Abstract
The Lineage graph model used by Capital One is used to organize all of the datasets and track the changes made over time, but it lacked a consistently optimal solution for extracting these lineage paths. For my technical project, I reported how I worked with one other person to hypothesize and investigate potential optimizations to the Lineage graph model at my summer internship in 2024. The code base utilized Gremlin queries built in Scala, and the memory was stored in AWS Neptune. Specifically, I worked on a model redesign that removed redundant intermediate steps in the graph traversals. After building locally and testing benchmarks, these design choices warranted an average of 30% decrease in memory usage. As an extra step, I had written an algorithm that refactors the original model into the optimized one and tested it on a larger scale. This, again, resulted in the same 30% decrease in the time taken for graph traversals. Because the scale of the consumer data at the company was much larger and data retrieval procedures are different, the next step would be to test these changes in AWS Neptune in a secure environment to further evaluate the business value and impact of the redesigns. As of November 2024, the results have been reproduced in an AWS environment, and the Lineage team groomed the work to move forward within the next year.
For my STS project, I researched what kind of impact TikTok’s algorithmic mechanisms may have on a user’s ability to form their own political opinion, specifically focusing on any factors such as algorithmic bias that may contribute to political polarization. At the backbone of social media algorithms is a recommender system, which converts content interactions into numeric rating vectors to recommend personalized content to a user – establishing algorithmic bias in generalizing without contextualizing information that impacts the content recommended to a user. Using discourse analysis, I read language used by social actors – users, creators, and government – to understand how TikTok’s social media algorithms play a role in political polarization. User testimony revealed how algorithmic bias can divide the Internet in the political space, and although race is a social issue, it often plays into political stances involving diversity, equity, and inclusion. Following the events of a temporary ban on January 18th, 2025, many users reported censorship within the app, further contributing to political polarization by restricting viewed content. I also found that because TikTok content creators are incentivized by views and popularity, they will produce more polarizing or conspiratorial content that is more likely to go viral and therefore will continue to spread this sort of thinking. Lastly, with government action against the application for being controlled by those in a foreign country, users are most at risk to experience censorship as the government controls what information is spread on the media, potentially stifling freedom of expression rights and allowing polarizing content to be amplified. All of these concerns regarding censorship contribute to political polarization, and media literacy taught at an early age can combat the spread of misinformation and teach citizens to balance the media from which they form political opinions.
