Table 3 Kappa values for the task-specific extraction using LLMs.

From: Unveiling the Spatiotemporal Dynamics of Global Brain Circulation: A Comprehensive Corpus (2000–2024)

Entity

Token Size

Recall

Precision

F1 Score

Origin

<200 tokens

0.927

0.929

0.927

Origin

200–1000 tokens

0.923

0.935

0.928

Origin

1000–3000 tokens

0.919

0.928

0.952*

Origin

>3000 tokens

0.907

0.935

0.921

Destination

<200 tokens

0.872

0.881

0.876

Destination

200–1000 tokens

0.893

0.919

0.905

Destination

1000–3000 tokens

0.917

0.922

0.919*

Destination

>3000 tokens

0.894

0.870

0.882

Timestamp-Year

<200 tokens

0.784

0.827

0.804

Timestamp-Year

200–1000 tokens

0.853

0.841

0.846

Timestamp-Year

1000–3000 tokens

0.883

0.912

0.897*

Timestamp-Year

>3000 tokens

0.859

0.849

0.854

Timestamp-Month

<200 tokens

0.841

0.885

0.862

Timestamp-Month

200–1000 tokens

0.877

0.933

0.904*

Timestamp-Month

1000–3000 tokens

0.883

0.925

0.903

Timestamp-Month

>3000 tokens

0.862

0.881

0.871

  1. *Best F1 scores for each entity are shown in bold.