My research interest lies in behavioral economics, experimental economics, and decision theory. More specifically, I am interested in using computational models including natural language processing, computer vision, and deep reinforcement learning to tackle important questions in behavioral economics.
Abstract: People have myriad choices about what news to read, and editors craft headlines to compete for their attention. We analyze how curiosity drives news consumption. Information-gap theory predicts that curiosity depends on the degree to which a question (or information gap) feels salient, important, and surprising. We apply natural language processing methods to create measures of these constructs in news headlines, which we validate experimentally. We then test these predictions, along with negativity bias, in the context of news consumption using a comprehensive dataset of over 100k news articles on WeChat. The evidence shows that people are more likely to consume news when: 1) the headline raises a salient question, indicated by the presence of a question mark or ellipsis in the headline; 2) the news appears more important, either because the headline is displayed in a higher position on the webpage or because the headline includes an exclamation mark for emphasis; 3) the topics mentioned in the headline are more surprising, as measured by the relative entropy (KL-divergence) of the distribution over topics referenced in the headline relative to a baseline distribution over topics, or as indicated by negation in the headline; and 4) the headline has lower (more negative) valence. Information-gap theory broadly predicts aggregate news consumption patterns. Yet we also find a slightly negative relationship between the number of people reading an article and the fraction liking it. Opening a salient information gap may spark curiosity about a particular news article, but might not generate long-run reader engagement.
Work in Progress