A framework to predict early news popularity using deep temporal propagation patterns. (1st June 2022)

• Record Type: Journal Article
• Authors: Saeed, Ramsha; Abbas, Haider; Asif, Sara; Rubab, Saddaf; Khan, Malik M.; Iltaf, Naima; Mussiraliyeva, Shynar
• Is Part Of: Expert systems with applications. Volume 195(2022)
• Publication Date: 2022-06-01
• Subjects: Popularity -- Long short-term memory -- Temporal propagation patterns -- Convolutional neural network; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2021.116496
• Languages: English
• ISSNs: 0957-4174

Abstract

Abstract: The increasing competition among the news industries puts editors under the pressure of posting news articles that should gain more user attention. News popularity is predicted using different content and metadata features. Some approaches use retweet paths formed on social media when a tweet is retweeted. However, before a piece of news spreads by retweeting, there are several initial tweets made by multiple different users that spread the same news. Retweeting behavior serves as the secondary features in this case while the initial tweets serve as the primary features. In this work, the popularity of a news item published on a certain website is predicted by exploiting the initial tweeting behavior of the news item on Twitter. The temporal characteristics of a news item are exploited as the news propagates via tweets. Additionally, other content and metadata features have also been used to predict news popularity. Data is extracted from different websites of cybersecurity news and Twitter. A deep neural network is proposed to predict early news popularity. The proposed model yields the macro averaged F-score of 92% which shows the effectiveness of temporal propagation patterns in predicting news popularity. The proposed model is compared with the baseline models and state-of-the-art techniques, and it is shown that the proposed model outperforms all the existing techniques. Highlights: A corpus containing data from cybersecurity news websites and Twitter is created. A model based on news temporal propagation patterns is proposed to predict its popularity. Content features, user features, and news source features are also used. A novel deep learning model is devised to predict early news popularity.