In the final week of my DREU program, we decided to change the survey platform we are using to be able to have more features we want for our study. We also had to put our Django server on AWS and connect that to our survey. We realized that with the new platform, we can significantly improve the survey experience and keep track of many more variables, which we hope will enhance the data we collect in the study.
My name is Olivia Figueira and I am a senior Computer Science and Engineering major with a minor in Mathematics at Santa Clara University in Santa Clara, California, graduating in 2021. When it comes to research, I am interested in the fields of security and privacy, human-computer interaction, machine learning, and humanitarian computing. In Summer 2020, I participated in the virtual CRA-WP DREU at the University of Illinois at Urbana-Champaign.
About My Mentors
Dr. Wang’s research covers a range of topics in Security and Privacy, Internet Measurement, Mobile Networks, and HCI, mostly from data-driven perspectives. Learn more about Dr. Wang’s work here.
Dr. Khurana’s research focuses on Cryptography and the foundations of Security, and more broadly in Theoretical Computer Science. Learn more about Dr. Khurana’s work here.
About My Project
Fake online social network (OSN) profiles created using natural language and image generation models are becoming increasingly sophisticated and thus harder to detect by traditional methods. Current research focuses on improving the detection of intra-field inaccuracies and artifacts, meaning those that exist within a deep fake image or within generated text in isolation. However, personas and OSN profiles are built on a collection of multi-modal fields that must maintain both intra-field and inter-field consistency. This project aims to determine how intra-field and inter-field inconsistencies affect perceived trust by OSN users and how those inconsistencies compare with each other leveraging a user study with fake LinkedIn profiles created using generative text and image algorithms. We had not begun the user study by the time my DREU program came to an end, so the results are forthcoming.
Read more about this research project in my final report.
In my second to last week in my DREU program, my project group decided to change a part of the implementation of our survey so that we can better control the conditions of what the survey participants see. To do this without having to pay for a lot more advanced features on popular survey sites, this week we spent a lot of time building our own server using Django. This took the whole week since we decided to track a lot more features since we could now control the environment of the LinkedIn profiles, meaning we could actually show them interactive websites instead of just screenshots.
My eighth week of my DREU program coincided with the final week of the UIUC CS REU program’s last week, so I participated in a virtual research showcase in which I created a video presenting my work from this past summer. It was great to watch other students’ videos and learn about all the other research projects going on in the CS department.
We worked on creating the pilot study this week and finalizing the questions and profiles we will use on it. Choosing and creating the profiles takes a lot longer than any of us had previously thought, but we wrote up the methodology behind creating the base profiles and variations this week to make sure everything is consistent.
This week we finalized the methodology for the pilot study. I worked with one of the Ph.D. students in the project group to create the survey platform and sample profiles to show the group what we are thinking of having for the pilot study. We also worked on finding and creating more base profiles and variations for the pilot study now that we have a concrete methodology and explicit criteria for how we will be choosing such profiles. Next week we will finalize the survey questions and introductory questions on the survey.
This week we worked on preparing base profiles and study methodology as we work on getting our pilot study up and running. We are planning to conduct a pilot study in order to get a baseline for the profiles and to see if users can even notice a different between our base profiles and variations of the profiles, and given the results of the study we will modify the overall study. This work involved filtering through the real resume data we have to find profiles that seem to be good on all counts, meaning they have a relatively well-written summary and matching educational and job experiences.
This week I worked on figuring out the methodology behind our user study. I worked closely with one of the PhD students in the group to discuss related papers on different methodologies for user studies and analyzing dependent variables in order to figure out how we will be setting up our user study. This also involved organizing the way in which we will be setting up the generated profiles. Since there are so many variables (artifacts and variations from a base profile) in a single profile, we need to make sure we are accounting for all differences between profiles and how those may lead to confounding variables.
This week I worked on writing a script to generate the social media profiles initially based on raw resume data. After realizing I had hardware incompatibilities in trying to run the deep fake generation open source algorithm, I took over this other task for most of this week. Writing the script involved using MongoDB data and Python to generate an HTML file based on a template HTML file taken from the LinkedIn website code that can be accessed directly on their website. I had never written Python that interacts with web programming, but I do have web programming background so it was a good skill to pick up.
This week I attended a meeting with the entire lab of my project mentor, Dr. Wang. The meeting was at 7am PDT for me (9am CT for the group), so this is an aspect of my virtual DREU that I had not thought about until the meeting, but it went well nevertheless! I met other PhD students in Dr. Wang’s lab and another undergraduate research assistant who will be working on another project this summer.
Prior to starting my DREU virtual program at UIUC, I met with my two mentors to discuss project ideas. I decided on a deep fake generation project, which is a project in Dr. Wang’s lab. Dr. Khurana said she will still stay connected to serve as a mentor for other topics, such as graduate school, so it is great to have two mentors!