Joint Entity Relationship Extraction for Software Bug Knowledge
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for entity and relationship extraction in software bug reports often ignore the connection between the two tasks, leading to potential errors and redundant information, especially when relying on manual feature extraction and separate processing of entities and relationships.
Innovation Solution
An entity and relationship joint extraction method using a transition-based model that processes text data through preprocessing, word embedding, and Bi-LSTM coding, followed by a transition system to generate a directed graph, allowing for the simultaneous extraction of entities and relationships and avoiding error propagation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If series extraction method is used to separate entity recognition and relationship recognition, then each module can be independently optimized and implemented with traditional models, but the connection between the two tasks is ignored and errors can propagate between stages
Solution Approach 1:
The patent merges entity recognition and relationship recognition into a single joint extraction model that processes text simultaneously for both tasks. This integration allows the model to capture contextual relationships between entities and their relationships in one unified framework, eliminating the errors that propagate in series processing while maintaining implementation feasibility through a cohesive architecture.
2Reliability
If joint extraction model is used to extract entities and relationships simultaneously, then connection between tasks is captured and extraction accuracy improves, but model complexity increases and manual feature extraction is required
Solution Approach 1:
The patent replaces manual feature extraction mechanisms with an end-to-end neural network model. The system automatically learns relevant features and representations through training on bug report data, eliminating the need for manual characteristic engineering while managing model complexity through modern deep learning architectures that handle joint extraction tasks effectively.
3Ease of manufacture
If end-to-end neural network model is used for joint extraction, then manual feature extraction work is reduced, but the model still processes entities and relationships separately leading to redundant information
Solution Approach 1:
The patent segments the joint extraction output into distinct entity representations and relationship representations while maintaining their interconnectedness through the unified model architecture. This segmentation allows the system to clearly distinguish between entities and their relationships, reducing information redundancy while preserving the benefits of joint processing and automated feature learning.
Data Source
AI summary
Provided is an entity and relationship joint extraction method oriented to software bug knowledge. The method includes collecting text data of an open-source bug library and preprocessing the text data to obtain a bug text data corpus; extracting, from the bug text data corpus, a statement S for describing a bug, and then processing S, and using the processed S as a subsequent input statement; constructing an entity and relationship joint extraction model; obtaining, in conjunction with the constructed entity and relationship joint extraction model based on a transition system, an entity set E and a relationship set R corresponding to the input statement; and outputting the entity set E and the relationship set R to complete joint extraction of entities and relationships.