Content-based lecture video retrieval

Distance education, which is frequently encountered as e-learning or online learning today, is a new generation education approach in which the instructor and the student are not physically in the same place during education. Various technologies are used to facilitate online student-instructor communication. This approach has become more common and available on the world wide web (www) with the coronavirus pandemic (COVID-19), especially regarding lecture videos. However, the high rate of increase in the number of lecture videos on the internet has made it very difficult for users who want to access a specific video with a certain content. In the context of developing a proposal for these challenges, this research deals with the content-based search method that aims to provide users with access to videos related to certain educational content. The videos need to be accurately categorized in order to make it easier for users to find the specific video content. From the technical point of view, in order to be able to make such categorization, the textual information of the videos should be first retrieved. In this context, two different indexing methods called optical character recognition (OCR) and automatic speech recognition (ASR) are adopted to extract the textual information of the videos in the research. These two methods and an analysis in which these two methods are used together, are handled in this thesis over a specific data set. A collection of 110 lecture videos is used as the dataset. Within this scope, by referring to a thesis that made analysis with OCR method using the same dataset, this time the same metric analyzes are made with the ASR method. Finally, various metric values are calculated using both OCR and ASR methods. Three different traditional machine learning methods are used for classification analysis of the dataset. The traditional machine learning methods used are Support Vector Machine (SVM), Naïve Bayes and Random Forest methods. Accordingly, metric analysis of different machine learning methods and different indexing methods on the same dataset are compared. As a result of the analysis, the strengths and weaknesses of traditional machine learning methods and indexing methods are explained in order to use the lecture videos in content-based search. In addition, the aspects of the method that can be improved for future studies on the same subject are emphasized and suggestions on this subject are presented. This thesis concludes with a discussion of how the comparative research can affect both the education and software industries.