Class-Imbalanced-Aware Distantly Supervised Named Entity Recognition.

Area of Science:

• Natural Language Processing
• Machine Learning
• Artificial Intelligence

Background:

• Distantly supervised named entity recognition (NER) automates model learning without manual data labeling.
• Positive unlabeled (PU) learning methods have shown success in distantly supervised NER.
• Existing PU learning methods struggle with class imbalance and require prior probability estimation, hindering performance.

Purpose of the Study:

• To propose a novel PU learning method for distantly supervised NER.
• To address the limitations of existing methods regarding class imbalance and prior estimation.
• To improve the performance of distantly supervised NER models.

Main Methods:

• A novel PU learning method is proposed for distantly supervised NER.
• The method automatically handles class imbalance.
• The method eliminates the need for class prior estimation.

Main Results:

• The proposed method achieves state-of-the-art performance in distantly supervised NER.
• Experimental results validate the theoretical analysis and demonstrate the method's superiority.
• The approach effectively manages class imbalance without prior estimation.

Conclusions:

• The novel PU learning method offers a significant advancement for distantly supervised NER.
• The method's ability to handle class imbalance and avoid prior estimation leads to superior performance.
• This work provides a more robust and effective solution for automated NER.