Evaluation object-oriented sentiment analysis method for multi-task joint learning

A technology for evaluating objects and sentiment analysis, applied in text database clustering/classification, special data processing applications, instruments, etc., can solve the problems of low accuracy of text useful information extraction and sentiment evaluation word extraction without considering relevance, etc. The effect of joint extraction task optimal and good extraction effect

Active Publication Date: 2020-06-26
HARBIN INST OF TECH
View PDF13 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to solve the problem that the existing text does not consider the relevance of the evaluation object, the evaluation object emotion, and the evaluation word extraction, resulting in the low accuracy of the text useful information extraction, and proposes a multi-task joint learning oriented evaluation object sentiment analysis method

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
  • Evaluation object-oriented sentiment analysis method for multi-task joint learning
  • Evaluation object-oriented sentiment analysis method for multi-task joint learning
  • Evaluation object-oriented sentiment analysis method for multi-task joint learning

Examples

Experimental program
Comparison scheme
Effect test

specific Embodiment approach 1

[0023] Specific implementation manner 1: The specific process of an evaluation object-oriented sentiment analysis method for multi-task joint learning in this implementation manner is:

[0024] Step 1: The text is represented by word vectors;

[0025] Step 2: Represent the word vector of the text based on step 1. Perform a preliminary abstraction and get a shared representation;

[0026] Step 3: Perform evaluation object and evaluation word extraction based on step 2, and obtain the label probability distribution of evaluation object and evaluation word extraction;

[0027] Step 4: Perform emotional judgment of the evaluation object based on Step 2 and Step 3, and obtain the label probability distribution of the evaluation object's emotional judgment;

[0028] Step 5: Perform information transmission based on steps 3 and 4, and get the updated probability distribution prob AE , Prob AS ;

[0029] Step 6: Based on the model structure determined in steps 1 to 5, use artificial annotation...

specific Embodiment approach 2

[0030] Specific embodiment two: this embodiment is different from specific embodiment one in that: in the step one, the text is represented by word vectors; the specific process is:

[0031] The text X = {w 0 ,w 1 ,...,w n } Use domain word vector E domain And general word vector E general Is represented and spliced ​​as

[0032] Among them, the domain word vector is trained on a large-scale unlabeled corpus in the target domain (for example, when the target domain is laptop, the data size used is 142.8M Amazon reviews on laptop);

[0033] The general word vector is trained on a large-scale unlabeled corpus in a large-scale indefinite field (text obtained from various websites containing 840B words (not limited to whether it is a comment));

[0034] E domain =Embedding domain (X) (1)

[0035] E general =Embedding general (X) (2)

[0036]

[0037] In the formula, X is the index of the input text, Embedding domain Is the domain word vector layer; E domain Represents the domain wor...

specific Embodiment approach 3

[0041] Specific embodiment three: this embodiment is different from specific embodiment one or two in that in the second step, the word vector of the text is expressed based on the step one Perform a preliminary abstraction to obtain a shared representation; the specific process is:

[0042] Use a single-layer CNN to represent the word vector of the text Perform a preliminary abstraction to get the shared representation output shared ;

[0043] The purpose is to block the sentence, so that the semantic structure such as the phrase is represented as a whole, and reduce the task amount of the subsequent task model;

[0044]

[0045] In the formula, output shared To abstract the shared representation, CNN shared It is a CNN layer used for preliminary abstraction to obtain a shared representation.

[0046] Other steps and parameters are the same as those in the first or second embodiment.

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 invention discloses an evaluation object-oriented sentiment analysis method for multi-task joint learning, and relates to an evaluation object-oriented sentiment analysis method for multi-task joint learning. The objective of the invention is to solve the problem of low extraction accuracy of useful information of texts due to the fact that existing texts do not consider relevance for extraction of evaluation objects, emotion of the evaluation objects and evaluation words. The method comprises the steps of 1, performing word vector representation on a text; 2, performing preliminary abstraction on the word vector representation of the text to obtain a shared representation; 3, performing evaluation object and evaluation word extraction based on the step 2 to obtain label probability distribution of evaluation object and evaluation word extraction; 4, performing evaluation object emotion judgment based on the step 2 and the step 3 to obtain label probability distribution of evaluation object emotion judgment; 5, obtaining updated probability distribution; and 6, obtaining an emotion analysis model, and completing emotion analysis on the evaluation object by adopting the emotionanalysis model. The method is applied to the field of joint extraction of evaluation objects, evaluation object sentiments and evaluation words.

Description

Technical field [0001] The invention relates to an evaluation object-oriented emotion analysis method for multi-task joint learning. Background technique [0002] With the development of the Internet in recent years, text, as a commonly used information carrier in the Internet, has shown an explosive growth trend, such as user comments on e-commerce platforms, barrage on video platforms, news, blogs, chats, etc. The text contains the information that users evaluate a certain object. This article will collectively call the text with this information a comment. The comments contain very valuable information such as user preferences and product defects. By summarizing the information contained in a large number of comments, it can help the platform better push content that meets user preferences, and can help manufacturers make timely corrections to product problems. Can help users summarize the experience of others. The present invention adopts the method of deep learning to auto...

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
IPC IPC(8): G06F40/216G06F40/289G06F16/35
CPCG06F16/35Y02D10/00
Inventor 赵妍妍王帅秦兵
Owner HARBIN INST OF TECH
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap