School of Engineering Department of Computer Science and Engineering 133 Gaining Insights from Noisy Data: Designing Transparent System for Online Decision Making Supervisor: MA Xiaojuan / CSE Student: NAN Xi / DSCT Course: UROP1100, Spring Online platforms such as Reddit's "Subreddit" and Weibo's "Super topic" serve as digital fandom communities where users can seek information and discuss their idols. The rapid development of social media increases the possibilities of potential members joining the fandom community. Celebrities can suddenly attract attention from the public because of certain events. For example, Psy, a South Korean singer, was known domestically for his humorous videos and stage performances, and internationally for his hit single "Gangnam Style". In this case, a sudden influx of newcomers joins the fandom community and these newcomers possibly influence the community culture and interaction patterns. In this project, we try to characterize the user behavior in online fandom communities during the event-induced period. Gaining Insights from Noisy Data: Designing Transparent System for Online Decision Making Supervisor: MA Xiaojuan / CSE Student: YIN Zhuohao / DSCT Course: UROP1100, Spring With various social media platforms developing at a rapid pace, people are able to access much more information and thus more likely to participate in discussions in serendipitous groups, especially when a viral event concerning certain fandom communities occurs. This research explores the effects of an event-induced sudden influx of newcomers into online fandom communities. Specifically, it studies the user behavior of the newcomers and existing fans and also the pattern of interaction among different types of fans. Gaining Insights from Noisy Data: Designing Transparent System for Online Decision Making Supervisor: MA Xiaojuan / CSE Student: WU Yifeng / QFIN Course: UROP1100, Summer An image conveys semantic information to the recipient, but there is a perceptive gap between what image the presenter wants to convey and how image the recipient tends to interpret. In the context of online medical crowdfunding, this gap befuddles campaign owners when they are choosing a cover image to leave first impression that has persuasive power on potential donors. This research attempts to model recipient’s first impression with an image through deep learning modelling. This paper reports research assistant’s contributions and takeaways throughout different stages of this research, including exploratory data analysis works that visualize collected data, survey on references that justify certain assumptions and operations made during the research, and data cleaning for better training results.