Investigating and Enhancing Online Software Development Resources: Automated Responses, Semantic Search, and Tagging in Video Tutorials

Abstract

The field of software development is rapidly evolving, requiring developers to continually refine their skills and adapt to new technologies. While video tutorials have become a popular medium for learning new concepts and techniques, challenges persist in interactive engagement, search functionality, and effective tagging. This dissertation explores innovative methods to enhance software development video tutorials by addressing these challenges using advanced large language models and transformer-based models.

Firstly, we explore developers' preferences in terms of online learning resources in the era of AI-driven chatbots like ChatGPT. Despite the rise of AI chatbots, which offer instant, personalized responses, video tutorials remain a preferred medium due to their visual and detailed explanations. However, our study also revealed opportunities for improving video tutorials by integrating interactive elements and leveraging AI technologies, setting the foundation for our subsequent projects.

Building on these insights, we introduce VidTutorAssistant, a system that leverages Generative Pre-trained Transformer (GPT) models to automate responses to video tutorial viewers' questions, thereby increasing interactive engagement in video tutorials and enhancing the learning experience.

Next, we present an improved video tutorial search method, ISM, that leverages transformer-based models to create semantically dense vectors from video data, enabling a more intuitive and efficient search experience. By capturing the contextual meaning of queries and video content, ISM surpasses traditional search methods, helping developers find the most relevant tutorials and specific content within them.

Finally, we introduce BM25-BERT, a hybrid approach for refining video tagging that combines traditional BM25F methods with transformer-based models. By re-ranking candidate tags initially generated by BM25F to improve context-awareness and accuracy, this method significantly enhances tutorial discoverability.

Through empirical studies and user evaluations, this dissertation advances software engineering and educational technology by offering innovative solutions to enhance video tutorials and examining the impact of AI tools on developers' learning preferences. By rigorously assessing the proposed methodologies, this research contributes to both academic research and practical applications.