Use Large Language Models for Named Entity Disambiguation in Academic Knowledge Graphs
Shaojun Liu, Yanfeng Fang · Atlantis Highlights in Computer Sciences/Atlantis highlights in computer sciences · 2023
This study investigates the application of large language models (LLMs) in disambiguating homonymous named entities in academic knowledge graphs.Current state-of-the-art methods rely on supervised learning techniques that often necessitate extensive annotated datasets, which may be scarce in specialized domains.For further exploration, we constructed an academic knowledge graph in the science and technology domain using publicly available data and extracted contrasting homonymous named entities from different projects to create a test dataset.We evaluated the performance of the ChatGPT model on this dataset using zero-shot, in-context, and chain-of-thought prompting strategies.The experimental results reveal that while LLMs achieve limited success in a zero-shot setting, chain-of-thought prompting can enhance their reasoning abilities.However, a performance gap persists when compared to supervised learning methods specifically trained on the dataset.These findings suggest that LLMs, such as ChatGPT, present a promising direction for assisting in knowledge graph construction for named entity disambiguation, particularly when labeled data is scarce.The utilization of LLMs could be especially beneficial for domains lacking extensive annotated datasets, offering a competitive alternative for disambiguating homonymous named entities.