Entity Relationship Pair Identification via Negative Sample Re-Sampling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing entity relationship identification methods in artificial intelligence fail to utilize implicit information, leading to inaccurate identification of entity relationships.
Innovation Solution
A method involving text classification, entity and relationship identification, construction of positive and negative example sample sets, and re-sampling of negative samples to improve accuracy, using models like BERT and Bi-LSTM+CRF for entity extraction and relationship determination.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional entity identification technology is used, then the process is simple and fast, but the identification accuracy is low due to inability to utilize implicit information
Solution Approach 1:
The patent segments the entity relationship identification process into multiple stages: text classification, entity identification, relationship identification, positive sample construction, and negative sample re-sampling. Each stage processes specific aspects of the data independently, allowing implicit information to be systematically extracted and utilized without overwhelming complexity
Solution Approach 2:
The patent performs preliminary text classification and entity identification before relationship identification. Positive example samples are constructed in advance based on identified entities and relationships, and negative samples are re-sampled based on these positive samples. This preliminary preparation enables the model to learn from both explicit and implicit information patterns, improving identification accuracy
2Measurement precision
If more implicit information is utilized, then identification accuracy improves, but computational complexity and processing time increase
Solution Approach 1:
The patent applies partial action by selectively processing texts based on classification results. Not all texts undergo full entity and relationship identification - only classified texts are processed further. The negative sample re-sampling also focuses only on samples that need adjustment based on positive sample patterns, avoiding unnecessary computational overhead on already-processed data
3Loss of information
If traditional entity identification is used, then processing is faster, but implicit information is not utilized leading to inaccurate results
Solution Approach 1:
The patent introduces positive example samples as an intermediary between identified entities and final relationship identification. These positive samples, constructed from high-confidence entity-relationship pairs, serve as a bridge that captures implicit information patterns. The model learns from these intermediaries to improve relationship identification accuracy without requiring complete re-processing of all input texts
Data Source
AI summary
The present disclosure discloses a method for identifying entity relationship pairs, including: acquiring a service text set, classifying service texts in the service text set based on text categories of the service texts to obtain a classified text set; performing entity identification and relationship identification on the classified texts in the classified text set to obtain an entity set and a relationship set; constructing a positive example sample set and a negative example sample set based on the entity set and the relationship set; and re-sampling the negative example sample set based on the positive example sample set to obtain a target entity relationship pair. The present disclosure also discloses an electronic device, an electronic apparatus, and a computer readable storage medium for identifying entity relationship pairs. The present disclosure can improve the accuracy of entity relationship identification.