2026-05-24T08:08:11Zhttps://keep.lib.asu.edu/oai/request

oai:keep.lib.asu.edu:node-1982662024-12-23T18:01:48Zoai_pmh:alloai_pmh:repo_items

198266 https://hdl.handle.net/2286/R.2.N.198266 http://rightsstatements.org/vocab/InC/1.0/ All Rights Reserved 2024 126 pages Doctoral Dissertation Academic theses Text eng Agrawal, Garima Liu, Huan Bertsekas, Dimitri P Davulcu, Hasan Deng, Yuli Arizona State University Partial requirement for: Ph.D., Arizona State University, 2024 Field of study: Computer Science Domain-specific Question-Answering (QA) systems are essential in fields like education, healthcare, and finance, where accurate, context-aware responses depend on specialized knowledge. However, developing knowledge-aware AI for these systems presents challenges, including the complexity of representing domain knowledge, handling evolving data, and ensuring reliable answers. This dissertation addresses these challenges across three main areas: Domain Knowledge Representation, Dynamic Knowledge Retrieval, and Effective Knowledge Augmentation. The focus is on transforming unstructured domain knowledge into structured representations, such as knowledge graphs, demonstrated through a case study in cybersecurity education. The dissertation introduces AISecKG, a cybersecurity ontology, develops a named-entity annotated dataset, and represents unstructured course material as cybersecurity knowledge graphs. A dynamic retrieval method is proposed to generate interpretable answers to user queries from evolving knowledge graphs using New Auction Path algorithms. These algorithms reuse learned node prices to guide the search for similar queries while assigning arbitrary prices to newly added nodes. Additionally, CyberGen is introduced as a method for augmenting knowledge graphs with large language models (LLMs) to generate the CyberQ dataset, addressing issues like hallucinations in LLMs by validating responses with knowledge graph ontologies. An analysis of critical failure points in existing knowledge graph-based retrieval augmented generation (KG-RAG) systems further leads to the proposal of Mindful-RAG, a more intent-driven and contextually-aware retrieval process to improve response accuracy. This work contributes to the development of more reliable, adaptive QA systems by integrating structured knowledge, dynamic retrieval, and augmentation techniques. Although focused on cybersecurity education, the methods can be generalized to other domains, enhancing the accuracy and reliability of AI-driven question-answering systems across specialized fields. Artificial Intelligence Educational technology Cybersecurity Knowledge Augmentation Knowledge Graphs LLM Hallucination Ontology Design Retrieval Augmented Generation (RAG) Knowledge-Aware AI for Domain-Specific Question-Answering Systems