PaperlayerView exact revision on arXiv

Source-linked AI summary

Measuring the impact of hits

Matúš Medo, Liudmila Rozanova

arXiv:2609.35104v1physics.soc-phcs.SI

TL;DR

It remains unclear whether real systems exhibit competition for limited attention, despite its central role in popularity models. The paper compares attention decay around high-attention hits with baseline periods, finding weak or negligible impacts across three datasets. It concludes that resources such as attention may vary endogenously, with implications for modeling and e-commerce.

  • Problem

    Although popularity models often assume limited attention and competition, the strength and presence of competition in real systems remain unclear.

    Source: Introduction · pp. 1–2

  • Method

    The framework compares attention decay in baseline periods with decay after a high-attention hit arrives to measure competition for collective attention.

    Source: Introduction · p. 2

  • Results

    Hits have a small or negligible impact across three datasets, although Reddit non-hit comment rates briefly drop by about 40% at peak hit activity and recover within minutes.

    Source: Discussion · p. 13

  • Takeaways & Limitations

    The findings support treating attention as potentially endogenous and suggest that measuring short- and long-term competitor impacts can inform movie-release decisions.

    Source: Introduction · p. 2, Discussion · p. 14

    The slowing-factor framework is especially suited to systems where hits reach high activity quickly; generalization to gradually building popularity remains open.

    Source: Discussion · p. 13

Abstract

from arXiv · show

Many real systems can be represented as growing networks where new nodes and links gradually emerge. The Barab\'asi-Albert model for growing networks, and many models inspired by it, are based on the idea that nodes compete for links. However, the strength and the very presence of this competition have not been tested. We propose a robust statistical approach to quantify how strongly nodes compete for links, and apply it to data from various real systems---commenting on online news and cinema attendance data. We find a range of possible behaviors, from the perfectly elastic case, where new entrants shape network growth in a way that leaves the rest of the system unaffected, to an intermediate case where new entrants measurably affect the rest. Perfect competition is never observed. These findings have direct implications for complex systems modeling and e-commerce applications.

Introduction

Popularity models often assume that items compete for limited attention or network links, but how strongly real systems exhibit this competition remains unclear. The paper introduces a statistical comparison of attention decay and finds weak or absent competition in three real systems.

Introduction

  • Growing-network models such as Barabási-Albert assume nodes compete for incoming links, while other popularity models treat item success as independent.Competition assumptions can shape system behavior, including criticality and heavy-tailed popularity distributions.Source: Introduction · p. 1
  • The proposed measurement compares attention decay when no successful new item arrives with decay after a high-attention hit appears.The comparison estimates collective-attention elasticity, and a tunable network-growth model tests the measurement across competition levels.Source: Introduction · p. 2
  • Both commenting datasets show no statistically significant competition, while US box office data shows only weak competition.The datasets cover BBC sports articles, Reddit worldnews posts, and 36 years of weekly US box office data.Source: Introduction · p. 2
  • The framework distinguishes systems where resources such as attention or funding are exogenous from those where they can vary endogenously.The authors report weak or absent competition in three real systems and identify direct commercial implications.Source: Introduction · p. 2

Results

Hits have no statistically significant sustained effect on other news items in either commenting dataset, and their effect on cinema revenues is weak over the five-week period studied.

Audience differences

  • Hits do not appear to reach high impact by activating only less-active users; they attract attention nearly uniformly across user activity levels.The analysis compares article–user group links with randomized data that preserve users’ total activity and articles’ total impact.Source: BBC Sport comment sections · p. 4

Elasticity validation

  • Synthetic FAE tests detect hit effects for most elasticity values, but not significantly for λ ≳0.96, supporting high elasticity as an explanation for the news-data null result.The model spans perfect elasticity (λ = 1), where hits do not affect other articles, and no elasticity (λ = 0), where hits sharply reduce their attention.Source: Figure 1 · p. 5, The effect of hits on the other… · p. 6

Reddit discussions

  • Reddit hit appearances raise total comment rate by ∼3.5×, driven almost entirely by the hit post, while non-hit posts remain near ∼20 comments per minute.The background ratio nevertheless drops to 0.60 at the hit peak versus 1.97 in baseline, a transient ∼40% suppression.Source: Reddit discussions · p. 6, Figure 2 · p. 7, Table 1 · p. 8

Results

Cinema box office data

  • In cinema, the baseline slowing factor is 0.58 (a 12.9-day decay timescale); hit-related differences become significant from week 3 but remain small over five weeks.Movies typically reach the majority of their box office income during the displayed five-week period.Source: Figure 3 · p. 8, Cinema box office data · p. 9

Methods

The methods compare activity around hit arrivals with no-hit baselines, estimate a slowing factor, and use a tunable synthetic model of attention elasticity. The analysis uses BBC and Reddit comments and weekly US box-office data.

Methods

  • The BBC dataset covers 3,087 sports articles, 852,400 comments, and 67,527 readers collected from October 2018 through June 2019, with comments timestamped to the minute.Articles were generally open for comments for 24–48 hours.Source: Empirical datatasets · p. 9
  • In worldnews, the 25 most-commented posts were treated as hits and paired with 25 baseline intervals, yielding 50 aligned activity trajectories; the top 10% of posts drew over 90% of comments.The dataset comprised approximately 56,000 posts and 2.3 million comments in the first quarter of 2018.Source: Empirical datatasets · pp. 9–10
  • The cinema analysis uses cleaned weekly box-office data for 14,000 movies released from 1983 to 2018, with hits selected by within-year gross-revenue percentile.Weekly aggregation was used to suppress attendance variation across the days of the week.Source: Empirical datatasets · p. 10
  • The estimator compares activity before and after hit arrivals while excluding windows with other arrivals, then uses Poisson likelihoods, maximum likelihood, and bootstrap estimates to assess the change.The hit article is excluded from the hit-scenario likelihood; the difference between baseline and hit estimates is assessed with a z-score and two-tailed p-value.Source: Measuring the effect of hits on other… · pp. 10–11
  • The FAE model tunes collective-attention elasticity λ from non-elastic competition at 0 to perfectly elastic independence at 1, while keeping data volume approximately unchanged.Article impacts decay exponentially with age, and the model’s λ = 1 expected comment rate for an article is independent of other articles’ fitness and appearance times.Source: The FAE model with Fitness, Aging, and… · pp. 11–12

Discussion

Newly emerging hits have small or negligible effects on the rest of the system overall, though Reddit shows a brief suppression of non-hit comments. The findings also delimit where the measurement applies and challenge the generality of widely observed popularity mechanisms.

Discussion

  • Hits have a small or negligible impact across the three datasets, despite often attracting more attention than all other items combined.The decay of attention after a hit is comparable to normal aging without new competition, suggesting weaker competition for collective attention than previously thought.Source: Discussion · pp. 13–14
  • At peak hit activity, Reddit’s non-hit comment rate drops by ∼40% and recovers within minutes, indicating competition on timescales under 5 minutes.The 10-minute slowing-factor windows do not capture this transient effect.Source: Discussion · p. 13
  • The slowing factor is best suited to platforms where hits reach high activity quickly; generalizing the framework to gradually growing popularity remains open.Gradual growth may result from preferential attachment or expectations building before an event.Source: Discussion · p. 13
  • For systems with different item types, focusing on items of the same type may improve sensitivity, but requires choosing category granularity and can increase statistical uncertainty.Source: Discussion · p. 13
  • The analysis covers three datasets; whether similarly elastic collective attention holds for other cultural products, economic goods, and services remains for future work.Source: Discussion · p. 13
  • The findings limit the generality of mechanisms such as preferential attachment, and the models and methods may inform online newspaper editors and content creators.For e-commerce, measuring both short-term and long-term compound impacts of strong competitors is valuable, including for movie-release timing.Source: Discussion · pp. 13–14

Related papers

Loading 2609.35104v1…