Document Classification Using Word Set Images for Noisy Documents
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing document classification methods face challenges with typographical errors in low-quality documents and require large neural network resources due to varying text lengths, leading to inefficiencies and reduced accuracy.
Innovation Solution
A method that extracts words from a document image, generates word sets based on similarity, creates word set images, and uses an image captioning model to identify important keywords for classification, minimizing similarity and reducing model complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If RNN is used to classify documents by successively analyzing extracted texts, then document classification can be performed with overall context understanding, but the network size must be determined according to text length, causing great variation in necessary resources
Solution Approach 1:
The patent extracts only the most important words from the document text using an importance scoring mechanism. Instead of feeding the entire text into the RNN, only the top-k important words are selected and fed into the classification model. This extraction process reduces the input size while preserving the most relevant information for classification, thereby reducing network resource requirements while maintaining accuracy.
Solution Approach 2:
The patent applies different processing strategies to different parts of the text based on their importance. Important words are preserved and fed into the RNN, while less important words are discarded. This local differentiation allows the model to focus computational resources on the most relevant parts of the document, reducing overall network size requirements while maintaining classification reliability.
2Measurement precision
If all text information is used for document classification, then comprehensive analysis is possible, but unnecessary information increases model complexity and resource requirements
Solution Approach 1:
The patent extracts and removes unnecessary information from the document text before classification. An importance scoring mechanism identifies and retains only the most relevant words, discarding the rest. This extraction process reduces model complexity and resource requirements while preserving the information necessary for accurate classification.
Solution Approach 2:
Instead of processing all text information, the patent processes only a partial subset - specifically the top-k most important words. This partial action approach is sufficient for achieving accurate classification without the overhead of processing unnecessary information, thereby reducing model complexity while maintaining precision.
3Ease of manufacture
If rule-based classification is used with complete text extraction, then simple classification can be performed, but typographical errors in low-quality documents require additional rules to handle
Solution Approach 1:
The patent introduces an intermediary importance scoring mechanism that processes the extracted text before classification. This intermediary layer identifies and prioritizes important words, making the classification process more robust to typographical errors in less important parts of the text. The scoring mechanism acts as a mediator between text extraction and classification, improving reliability while maintaining simplicity.
Data Source
AI summary
The present disclosure relates to a text-based document classification method and a document classification device. A text-based document classification method according to an embodiment of the present disclosure is performed by a processor inside a computing device, and may include: extracting, from a document image that has been input, words included in the document image; generating, based on a degree of similarity between the words, a word set including a configured number of words; generating a word set image by individually turning the word set into an image; extracting an important keyword used for document classification among words included in the word set image; and classifying a type of the document image by using the important keyword.