Method for training a neural network for processing text and method for processing text

By identifying the first sentiment word and the second sentiment words of different sentiment categories in the sample text in intelligent customer service, and adjusting the neural network parameters, the problems of large computational load and long training time in the existing technology are solved, and higher recognition accuracy and training efficiency are achieved.

CN114861660BActive Publication Date: 2025-11-25BEIJING BAIDU NETCOM SCI & TECH CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202210589282.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-05-26
Publication Date
2025-11-25
Estimated Expiration
2042-05-26

AI Technical Summary

Technical Problem

Existing technologies for identifying user emotions in intelligent customer service suffer from problems such as high computational load and long training time for neural networks. Furthermore, directly selecting emotion words from a vocabulary list may lead to the learning of incorrect knowledge, affecting the accuracy of identification.

Method used

By identifying the first sentiment word and the second sentiment word that differs from its sentiment category in the sample text, a neural network is used to predict whether these sentiment words belong to the sample text. The neural network parameters are adjusted to train the neural network, reducing the selection range and thus reducing the computational load and training time.

Benefits of technology

This improved the recognition accuracy of neural networks in downstream tasks, reduced computational load, decreased reliance on errors in the sentiment dictionary, avoided loss of emotional information, and improved training efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114861660B_ABST
    Figure CN114861660B_ABST
Patent Text Reader

Abstract

The disclosure provides a training method of a neural network for processing text and a method for processing text, and relates to the field of artificial intelligence, in particular to machine learning technology, natural language processing technology and deep learning technology. The training method comprises the following steps: determining a first sentiment word in sample text and its real affiliation and at least one second sentiment word different from the sentiment category of the first sentiment word and its respective real affiliation; determining a corresponding comprehensive feature vector based on the first sentiment word and the sample text; determining a corresponding comprehensive feature vector based on each second sentiment word and the sample text; obtaining the predicted affiliation of the first sentiment word and the predicted affiliation of each of the at least one second sentiment word based on the comprehensive feature vector corresponding to the first sentiment word and the comprehensive feature vector corresponding to each of the at least one second sentiment word; and training a neural network based on the predicted affiliation and the real affiliation of the first sentiment word and the at least one second sentiment word.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Method for classifying text emotion fusing emotion characteristics and semantic characteristics

    CN108536870A

  • A commodity review emotion analysis method based on a deep learning model

    CN108984523A