Support Vector Machine with Ensemble Tree Kernel for Relation Extraction

Area of Science:

• Natural Language Processing
• Machine Learning
• Information Extraction

Background:

• Traditional semisupervised relation extraction methods struggle with semantic variation, leading to inaccuracies.
• Semantic variation poses a significant challenge in accurately identifying relationships between entities in text.

Purpose of the Study:

• To propose a novel semisupervised relation extraction algorithm (LXRE) that overcomes the limitations of semantic variation.
• To enhance the precision and recall of relation extraction through ensemble learning.

Main Methods:

• Developed a novel semisupervised relation extraction algorithm named LXRE, utilizing ensemble learning.
• Integrated two types of tree kernel-based support vector machine classifiers.
• Employed a constrained extension seed set strategy to improve model robustness.

Main Results:

• The LXRE algorithm demonstrated superior performance compared to two common relation extraction methods.
• LXRE achieved better results across Precision, Recall, F-measure, and Accuracy on benchmark datasets (PropBank and AIMed).
• The proposed method effectively mitigates inaccuracies caused by semantic variation.

Conclusions:

• The LXRE algorithm offers a robust and effective solution for semisupervised relation extraction.
• Ensemble learning and constrained seed set extension significantly improve relation extraction capabilities.
• LXRE shows strong potential for practical applications in information extraction.