LLM-Based Fine-Tuning Data Generation for Relation Triplet Extraction with Expert Ensemble and Demonstration Selection

The task of relation triplet extraction is crucial in the field of natural language processing as it enables the extraction of meaningful relational information from text. However, the current lack of high-quality fine-tuning data severely limits the performance of this task. This paper introduces a methodology for LLM-based fine-tuning data generation aimed at enhancing relation triplet extraction using expert ensemble and demonstration selection. Despite the zero-shot capabilities of Large Language Models (LLMs), these models often struggle with complex tasks such as relation triplet extraction. To address this, we propose a multi-step method: guiding ChatGPT to generate documents with relevant relational information, em-ploying demonstration selection to choose pertinent examples, guiding ChatGPT to extract relation triplets by Chain of Thought (CoT) prompting method and fine-tuning the LLAMA model using Low-Rank Adaptation (LoRA). Our results show significant improvements in performance compared to baselines.