search

deepkashiwa20 / DeepUrbanEvent

[ACM TIST 2021] [KDD 2019 Paper Applied Data Science Track] DeepUrbanEvent: A System for Predicting Citywide Crowd Dynamics at Big Events
MIT License
3 stars 0 forks source link

Applied application cases. #2

Open aPhoneixCat opened 8 months ago

aPhoneixCat commented 8 months ago

I am conducting research on the crowd management for HK during typhoon. Your paper inspires me a lot. Just curious, yet, if there're any applied application in Japan regarding this paper’s theory?

deepkashiwa20 commented 7 months ago

I am conducting research on the crowd management for HK during typhoon. Your paper inspires me a lot. Just curious, yet, if there're any applied application in Japan regarding this paper’s theory?

Thank you so much for your kind message. We are trying to build the prototype system for Tokyo Metroplitan Government, but not released yet. And the main scenario may still be earthquake for Japan.

deepkashiwa20 commented 7 months ago

Also welcome to check our recent study that utilizes multimodal data for nowcasting people flow during typhoon.

[WWW23] R. Jiang, Z. Wang, Y. Tao*, C. Yang, X. Song#, R. Shibasaki, S. Chen#, M. Shyu, "Learning Social Meta-knowledge for Nowcasting Human Mobility in Disaster", Proc. of the ACM Web Conference (WWW), 2023.

aPhoneixCat commented 7 months ago

Thank you so much for your provided information. Such a good paper which helps a lot. We're also considering crawling social media data to enhance the model but we're still struggling how to obtain stable and real-time GPS data. Currently we're using GPS traces from OpenStreetMap, which is less real-time and stable.

aPhoneixCat commented 7 months ago

Also welcome to check our recent study that utilizes multimodal data for nowcasting people flow during typhoon.

[WWW23] R. Jiang, Z. Wang, Y. Tao*, C. Yang, X. Song#, R. Shibasaki, S. Chen#, M. Shyu, "Learning Social Meta-knowledge for Nowcasting Human Mobility in Disaster", Proc. of the ACM Web Conference (WWW), 2023.

Btw, I've read this paper, the datasets used for Japan and US is GPS trajectory and POI visitation data. Would be very curious why using different type of datasets? Will that impact the model evaluation performance？ Assume GPS trajectory might be the most reliable dataset for HMD research

deepkashiwa20 commented 7 months ago

Thanks, there are two reasons: 1) we are very data rich in Japan, however our collaborator in US is not that rich, perhaps due to more strict regulation on GPS trajectory data. So, we use an open dataset called SafeGraph that records the POI-specific visitation number, which is also of great quality, even for crowd management. 2) we would like to prove our proposed methodolog could be applied to various data sources.

aPhoneixCat commented 7 months ago

Hope u have a great day! Really great thanks for the explanation. Totally understood. We share the same problem with u for GPS data in Hong Kong. Currently, we're working on a project for HK government on Human Mobility Management, struggling to explore possible datasets to support our research in HK Lol.

deepkashiwa20 commented 7 months ago

Perhaps you can try to negotiate with 1) taxi company or ride-sharing service company 2) bus company 3) metro company 4) mobile service providers such as CMHK, PCCW, UnicomHK, smartone, etc. Usually, 1~3 might be easier than 4. Anyway, wish you could find the right data to deploy the simiar AI system in HK. ^_^

aPhoneixCat commented 7 months ago

Really appreciate. Best wishes to your paper/project's success in JP too.