Metaphor-Aware Sentiment Classification for Implicit Emotion Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing text sentiment classification techniques fail to accurately capture implicit sentiment expressions such as metaphors, leading to incomplete extraction of text sentiment information and reduced accuracy and credibility.
Innovation Solution
A sentiment classification model is constructed by integrating metaphor identification, using pre-trained language models and a K-fold cross-validation method to analyze both explicit and implicit sentiment expressions, combining characteristic values through a feedforward neural network and binary classification softmax layer, and optimizing with a BP algorithm.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional text sentiment classification techniques focus only on obviously emotionally-charged words, then the classification process is simple and fast, but the extraction of text sentiment information is incomplete and accuracy is reduced
Solution Approach 1:
The patent segments the sentiment classification task into two distinct parts: explicit sentiment classification (using pre-trained language models to analyze emotionally-charged words) and metaphor identification (using specialized models to detect implicit metaphorical expressions). This segmentation allows each component to specialize in its strength while maintaining overall system manageability.
Solution Approach 2:
The patent implements a nested architecture where the metaphor identification model is integrated within the overall sentiment classification system. The metaphor identification component processes text and feeds its results into the explicit sentiment classification model, creating a nested structure where one model operates within the framework of another to achieve comprehensive sentiment analysis.
2Loss of information
If implicit sentiment expressions such as metaphors are analyzed, then sentiment extraction completeness is improved, but the processing complexity and time increase
Solution Approach 1:
The patent employs pre-trained language models that have been previously trained on large corpora to capture linguistic patterns and sentiment expressions. This preliminary training allows the models to quickly process new text without requiring extensive computation during actual sentiment classification, thus reducing processing time while maintaining comprehensive analysis capability.
Solution Approach 2:
The patent uses pre-trained models that have learned from extensive training data, effectively copying successful patterns of sentiment recognition. These pre-trained models can be applied to new text without retraining, enabling fast processing while maintaining high accuracy in detecting both explicit and implicit sentiment expressions.
3Measurement precision
If multiple characteristic values (explicit and metaphor information) are combined, then the sentiment classification accuracy is improved, but the computational complexity increases
Solution Approach 1:
The patent merges the output of the metaphor identification model with the output of the explicit sentiment classification model to create a comprehensive sentiment classification result. This combining approach integrates multiple sources of sentiment information (explicit emotional words and implicit metaphorical expressions) into a unified classification outcome, improving accuracy while maintaining model structure simplicity through straightforward integration.
Data Source
AI summary
A method for constructing a sentiment classification model based on metaphor identification is disclosed, including: constructing a metaphor training corpus, and training a first pre-trained language model using the metaphor training corpus to obtain a metaphor recognition model; constructing and inputting a sentiment classification corpus into a second pre-trained language model and the metaphor recognition model to extract an explicit characteristic value and a metaphor information characteristic value of a text in the sentiment classification corpus; combining the explicit characteristic value and the metaphor information characteristic value to obtain a comprehensive characteristic value which is input to a feedforward neural network and a binary classification softmax layer to obtain a sentiment classification result; and performing optimization training based on the sentiment classification result using BP algorithm to obtain an optimal sentiment classification model. Explicit and implicit sentiment expressions are comprehensively considered, and more accurate and credible evaluation results are provided.