A Novel Approach to Determining the Quality of News Headlines

Headlines play a pivotal role in engaging and attracting news readers since headlines are the most visible parts of the news articles, especially in online media. Due to this importance, news agencies are putting much effort into producing high-quality news headlines. However, there is no concise definition of headline quality. We consider headlines as high quality if they are attractive to readers, and highly related to the article contents. While almost all the previous studies considered headline quality prediction as either clickbait detection (which is a binary text classification problem), or popularity prediction (which is a regression problem), our model employs four quality indicators to incorporate these two factors. In this paper, we first discuss the previous works on the news headline quality detection. We then propose a machine learning-based model to predict the quality of a headline based on four quality indicators before the publication of the news article. The proposed model is an extended version of our previously proposed model in a way that it considers sentiment features of headlines as well. We conduct experiments on a news dataset and compare our method with the state-of-the-art NLP models.