search

benedekrozemberczki / pytorch_geometric_temporal

PyTorch Geometric Temporal: Spatiotemporal Signal Processing with Neural Machine Learning Models (CIKM 2021)
MIT License
2.61k stars 367 forks source link

I find an error in "Twitter Tennis RG" datasets processing step. #200

Closed Sebastianyang666 closed 1 year ago

Sebastianyang666 commented 1 year ago

In eassy the dataset of "Twiiter Tennis RG and UO" are belong to Dynamic Graph Static Signal,but they are Dynamic Graph Temporal Signal in codes and office web pages.Can you check it again.If just change the Sinal Processing Step,codes can't run successfully.

benedekrozemberczki commented 1 year ago

Do you want to Pull Request it?

ferencberes commented 1 year ago

I'll look into it this week.

Benedek Rozemberczki @.***> ezt írta (időpont: 2022. nov. 20., Vas 18:13):

Do you want to Pull Request it?

— Reply to this email directly, view it on GitHub https://github.com/benedekrozemberczki/pytorch_geometric_temporal/issues/200#issuecomment-1321189615, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABVK23HBCMH55ZDPVNCQCLDWJJL2HANCNFSM6AAAAAAR62FUNY . You are receiving this because you are subscribed to this thread.Message ID: <benedekrozemberczki/pytorch_geometric_temporal/issues/200/1321189615@ github.com>

ferencberes commented 1 year ago

Dear @Sebastianyang666,

Thank you for spotting this error in the paper. "Twitter Tennis RG and UO" are indeed marked as Dynamic Graph Static Signal datasets in the paper but due to the underlying nature of these Twitter datasets, the signal is also dynamic as it represents the number of mentions received by a given node (Twitter account) for the next snapshot. So to the best of my knowledge the implementation is correct.

Originally, these datasets were donated to this project by me so let me know if you need more information on this issue!

Best, Ferenc

benedekrozemberczki commented 1 year ago

Thanks @ferencberes, closing for now!

Sebastianyang666 commented 1 year ago

Dear @Sebastianyang666,

Thank you for spotting this error in the paper. "Twitter Tennis RG and UO" are indeed marked as Dynamic Graph Static Signal datasets in the paper but due to the underlying nature of these Twitter datasets, the signal is also dynamic as it represents the number of mentions received by a given node (Twitter account) for the next snapshot. So to the best of my knowledge the implementation is correct.

Originally, these datasets were donated to this project by me so let me know if you need more information on this issue!

Best, Ferenc

@

I'll look into it this week. Benedek Rozemberczki @.***> ezt írta (időpont: 2022. nov. 20., Vas 18:13): … Do you want to Pull Request it? — Reply to this email directly, view it on GitHub <#200 (comment)>, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABVK23HBCMH55ZDPVNCQCLDWJJL2HANCNFSM6AAAAAAR62FUNY . You are receiving this because you are subscribed to this thread.Message ID: <benedekrozemberczki/pytorch_geometric_temporal/issues/200/1321189615@ github.com>

Thanks @ferencberes . I see. However, if I keep using the current source code “classTwitterTennisDatasetLoader(event_id='rg17', N=None, feature_mode='encoded', target_offset=1)”, it cannot reproduce the results of “Twitter Tennis RG” presented in Table 5 (https://arxiv.org/abs/2104.07788) (because the results were obtained based on the setting that Twitter Tennis RG is Static signal and dynamic Graph, rather than a temporal signal, dynamic graph). Therefore,could you please fix this issue or provide me with any guidance how to reproduce the records for “Twitter Tennis RG” in Table 5? In other words, if we would like to consider the Static signal and dynamic Graph setting, how could I refine/modify the current source code “classTwitterTennisDatasetLoader(event_id='rg17', N=None, feature_mode = 'encoded', target_offset=1)”?

Thanks,