LMCBert: An Automatic Academic Paper Rating Model Based on Large Language Models and Contrastive Learning.

Area of Science:

• Artificial Intelligence
• Natural Language Processing
• Scholarly Communication

Background:

• Academic paper acceptance relies on a resource-intensive, bias-prone peer-review process.
• Automated academic paper rating (AAPR) methods exist but often use full content, leading to inefficiency and redundancy.
• Existing models like BERT struggle with AAPR due to domain-specific language discrepancies.

Purpose of the Study:

• To develop a more efficient and accurate automated system for predicting academic paper acceptance.
• To address the limitations of existing AAPR methods, including inefficiency and suboptimal performance of pre-trained models.
• To propose LMCBert, a novel model combining large language models (LLMs) and momentum contrastive learning (MoCo).

Main Methods:

• Utilizing LLMs to extract core semantic content from academic papers, reducing redundancy.
• Implementing momentum contrastive learning (MoCo) to optimize BERT training for improved semantic differentiation.
• Developing the LMCBert model integrating LLMs and MoCo for enhanced AAPR.

Main Results:

• LMCBert effectively extracts core semantic information, improving understanding of academic texts.
• MoCo optimization enhances BERT's semantic representation differentiation for AAPR.
• Empirical evaluations confirm LMCBert's effective performance on the evaluation dataset.

Conclusions:

• LMCBert offers a valid and effective approach to automated academic paper rating.
• The integration of LLMs and MoCo significantly improves the accuracy of predicting paper acceptance.
• The proposed method addresses key limitations in current automated academic paper assessment techniques.