Intelligent Agents in Education and Training

Permanent URI for this collectionhttps://hdl.handle.net/10125/112525

Browse

Recent Submissions

Now showing 1 - 5 of 5
  • Thumbnail Image
    Item type: Item ,
    Enhancing Fake News Detection Using GPT-2 with a Hybrid Deep Learning Approach
    (2026-01-06) Govinda, Priyadarshini; Govinda, Priyanka
    Nowadays, the spread of fake news represents a growing problem. To overcome this issue, it is essential to develop effective fake news detection systems. In this paper, we augment the process of fake news detection using GPT-2 alongside Convolutional Neural Networks (CNN) or Long Short-Term Memory (LSTM) networks. We use a dataset from Kaggle comprising of real and fake news articles. We compare the performance of the existing state-of-the-art real and fake news detection algorithms with our proposed hybrid model in Table 1. We blend GPT-2, known for contextual understanding, with either CNN or LSTM networks to capture more syntactical and semantic features from news articles and outperform the baseline algorithms. The preliminary results show that our hybrid model outperforms all the baseline algorithms on detecting real from fake news. By mixing generative pre-trained transformers with traditionally deep learning models, the robustness of misinformation detection systems can be significantly enhanced.
  • Thumbnail Image
    Item type: Item ,
    Video-Based Large Language Model for Enhanced Temporal Understanding
    (2026-01-06) Gantla, Devi Priyanka; Chintala, Gowri Shankar; Raghupatrini, Saichand
    This paper presents a novel video-based large language model (LLM) designed to tackle key challenges in temporal reasoning and multimodal integration for video understanding. Unlike existing approaches, our model introduces Enhanced Temporal Position Embeddings (ETPE) to capture long-range temporal dependencies and a Modal Fusion Bridge (MFB) to dynamically integrate visual and auditory modalities through bidirectional attention. Additionally, we implement an adaptive frame sampling strategy that minimizes redundancy while retaining motion-rich segments to enhance computational efficiency. Our system achieves state-of-the-art performance, recording 92.6% accuracy on Kinetics-400 and 90.4% on Kinetics-600—using only 5% of the training data. These innovations lead to significant gains in efficiency and accuracy across tasks such as video question answering, event localization, and real-time processing. We further demonstrate the practical applicability of our model through a Gradio-based interactive interface, enabling real-world use cases such as sports video analysis and accessibility support. Overall, this work establishes a scalable and efficient framework for multimodal video understanding systems.
  • Thumbnail Image
    Item type: Item ,
    Social Skills Scenarios and Emotional Affect Training Via Chatbot for Secondary Students with Autism Spectrum Disorder
    (2026-01-06) Suh, Sang; Hattensty, Emily; West, Nathaniel; Belvin, Chanel
    Chatbots show promise in making conversational support therapy more accessible while providing a safe environment for developing social skills without judgment. This research aims to create a chatbot that generates social scenarios, ranked by difficulty customized for each user. Unlike previous approaches that relied solely on rules-based decision trees, this system incorporates natural language processing (NLP) techniques, including a two-layer neural network for word vector space creation. This advanced NLP approach enables more sophisticated feedback and unprecedented customization in social skills training. The system continuously adapts to challenge users appropriately, addressing a common limitation of existing platforms that often provide overly simplistic training scenarios. In future work, the social scenarios and trainings would be adapted to more successfully address end user accessibility
  • Thumbnail Image
    Item type: Item ,
    Designing Conversational Agents to Support Learning from Scientific Graphs
    (2026-01-06) Siedenburg, Tim; Kuhlmann, Lea; Staudt, Philipp
    The need for accountability in research and an informed policy-making and society has increased demand for publicly accessible research data. However, the tendency in the scientific context to rely on graphs to illustrate findings can be challenging for users lacking domain expertise. Rooted in Conversation theory, this study explores using conversational agents to help users interpret such graphs. Using design science research, we develop and test a prototype conversational agent, addressing previously identified challenges like accessibility as well as language and education barriers in graph interpretation. While users with conversational agent access in a large-scale experiment do not demonstrate improvements in objective learning success, they report higher perceived learning, perceived empowerment and user experience, with reduced cognitive load. This research highlights the potential of large language model-based agents to improve access to complex data, offering insights into science communication as well as educational agent design and their limitations.
  • Thumbnail Image
    Item type: Item ,