Federated deep blockchain-based system for secure verification of academic transcripts and matching study plans in...

Area of Science:

• Computer Science
• Artificial Intelligence
• Blockchain Technology

Background:

• Global education market expansion increases challenges in verifying academic document authenticity.
• Matching academic records between institutions requires secure and efficient verification methods.

Purpose of the Study:

• To introduce a deep Blockchain-based system for verifying, transferring, and matching academic documents.
• To leverage a pre-trained large language model (LLM) for semantic representation extraction from academic transcripts.
• To enhance privacy and collaboration using Federated Learning (FL) within a blockchain framework.

Main Methods:

• Utilized ARABBERTV2, a pre-trained LLM, for extracting semantic representations from academic transcripts.
• Applied dimensionality reduction and classification for detecting equivalences and mismatches in study plans.
• Implemented Federated Learning (FL) for privacy-preserving model fine-tuning and blockchain-coordinated aggregation.
• Evaluated the system on 661 study plans from Saudi universities.

Main Results:

• The proposed model achieved 98.13% classification accuracy and a Kappa statistic of 0.9784 after dimensionality reduction.
• Integrating Federated Learning improved accuracy from 93.5% to 95.6% and AUC-ROC from 0.947 to 0.972.
• The system demonstrated reduced inter-university performance variance and highlighted differences in Saudi universities' primary study plans.
• Deep LLM characteristics contributed to perceptive categorization conclusions.

Conclusions:

• The developed blockchain system efficiently verifies and matches academic documents, crucial for a public framework in academic environments.
• Federated Learning enhances privacy and collaboration in academic document verification.
• The study underscores the potential of LLMs and blockchain in securing and streamlining academic record management.