Source-linked AI summary
Artificial Intelligence-Enabled Intelligent Assistant for Personalized and Adaptive Learning in Higher Education
Ramteja Sajja, Yusuf Sermet, Muhammed Cikmaz, David Cwiertny, Ibrahim Demir
TL;DR
Higher education needs more scalable, adaptable, and integrated support than existing AI-based virtual teaching assistants provide. This paper presents an LMS-connected AIIA framework with personalized assistance, assessment, and learning tools, contributing a broader educational support system while identifying implementation limitations.
Problem
Research on AI-based virtual teaching assistants remains limited in scalability, adaptability across learning contexts, LMS integration, and mechanisms for preventing academic dishonesty.
Method
The VirtualTA chatbot generates a knowledge base by extracting, parsing, and encoding course resources acquired from learning management systems.
Results
The framework integrates question answering, flashcards, automated assessment, personalized support, context-aware conversation, and an embedded coding sandbox.
Takeaways & Limitations
The VirtualTA system offers a more comprehensive and adaptable educational support approach with integrated functionality, context awareness, academic-integrity emphasis, and learning analytics.
Takeaways & Limitations
The system lacks OCR functionality for accurately parsing scanned PDF content, limiting its handling of unstructured PDF resources.
Abstract
from arXiv · showhide
This paper presents a novel framework, Artificial Intelligence-Enabled Intelligent Assistant (AIIA), for personalized and adaptive learning in higher education. The AIIA system leverages advanced AI and Natural Language Processing (NLP) techniques to create an interactive and engaging learning platform. This platform is engineered to reduce cognitive load on learners by providing easy access to information, facilitating knowledge assessment, and delivering personalized learning support tailored to individual needs and learning styles. The AIIA's capabilities include understanding and responding to student inquiries, generating quizzes and flashcards, and offering personalized learning pathways. The research findings have the potential to significantly impact the design, implementation, and evaluation of AI-enabled Virtual Teaching Assistants (VTAs) in higher education, informing the development of innovative educational tools that can enhance student learning outcomes, engagement, and satisfaction. The paper presents the methodology, system architecture, intelligent services, and integration with Learning Management Systems (LMSs) while discussing the challenges, limitations, and future directions for the development of AI-enabled intelligent assistants in education.
1. Introduction
Higher education’s changing landscape has exposed limits in personalized support and instant feedback, motivating interest in AI, NLP, and web-based virtual teaching assistants. This study introduces AIIA, a web-based AI-enabled framework designed to reduce cognitive load and provide interactive, personalized learning support.
- Motivation: Traditional teaching methods often struggle to provide personalized support and instant feedback for diverse, globally distributed students.The introduction links these challenges to rapid digital transformation and evolving student needs.
- Technological context: AI, NLP, LLMs, chatbots, and web platforms offer tools for communication, knowledge generation, feedback, visualization, and support across specialized educational domains.The introduction describes applications spanning engineering modeling, coding, mathematics, statistics, visual inputs, and ethical reflection.
- Study contribution: The study introduces AIIA, a NodeJS-backed web framework using AI and NLP to create an interactive and engaging platform for qualitative disciplines.AIIA is presented as an AI-enabled Virtual Teaching Assistant intended to enhance student learning.
- Study contribution: AIIA reduces cognitive load through accessible information, knowledge assessment, inquiry responses, quiz and flashcard generation, and personalized support tailored to learners’ needs and styles.The framework is intended to support the integration of AI-enabled technologies and web systems into education.
- Study significance: The research aims to inform the design, implementation, and evaluation of AI-enabled VTAs and the development of tools that enhance learning outcomes, engagement, and satisfaction.The introduction positions the findings as contributions to broader discourse on integrating AI and NLP in education.
2. Related Work
Prior research shows that AI-based virtual teaching assistants can support higher education while raising ethical, pedagogical, scalability, and academic-integrity concerns. This study addresses identified gaps through an AI-enabled Intelligent Assistant designed for personalized and adaptive learning, LMS integration, richer instructional data, and support for both students and instructors.
- Prior AI and VTA research: Existing literature reviews AI applications in education, including adaptive learning, teaching evaluation, virtual classrooms, NLP, and ChatGPT integration in higher education.The reviewed work spans conceptual discussions, empirical chatbot studies, educational dialogue evaluation, and domain-specific applications.
- Prior AI and VTA research: Research identifies both the potential benefits and challenges of AI-based VTAs, including personalized learning, automated assessment, engagement, ethical risks, and pedagogical limitations.Conversational agents perform well on conversational uptake but are quantifiably worse than real teachers on several pedagogical dimensions, especially helpfulness.
- Research gaps: Key gaps concern scalability and adaptability across diverse learning contexts, future effects on higher education, and integration with other educational systems.The literature has not fully addressed how AI-based VTAs can generalize across learning contexts and integrate into higher-education environments.
- Research gaps: Most existing AI-based solutions omit class recordings and interactions and do not adequately balance personalized assistance for students with support for instructors.Incorporating these instructional sources could enrich VTA knowledge bases and provide a more comprehensive learning experience.
- Study contribution: The proposed AIIA addresses these gaps by integrating with existing LMSs, using class recordings and interactions, serving students and instructors, and incorporating measures to prevent cheating.The study designs, implements, and evaluates the assistant for personalized and adaptive learning in higher education while addressing academic integrity.
3. Methodology
The methodology develops AIIA as a conversational-AI framework for inclusive, personalized, and adaptive higher-education learning. It combines language models, course-material embeddings, speech transcription, reliability safeguards, and a four-component system architecture to support information access, assessment, feedback, and learning pathways.
- Research objective: AIIA applies conversational AI and NLP to address diverse learner needs, bridge learning gaps, and support inclusive, equitable, and engaging postsecondary education.The framework is intended to improve learning experiences and outcomes while facilitating continuous learning.
- Language-model implementation: GPT-3.5 was selected for its user-friendly API and advanced NLP capabilities, with text-davinci-003, gpt-3.5-turbo, fine-tuned Davinci, and Fine-Tuned Curie used for different language tasks.The selected models support text completion, query classification, and open-ended question generation through few-shot learning and fine-tuning.
- Knowledge representation and retrieval: OpenAI’s text-embedding-ada-002 converts assignments, announcements, lecture notes, forum posts, and recordings into 1,536-dimensional embeddings for cosine-similarity search.The embeddings support retrieval of semantically relevant information from long classroom documents.
- Reliability and system architecture: Prompt engineering and uncertainty responses are used to reduce hallucinations by withholding answers when the model lacks confidence and communicating uncertainty instead.This mechanism is intended to prevent incorrect or conflicting information from reaching users.
- Speech-data integration: Whisper transcribes recorded classes into textual data, expanding the analysis and query-resolution data pool and improving response comprehensiveness and knowledge transfer.The approach incorporates underused lecture and class recordings made increasingly prevalent during the shift to remote learning.
Resource
The VirtualTA builds a dynamic knowledge base by acquiring LMS resources, parsing applicable files into coherent 800-character blocks, and encoding them as text embeddings. Its response pipeline classifies queries, matches embedded questions to course content, generates responses or learning materials, and integrates with centralized infrastructure and LMS interfaces.
- Knowledge Base Generation: VirtualTA acquires lecture files, lecture recordings, and reading material from an LMS for knowledge-base generation.The process includes resource extraction, parsing, and encoding.
- Knowledge Base Generation: Applicable file-format resources are parsed into approximately 800-character chunks while preserving intact words and sentences.Some resources, including announcements, bypass parsing.
- Knowledge Base Generation: The resulting 800-character blocks are encoded into high-dimensional text embeddings that form VirtualTA’s fundamental knowledge base.The embeddings preserve and enhance the semantic richness of course content.
- Query Processing: The response pipeline classifies query type and intent, converts the question into an embedding, and uses cosine similarity to select the ten most correlated course documents.This multistage process prepares focused context before response generation.
- Response Generation: The system uses fine-tuned models and prompt engineering for question generation and hallucination mitigation, then executes classified requests such as answers, summaries, code, questions, and essay outlines.A confidence-based mechanism can withhold answers when the model is unsure, communicating a message such as “I'm not sure.”
- System Infrastructure: A centralized web-based cyberinfrastructure combines NGINX, NodeJS, PostgreSQL, caching, and management modules with Canvas-integrated interfaces and LMS data-processing support.The system also includes a Python Flask Speech API based on WHISPER and pyannote for transcription services.
4. Results
The results section presents intelligent services and system enhancements designed to support both students and instructors. These features aim to improve students’ learning experiences while assisting instructors with instructional tasks and assessments.
- Results: The implemented intelligent services and enhancements address the needs of both students and instructors.The system provides tools for students and support for instructors’ instructional and assessment activities.
- Student-oriented enhancements: Student-oriented enhancements include Dynamic Flashcard Integration.This feature is presented as one of the tools intended to provide students with valuable learning resources.
- Student-oriented enhancements: Student-oriented enhancements also include Automated Assessment: Intelligent Quiz Generation.The passage identifies intelligent quiz generation as part of the system’s automated assessment capabilities.
4.1. Student-Oriented Enhancements
The system integrates student-oriented tools for studying, assessment, information access, coding practice, summarization, and context-aware support. These enhancements aim to make learning more interactive, accessible, personalized, and responsive to student needs.
- Enhancement Overview: Student-oriented enhancements include flashcards, auto-graded quizzes, automated question-answering, an embedded coding sandbox, topic summarization, and context-aware conversation.Together, these capabilities support preparation, self-assessment, information access, code experimentation, condensed learning materials, and personalized interaction.
- Flashcards: Flashcards support course preparation with true/false and open-ended questions, answers, and detailed explanations or reasoning.Students can request flashcards on any course topic, with the answer revealed by flipping the card.
- Automated Assessment: Quizzes combine true/false and open-ended questions with auto-grading to help students test their knowledge and understanding.The self-assessment feature promotes active participation, enables progress monitoring, identifies improvement areas, and reinforces understanding through explanations.
- Question-Answering: Automated question-answering provides prompt responses to administrative and course-content inquiries using information from the system’s knowledge base.The feature is intended to streamline information access and support students’ independent learning.
- Coding Sandbox: An integrated coding sandbox lets students seek programming guidance, execute basic code, and clarify code snippets directly within the chatbot interface.The environment supports beginners by enabling quick testing and interaction with programming concepts, including in non-CS courses.
- Summarization and Context-Aware Conversation: The system generates concise course-topic summaries with GPT-3.5 and adapts responses to student communication styles while maintaining relevant conversational context.Summarization helps students grasp key points quickly, while context-aware replies can acknowledge difficulties, provide empathy, and indicate confidence and information sources through disclaimers.
4.2. Instructor-Focused Enhancements
The instructor-focused enhancements streamline grading, homework handling, and assessment-question generation. They include an auto-evaluator, automatic homework detection, color-coded grading, and configurable question generation for exams or quizzes.
- Core enhancements: The system combines an auto-evaluator, automatic homework detection, and automated assessment-question generation to support instructors’ instructional and assessment tasks.These enhancements aim to streamline grading, promote independent learning and academic integrity, and reduce the effort required to create assessments.
- Automated grading: The auto-evaluator compares student responses with instructor-provided questions and correct answers, including solutions uploaded as PDF files, and assigns scores.It also provides explanations for each question; although scoring may occasionally be lenient, the output supports instructors’ final grading decisions.
- Automated grading: Scores of less than or equal to two appear red, scores of less than or equal to five appear yellow, and scores greater than five appear green.This color-coded scheme helps instructors quickly identify each question’s performance level during evaluation.
- Homework detection: When instructors mark assignments as offlimits for direct answers, VirtualTA withholds direct responses and guides students toward appropriate resources.The mechanism detects whether questions resemble homework or assignments, encouraging independent learning, critical thinking, and academic integrity.
- Question generation: Instructors can request exam or quiz questions by selecting True/False, Multiple Choice, or Open-Ended formats.This feature is intended to save time and effort in the otherwise laborious process of generating assessment questions.
5. Discussions
The discussion presents VirtualTA as a comprehensive and adaptable educational support system while identifying implementation challenges and future directions for expanding its capabilities and impact.
- Contributions: Its integration of diverse functions, support for different question types, context awareness, academic integrity, and learning analytics contributes to a sophisticated educational support system.The discussion characterizes VirtualTA as more comprehensive and adaptable than existing educational chatbot systems.
- Contributions: VirtualTA extends traditional chatbot support with flashcards, quizzes, automated homework evaluation, a coding sandbox, and summary generation.These features create a broader learning support ecosystem than basic question answering.
- Challenges and Limitations: PDF processing, LMS integration, and evolving underlying models created development challenges requiring structured-information extraction, a custom LMS library, and continual API adaptation.Scanned or unstructured PDFs complicated extraction, LMS platforms lacked standardized data-request methods, and model updates required ongoing upgrades.
- Future Directions: Future work includes improving natural language understanding, personalization, adaptive learning, multimodal support, evaluation, multi-LMS integration, and real-time video interaction.The proposed directions include semantic parsing, entity recognition, sentiment analysis, multimedia resources, stakeholder feedback, broader LMS compatibility, and immediate AI or instructor feedback.
- Future Directions: Further directions address instructor-assistant collaboration, gamification, longitudinal assessment, data privacy, algorithmic bias, and the human role in education.The discussion also proposes studying student performance, retention, and overall academic outcomes over time.
6. Conclusions
The conclusion presents the AIIA/VirtualTA system as an AI- and NLP-enabled framework for personalized and adaptive learning in higher education. It highlights the system’s potential contributions while acknowledging development challenges and identifying future advancement directions.
- Contributions: The research designed, implemented, and evaluated an AIIA system integrating advanced AI and NLP to support personalized and adaptive learning in higher education.The system aims to enhance learning outcomes and student engagement while addressing diverse learner needs in qualitative disciplines.
- Contributions and limitations: VirtualTA provides a comprehensive suite of tools and resources with potential to expand AI’s role in higher education.The conclusion also acknowledges development challenges and limitations that future iterations may address.
- Contributions: The AIIA framework supports self-regulated learning, student-faculty communication, and expanded access to learning resources.These capabilities are intended to enhance learning support effectiveness and shape higher education’s future trajectory.
Funding
The project was funded by NOAA through CIROH and The University of Alabama, with additional support from the National Science Foundation.
- Funding: Funding came from NOAA through CIROH under its cooperative agreement with The University of Alabama, alongside National Science Foundation award #2230710.The NOAA cooperative agreement is identified as NA22NWS4320003.