2024-01-30: Disinformation Detection and Analytics REU Program - Final Presentations

By -


Adam Martin presenting his work at 2023 REU Final Presentation Session

The Research Experiences for Undergraduates (REU) offers practical learning and research opportunities for undergraduates concentrating on disinformation detection and analytics. Commencing on June 5th, 2023, this 10-week summer initiative takes place at Old Dominion University, hosted by the Web Science and Digital Libraries (WS-DL) Research Group within the Department of Computer Science in collaboration with Virginia Modeling, Analysis, and Simulation Center (VMASC) and the Department of Communication Disorders and Special Education. The REU program is funded by the 2023 National Science Foundation (NSF) Research Experiences for Undergraduate Site. On August 3rd, 2023, the endpoint of this program was marked by the final presentations delivered by the six participating students. Having previously provided updates on their mid-summer progress, these students showcased the outcomes of their research. The final presentation session, attended by mentors, other faculty members, and students from the CS department, witnessed both in-person and virtual participation, offering an opportunity for a comprehensive exploration of their research findings.

The final presentations of the REU program started with a brief introduction from Dr. Jian Wu, Co-PI of the program.

Michael Evans (@mevansci) advised by Dr. Jian Wu (@fanchyna)

Michael kicked off the final presentation session of the REU program with a presentation on "Scientific News Verification with GPT". In this project, Michael explored building scalable and domain-adaptable scientific claim verification methods using LLMs (GPT 3.5). During the REU program, he contributed to expanding the Multi-Domain Scientific Claim Verification Evaluation Corpus (MSVEC) dataset, which includes true and false evidenced scientific claims in multiple domains. 

Adam Martin (@AdamMartinTech) advised by Dr. Erika Frydenlund (@ErikaFrydenlund)

In the following presentation, Adam presented “Training a Machine Learning Model to Detect Russia-Ukraine War Disinformation.” In his study, Adam utilized large language models to identify Russian disinformation given a training dataset of news articles. In his approach, he used data from a database that includes Russian disinformation articles and he collected and translated pro-information articles from legitimate sources using a web-scraper. Additionally, his methodology included preprocessing and labeling the data, constructing a model for disinformation identification, and subsequently evaluating the model's effectiveness.

Rachel Zheng (@rachel_z_03) advised by Dr. Michele Weigle (@weiglemc)

In the REU final presentations, Rachel presented her work on “Exploring the Challenges in Archiving Instagram”. During the REU program, she worked on pinpointing when redirects to Instagram’s login page became prominent. She wrote a script to scrape archived Instagram to get the page sources that provide information that is not visible on the replayed mementos. She further discussed the major changes in Instagram she identified through the scraped pages.

Isabelle Puwo (@IsabellePuwo14) advised by Dr. Anne Perrotti

Following Rachel's presentation,  Isabelle Puwo presented her work on "Exploring TikTok as an appropriate teaching tool for general education teachers, special education teachers, and speech-language pathologists". She analyzed the varied purposes behind TikTok content produced by teachers in general education, special education (SPed), and speech-language pathology (SLP). Her results showed that teachers have the highest general and professional development purpose, while SLPs have the highest demonstration and material purpose.​ Isabelle concluded her presentation by stating that the SPed, general education teachers, and SLPs can create informative content related to process and professional development.

Johnovon Richards (@Johno_RichCS) advised by Dr. Faryaneh Poursardar (@Faryane)

Johnovon presented his presentation titled “Leveraging Data Analysis and Machine Learning to Authenticate Yelp Reviews through User Metadata Patterns”. He discussed different factors that provide indications of fake reviews such as no profile image of the reviewer, low friend/review count, incoherent text/text not matching rating, and uninformative text/personal bias. In this work, Johnson trained a Logistic Regression model on labeled Yelp Review data. His results indicate that this model model shows the potential to effectively distinguish between fake and real reviews based on user metadata, however, he discussed how future research can explore more advanced machine learning algorithms and feature engineering techniques for this task.

Parker Story (@ParkerStory_) advised by Dr. Vikas Ashok (@vikas_daveb)

Parker concluded the REU 2023 final presentations event by presenting his work titled “Dark Pattern Webpage Ads: Impact on Screen Reader Users”.  He highlighted the challenges faced by screen reader users when dealing with ads, including difficulties in using ad-blockers, increased browsing time, deceptive ad placement, lack of effective narration, and the need for external support in assessing ad safety. During the REU program, Parker built a dataset containing deceptive ads and non-deceptive ads, trained a multi-modal classifier, and evaluated the performance of the classifier. He concluded his presentation by discussing their plan to improve the dataset and model and submitting a Journal paper. 

The presentations provided diverse viewpoints on various types of disinformation available on the web and different approaches for disinformation detection and analytics. The discussion highlighting the further research avenues on disinformation detection, made for a great ending to the 2023 ODU CS REU summer internship program.

-- Yasasi Abeysinghe (@Yasasi_Abey)

Comments

Post a Comment