[Paper Review] Expecting to be HIP: Hawkes Intensity Processes for Social Media Popularity

Marian-Andrei Rizoiu, Lexing Xie+4 more|DIGITAL.CSIC (Spanish National Research Council (CSIC))|Feb 19, 2016

Diffusion and Search Dynamics29 references21 citations

TL;DR

This paper proposes the Hawkes Intensity Process (HIP), a novel mathematical model that quantifies how external promotions (e.g., tweets) drive video popularity on platforms like YouTube. By modeling the interplay between exogenous stimuli and endogenous virality, HIP improves popularity forecasting accuracy by 28.6% over history-only baselines and enables identification of videos with high viral potential or low promotional sensitivity.

ABSTRACT

Modeling and predicting the popularity of online content is a significant problem for the practice of information dissemination, advertising, and consumption. Recent work analyzing massive datasets advances our understanding of popularity, but one major gap remains: To precisely quantify the relationship between the popularity of an online item and the external promotions it receives. This work supplies the missing link between exogenous inputs from public social media platforms, such as Twitter, and endogenous responses within the content platform, such as YouTube. We develop a novel mathematical model, the Hawkes intensity process, which can explain the complex popularity history of each video according to its type of content, network of diffusion, and sensitivity to promotion. Our model supplies a prototypical description of videos, called an endo-exo map. This map explains popularity as the result of an extrinsic factor - the amount of promotions from the outside world that the video receives, acting upon two intrinsic factors - sensitivity to promotion, and inherent virality. We use this model to forecast future popularity given promotions on a large 5-months feed of the most-tweeted videos, and found it to lower the average error by 28.6% from approaches based on popularity history. Finally, we can identify videos that have a high potential to become viral, as well as those for which promotions will have hardly any effect.

Motivation & Objective

• To model the complex dynamics of online content popularity under continuous external influence.
• To quantify the relationship between external promotions (e.g., on Twitter) and endogenous popularity growth (e.g., on YouTube).
• To develop a forecasting model that incorporates both exogenous stimuli and intrinsic viral potential.
• To identify videos with high viral potential or low sensitivity to promotion.

Proposed method

• Proposes the Hawkes Intensity Process (HIP), an extension of the Hawkes point process that models expected event volumes rather than event times.
• Uses expectation over stochastic event histories to derive a deterministic intensity function for popularity prediction.
• Introduces two key metrics: endogenous response (inherent virality) and exogenous sensitivity (response to promotion).
• Employs the endo-exo map, a 2D visualization tool combining sensitivity and endogenous response to classify video potential.
• Applies HIP to a 5-month dataset of 81.9M YouTube videos linked to 1.06B tweets, using #shares and #tweets as exogenous inputs.
• Uses paired and two-sample T-tests to validate forecasting performance against baseline MLR models.

Experimental results

Research questions

• RQ1 How does continuous external promotion influence the popularity dynamics of online content?
• RQ2 Can we predict future popularity of a video given planned external promotions?
• RQ3 Which videos are most likely to go viral, and which are insensitive to promotion?
• RQ4 How do different external stimuli (e.g., #shares vs. #tweets) compare in forecasting power?
• RQ5 Can we model the interplay between intrinsic virality and external promotion using a unified mathematical framework?

Key findings

• HIP reduces average forecasting error by 28.6% compared to popularity history-based models like MLR.
• The model achieves a median absolute forecasting error of 3.25% on videos with high exogenous shocks, significantly outperforming MLR’s 6.5% median error.
• No statistically significant difference was found between using #shares or #tweets as external stimuli, as both showed near-identical performance (Cohen’s d = -0.05 for HIP, 0.00 for MLR).
• The correlation between #shares and #tweets time series was high across videos, with a mean Pearson’s r of 0.78 and median of 0.87.
• HIP significantly outperforms MLR in forecasting, with p-values < 10−95 and effect sizes (Cohen’s d) of 0.197–0.253, indicating strong statistical significance.
• The endo-exo map successfully identifies videos with high viral potential (high sensitivity and high endogenous response), enabling targeted promotion strategies.