Date of Submission
Spring 2024
Academic Program
Computer Science
Project Advisor 1
Rose Sloan
Abstract/Artist's Statement
This work presents LociGraph, an artificial intelligence agent that can autonomously search information on the non-public web, such as email inboxes, online communities, social media, or web applications not searchable on a public search engine. With a given query, the agent will browse the web using the keyboard and mouse to find a webpage containing the relevant information and extract the information in a structured format. For example, if the agent is given the query [Alex, studied at, ?] on your email inbox, the agent will start by typing “Alex” into the search bar, click on email related to Alex, read the content “Alex went to Bard College” and return [Alex, studied at, Bard College]. The framework consists of two parts: an agent pipeline, where a group of agents analyze the webpage content and suggest the next action, and a browser extension, where the user can enter the query and execute the suggested action. Preliminary evaluations show that Large Language Model (LLM) agents can navigate and extract information from real-world websites, but struggle to extract information from indirectly relevant content. All code, benchmarks, and results are available at https://github.com/ntcho/LociGraph.
Open Access Agreement
Open Access
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Cho, Nathan, "LociGraph: AI Agent Framework for Browser-Based Knowledge Graph Construction" (2024). Senior Projects Spring 2024. 46.
https://digitalcommons.bard.edu/senproj_s2024/46
This work is protected by a Creative Commons license. Any use not permitted under that license is prohibited.
Included in
Artificial Intelligence and Robotics Commons, Databases and Information Systems Commons, Software Engineering Commons