Enhanced Peer Review Analysis using transformer based sentiment distillation

This thesis investigates the utilization of transformer-based models to optimize the automation and accuracy of peer review analysis and decision-making in scholarly publishing. By integrating cutting-edge techniques, we have designed prompt to extract influential sections from peer reviews and the implementation of sentiment analysis to forecast paper acceptance determinations. The methodology encompasses a meticulous process of generating a “Critique Summary” that distills the sentimental polarity with influential excerpts, incorporating both review ratings and textual content. Subsequently, a state-of-the-art transformer model, such as Llama 2, undergoes fine-tuning using the synthesized dataset to refine its predictive capabilities for paper acceptance decisions. Through this innovative approach, we have tried to streamline the peer review workflow, markedly diminishing the duration between initial reviews and final editorial decisions, thus enhancing the overall efficacy and efficiency of academic publishing endeavors.