Advances in Knowledge Graph-Based Natural Language Processing

The field of natural language processing is witnessing a significant shift towards the integration of knowledge graphs to enhance factual accuracy and reduce hallucination in large language models. Researchers are exploring innovative approaches to align knowledge graphs with language models, such as infusing entity embeddings into the latent space of language models. This direction has shown promising results in improving the factuality of language models and reducing hallucination. Another area of focus is the development of methods to generate SPARQL queries from natural language questions, with a emphasis on evaluating the quality of large language models and estimating the influence of training data on question answering quality. Noteworthy papers in this area include:

  • ALIGNed-LLM, which introduces a simple yet effective approach to improve language models' factuality via a lean strategy to infuse KGs into the latent space of language models.
  • WebShaper, which proposes a formalization-driven IS data synthesis framework to construct a dataset for training IS agents.
  • mKGQAgent, which introduces a human-inspired framework that breaks down the task of converting natural language questions into SPARQL queries into modular, interpretable subtasks.

Sources

Aligning Knowledge Graphs and Language Models for Factual Accuracy

SPARQL Query Generation with LLMs: Measuring the Impact of Training Data Memorization and Knowledge Injection

WebShaper: Agentically Data Synthesizing via Information-Seeking Formalization

mRAKL: Multilingual Retrieval-Augmented Knowledge Graph Construction for Low-Resourced Languages

Text-to-SPARQL Goes Beyond English: Multilingual Question Answering Over Knowledge Graphs through Human-Inspired Reasoning

Built with on top of