is achieved.
Inventors:
CAO CHENJIE (CN)
XU GUOQIANG (CN)
XU GUOQIANG (CN)
Application Number:
PCT/CN2020/103933
Publication Date:
October 14, 2021
Filing Date:
July 24, 2020
Export Citation:
Assignee:
ONE CONNECT SMART TECH CO LTD SHENZHEN (CN)
International Classes:
G06F16/33
Foreign References:
| CN111177326A | 2020-05-19 | |||
| CN110442691A | 2019-11-12 | |||
| CN107436900A | 2017-12-05 | |||
| CN110888966A | 2020-03-17 | |||
| EP3627398A1 | 2020-03-25 |
Attorney, Agent or Firm:
GRANDER IP LAW FIRM (CN)
Download PDF: