search

zjukg / AdaMF-MAT

[Paper][LREC-COLING 2024] Unleashing the Power of Imbalanced Modality Information for Multi-modal Knowledge Graph Completion
18 stars 3 forks source link

[ASAP] About raw data used in paper code #3

Closed revoluzionario closed 4 months ago

revoluzionario commented 4 months ago

Hello, I have some questions to ask about the raw data used in this paper:

  1. Did you use the embeddings which is provided beforehand, or generate it from the raw data?
  2. If you use raw data, where did you retrieve it, and did you check it before generating embeddings, because I checked and realized that the number of entities in the dataset you referenced and the statistics is not similar.
  3. I tried to scrap data from the links you provided in /benchmarks but MKG-W data is broken, meanwhile MKG-Y doesn't consist links to scrap. Can you check it out?

Thanks! Hope to receive your answer!

Zhang-Each commented 4 months ago

Hello, here are the answers for your questions.

  1. The embeddings of MKG-W/MKG-Y are provided by https://github.com/quqxui/MMRNS, the embeddings of DB15K are provided by https://github.com/mniepert/mmkb. We reuse their released entity features.
  2. The statistical information of the datasets are the same as https://github.com/quqxui/MMRNS. MKG-W/MKG-Y have 15000 entities and DB15K has 12842 entities.
  3. I can not get your point about "broken data" and the MKG-Y embeddings can be clearly obtained in our google driver link. We have already upload the embeddings of the three dataset. We have update the embeddings by a second-time upload.