Entity Word Recognition via Question-Answer Model
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for recognizing entity words in professional documents suffer from a low recall rate due to the limitations of professional terminology tables, which restrict the coverage and accuracy of entity word mining.
Innovation Solution
A method and apparatus utilizing a question-answer model to generate entity word questions based on categories, segment documents, and input them into a pre-trained model to improve recall and accuracy, enabling high scalability and low-cost entity word recognition for knowledge map construction and retrieval.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If professional terminology tables are used for entity word recognition, then the method is simple and easy to implement, but the recall rate is low due to limited coverage
Solution Approach 1:
The patent replaces the traditional mechanical approach of using professional terminology tables with a deep learning-based question-answer model. This substitution enables the system to automatically understand and extract entity words from unstructured text, significantly improving recall rate without requiring manual vocabulary construction or maintenance of terminology tables.
Solution Approach 2:
The patent transforms the entity word recognition task from a traditional matching problem into a question-answering task by parameterizing the input as natural language questions. This parameter change allows the model to handle diverse entity types and relationships flexibly, improving both recall and adaptability while maintaining reasonable system complexity through pre-trained models.
2Measurement precision
If manual vocabulary construction is performed to improve entity word coverage, then the recall rate increases, but the time consumption and cost increase significantly
Solution Approach 1:
The patent employs pre-trained question-answer models that have been preliminarily trained on large-scale corpora before deployment. This preliminary action enables the model to possess general entity recognition capabilities upfront, eliminating the need for time-consuming manual vocabulary construction while maintaining high entity word coverage through transfer learning and adaptation to specific domains.
Solution Approach 2:
The system enables automatic entity word extraction through the question-answer model without requiring manual intervention for vocabulary construction or maintenance. The model self-adapts to different domains and entity types by processing training data automatically, significantly reducing both time consumption and labor costs while achieving comprehensive entity word coverage.
3Adaptability or versatility
If traditional entity word recognition methods are used, then the system complexity is low, but the intelligence and adaptability of recognition are insufficient
Solution Approach 1:
The patent replaces simple keyword matching and rule-based methods with an intelligent deep learning question-answer model. This substitution dramatically improves recognition intelligence and adaptability, enabling the system to understand contextual meanings, handle ambiguous entities, and adapt to different domains automatically, while the use of pre-trained models keeps the deployed system complexity manageable.
Data Source
AI summary
The disclosure discloses a method and an apparatus for recognizing an entity word. The method includes: obtaining an entity word category and a document to be recognized; generating an entity word question based on the entity word category; segmenting the document to be recognized to generate a plurality of candidate sentences; inputting the entity word question and the plurality of candidate sentences into a question-answer model trained in advance to obtain an entity word recognizing result; and obtaining an entity word set corresponding to the entity word question based on the entity word recognizing result.