Skip to main content Skip to navigation

One-shot Learning for Question-Answering in Gaokao History Challenge

Project Overview

The document explores the application of generative AI in education through a deep question-answering (QA) model designed for the Gaokao history exams in China, addressing the complexities of answering intricate historical questions. It introduces a hybrid neural model that employs one-shot learning techniques, combining a cooperative gated neural network (CGNN) with a neural Turing machine (NTM) to optimize answer retrieval and representation learning despite the constraints of limited training data. The findings indicate substantial performance gains compared to conventional methods, showcasing the transformative potential of AI in improving educational assessments. This application not only enhances the accuracy and efficiency of answering complex questions but also underscores the broader implications of integrating advanced AI technologies into educational environments, paving the way for more personalized and adaptive learning experiences.

Key Applications

Hybrid neural model for deep question-answering

Context: University admission exams (Gaokao) in China, primarily targeting high school students preparing for history exams.

Implementation: Utilizes a cooperative gated neural network (CGNN) combined with a neural Turing machine (NTM) for one-shot learning and feature representation.

Outcomes: Achieved substantial performance gains in question-answering accuracy, effectively handling complex historical questions with a small amount of training data.

Challenges: Limited availability of training data, difficulty in accurately capturing semantic relationships in lengthy answers, and the indirect nature of exam questions.

Implementation Barriers

Data Availability

Insufficient training data for the QA task poses a significant challenge.

Proposed Solutions: Implement one-shot learning techniques to optimize performance with minimal training samples.

Semantic Complexity

The complexity and length of historical questions and answers hinder effective feature extraction.

Proposed Solutions: Utilize advanced models like CGNN and NTM to enhance semantic representation and improve retrieval accuracy.

Project Team

Zhuosheng Zhang

Researcher

Hai Zhao

Researcher

Contact Information

For information about the paper, please contact the authors.

Authors: Zhuosheng Zhang, Hai Zhao

Source Publication: View Original PaperLink opens in a new window

Project Contact: Dr. Jianhua Yang

LLM Model Version: gpt-4o-mini-2024-07-18

Analysis Provider: Openai

Let us know you agree to cookies