Text classification model based on dual-channel feature extraction

The invention discloses a text classification model based on dual-channel feature extraction, deep learning development enables a pre-training model to obtain a significant achievement in text representation, a pre-training + fine tuning mode becomes a mainstream practice of text classification, and...

Full description

Saved in:
Bibliographic Details
Main Authors CHENG ZHAOLIN, ZENG MENGJIA, WANG YUJIA
Format Patent
LanguageChinese
English
Published 25.06.2024
Subjects
CALCULATING
COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
COMPUTING
COUNTING
ELECTRIC DIGITAL DATA PROCESSING
PHYSICS
Online AccessGet full text

Cover

More Information
Summary:The invention discloses a text classification model based on dual-channel feature extraction, deep learning development enables a pre-training model to obtain a significant achievement in text representation, a pre-training + fine tuning mode becomes a mainstream practice of text classification, and in order to optimize feature extraction capability, a dual-channel feature extraction scheme is provided. Wherein one channel uses a dynamic pooling strategy to improve a pooling layer of the convolutional neural network, establishes a relation between the number of reserved feature values and the number of network layers and dynamically reserves text feature information, and the other channel corresponds sequence information of neurons to syntactic information in a natural language to achieve an effect of sequencing the neurons. The hierarchical level is identified by calculating the information hierarchy and controlling the hierarchy updating frequency, the purpose of learning the hierarchical structure of the n
Bibliography:Application Number: CN202410200268