English-focused CL-HAMC with contrastive learning and hierarchical attention for multiple-choice reading...

Area of Science:

• Artificial Intelligence
• Natural Language Processing
• Educational Technology

Background:

• Multiple-choice questions are vital for assessing language proficiency in standardized tests.
• Manual analysis of test items for educational content is time-consuming and labor-intensive.
• AI-powered Machine Reading Comprehension for Multiple Choice (MCRC) offers a solution for automated content generation.

Purpose of the Study:

• To address limitations in existing MCRC models, specifically misclassification of similar distractors and handling questions requiring indirect reasoning.
• To propose a novel AI model, the Contrastive Learning-driven Hierarchical Attention Model for Multiple Choice (CL-HAMC), for enhanced MCRC.
• To improve the accuracy and efficiency of generating auxiliary educational content for English language tests.

Main Methods:

• Developed a hierarchical attention model utilizing multi-head attention to simulate human multi-layered reasoning.
• Integrated a contrastive learning strategy to enhance the model's ability to distinguish subtle semantic differences in answer options.
• Employed hierarchical modeling of passage-question-option interactions to capture complex relationships.

Main Results:

• The CL-HAMC model achieved state-of-the-art (SOTA) performance on the RACE, RACE-M, and RACE-H benchmarks.
• Demonstrated substantial and consistent performance gains across multiple datasets.
• Showcased competitive results on the DREAM dataset, indicating broad applicability.

Conclusions:

• The proposed CL-HAMC model effectively addresses challenges in MCRC, particularly with difficult distractors and implicit answers.
• This AI approach offers a significant advancement in the automated processing of multiple-choice questions for English language learning.
• The study provides a robust solution for creating high-quality, AI-generated educational materials, reducing manual effort.