Skip to the content.

TPIC2017 Dataset

A social media dataset for temporal popularity prediction

For temporal prediction on popularity, TPIC is an temporal image popularity dataset with 680K photos and corresponding anoynimized photo-sharing records on Flickr ranges of about 36 monthes. Specifically, TPIC is also a multi-faceted data collection, which contains photo image, user profile and photo metadata. We provide the popularity as the normalized views of each photo sharing behaviors. In order to protect the private information of sharing behaviors, we convert the timestamps to segmented time information and indexed of it by integer numbers.

Download Data

Data Organization

Each row of data has a unique picture id(pid) along with user id(uid). All the CSV files listed above have data header that demonstrate the the meaning of the column.


The file organization inside the file contains picture id, user id, comment count, has people, title length, description length, tag count, average view, group count, average member count information:

pid uid commentcount haspeople titlelen deslen tagcount avgview groupcount avgmembercount  
304582	50@N31	0	0	15	0	14	199.32	1188	6601
304592	142@N94	0	0	11	9	0	615.61	67	21637

THe data is crawled from Flickr with user id and photo id anonymized


Data organized inside the file are the phtoto urls correspond to given photo id and user id:

pid uid url
9624	25@N92
665085	275@N38


In order to use temporal information from dataset while protecting the user privacy, we extract year, month, day, and hour index with corresponding photo and user from dataset:

pid uid year month day hour_index
311862	11@N30	2007	3	16	4
311863	89@N59	2007	3	16	4

The definition of hour index is defined below:


The label file contains the popularity (log-views), picture id with associate user id:

pid uid logview
9624	25@N92	3.2
665085	275@N38	2.3


  title={Sequential Prediction of Social Media Popularity with Deep Temporal Context Networks},
  author={Wu, Bo and Cheng, Wen-Huang and Zhang, Yongdong and Qiushi, Huang and Jintao, Li and Mei, Tao},
  location = {Melbourne, Australia}}
  author = {Wu, Bo and Mei, Tao and Cheng, Wen-Huang and Zhang, Yongdong},
  title = {Unfolding Temporal Dynamics: Predicting Social Media Popularity Using Multi-scale Temporal Decomposition},
  booktitle = {AAAI}
  year = {2016},
  location = {Phoenix, Arizona}}


  1. Bo Wu, Wen-Huang Cheng, Yongdong Zhang, Qiushi Huang, Jintao Li, and Tao Mei. 2017. Sequential prediction of social media popularity with deep temporal context networks. In Proceedings of the 26th International Joint Conference on Artificial Intelligence (IJCAI’17). AAAI Press 3062-3068, 19-25 August, 2017, Melbourne, Australia.

  2. Bo Wu, Wen-Huang Cheng, Yongdong Zhang, and Tao Mei. 2016. Time Matters: Multi-scale Temporalization of Social Media Popularity. In Proceedings of the 2016 ACM on Multimedia Conference (MM ‘16). ACM, New York, NY, USA, 1336-1344

  3. Bo Wu, Tao Mei, Wen-Huang Cheng, and Yongdong Zhang, “Unfolding Temporal Dynamics: Predicting Social Media Popularity Using Multi-scale Temporal Decomposition,” In Proceedings of the Thirtieth AAAI Conference on Artificial Intelligence (AAAI’16). AAAI Press 272-278, 12-17 February, 2016, Phoenix, USA.

Tag Reference

Jang, Jin Yea, et al. “Generation like: Comparative characteristics in instagram.” Proceedings of the 33rd Annual ACM Conference on Human Factors in Computing Systems (CHI’15), 2015.


Bo Wu (Chinese Academy of Sciences, Microsoft Research Aisa)