Korean anaphora recognition system to develop healthcare dialogue-type agent

Area of Science:

• Natural Language Processing
• Computational Linguistics
• Artificial Intelligence

Background:

• Anaphora recognition is crucial for dialogue agents, identifying noun-pronoun relationships.
• Existing methods include rule-based, machine learning-based, and semantic-based approaches.
• Combining these methods can potentially enhance anaphora recognition performance.

Purpose of the Study:

• To develop a novel hybrid anaphora recognition system.
• To integrate the strengths of rule-based, machine learning-based, and semantic-based techniques.
• To achieve optimal capacity and efficiency in anaphora resolution.

Main Methods:

• A rule-based system was created using expression traits and context.
• A semantic database was built to link referent candidates with predicate semantics.
• A machine learning model was developed using training data of expressions and referents.
• These three methods were combined into a single hybrid system.

Main Results:

• The rule-based system achieved a precision rate of 54.9%.
• The hybrid-based system demonstrated a significantly higher precision rate of 63.7%.
• The hybrid approach proved to be the most efficient method.

Conclusions:

• The hybrid anaphora recognition system integrates rule-based and machine learning methods.
• This novel system exhibits enhanced functional capabilities compared to individual methods.
• The hybrid approach offers improved performance for anaphora resolution in dialogue systems.