Characterizing the life cycle of online news stories using social media reactions

Abstract

This paper presents a study of the life cycle of news articles posted online. We describe the interplay between website visitation patterns and social media reactions to news content. We show that we can use this hybrid observation method to characterize distinct classes of articles. We also find that social media reactions can help predict future visitation patterns early and accurately.We validate our methods using qualitative analysis as well as quantitative analysis on data from a large international news network, for a set of articles generating more than 3,000,000 visits and 200,000 social media reactions. We show that it is possible to model accurately the overall traffic articles will ultimately receive by observing the first ten to twenty minutes of social media reactions. Achieving the same prediction accuracy with visits alone would require to wait for three hours of data. We also describe significant improvements on the accuracy of the early prediction of shelf-life for news stories.

Learning from the News: Predicting Entity Popularity on Twitter

“…We perform this evaluation process, rolling the training and test set until December 2015, resulting in 365 days under evaluation. The process is applied for each one of the six entities, for different time of predictions t p and for different values of the decision boundary k. We test tp = 0, 4,8,12,16,20 and k = 0.5, 0.65, 0.8. Therefore, we report results in Section 5 for 18 different experimental settings, for each one of the six entities.…”

Modeling and predicting the popularity of online news based on temporal and content-related features

“…In particular, different pieces of content may display a very similar popularity at an early stage, yet exhibit a diverse popularity behavior afterwards. In other words, despite the observations in [10], online content may experience very different popularity evolution patterns [15,5]. Therefore, the authors of [15,12] investigated whether the use of the historical popularity values of online content between the publication time and an early reference time leads to more accurate predictions of the total popularity at a future target time.…”

“…Oghina et al [14] trained a linear regression model based on several textual features extracted from Twitter, as well as various statistics from Youtube, to predict movie ratings on IMDb. The authors of [5] proposed a second-order multiple linear regression model to predict the number of views of online news articles after 7 days. For a given reference time, the model used the total number of views, Facebook shares and Twitter posts of the article, in addition to Twitter statistics such as the average number of followers of people sharing on Twitter and the entropy of the tweets.…”

“…As input, the popularity trends during the first hour were given. The winner [9] of the challenge combined different ideas of the models proposed in [5] and [15]. Similar to [5], they used secondorder multiple linear regression models based on several popularity metrics to predict the number of views.…”

“…As the total number of views consists of easily identifiable components related to the origin of the views (i.e., direct views, Facebook views, Twitter views), we train different regressors to respectively predict the behavior for each of these components. Existing ap-proaches train linear regressors using features based on historical popularity values of the articles [17,15,5,9]. We investigate three ways to improve upon these baseline methods: (a) We explicitly make use of our proposed temporal model underlying the historical view pattern of the considered article, and use its parameters as additional features for the regressors.…”