Assessment of Large Language Models for Wildcard Match Identification

Claire Han; Abel Andres Reyes-Angulo; Sidike Paheding

Download Paper | Permalink

Conference: 2025 ASEE Annual Conference & Exposition
Location: Montreal, Quebec, Canada
Publication Date: June 22, 2025
Start Date: June 22, 2025
End Date: August 15, 2025
Conference Session: Computing and Information Technology Division (CIT) Technical Session 4
Tagged Division: Computing and Information Technology Division (CIT)
Tagged Topic: Diversity
Page Count: 12
Permanent URL: https://peer.asee.org/55487
Download Count: 4

Paper Authors

author page

Claire Han

biography

Abel Andres Reyes-Angulo Michigan Technological University orcid.org/0000-0003-0332-8231

visit author page

ABEL REYES-ANGULO is a Ph.D. student in Computational Science and Engineering at Michigan Technological University in Houghton, MI. He received his B.S. degree in Computer Engineering and his M.S. degrees in Electrical and Computer Engineering from Purdue University Northwest in Hammond, IN, in 2018 and 2021, respectively. His research interests include, but are not limited to, machine learning, deep learning in computer vision, medical image analysis with AI, software development, and virtual reality.

visit author page

biography

Sidike Paheding Fairfield University orcid.org/0000-0003-4712-9672

visit author page

Assistant Professor in the Department of Computer Science and Engineering.

visit author page

Download Paper | Permalink

Abstract

Throughout its development, the internet has become a massive part of everyday life. Convenience in communication, entertainment, exercise, and various other factors have provided positive impacts on people around the world. However, with these milestones, the internet has simultaneously become a haven for malicious actors. These malicious actors continuously compromise private information and software systems through phishing emails, hacking, and WiFi networks. As these attackers evolve their tactics, cybersecurity is becoming a key field in computer science and must advance alongside new problems. This paper investigates the impact that Artificial Intelligence (AI) has on web browser cookies and wildcard features, which are special symbols that can represent other characters and sort files. Since wildcards organize cookies into groups that have similar data and function trends, browsers can determine which cookies are problematic earlier on. AI is an expert on observing and analyzing trends in data; including AI in the wildcard features would provide an even faster solution for web browsers to see which cookies are secure. Currently, there is plenty of research focused on cybersecurity, including the concern of fingerprinting, phishing, and cross-site attacks, but web browsers, and the cookies in those browsers, are an overshadowed topic. Since cookies often provide security and user convenience in various websites, they are critical to ensure search history and advertising privacy for everyday users. On the contrary, if cookies are not well-researched or protected, malicious actors can take advantage of vulnerable marketing, analytics, and third-party cookies to steal personal information and expose online profiles. Furthermore, in today’s society, AI is one of the most useful tools in cybersecurity. Though developments in AI are still in progress, it proves to be extremely important in learning new hacking techniques, defending systems against attacks, and refining algorithms for machines and security software. Even in the world of browsers, AI is beneficial, as it can observe patterns in dangerous cookies to preserve user privacy, extract cookie features to analyze individually, and examine wildcards to expedite data requests and organization. This project assessed web browser security through the use of Large Language Models (LLMs), such as GPT-2, T5, and Alpaca, as AI algorithms for identifying wildcard features in cookies. By inputting a long sequence of information about a cookie, such as stating its domain, function, and retention period, the LLM responds back with a yes or no answer about whether the cookie is a wildcard match. The algorithm was trained on a diverse dataset from the Open Cookie Dataset, showing a potential breakthrough for cybersecurity and AI to secure online safety. Preliminary results provide a promising performance of above 91% of accuracy in cookie wildcard match identification.

Citation
Format

Han, C., & Reyes-Angulo, A. A., & Paheding, S. (2025, June), Assessment of Large Language Models for Wildcard Match Identification Paper presented at 2025 ASEE Annual Conference & Exposition , Montreal, Quebec, Canada . https://peer.asee.org/55487

TY - CPAPER
AB - Throughout its development, the internet has become a massive part of everyday life. Convenience in communication, entertainment, exercise, and various other factors have provided positive impacts on people around the world. However, with these milestones, the internet has simultaneously become a haven for malicious actors. These malicious actors continuously compromise private information and software systems through phishing emails, hacking, and WiFi networks. As these attackers evolve their tactics, cybersecurity is becoming a key field in computer science and must advance alongside new problems. This paper investigates the impact that Artificial Intelligence (AI) has on web browser cookies and wildcard features, which are special symbols that can represent other characters and sort files. Since wildcards organize cookies into groups that have similar data and function trends, browsers can determine which cookies are problematic earlier on. AI is an expert on observing and analyzing trends in data; including AI in the wildcard features would provide an even faster solution for web browsers to see which cookies are secure. Currently, there is plenty of research focused on cybersecurity, including the concern of fingerprinting, phishing, and cross-site attacks, but web browsers, and the cookies in those browsers, are an overshadowed topic. Since cookies often provide security and user convenience in various websites, they are critical to ensure search history and advertising privacy for everyday users. On the contrary, if cookies are not well-researched or protected, malicious actors can take advantage of vulnerable marketing, analytics, and third-party cookies to steal personal information and expose online profiles. Furthermore, in today’s society, AI is one of the most useful tools in cybersecurity. Though developments in AI are still in progress, it proves to be extremely important in learning new hacking techniques, defending systems against attacks, and refining algorithms for machines and security software. Even in the world of browsers, AI is beneficial, as it can observe patterns in dangerous cookies to preserve user privacy, extract cookie features to analyze individually, and examine wildcards to expedite data requests and organization. This project assessed web browser security through the use of Large Language Models (LLMs), such as GPT-2, T5, and Alpaca, as AI algorithms for identifying wildcard features in cookies. By inputting a long sequence of information about a cookie, such as stating its domain, function, and retention period, the LLM responds back with a yes or no answer about whether the cookie is a wildcard match. The algorithm was trained on a diverse dataset from the Open Cookie Dataset, showing a potential breakthrough for cybersecurity and AI to secure online safety. Preliminary results provide a promising performance of above 91% of accuracy in cookie wildcard match identification.
AU - Claire Han
AU - Abel Andres Reyes-Angulo
AU - Sidike Paheding
CY - Montreal, Quebec, Canada
DA - 2025/06/22
PB - ASEE Conferences
TI - Assessment of Large Language Models for Wildcard Match Identification
UR - https://peer.asee.org/55487
ER -