Issue with fine classes in trec dataset

[albertvillanova]

Describe the bug

According to their paper, the TREC dataset contains 2 kinds of classes:

• 6 coarse classes: TREC-6
• 50 fine classes: TREC-50

However, our implementation only has 47 (instead of 50) fine classes. The reason for this is that we only considered the last segment of the label, which is repeated for several coarse classes:

• We have one desc fine label instead of 2:
  • DESC:desc
  • HUM:desc
• We have one other fine label instead of 3:
  • ENTY:other
  • LOC:other
  • NUM:other

From their paper:

We define a two-layered taxonomy, which represents a natural semantic classification for typical answers in the TREC task. The hierarchy contains 6 coarse classes and 50 fine classes,

Each coarse class contains a non-overlapping set of fine classes.