Academic literature is built upon citations and supporting sources–and how they are used to support or refute claims. However, it is imperative to understand when these sources are correctly summarized versus when they are exaggerated or misconstrued. Our research project focuses on this idea by using leading edge natural language processing (NLP) tools to evaluate citations in academic literature. More specifically, we are studying citation contexts, exaggeration, and linguistic modality, as pertains to scientific claims made in specific papers in the academic literature. This is crucial to understand how information is warped even in controlled and reviewed academic contexts. Prior work has demonstrated that scientific claims may be exaggerated when scientific results are cited in another source such as news media or social media. By analogy, we are investigating the extent to which this phenomenon exists in citations in the scientific literature, by adapting and developing NLP techniques for sentiment and exaggeration detection over citation contexts. We collected a dataset of citation contexts extracted from one million scientific papers, and have aligned these contexts to the scientific claims made in the original cited publications using the language model SBERT to calculate similarity scores between sentences. Our next step in this quantitative approach will be architecting our own NLP model to determine the distortion of the information from the original claim to the citation. Our preliminary findings suggest that this citation claim drift does indeed occur, but we have yet to complete our analysis on the extent of the issue. Developments in this field of study will help us better understand how information is misconstrued in an academic setting. This is vital in understanding the validity and veracity of research papers and their supporting sources.

While there is much political science research focusing on Special Interest Groups (SIGs) and how they affect decision making on federal policy there is limited research focusing on municipal policy. Local interest groups play an important role in decision making in city council meetings as they lobby, provide public comment, and help craft legislation. This research aims to study how local SIGs are referenced during city council meetings and to measure and understand the impact they have on decision making and public policies. To do so, we are using Natural Language Processing (NLP) techniques, including training our own span categorization model, to process a dataset of city council meeting transcripts to enable the extraction of specific references to local interest groups from the text transcript of the meeting.. We then tie the detected references of SIGs to legislative outcomes in order to measure the impact that special interest groups have on the local municipal process.My contributions to the research include: experimenting with various Named Entity Recognition (NER) models to identify SIG references, annotating meeting transcripts, and conducting analysis to answer specific research questions. The preliminary result indicates that the NLP approach to this problem is accomplishable, but we will need a customized Span Categorization model to improve the accuracy for SIG identification. This research will have significant implications for local policymakers, researchers, and the general public, due to our improved understanding of how municipal public policy is created and who is involved in the process.

Search engines are the primary gateways of information. However, the veracity of search results is often not considered before the search results are promoted to users. As the most popular video search engine, YouTube recommends misinformation on the treatment, spread, and origins of COVID-19, undermining public health efforts. Despite the global effects of COVID-19 misinformation, the majority of research is confined to the Global North, leaving the Global South behind. My research aims to qualitatively and quantitatively compare the exposure to COVID-19 misinformation on YouTube between a country in the Global North and Global South. I focused on the United States and South Africa, both of which have been heavily affected by the pandemic. Using 48 curated COVID-19 misinformation search queries, I systematically audited YouTube search results for 10 consecutive days with 12 programmed bots emulating “real” users in South Africa and the United States. This sock-puppet audit method ethically prevented harmful exposure to misinformation to real users and provided a scalable, controlled way to collect data. Then, I fact-checked the collected video results and trained a machine-learning model to scale the annotation process, allowing for the measurement of misinformation prevalence in different geolocations. My preliminary findings showed that YouTube search results differ by up to 20% between South Africa and the United States. Based on these early findings, I expect to see potential differences in the veracity and rankings of COVID-19 misinformation search results between users in South Africa and the United States. This research is the first to report a comparative investigation of COVID-19 misinformation on YouTube between a country in the Global North and Global South. It also establishes a novel method of conducting geolocation audits in different countries, paving the way for further audit research in the Global South and ensuring accountability for social media platforms.

The amount of online information is rapidly increasing, including an abundance of potentially misleading content. However, fact-checkers rely heavily on manual search methods to identify this content, leading to a significant investment of time and resources. While social media monitoring tools exist for platforms like Twitter and Facebook, such tools still need to be improved for video search platforms like YouTube. We collaborated with Africa's largest indigenous fact-checking organization – PesaCheck, during the 2022 Kenyan general election. The spread of rumors aimed at voter suppression gave an advantage to the tally of a specific presidential candidate, which created an urgent need for automated fact-checking systems. To address this issue, a team of undergraduates from the information school and computer science department developed a tool to automatically generate search queries related to significant events and topics for fact-checkers to monitor while providing them the flexibility to modify or create their queries. I was in charge of developing a method to identify video ids, extract the keywords with Natural Language Understanding (NLU), and the function that allowed users to choose keyword tags to characterize videos and then link them to the top search queries of the day. During the crucial period of the Kenyan election, ten full-time fact-checkers were using our tool to check and report falsities on the internet. During the four months of deploying this fact-checking tool, 42 misinformation discovery annotations were added by various fact-checkers, and over 500 misinformation queries were generated by the tool. As a result, the new Kenyan president emerged from one of the most competitive elections in the nation's history. Our research project will also be published as a case study paper to discover value-sensitive fact-checking systems through participatory design.

Climate change coupled with inadequate infrastructure has disproportionately impacted people from underserved communities, as is the case for one village located in rural Alaska. Due to the increasing frequency of microbial water contamination exacerbated by temperature rise, this community experiences water insecurity. Previous research also illustrates that water samples collected from the Kuskokwim River, which is used for subsistence, had excessively high mercury levels compared to the regional baseline levels. Furthermore, our team's preliminary findings from informal interviews with community members indicate that their use of a rainwater cistern is well received by many households. However, community leaders have questioned the cistern's operation from September to March. When the temperature falls below zero degrees Fahrenheit in the winter, the cistern must be shut down to avoid damage to the system, and because it is not possible to collect water. Our work will develop and administer an online household-level survey to understand perceptions about local drinking water conditions. We implement principles derived from Indigenous research methodologies and Community-Based Participatory Research approaches. * Preliminary results gathered from conversations with the tribal council suggested a positive perception of the water quality from the installed water cistern, but we hope to quantify the water quality from survey data. * Our research focused on two areas: we will evaluate the utility of an online survey in collecting data from one remote community to understand their water needs and use the data to help guide future decision-making to improve the cistern operation under extreme temperatures. This work is part of a larger initiative to underscore the self-determination of under-resourced communities in Alaska, offering insights into Indigenous research methods and inspiring future design in the field. Note: * means revison per feedback, but will still need to get tribal approval as they approved for previous version as is