Virality Prediction for News Tweets Using RoBERTa

Christian E. Maldonado-Sifuentes, Jason Angel, Grigori Sidorov, Olga Kolesnikova, Alexander Gelbukh

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

3 Scopus citations

Abstract

The virality of a tweet is essential to convey its message to a broader audience and, eventually, to generate influence. This is especially important for news outlets as they struggle to transition from traditional media to online formats. As their usual readers will not migrate directly to digital news outlets need to gather new audiences from the spaces where real-time information and discussions are happening; this is Social Media and in particular Twitter. Since the news websites and Twitter languages differ greatly news outlets need to write their tweets properly to maximize their impact on Twitter. We propose a method to predict if a tweet will be influential or not influential based on its text using a variant of Google BERT named RoBERTa, and a corpus of 5000 high-quality and automatically labeled highly-influential and non-influential tweets to train and classify tweets in these categories. Our method reaches an F1 of 0.873, improving 4 and 9 over approaches using LSTMs and n-grams respectively.

Original languageEnglish
Title of host publicationAdvances in Soft Computing - 20th Mexican International Conference on Artificial Intelligence, MICAI 2021, Proceedings
EditorsIldar Batyrshin, Alexander Gelbukh, Grigori Sidorov
PublisherSpringer Science and Business Media Deutschland GmbH
Pages81-95
Number of pages15
ISBN (Print)9783030898199
DOIs
StatePublished - 2021
Event20th Mexican International Conference on Artificial Intelligence, MICAI 2021 - Mexico City, Mexico
Duration: 25 Oct 202130 Oct 2021

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume13068 LNAI
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference20th Mexican International Conference on Artificial Intelligence, MICAI 2021
Country/TerritoryMexico
CityMexico City
Period25/10/2130/10/21

Keywords

  • Applied deep learning
  • BERT
  • RoBERTa
  • Social media
  • Twitter influence
  • Twitter popularity
  • Twitter virality

Fingerprint

Dive into the research topics of 'Virality Prediction for News Tweets Using RoBERTa'. Together they form a unique fingerprint.

Cite this