Skip to main content

Table 5 Comparison of training Level 1 category predictions with and without cost sensitivity.

From: Cost sensitive hierarchical document classification to triage PubMed abstracts for manual curation

Number of references

No cost

Cost sensitive

Classified as high priority

13722

15020

   Correct classification

12515

12978

   Incorrect, should be...

1207

2042

Other high priority

407

464

Low priority

800

1578

Classified as low priority

9111

7813

   Correct classification

7799

7112

   Incorrect, should be...

1312

701

Other low priority

325

234

High priority

987

467

  1. The number of references predicted into the Level 1 categories with and without cost sensitivity. In the cost sensitive scenario, there was a decrease in the number of high priority references misclassified into low priority categories.