Abstract

Some Question Answering (QA) systems rely on knowledge bases (KBs) to provideaccurate answers. Entity Linking (EL) plays a critical role in linking naturallanguage mentions to KB entries. However, most existing EL methods are designedfor long contexts and do not perform well on short, ambiguous user questions inQA tasks. We propose an entity linking agent for QA, based on a Large LanguageModel that simulates human cognitive workflows. The agent actively identifiesentity mentions, retrieves candidate entities, and makes decision. To verifythe effectiveness of our agent, we conduct two experiments: tool-based entitylinking and QA task evaluation. The results confirm the robustness andeffectiveness of our agent.

 

Quick Read (beta)