Entity Linking

Entity linking finds entity mentions in text or tables and connects them to the right knowledge-base entry, or marks them as unknown when no suitable entry exists. Systems commonly retrieve candidates using aliases, dense or token-level representations, and knowledge-graph structure, then rank them with contextual models, cross-encoders, or large language models; some pipelines also detect mentions end to end. Evaluation spans biomedical, multilingual, historical, product, and table data, with benchmarks measuring linking accuracy, candidate recall, and performance on rare or ambiguous entities. A central tension is balancing accuracy and coverage against inference cost, while candidate retrieval, knowledge-base quality, and changing terminology remain important sources of error.