Deep Learning Text Classification via Adaptive Model Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional text classification methods, both based on machine learning and deep learning, face challenges in handling Chinese text due to its complex character set and semantics, leading to poor performance in scenarios with overlapping features among category labels and high classification granularity, where the accuracy and recall rate decrease with increasing category labels and text length.
Innovation Solution
A deep learning-based text classification method that processes a training corpus to construct a word weight table, screens key clauses based on clause weights, and trains subsample sets across different word length intervals to adaptively select text classification models, preserving the original sentence structure and learning semantic features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional machine learning or deep learning methods are used for text classification, then the basic classification function is achieved, but the classification accuracy and recall rate decrease when the number of category labels and text length increase
Solution Approach 1:
The patent segments text classification into multiple models based on text length intervals. Each model is trained on subsample sets corresponding to specific word length intervals, allowing the system to adaptively select the appropriate model based on the input text length. This segmentation resolves the contradiction by providing specialized models for different text lengths rather than using a single general model.
Solution Approach 2:
The patent implements dynamic model selection based on text length. The system automatically determines which classification model to use by comparing the input text length against predefined intervals, enabling the classification accuracy to be optimized for each specific text length scenario.
2Productivity
If traditional text classification methods are used, then the processing speed is maintained, but the ability to handle complex Chinese text semantics and overlapping features among category labels is insufficient
Solution Approach 1:
The patent performs preliminary actions by pre-processing training corpora to construct word weight tables and pre-training multiple classification models for different text length intervals before actual classification. This allows the system to quickly select and apply the most appropriate pre-trained model for the given input text, maintaining processing speed while improving accuracy.
Solution Approach 2:
The patent changes the parameter of model selection based on text length. By training separate models for different word length intervals and selecting the appropriate model based on the input text length parameter, the system optimizes classification accuracy for complex Chinese text while maintaining efficient processing.
3Device complexity
If a single text classification model is trained on all text lengths, then the model structure is simple, but the classification performance deteriorates for texts with varying lengths
Solution Approach 1:
The patent divides the text classification task into multiple models based on text length intervals. Instead of using a single model for all text lengths, the system creates specialized models for different intervals (e.g., short texts, medium texts, long texts), which significantly improves classification accuracy for varying text lengths while keeping each individual model relatively simple.
4Measurement precision
If key clauses are screened based on clause weights, then the classification accuracy is improved, but the computational complexity increases
Solution Approach 1:
The patent performs preliminary screening of key clauses during the training phase by computing clause weights and identifying important clauses in training corpora. This pre-processing creates optimized training datasets that improve model learning efficiency and classification accuracy without significantly increasing the complexity of the actual classification process.
Data Source
AI summary
Disclosed of the present application is relation to deep learning based text classification. The training corpus is screened by key clauses according to the weights of clauses in the training corpus, so as to keep the complete sentence and the original word order as much as possible according to the language habits. Thus, the deep learning model can learn normal semantic features. In addition, the subsample sets corresponding to different preset word length intervals is obtained from the training sample set, and each subsample set is putted into the deep learning model for training, so that several text classification models corresponding to different preset word length intervals can be obtained for text classification. Therefore, the deep learning models can be self-adaptively selected to classify texts based on the above mentioned multiple word length intervals and multi-model training method, to improve text classification accuracy.