Task: Temporal Popularity Prediction

The task is focused on predicting the impact of new social media posts (photos, videos or news) before they were shared on social media. Given a new post with the temporal multimedia context of a publisher, the popularity computed by the normalized score of clicks or visits of the post (e.g., tweet count for Twitter, view count for Flickr, etc).

The contestants are asked to develop their temporal prediction models based on the SMP dataset provided by the Challenge (as training data), plus possibly additional public/private data, to address one or both of the given tasks. For the evaluation purpose, a contesting system is asked to produce prediction results of popularity. The accuracy will be evaluated by pre-defined quantitative evaluation. The contestants need to introduce their systems and datasets in the conference.

Evaluation Criteria

The evaluation provided here can be used to obtain performances on the testing set of SMP Challenge. It contains multiple common metrics, including Spearman’s Rho (SR), Mean Absolute Error (MAE), Mean Squared Error (MSE).


(Update) Each team is allowed to submit the results of at most 20 runs, and we will select the latest one as the final (we do not guarantee to evaluate additional runs), which will be measured for performance comparison across teams.

Each submission is required to be formatted in a JSON File as follows.

      "version": "VERSION 1.2",
      "result": [
              "post_id": "post6374637",
              "popularity_score": 2.1345
              "post_id": "post3637373",
              "popularity_score": 3.1415
      "external_data": {
          "used": "true",
          "details": "VGG-19 pre-trained on ImageNet training set"

Copyright © 2019. SMP Challenge Organization Committee. All rights reserved.