Clickbait Classification Model on Online News with Semantic Similarity Calculation Between News Title and Content

Abstract

Clickbait is a sensational title that makes us click internet links to an article, image, or video. Online content providers use clickbait to gain user traffic, that leads to increasing income from the placed ads in their page. To attract more and more traffic, online content providers write sensational and hyperbolic titles, and even misleading and not telling the whole story. This can give us, the internet consumer, wrong perspective, and half-truth. And nowadays, clickbait titles are worse than ever. Modern clickbait titles are not hyperbolic nor ambiguous enough, and sometimes very hard to identify. This paper aims to classify clickbait titles, to help humans identify clickbait and stop sharing more online content that contains clickbait and misleading titles. This model classifies clickbait by calculating semantic similarity between the article title and the summary of the article content. The article content is summarized by T5 (Text-to-text Transfer Transformer) model. IndoBERT is then used to calculate semantic similarity score between generated summary and the article title. The article title, content, summary, and semantic similarity score are used for clickbait classification with various algorithms. The result shows that by adding article content alongside article title in the classification process improves F1-score by 7% when classified with IndoBERT. In another future research, this model can be integrated with another application such as twitter or telegram bot to send us warning every time a user consumes online content with clickbait title. Thus, it can prevent online communities from sharing misleading information caused by clickbait