The 2018 ICWSM conference is starting, and the WuLab is presenting our research on improving content quality on social media websites. This research was led by Hamed Nilforoshan, who is a rising Senior at Columbia University. He will present 2:40PM Tuesday June 26 in Paper Session III: Discourse Analysis. Come by the talk and say hi!
The paper, titled Leveraging Quality Prediction Models for Automatic Writing Feedback, addresses the challenge of providing timely and useful writing feedback on the web. Many of the services that we use on the internet — such as Amazon, Airbnb, Quora, Reddit, Twitter — rely on the users themselves to write the content that we read and use to make decisions. For example, when we purchase bluetooth headphones from Amazon, we read the product reviews to understand if the sound quality is good, if it is easy to connect, or if its battery life is as advertised. High quality, well-written content helps us make more informed decisions, while low quality content misleads us or adds to our confusion.
Unfortunately, it is easy for many well-intentioned users inadvertently contribute low-quality content. The community guidelines on most websites are too long and complex for a busy user to carefully study, and even then, they may have trouble applying them because norms can differ by community. For example, high quality product reviews prioritize informativeness, whereas high quality Airbnb profiles prioritize trustworthiness. Instead, it would be helpful to provide real-time suggestions during the user’s writing process. The challenge is to automatically generate useful feedback.
Our main observation is that there is already a research community focused on developing accurate quality prediction models for many different social networks. Thus, our paper proposes and develops techniques to leverages these quality prediction models in order to generate feedback based on what will most improve the model’s predicted quality. We find that this is a very promising direction that is simple to adapt to different application domains, and significantly improves upon existing approaches towards writing feedback.
We want to thank James Sands, Kevin Lin, Rahul Khanna, Jiannan Wang, Lydia Chilton, Eric Riesel, Markus Krause, Xiao Ma, Mor Naaman, Julian McAuley, Niloufar Salehi, Robert Netzorg, Ali Alkhatib, and Philippe Cudre-Mauroux for helping us with this work. This work was supported by an Amazon Research Award; NSF #1527765 and its REU.
