LLMAEL: Large Language Models are Good Context Augmenters for Entity Linking
Amy Xin, Yunjia Qi, Zijun Yao, Fangwei Zhu, Kaisheng Zeng, Xu Bin, Lei Hou, Juanzi Li
2024-07-09
Summary
This paper talks about LLMaEL, a new method that uses large language models (LLMs) to improve entity linking, which is the process of connecting words in text to their corresponding real-world entities, like people or places.
What's the problem?
The main problem is that traditional entity linking models have difficulty identifying less common entities because they don't have enough training data for these 'long-tail' entities. While LLMs are better at understanding these uncommon mentions, they struggle with generating the correct IDs for these entities and training them can be very expensive.
What's the solution?
To solve this issue, the authors developed LLMaEL, which combines the strengths of LLMs and traditional entity linking models. Instead of making LLMs do all the work, LLMaEL uses them to enhance the context around entity mentions by generating additional descriptions. This enriched context is then fed into a standard entity linking model to improve its accuracy. The authors tested LLMaEL on six different datasets and found that it performed better than traditional models in most cases, setting new records for accuracy.
Why it matters?
This research is important because it shows how combining advanced language models with traditional methods can lead to better performance in tasks like entity linking. By improving how entities are identified and linked in text, this approach can enhance various applications in natural language processing, such as search engines, chatbots, and information retrieval systems.
Abstract
Entity Linking (EL) models are well-trained at mapping mentions to their corresponding entities according to a given context. However, EL models struggle to disambiguate long-tail entities due to their limited training data. Meanwhile, large language models (LLMs) are more robust at interpreting uncommon mentions. Yet, due to a lack of specialized training, LLMs suffer at generating correct entity IDs. Furthermore, training an LLM to perform EL is cost-intensive. Building upon these insights, we introduce LLM-Augmented Entity Linking LLMAEL, a plug-and-play approach to enhance entity linking through LLM data augmentation. We leverage LLMs as knowledgeable context augmenters, generating mention-centered descriptions as additional input, while preserving traditional EL models for task specific processing. Experiments on 6 standard datasets show that the vanilla LLMAEL outperforms baseline EL models in most cases, while the fine-tuned LLMAEL set the new state-of-the-art results across all 6 benchmarks.