EDU-AI: Enhancing Data Retrieval and Dialogue with Retrieval-Augmented Generation and NLP Techniques
Abstract
This study introduces EDU-AI (Educational Data Utility with AI), an advanced platform designed to provide accurate, context-aware and conversational response to academic queries within institutions. It excels at responding to queries such as “What is the minimum GPA for graduate admission to ETSU’s CS program?” by integrating real-time information retrieval with sophisticated language generation. The system searches a vast knowledge base to provide clear and concise responses in seconds. EDU-AI integrates Retrieval-Augmented Generation (RAG) architecture with large language models (LLMs) and Natural Language Processing (NLP) techniques. It employs techniques such as topic modeling, coreference resolution, similarity search, document ranking, query-rewriting, and prompt engineering to deliver accurate, contextually relevant, and conversational responses. To evaluate its effectiveness, EDU-AI was implemented as a prototype for East Tennessee State University (ETSU) publicly available web data to build an intelligent question-answering system for the ETSU scholars. The system achieved an 83% retrieval accuracy in document extraction from a pool of 5000 documents, and with average response time ~3 seconds, demonstrating reliability and timely information. A perplexity score of 26, measured via GPT2LMHeadModel, indicates sensible, confident responses, while an 88% coherence with ground truth, assessed using all-mpnet-base-v2, reflects human-like relevance. Qualitative feedback from ETSU focus groups praised its accuracy, context-awareness, and inclusivity (e.g., voice command support), but noted areas for improvement: reducing context carryover, expanding domain coverage, and enhancing accessibility with multi-language support and mobile compatibility. The study’s key contributions include the improvement of information accessibility within academic environments through development of a scalable AI-driven platform for educational data retrieval, presentation, and inclusivity features. Future enhancements will focus on refining conversation management, broadening knowledge coverage, and expanding accessibility features. Keywords: Artificial Intelligence, Retrieval-Augmented Generation (RAG), Education, Natural Language Processing (NLP), Large Language Model (LLM)
Start Time
16-4-2025 10:00 AM
End Time
16-4-2025 11:00 AM
Room Number
303
Presentation Type
Oral Presentation
Presentation Subtype
Grad/Comp Orals
Presentation Category
Science, Technology and Engineering
Faculty Mentor
Ahmad Al Doulat
EDU-AI: Enhancing Data Retrieval and Dialogue with Retrieval-Augmented Generation and NLP Techniques
303
This study introduces EDU-AI (Educational Data Utility with AI), an advanced platform designed to provide accurate, context-aware and conversational response to academic queries within institutions. It excels at responding to queries such as “What is the minimum GPA for graduate admission to ETSU’s CS program?” by integrating real-time information retrieval with sophisticated language generation. The system searches a vast knowledge base to provide clear and concise responses in seconds. EDU-AI integrates Retrieval-Augmented Generation (RAG) architecture with large language models (LLMs) and Natural Language Processing (NLP) techniques. It employs techniques such as topic modeling, coreference resolution, similarity search, document ranking, query-rewriting, and prompt engineering to deliver accurate, contextually relevant, and conversational responses. To evaluate its effectiveness, EDU-AI was implemented as a prototype for East Tennessee State University (ETSU) publicly available web data to build an intelligent question-answering system for the ETSU scholars. The system achieved an 83% retrieval accuracy in document extraction from a pool of 5000 documents, and with average response time ~3 seconds, demonstrating reliability and timely information. A perplexity score of 26, measured via GPT2LMHeadModel, indicates sensible, confident responses, while an 88% coherence with ground truth, assessed using all-mpnet-base-v2, reflects human-like relevance. Qualitative feedback from ETSU focus groups praised its accuracy, context-awareness, and inclusivity (e.g., voice command support), but noted areas for improvement: reducing context carryover, expanding domain coverage, and enhancing accessibility with multi-language support and mobile compatibility. The study’s key contributions include the improvement of information accessibility within academic environments through development of a scalable AI-driven platform for educational data retrieval, presentation, and inclusivity features. Future enhancements will focus on refining conversation management, broadening knowledge coverage, and expanding accessibility features. Keywords: Artificial Intelligence, Retrieval-Augmented Generation (RAG), Education, Natural Language Processing (NLP), Large Language Model (LLM)