[Paper Review] To Post or Not to Post: Using Online Trends to Predict Popularity of Offline Content

Sofiane Abbar, Carlos Castillo+1 more|arXiv (Cornell University)|Jul 17, 2018

Complex Network Analysis Techniques37 references3 citations

TL;DR

This paper proposes a novel method to predict the offline popularity of news articles before publication by leveraging time series forecasting of topic popularity and similarity to recently published, similar articles. By combining predicted topic volume with observed visit patterns of similar articles, the approach achieves a mean average percentage error (MAPE) of 11.47%, rivaling models that rely on early post-publication metrics.

ABSTRACT

Predicting the popularity of online content has attracted much attention in the past few years. In news rooms, for instance, journalists and editors are keen to know, as soon as possible, the articles that will bring the most traffic into their website. The relevant literature includes a number of approaches and algorithms to perform this forecasting. Most of the proposed methods require monitoring the popularity of content during some time after it is posted, before making any longer-term prediction. In this paper, we propose a new approach for predicting the popularity of news articles before they go online. Our approach complements existing content-based methods, and is based on a number of observations regarding article similarity and topicality. First, the popularity of a new article is correlated with the popularity of similar articles of recent publication. Second, the popularity of the new article is related to the recent historical popularity of its main topic. Based on these observations, we use time series forecasting to predict the number of visits an article will receive. Our experiments, conducted on a real data collection of articles in an international news website, demonstrate the effectiveness and efficiency of the proposed method.

Motivation & Objective

• To enable early prediction of news article popularity before publication, supporting editorial decision-making.
• To address the limitation of existing models that require early post-publication popularity data.
• To improve prediction accuracy by integrating topic popularity trends and article similarity.
• To develop a forecasting model that combines content-based and time-series approaches for higher accuracy.
• To demonstrate that topic popularity forecasts can enhance article popularity prediction without relying on early metrics.

Proposed method

• The method uses time series forecasting to predict future popularity of a news article's topic based on historical visit patterns of that topic.
• It identifies recent articles similar to the target article using a cosine similarity threshold (θ = 0.1) on title and content embeddings.
• The model aggregates visit counts from these similar articles over a sliding window (δ = 1, 2, 3, or 4 days) as a predictor for the target article’s future popularity.
• It combines two predictors: (1) visit counts of similar articles and (2) historical visit volume of the article’s topic, both used in a linear regression model.
• The model further improves performance by incorporating a predicted future topic popularity, estimated using the best-performing forecasting method from Section 4.4.
• The final ensemble model integrates observed similarity trends, observed topic volume, and predicted topic volume in a single regression framework.

Experimental results

Research questions

• RQ1 Can topic popularity trends be used to predict the future popularity of individual news articles before publication?
• RQ2 How does the popularity of similar recently published articles correlate with the future popularity of a new article?
• RQ3 Can combining article similarity and topic popularity forecasts improve prediction accuracy beyond using either alone?
• RQ4 How does the performance of a pre-publication prediction model compare to models that use early post-publication popularity data?
• RQ5 What is the optimal time window (δ) for aggregating recent visit data to maximize prediction accuracy?

Key findings

• The proposed method achieves a mean average percentage error (MAPE) of 11.47% in predicting article popularity, which is comparable to models using early popularity data.
• The model outperforms single-predictor approaches: using both similarity trends and topic volume yields lower error than using either alone.
• Incorporating predicted topic volume improves performance slightly and consistently, especially when δ = 3 or δ = 4 days are used.
• The method achieves a MAPE of 11.47% without requiring any early popularity measurements, making it suitable for pre-publication editorial decisions.
• The performance of the model is comparable to models using 5 minutes, 1 hour, or 6 hours of early post-publication data, which achieve MAPEs of 9.59%, 6.83%, and 4.75% respectively.
• The best results are obtained when combining observed similarity trends, observed topic volume, and predicted topic volume, with δ = 3 or δ = 4 providing the most accurate forecasts.