Development and testing of a temporal topic detection algorithm applied for enhancing e-learning in high-school and college students

Abstract

This thesis presents a comprehensive study on using advanced software tools for audio tran scription from video files, combined with innovative approaches to topic segmentation and comprehension within video content. The focus is on designing two distinct framework ar chitectures for performing Topic Segmentation and Comprehension from video transcripts. Thefirst framework explores the use of Large Language Models with YouTube APIs and Streamlit library as other tools in order to analyze every link from a file filled with YouTube links of educational videos The second framework leverages the use of Multi-RAG Agents to explore how it can be used and implemnted different agents using different LLMs for topic segmentation in tran scribed texts. By using variations of Claude 3.5 and GPT-4 APIs of the models and the Lang Chain library, which enables the integration of multiple LLMs and multi agent systems, this approach optimizes responses, enhancing the segmentation accuracy. Each architecture’s performance was evaluated based on if it was able to achieve the results of getting the recommendations for the educational videos from the search input by the user. The transcribed texts undergo rigorous topic segmentation analysis, using state of-the-art NLP and LLM techniques, leading to significant improvements in identifying and categorizing topics within video content. This facilitates a more nuanced and detailed content analysis process, supported by a clean and scalable infrastructure designed to meet the needs of content analysts and researchers. Through extensive testing and validation, this system demonstrates substantial improve ments in the precision of topic searches within videos, establishing a new benchmark for video content analysis. This work highlights the potential of integrating transcription tools with cutting-edge NLP techniques, paving the way for enhanced insights in video content analysis.