SparkRA: A Retrieval-Augmented Knowledge Service System Based on Spark Large Language Model

TL;DR

This work tackles the gap in LLMs for scientific literature services by building SciLit-LLM through continual pre-training and supervised fine-tuning on scholarly texts, based on the iFLYTEK Spark LLM. It then deploys SparkRA, a three-function knowledge service (literature investigation, paper reading, academic writing) that operates in English and Chinese. Experimental results show SparkRA outperforms baselines including GPT-3.5 and Llama3-8B across tasks, and even surpasses GPT-4 in paper-polishing quality, with strong translation and polishing metrics. The system demonstrates real-world impact, with over 50,000 registered users and over 1.3 million total interactions as of mid-2024.

Abstract

Large language models (LLMs) have shown remarkable achievements across various language tasks.To enhance the performance of LLMs in scientific literature services, we developed the scientific literature LLM (SciLit-LLM) through pre-training and supervised fine-tuning on scientific literature, building upon the iFLYTEK Spark LLM. Furthermore, we present a knowledge service system Spark Research Assistant (SparkRA) based on our SciLit-LLM. SparkRA is accessible online and provides three primary functions: literature investigation, paper reading, and academic writing. As of July 30, 2024, SparkRA has garnered over 50,000 registered users, with a total usage count exceeding 1.3 million.

Figures (4)

• Figure 1: The process of building SparkRA system.
• Figure 2: The system architecture of SparkRA integrates iFLYTEK Spark LLM and Scientific Literature LLM to facilitate literature investigation, paper reading, and academic writing.
• Figure 3: The architecture of RAG-based literature investigation.
• Figure 4: Literature investigation page.