RLAS-BIABC: A Reinforcement Learning-Based Answer Selection Using the BERT Model Boosted by an Improved ABC...

Area of Science:

• Natural Language Processing
• Artificial Intelligence
• Machine Learning

Background:

• Answer selection is crucial for open-domain question answering.
• Existing methods struggle with imbalanced datasets (more negative than positive answer pairs).
• This imbalance significantly degrades classifier performance.

Purpose of the Study:

• To propose RLAS-BIABC, a new method for answer selection.
• To address data imbalance issues in answer selection tasks.
• To improve the overall performance of question answering systems.

Main Methods:

• Utilizes attention mechanism-based Long Short-Term Memory (LSTM) and Bidirectional Encoder Representations from Transformers (BERT) word embeddings.
• Employs an improved Artificial Bee Colony (ABC) algorithm for pretraining and a reinforcement learning (RL) algorithm for training.
• Introduces a mutual learning technique to enhance the ABC algorithm by modifying candidate solutions based on fitness.

Main Results:

• The proposed RLAS-BIABC method effectively handles imbalanced classification by framing it as a sequential decision-making process.
• The RL agent is trained with rewards favoring minority classes, optimizing policy weights initialized by the improved ABC algorithm.
• Evaluated on LegalQA, TrecQA, and WikiQA datasets, RLAS-BIABC demonstrates state-of-the-art performance.

Conclusions:

• RLAS-BIABC offers a robust solution for answer selection, particularly in scenarios with imbalanced data.
• The combination of BERT, LSTM, RL, and an enhanced ABC algorithm leads to superior performance.
• The method represents a significant advancement in open-domain question answering technology.