Research Projects

Download my research portfolio here.

Crowdsourcing Change: Design opportunities for online petitioning platforms

    

Motivation

Online petitioning platforms such as Change.org and Care2 petitions are mobilizing people around specific concerns, short-circuiting structures such as organizational hierarchies or elected officials. Such online mediated processes allow consumers and concerned citizens to voice their opinions, often to businesses, other times to civic groups or other authorities. Here, I examine online petitioning platforms through the lens of crowdsourcing by conceptualizing the platforms as specialized cases of crowdsourced systems (see spectrum below). They require a small of amount of 'work' by petition signers to specifically promote social change. 

 

Research steps

Used a heterogenous purposeful sampling approach to select four petitioning platforms that showed substantial diversity along four task dimensions that map onto users' interaction trajectory. The four dimensions included  i) starting petitions, ii) updating petitions, iii) signing petitions and iv) petition analysis. 

Conducted a comparative analysis of Change.org, Care2 petitions, MovenOn.org and U.S. government's We, the people petitioning platform.

 

Insight  

Borrowing from crowdsourcing literature in collaborative work, user studies, humanitarian work, four design opportunities were presented:  (i) motivations to participate,  (ii) reputation building for petition starters,

(iii) achieving critical mass for collective action, and (iv) audience segmentation through message framing.

Online Petitioning Platforms

eg. Change.org, Care2 Petitions

Volunteer peer production

eg. Wikipedia, Yahoo! Answers, YouTube

Paid Crowdsourcing

eg. Amazon mTurk

Incentives

We want you to have the best experience. Update your password today and we’ll do the rest to ensure you have a safe environment to enjoy shopping on Pixel.

Default

We generated a password for you at the beginning of the survey. In a few weeks your password will expire. Our system will automatically generate another one for you. If you would like to create your own password, you must opt-out of our password generation.

Ego

You're a savvy netizen and you get it's important to update security settings. Keeping your personal data safe means being smart and proactive. Changing your password is one way to do that.

Norm

Don't be left behind! Pixel generated a password for you. 60% of Pixel account holders changed their password after their first use. Be a part of the secure online movement now!

Salience

Your personal information could be at risk for hackers to exploit. At this moment, thousands of hackers are combing the Internet for personal data. Changing your password could prevent your data from reaching them.

salience.png

Findings

Results show the salience nudge most effectively changed attitudes towards password management. It most significantly reduced one’s comfort level with keeping the auto-generated password, even more than the default nudge.

Personas for self-driving car experience

Background 

Self-driving experience is a unique opportunity for travelers. While it allows for autonomy and robotic chauffeuring, the technology changes how we experience travel. For the industry to become mainstream  self-driving technology should address efficient navigation in complex urban environments while ensuring utmost safety. This experience can vary for people traveling with families and those that are traveling for business. This is the design opportunity that I seeked to address while learning how to develop personas for UX research.

Target Audience

The target audience for this project is travelers in urban cities, specifically business professionals, both men and women.

Developing Personas

I started out by researching about the demographics of potential early adopters of self-driving cars. This was my inspiration. I further developed and refined the characters of James and Vanya, by making salient their day to day activities and daily travel encounters. I knew I had to give them pictures and the ones below captured the essence of what I was trying to portray. I finally came up with these:

James is a financial consultant in J & M Associates. He graduated from Columbia Business School. He lives with his family in Boston and works with different clients all over the country. When he is traveling, he usually prefers to take a taxi to drive to the meeting location. He sometimes also has business meetings he attends via conference calls. The sheer number of travel hours that James accrues tires him out. Occasionally, his family travels with him to other cities. When they accompany him, he prefers to rent a car. But his inherent nature also makes him fun loving and adventurous. He is always intrigued by new technology and never misses an opportunity to witness demos and follow technology releases. He is not really a car enthusiast but is always concerned about efficiencies.

self driving.jpg
selfdriving-vanya.jpeg

Vanya is a working mother. Her schedule is always packed. On weekday mornings she is rushing to get to work, in the evening she is usually driving her kids to after school programs. Over the weekends she is even busier, hopping from one location to other for kids classes and workshops. Reaching the destination on time and also remembering to take everything with her during the day is important. She always wishes she had an assistant for her chores. 

AQUAlity

combine4.jpg

Background

Communities all around the world are facing challenges in maintaining and using natural resources. One such important resource is water in the community. Engaged citizenry is a hallmark of a vibrant society. In Pennsylvania, there are several groups in the Center County region that are already engaged in monitoring water quality. Specifically,  PaSEC (Pennsylvania Senior Environment Corps) group is an exemplar of how senior citizens volunteer their time towards establishing baseline water quality data.

 

Motivation

To better understand member practices, I visited field sites where PaSEC members measure stream parameters. I also conducted individual interviews with PaSEC members. PaSEC group members (senior citizens) collaborate in teams of 5-6. While two of them go down into streams, one person stays on the banks and jots down notes on paper and others run tests from water samples at the base station. At the end of water sample and stream parameters data collection (eg. flow, depth, nitrates, phosphates, pH, dissolved oxygen), a team member transfers the readings from paper to Google sheets. While this has become a routine for members, this process in itself is very error-prone. Group members repeated readings multiple times to each other making sure the correct values were recorded.

To streamline this data monitoring process, increase accuracy and efficiency, I prototyped a mobile app for  PaSEC's group members. Some screenshots from the prototype are shown below. 

combined.jpg

Nudging for better password management?

Problem

Nudges have been widely used in government and health interventions for the subtlety with which they positively influence choices. Can a gentle nudge influence users to exhibit secure online account behaviors? And further, which of the nudge types is most effective?

 

Research steps

Systematically examined the effectiveness of five types of nudges: incentives, norms, default, salience, ego in reducing one’s comfort level with keeping the auto-generated password and intentions of creating a new password. Designed a between-subjects online experiment with 5 nudge messages (see grid below) and one control condition to examine password changing attitudes and practices.

 

Designing wearables for team communication research

Background 

Recent technological advances in sensor development have provided an avenue to explore complex communication patterns among team members less obtrusively. While some commercial solutions exist to monitor turn taking, speaker dominance, and individual speaking duration within a team, data from these existing devices are unreliable. Also, researchers are unable to know if the devices are working properly during the recording, sometimes making the whole recording unusable. The design challenge for one of my graduate level interaction design class was to develop a less obtrusive form factor for a wearable sensor and user-friendly desktop UI to monitor team communications.

Target Audience

The target audience for this project were researchers who study teams and have previously used team communication monitoring sensors.  

Approach

I led the user interview study, paper prototype development and usability studies for the UI. I also collaborated on the design of the form factor for the wearable sensor and acted as the liaison between different stakeholders- experts, developers and designers. Through in-depth market research and user interviews, we uncovered pain researchers (users) points that directly led to the design of the live view that shows the status of each of the teams and the selection of teams for each recording.

Result 

A sophisticated, frustration-free team communication monitoring experience was designed for team communication researchers. This was a direct result of iterating through multiple designs and evaluating their usability with potential users.

Explainable Artificial Intelligence for Natural Language Processing, IBM Research

Motivation

On the one hand, deep learning based AI applications are becoming widespread; on the other hand human comprehension and understanding of how AI models operate under the hood remains limited.  To incorporate AI infused platforms and services as everyday tools, there is a compelling need to develop a framework for models so that they are ‘right for the right reasons’ (Ross et. al 2017) while also unpacking the underlying hidden complexities for their users. This project was motivated by research question 'what are interpretability/explainability concerns that arise in real world text analytic projects'?

 

Research Methods

I first deep dove into the literature at the intersection of explainability of AI models and natural language processing. This eventually led to the development of a taxonomy that better maps out techniques and their applications as they relate to text processing. With this deep knowledge of body of work as a background, I conducted an interview study with 30 machine learning researchers, practitioners, strategists, designers and leadership working on NLP projects.

 

Insights

Immediately apparent from the interviews I conducted was that explanations are not one-size-fit all. AI models ‘touch’ different stakeholders at different stages in the pipeline. They encompass explanations about training data, base model mechanics and customized renderings. A data driven ecosystem needs to bridge the gap between AI capabilities and situated user interactions by not only answering what the model does at the global level, but also for each local instance so as to support actionability.

This research project provided the impetus for a systematic large scale endeavor at IBM Research in explainability in relation to NLP. 

 

Ross, A. S., Hughes, M. C., & Doshi-Velez, F. (2017). Right for the right reasons: Training differentiable models by constraining their explanations. arXiv preprint arXiv:1703.03717.

Data Empowered Teaching and Learning

Background

Data driven tools to support teaching and learning are increasingly being anchored by AI. What do stakeholders (students, faculty, administrators) think about this new way of supporting university academic workflows? In what way can needs and concerns be incorporated in user interfaces to black box algorithms? These overarching questions guide my work in multiple projects within this space.

 

Research threads and work

I have been intimately involved in leading the design of a faculty facing reflective teaching tool that uses machine learning to analyze lecture content. I have created iterated from lo-fi mocks to hi-fi prototypes to refine the use case of the innovative technology being built my data science engineers

 

I have been supporting focus group discussions to better inform design for student academic advising. Contributed to discussions about affordances of different explainability techniques that can aid in opening the black box of student risk estimation scores.

 

I am leading the design and development of a student facing interface of AI powered lecture content analyzing tool to support learning. This involves surveys, set-structured interviews and focus groups

 

Takeways

These projects have helped me see project grow from infancy to continuous usecase refinement for data and AI backed tools in academic settings. I have developed deep understanding of challenges and concerns with protected data at the university level and have gained hands-on experience in shaping a new technology while working alongside multiple stakeholders.

Market research

Hand-off to developers

Finalize design with experts

User interviews

.design.

Prototype

Usability testing

Consult with developers

kickoff.png

Leading initial kick-off meeting between developers and design team

IMG_20180404_162553.jpg

Deep diving into user research with interviews

Paper prototype.png

Desktop UI paper prototype

formfactor.png

Form factor of the wearable sensor