Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

Natural language multi-task modeling and prediction method and system with dependencies

A natural language and prediction method technology, applied in the field of multi-task learning, can solve problems such as inconsistent prediction results, ignoring strong logical associations of prediction results, inconsistent predictions of related laws and crimes, etc., to achieve robustness improvement, reasonable prediction, The effect of reducing the error accumulation problem

Active Publication Date: 2021-09-07
上海旻浦科技有限公司
View PDF6 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, these hierarchical multi-task models only consider the cascading of neural network encoders, but ignore the strong logical correlation between prediction results, resulting in inconsistent prediction results between various tasks, which limits the use of machine learning models. Applications in actual scenarios, such as the application of judicial judgment prediction based on judgment documents, will cause inconsistencies between the predicted relevant laws and crimes

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Natural language multi-task modeling and prediction method and system with dependencies
  • Natural language multi-task modeling and prediction method and system with dependencies
  • Natural language multi-task modeling and prediction method and system with dependencies

Examples

Experimental program
Comparison scheme
Effect test

example 1

[0169] In the task of setting artificial intelligence judicial judgment prediction, there are three sub-tasks. According to the description of the found facts in the judgment document, the relevant law prediction, crime prediction, and sentence prediction are performed. In the civil law system (such as China's judicial system), there is a logical dependency between these three subtasks. The charges and sentences depend on the relevant laws, and the sentences also depend on the charges. Therefore, the logical order of the three subtasks is: relevant laws Article prediction→crime prediction→sentence prediction. The facts found in the following judgment documents:

[0170] The People's Procuratorate of the Third District of Dongguan City charged that at about 4 o'clock on July 17, 2014, the defendant Jiang Mou and others went to the residence of the victim Zhong Mou at No. 35 Dongjin Road, Tianmei Village, Huangjiang Town, Dongguan City, and forcibly took Zhong Mou to Room 317, ...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The present invention provides a natural language multi-task modeling and prediction method and system with dependencies. The hierarchical encoder performs encoding representation of different levels of input word embedding for different levels of tasks; the label embedding layer converts the labels of different tasks Embedded in the same pull space; the label migrator transfers the embedded label; the predictor predicts the probability distribution of each task according to the encoding result and migration result of each task; the Gumbel sampling layer predicts for each task Gumbel sampling is performed on the probability distribution of the given probability, and the counterfactual value is obtained with the set probability, so as to perform counterfactual inference. If there is a causal relationship between tasks, the causal effect is obtained, and the multi-task model is jointly optimized. According to the causal correlation, the present invention can obtain returns from high-level tasks for low-level tasks, thereby making the prediction results of the optimized model more accurate for low-level tasks and improving the prediction accuracy of high-level tasks.

Description

technical field [0001] The present invention relates to a multi-task learning technology in the technical field of natural language processing, in particular to a natural language multi-task modeling and prediction method and system with dependencies. Background technique [0002] In the field of machine learning, multi-task learning is an important learning method because it allows the use of knowledge about related tasks to improve the effect of machine learning. In recent years, some studies have proposed a hierarchical multi-task model for tasks with dependencies. Since the potential dependencies between tasks can be utilized, its effect is generally better than that of a flat multi-task framework. However, these hierarchical multi-task models only consider the cascading of neural network encoders, but ignore the strong logical correlation between prediction results, resulting in inconsistent prediction results between various tasks, which limits the use of machine learn...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Patents(China)
IPC IPC(8): G06N3/04G06N3/08
CPCG06N3/08G06N3/045
Inventor 姜华陈文清田济东
Owner 上海旻浦科技有限公司
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products