# Popularity Prediction for Social Media over Arbitrary Time Horizons

**Authors:** Daniel Haimovich, Dima Karamshuk, Thomas J. Leeper, Evgeniy Riabenko, Milan Vojnovic  
**arXiv:** [2009.02092](https://arxiv.org/abs/2009.02092)  
**Date:** 2020-09-04  
**Summary page:** https://www.summarizepaper.com/en/paper/2009.02092v2/

## Key points

- The paper addresses the challenge of predicting the popularity of social media content in real-time.
- Accurate prediction of content popularity is crucial for various applications, including detecting and moderating harmful viral content.
- The prediction task is complex due to interactions between user interests, content features, resharing, feed ranking, and network structure.
- The proposed approach involves a feature-based approach based on a self-excited Hawkes point process model.
- The approach considers static features of both the content and users as well as observed popularity growth up to the prediction time.
- By combining predictions of popularity at reference horizons and an effective growth parameter, the method offers scalability and achieves high accuracy compared to leading baselines.
- Experiments conducted on a dataset consisting of public page content on Facebook over a two-month period demonstrate that the model outperforms separately trained models for specific prediction time horizons.
- Overall, the paper presents a highly scalable method for predicting social media content popularity over arbitrary time horizons that considers various factors influencing popularity and achieves competitive prediction accuracy.

## Summary

The paper titled "Popularity Prediction for Social Media over Arbitrary Time Horizons" addresses the challenge of predicting the popularity of social media content in real-time. The ability to accurately predict content popularity is crucial for various applications, including the detection and moderation of harmful viral content. The prediction task is complex due to the interactions between user interests, content features, resharing, feed ranking, and network structure. To achieve high accuracy for different prediction time horizons, it is essential for models to consider static features of both the content and users as well as observed popularity growth up to the prediction time. In this study, the authors propose a feature-based approach based on a self-excited Hawkes point process model. This approach involves predicting the popularity of content at one or more reference horizons while also predicting an effective growth parameter that reflects the timescale of popularity growth. By combining these predictions, the proposed method offers scalability and achieves a high degree of accuracy compared to several leading baselines. To evaluate their approach, the authors conducted experiments on a dataset consisting of public page content on Facebook over a two-month period. The dataset covered billions of content views and hundreds of thousands of distinct content items. The results demonstrate that their model outperforms separately trained models for specific prediction time horizons. Overall, this paper presents a highly scalable method for predicting social media content popularity over arbitrary time horizons. The proposed approach considers various factors influencing popularity and achieves competitive prediction accuracy. This research contributes to improving timely content moderation and enhancing our understanding of social media dynamics.