Ontology-Free General-Domain Knowledge Graph-to-Text Generation Dataset Synthesis using Large Language Model

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, medium

Summary

Daehui Kim, Deokhyung Kang, Sangwon Ryu, and Gary Lee introduce WikiOFGraph, a new large-scale dataset designed to address the scarcity of high-quality, general-domain Knowledge Graph-to-Text (G2T) generation datasets. G2T generation converts structured knowledge graphs into natural language text, and its performance with Pretrained Language Models (PLMs) depends on precise graph-text alignment. WikiOFGraph contains 5.85M general-domain graph-text pairs, synthesized using a novel method that combines Large Language Models (LLMs) and Data-QuestEval. This dataset achieves high graph-text consistency without relying on external ontologies. Experimental results confirm that PLMs fine-tuned on WikiOFGraph surpass those trained on alternative datasets across multiple evaluation metrics, establishing the method as a scalable and effective solution for G2T data generation.

Key takeaway

For NLP Engineers developing Knowledge Graph-to-Text (G2T) systems, this research offers a critical resource. You should consider fine-tuning your Pretrained Language Models (PLMs) on the WikiOFGraph dataset, as it demonstrably improves performance across various evaluation metrics. This approach allows you to build more robust general-domain G2T models without the complexities of external ontologies, accelerating development and deployment of high-quality text generation from structured data.

Key insights

A new LLM-driven method synthesizes a large, ontology-free G2T dataset, significantly improving PLM performance in general-domain text generation.

Principles

• High-quality G2T data requires precise graph-text alignment.
• Ontology-free datasets enhance general-domain applicability.
• LLMs can synthesize large, consistent G2T datasets.

Method

The method synthesizes G2T datasets by combining Large Language Models (LLMs) with Data-QuestEval to generate 5.85M graph-text pairs, ensuring high consistency without external ontologies.

In practice

• Fine-tune PLMs on WikiOFGraph for G2T tasks.
• Use LLMs for scalable dataset synthesis.
• Develop G2T systems without ontology constraints.

Topics

• Knowledge Graph-to-Text
• Large Language Models
• Dataset Synthesis
• Pretrained Language Models
• Natural Language Generation
• WikiOFGraph

Best for: Research Scientist, AI Scientist, NLP Engineer, Machine Learning Engineer

Related on AIssential

• Construction of Knowledge Graph based on Language Model — PLMs significantly enhance Knowledge Graph construction by automating entity and relation extraction from text.
• Domain-Specific Data Synthesis for LLMs via Minimal Sufficient Representation Learning — DOMINO inductively synthesizes domain-specific data for LLMs by learning minimal representations from reference examples, bypassing explicit domain descriptions.
• Building LLM From Scratch: Understanding How Large Language Models Work — Large Language Models fundamentally predict the next token by processing numerical representations of text through attention mechanisms.
• Specializing a Small Language Model for Closed-Domain Portuguese RAG using Knowledge Graph Supervision — Fine-tuning SLMs with knowledge graph supervision significantly enhances closed-domain RAG performance, surpassing larger LLMs.
• Semantic adapters in text-to-SQL for low-resource languages: the importance of semantic information — Injecting semantic structural knowledge into LLMs significantly boosts Text-to-SQL performance for low-resource languages.
• System Report for CCL25-Eval Task 5: New Dataset and LoRA-Fine-Tuned Qwen2.5 — Domain-specific datasets and LoRA fine-tuning significantly enhance LLM performance in specialized tasks like classical poetry analysis.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.