Figure (by Diana Zhang): The lexical diversity MTLD by number of reviews received by the author. The size of each circle indicates the amount of data in each group.
Human Centered Data Science Lab, January 2016 - Present
Collaborators: Dr. Cecilia Aragon, Dr. Sarah Evans, Jihyun Lee, Diana Zhang, Ruby Davis.
Synopsis: Online fanfiction repositories attract millions of writers and readers worldwide. The largest repository, Fanfiction.net, accumulated a rich corpus of about 61.5 billion words of fiction over twenty years, rivaling even the Google Books fiction corpus. This website is an important learning environment for young writers, as the site affords networked giving and receiving of feedback, termed distributed mentoring. One way to measure author learning is with the Measure of Textual Lexical Diversity (MTLD), which captures the author's range of vocabulary usage. In a massive longitudinal study of texts by 1.5 million authors, I implemented MTLD in Python and performed statistical analyses in R to find a correlative link between the cumulative number of reviews an author has received and the lexical diversity score of their writing.
Screenshot: An answer rating from one of the experimental conditions.
Prosocial Computing Lab, September 2016 - Present
Collaborators: Dr. Gary Hsieh, Dr. Erin Walker
Synopsis: Social Question and Answer (Q&A) websites offer users a place to post questions that are then answered by other users. Encouraging high-quality contributions to these sites can benefit askers, answerers, and others who use the answer archive. In this study, I examined the utility of providing crowdsourced feedback to answerers on a student-centered Q&A website, Brainly.com. In an experiment with 55 Brainly answerers, I compared perceptions of the current 5-star rating system against two feedback designs that used three explicit criteria (Appropriate, Understandable and Generalizable). Contrary to the experimental hypotheses, the designs resulted in lower perceptions of utility. I investigated these results using interviews and derived a set of implications for the design of feedback for answerers in online Q&A.
Figure (by Niharika Sharma): A connected mentorship network. The circles at each epicenter represent authors, while the connected circles clustered around them are reviewers. The color of each circle represents each user\s top fandom.
Human Centered Data Science Lab, September 2017 - Present
Collaborators: Dr. Cecilia Aragon, Dr. Sarah Evans, Ruby Davis, Niharika Sharma.
The concept of Dunbar's number -- a cognitive limit to the number of social relationships an individual can comfortably maintain -- has been well-studied in online and offline contexts. Analyses of twitter and facebook show that there are distinct levels of relationship investment, which differ between a person's smallest group of close friends and increasingly larger groups of friends and acquaintances. Do sites of distributed mentoring show a similar structure? We replicated Dunbar's methods on review metadata from Fanfiction.net. Examining relationships from reviewers to authors, we consider Dunbar's number in digital contexts. Using clustering techniques implemented in Python, we discovered two to three relationship layers in the fanfiction dataset, shedding light on the structure of the distributed mentoring community. The results characterize the relationships that occur between writers and readers of fanfiction and may lead to more effective platforms for distributed mentoring.
Photo (by Kai Lukoff): The front desk at the ASUW Bike Shop
ASUW Bike Shop, March - June 2017
Collaborators: Sarah Inman, Kai Lukoff
The Associated Students of the University of Washington (ASUW) bike shop is a hub for students and community members at the University of Washington (UW) who ride bikes. This student-run, nonprofit repair shop offers bike servicing, classes, and opportunities to work and volunteer to students and faculty. Setting it apart from most other repair shops, the ASUW bike shop encourages students to work on their own bikes by allowing them to use the space and the tools for free. Getting help and advice from the mechanics is also free -- as a result, the shop is a space where beginner and expert bike mechanics collaborate and share knowledge. Previous study of collaborative bike repair has shown that the loss of co-presence when collaboration occurs over a video call drastically reduces the efficacy of cooperation . We expand the scope by showing how the site of repair has function that goes beyond providing co-presence during the collaboration and beyond the experience of a single beginner-expert pair. Understanding the practices of knowledge sharing in this informal learning environment requires a deep examination that looks outside of a single controlled interaction. In the present study, we examine the practice of knowledge sharing at the ASUW bike shop using ethnographic methods to illuminate our research question: how do people share knowledge in the ASUW bike shop? The results of our investigation show how the values and interests of students and mechanics at the shop drive a culture of knowledge sharing. By examining the different kinds of knowledge present at the shop, we reveal how the nature of knowledge shapes the interactions occur in the process of sharing. We also come to a new understanding of the self-sufficiency culture of do-it-yourself (DIY) that we call do-it-together (DIT). Finally, we suggest research and design implications for communities of practice such as a bike repair shop.
Figure (by Christopher Collins): Interaction map demonstrating the flow of interaction through educative.io.
University of Washington, January 2018 - Present
Collaborators: Samantha West, Jihyun Lee, Christopher Collins
Educative is in the early phases of personalizing their courses for individual students' needs. Our group is working with Educative to identify user needs and uncover metrics that can help Educative appropriately adapt to individual differences. The primary users of Educative.io have introductory or intermediate understanding of software development and have little to moderate experience with online courses. Some variation in past experiences these will help us evaluate how users of different skill levels interact with the courses. We are in the process of recruiting participants and planning the usability study.
Sketch (by Samantha West) of the wearable homework tracking device.
University of Washington, September - December 2017
Collaborators: Samantha West
Middle school students struggle to keep track of their homework assignments. There are several reasons why this is the case: First, students often try to keep track by memory, and may forget assignments. Second, for students who use handwritten notes or journals, writing down the assignment and track of the notes can be burdensome. Third, a handwritten journal does not remind them when it's time to get started on the homework. And finally, technological solutions present the opportunity to better connect students and teachers. Design Question: How can we support elementary school children with the tracking and completion of their homework assignments using a conversational interface? In order to test our design direction, we created a hi-fidelity prototype that used the wizard of oz method to control its functionality. The wearable was created with an existing fitness tracking band, a strip of color changing, wireless, remote controlled LEDs, two buttons made of tape, and pre recorded audio feedback on a laptop that was played in response to users commands. One major design direction present in this specification of the TellMe device is screenlessness. During the course of our user research, we found that parents in many cases will restrict their childrens' use of screened technology such as smartphones, in favor of a low-tech cell phone. TellMe should be inclusive for all students. Our screenless design is non-distracting, minimal, and does not conflict with the values of parents.
Figure (by Mike Harwell): Wireframe of app screens for transcribing speech, filtering voices, and reviewing transcriptions.
University of Washington, September - December 2017
Collaborators: Samantha West, Mike Harwell, Jessica Son, Sandy Tsai
Working with a group of students, I designed an auto-captioning mobile app to provide real-time captions and translation. The app uses a phone's built-in omnidirectional microphones and combines automatic speech recognition (ASR) with machine learning to determine which voices may be of interest to a person and captions only those sounds. Auto-Captions features speaker identification and filtering, so captions are labeled and only relevant speakers are included. Auto-Captions eliminates the need for a professional stenographer, and allows the user to freely capture speech-to-text directly from their smartphone. Auto-Caption can filter out ambient noise and other distracting noises, providing a clear transcript of any conversation right to the user's smartphone in real-time. Finally, Auto-Captions features conversation storage, search, and retrieval, allowing users to look into their past conversations for important information.
iOS on Windows.
Microsoft, November 2015 to July 2016
Collaborators: The WinObjC team at Microsoft.
The Windows Bridge for iOS, or WinObjC, is an open-source project at microsoft that provides an Objective-C development environment on Windows, as well as support for iOS APIs. Working with a team of talented engineers, I implemented a series of compatibility libraries to accept iOS API calls and "bridge" them to Windows APIs as necessary. I was integral in the development of NSUserDefaults, NSFileManager, NSURLResponse, and NSValue, among other features. I coordinated with the other engineers, fixed bugs, reviewed pull requests, and wrote test apps as part of my work.
Android on Windows.
Microsoft, July 2014 to October 2015
Collaborators: The Astoria team at Microsoft.
The Windows Bridge for Android, codenamed Astoria, was a project at Microsoft designed to make it easy for developers to bring android apps to Windows. Although it was decomissioned in 2015, the legacy of Astoria lives on in linux compatibility on windows and in the skills honed by engineers working on this technical behemoth of a project. I implemented bridging features such as the contacts database, location, sensors, and process lifetime events.
Humm.ly: music and wellness app.
Humm.ly, February 2018 - Present
Collaborators: The Humm.ly team.
Humm.ly is a mental health and wellness app that draws from music therapy techniques to help people overcome anxiety and stress. I spearheaded the development and release of Humm.ly on Android, managed a team of engineers and external contractors and oversaw the production and release on Google Play.