How do the clickbaits work? We scraped nearly 3,000 most popular articles on the Weixin platform, and used an AI headline-scoring tool and semantic tool to measure. We find that the headlines of most articles are more emotional than the real content, and in recent years, headlines have even become more emotional. At the same time, titles with prejudiced labels can be more popular. The story introduces a more creative perspective, with detailed data and rigorous analysis.
The story has been widely disseminated on the Internet, and has been praised and recognized by professionals. At the same time, it was selected as an annual case of the 2022 China Data Content Competition.
We used Python to scrap the historical articles on WeChat, and used Baidu’s sentiment AI, article title generation AI, and headline-scoring AI assistant for analysis.
We also used Illustrator 2021 for designing.
Context about the project:
Finding the right way to disassemble a title is not easy. The story tries to use AI title-scoring tools to reflect people’s unconscious prejudices, what dimension to choose, and how to modify the title to answer questions more directly. It took a lot of time to explore the best way to answer the questions above.
What can other journalists learn from this project?
We took advantage of our understanding of the AI model training mechanism and used AI in a relatively lightweight way: since the AI scoring tool is trained with the historical traffic data of the articles, its scoring actually reflects how popular the title will be based on past experience- that is, what kind of headlines people like to read. This can inspire data journalists to learn how to use AI- without directly training a model, you can use existing AI models in a new way.