Construction and Application of Text Entity Relation Joint Extraction Model Based on Multi-Head Attention Neural...

Area of Science:

• Natural Language Processing
• Information Extraction
• Machine Learning

Background:

• Entity relationship extraction is crucial for information extraction and natural language processing.
• Existing models face challenges in accurately identifying and classifying entities and their relationships within text.

Purpose of the Study:

• To propose a combined extraction model using a multi-headed attention neural network for enhanced entity and relation extraction.
• To improve characteristic extraction capacity by integrating naming entity features, terminology labeling, and improved neural structures.

Main Methods:

• Developed a model based on the BERT architecture, incorporating a multi-headed attention mechanism.
• Optimized multi-head attention parameters (h=8, dv=16) for best classification performance.
• Integrated naming entity recognition and terminology labeling characteristics.

Main Results:

• Achieved a peak accuracy of 87.7% on the accuracy indicator, demonstrating enhanced feature extraction.
• Showcased strong generalization ability with training and verification set curves reaching 98% and 96% respectively.
• Completed test set extraction in 1005 ms, indicating efficient processing.

Conclusions:

• The proposed multi-headed attention neural network model significantly enhances entity relationship extraction performance.
• The model exhibits robust accuracy, generalization, and processing speed, offering practical value for information extraction tasks.