Unregistered biological words recognition by Q-learning with transfer learning.

Area of Science:

• Bioinformatics
• Natural Language Processing
• Machine Learning

Background:

• Identifying unregistered biological words is crucial for biological text mining.
• Current methods for recognizing out-of-vocabulary biological terms show unsatisfactory performance.

Purpose of the Study:

• To develop an improved algorithm for detecting unregistered biological words.
• To leverage transfer learning and Q-learning for enhanced biological term recognition.

Main Methods:

• A Q-learning algorithm was employed, viewing word identification as a Markov process.
• Transfer learning was integrated to optimize knowledge transfer between related tasks.
• A reinforcement learning framework automatically handled feature mapping for transfer learning.

Main Results:

• The proposed Q-learning with transfer learning approach demonstrated superior performance on GENIA corpus and JNLPBA04 datasets.
• Precision, recall, and F-score significantly surpassed conventional methods and Q-learning without transfer learning.

Conclusions:

• The developed algorithm effectively enhances the recognition of unregistered biological words.
• Combining Q-learning with transfer learning offers a promising direction for biological NLP tasks.