Question about relation2id.txt in TIVA benchmark

[broiron]

Hello, and thank you for sharing such excellent work!

While exploring the benchmarks, I noticed that in benchmarks/TIVA/relation2id.txt each “relation name” is simply a number that maps to the same numeric ID (e.g., 0 0, 1 1, 4 2, …). In the other benchmark folders, relation2id.txt maps human‑readable relation names to IDs.

Could you clarify why TIVA uses this number‑to‑number mapping? Is there a separate file (or script) that links these numeric IDs back to the original textual relation names?

Thank you in advance for your help!

[Zhang-Each]

Hi, thanks for your interest in our work. This is because the format of the raw TIVA data is complex, and we used this simple notation to label the different relations during pre-processing. For the processed TIVA dataset, we are currently unable to find the original text-to-id mapping. In our work, we also do not use the text information of relations in our model.