Joint Extraction for Tibetan Medicine Entities
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current joint extraction methods for entity relationships in traditional Tibetan medicines, such as NovingTagging, struggle to label nested entities and have limited feature learning capabilities, leading to poor extraction effects due to error accumulation and limited interaction between subtasks.
Innovation Solution
A joint extraction system and method that includes a word embedding layer, a class feature static fusion layer, and a binary dynamic model to enhance input features, label nested entities, and improve robustness, featuring a sequence of steps for text sample conversion, classification, dynamic feature learning, and loss calculation to update the model parameters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a sequence labeling method (NovingTagging) is used for joint extraction, then interaction between subtasks is enhanced and error accumulation is reduced, but nested entities cannot be labeled and feature learning capacity is limited
Solution Approach 1:
The patent segments the feature extraction process into multiple independent modules: word embedding module, class feature module, and dynamic feature module. Each module processes specific aspects of the input independently before combining results, allowing nested entities to be handled through hierarchical feature integration without constraining the overall sequence labeling framework
Solution Approach 2:
The patent introduces a new dimension to feature representation by creating a three-dimensional feature fusion structure: word vectors (dimension 1), class features (dimension 2), and dynamic features (dimension 3). This multi-dimensional approach enables the model to capture nested entity relationships through feature space interactions while maintaining sequence labeling capabilities
2Device complexity
If a pipeline method is used for joint extraction, then the process is simple and modular, but lack of interaction and error accumulation affect model effectiveness
Solution Approach 1:
The patent merges multiple independent feature extraction modules (word embedding, class feature, dynamic feature) into a unified joint extraction framework. The modules interact through feature fusion operations, allowing information to flow bidirectionally between entity recognition and relationship extraction subtasks, thereby eliminating error accumulation while maintaining modular design benefits
Solution Approach 2:
The patent implements feedback mechanisms through the dynamic feature module that continuously refines predictions based on context from other modules. The loss function incorporates feedback from both entity and relationship prediction errors, allowing the model to adjust features dynamically and improve overall effectiveness
3Extent of automation
If NovingTagging model is used, then joint extraction is achieved through end-to-end decoding, but single input features limit learning capacity and result in poor extraction effects
Solution Approach 1:
The patent creates a composite feature representation by combining multiple feature types (word embeddings, class features, dynamic features) into a unified feature vector space. This composite approach enriches the input representation without sacrificing the end-to-end decoding automation, allowing the model to learn complex patterns across multiple feature dimensions simultaneously
Data Source
AI summary
The present invention proposes a joint extraction system and method for an entity relationship in the field of traditional Tibetan medicines, and relates to the field of artificial intelligence. The joint extraction method for an entity relationship in the field of traditional Tibetan medicines includes acquiring training samples; converting the training samples into word vectors; classifying the training samples, and fusing a classifying result with the word vectors to obtain static fusion features; constructing a binary dynamic model, and feeding the static fusion features into the binary dynamic model to obtain a final predicted tag sequence; calculating a loss value of the binary dynamic model, and updating parameters to obtain an updated binary dynamic model; and performing joint extraction of an entity relationship by using the updated binary dynamic model. According to the present invention, nested entities can be labeled, and prediction accuracy can be improved; and entity boundaries are enhanced.