Title:
BERT-BASED TEXT CLASSIFICATION METHOD AND APPARATUS, COMPUTER DEVICE, AND STORAGE MEDIUM
Document Type and Number:
WIPO Patent Application WO/2022/095354
Kind Code:
A1
Abstract:
A BERT-based text classification method and apparatus, a computer device, and a storage medium. The method comprises: receiving a session request sent by a user by means of a request terminal and carrying consultation data (S101); in response to the session request, inputting the consultation data into a BERT network trained on the basis of a triplet loss function to perform a feature conversion operation, so as to obtain a session feature vector (S102); inputting the session feature vector into a dense classification layer to perform a category prediction operation, so as to obtain a prediction classification result (S103); reading a verbal skill database, and acquiring verbal skill reply information corresponding to the prediction classification result in the verbal skill database (S104); and sending the verbal skill reply information to the request terminal to complete consultation session reply (S105). In addition, the method also relates to a blockchain technology, and the consultation data of the user can be stored in a blockchain. According to the method, the influence of a noise label can be greatly reduced, and the problem that a basic classification structure of a traditional BERT is easily influenced by a noise label is effectively solved.
More Like This:
Inventors:
WANG JING (CN)
Application Number:
PCT/CN2021/090505
Publication Date:
May 12, 2022
Filing Date:
April 28, 2021
Export Citation:
Assignee:
PING AN TECH SHENZHEN CO LTD (CN)
International Classes:
G06F16/335; G06F16/35
Foreign References:
| CN112328786A | 2021-02-05 | |||
| CN110222167A | 2019-09-10 | |||
| CN110196913A | 2019-09-03 | |||
| US20200279105A1 | 2020-09-03 |
Attorney, Agent or Firm:
SL INTELLECTUAL PROPERTY CO., LTD. (CN)
Download PDF: