Detecting clickbaits using two-phase hybrid CNN-LSTM biterm model
作者:
Highlights:
• A ground dataset has been prepared from Facebook page and Reddit website.
• Extraction of text from non-textual (image-based) data using pre-processing approach.
• Automatic identification of the eight types of clickbait is done.
• A novel approach is proposed which works under two-phase structure.
• Shocking/Unbelievable, Hypothesis/Guess, Reaction types of clickbait are published maximum on social media.
摘要
•A ground dataset has been prepared from Facebook page and Reddit website.•Extraction of text from non-textual (image-based) data using pre-processing approach.•Automatic identification of the eight types of clickbait is done.•A novel approach is proposed which works under two-phase structure.•Shocking/Unbelievable, Hypothesis/Guess, Reaction types of clickbait are published maximum on social media.
论文关键词:Clickbait,News,Classifier,Features,Social media
论文评审过程:Received 27 August 2019, Revised 27 February 2020, Accepted 27 February 2020, Available online 28 February 2020, Version of Record 8 March 2020.
论文官网地址:https://doi.org/10.1016/j.eswa.2020.113350