Patent Text Recognition Using Overlapping Word-Vector Voting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing text recognition methods, particularly for patent documents, suffer from low accuracy and efficiency due to broad bibliographic information recognition and reliance on rule indexing or manual indexing, which often misses important patent information.
Innovation Solution
A text recognition method that determines overlapping character strings, converts them into word vectors, and uses a voting mechanism to identify effect text or non-effect text, utilizing an effect recognition model trained on multiple language training texts to enhance accuracy and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If rule indexing is used to determine effect text, then the method is easy to implement, but important patent information is omitted leading to low accuracy
Solution Approach 1:
The text is segmented into multiple overlapping character strings with different strides. Instead of processing the entire text as one unit, the method divides it into smaller segments that are processed individually and then combined through voting, thereby improving recognition accuracy while maintaining implementation simplicity
Solution Approach 2:
The system implements a feedback mechanism through voting operations. Multiple character strings that overlap and contain the same effect text will vote for the same result, and the voting mechanism aggregates these feedback signals to determine the final recognition result, reducing omissions and improving accuracy
2Measurement precision
If manual indexing is used to determine effect text, then text recognition accuracy is high, but a lot of manpower is required resulting in low efficiency
Solution Approach 1:
The system performs self-service by automatically processing text recognition without requiring manual indexing. The automated pipeline includes text segmentation, word vector conversion, and voting-based result determination, achieving both high accuracy and high efficiency by eliminating manual intervention entirely
Solution Approach 2:
The patent replaces the mechanical manual indexing process with an automated computational system. Manual human effort is substituted by an automated pipeline involving text segmentation, word vector conversion, and voting operations, thereby maintaining high accuracy while dramatically improving efficiency
3Adaptability or versatility
If broad bibliographic information is used for text recognition, then the scope of recognition is comprehensive, but the accuracy of recognizing effect text is low
Solution Approach 1:
The patent applies local quality by treating different regions of the text differently. Instead of uniform processing, it segments the text into character strings with varying strides, allowing different local regions to be processed with appropriate granularity. This enables comprehensive scope while improving effect text recognition accuracy through localized focused processing
Data Source
AI summary
A text recognition method, a text recognition apparatus, a computer-readable storage medium, and an electronic device are provided. In the text recognition method, adjacent character strings in a plurality of character strings partially overlap, so that the plurality of character strings may reflect the relationship between the contexts of a text to be recognized, then word vector conversion is performed on the plurality of character strings to obtain a plurality of word vectors, word vector recognition results respectively corresponding to the plurality of word vectors are generated on the basis of the plurality of word vectors to determine whether a text corresponding to the word vectors is an effect text or a non-effect text, and the plurality of word vector recognition results are synthesized to determine a text recognition result of the text to be recognized.