Ionian University
Ανάλυση εκπαιδευτικών βίντεο διαλέξεων βασισμένη σε κοινωνικά και γλωσσολογικά μεταδεδομένα
Abstract
dc:descriptionEducational video lectures are nowadays one of the main types of e-learning course material, and are provided in almost all distance educational systems. The significant increase in educational videos on-line was caused by the rise of Massive Open Online Courses that are available through open internet access, and without limitation to the number of participants who wish to attend them. In addition to the large increase in the number of educational video lectures, there is a growing interest on the part of the scientific community, which is triggering new studies in the field. Therefore, in-depth analysis of educational video lectures is necessary. Access to educational video lectures is possible through content based social online networks. Thus, beyond the videos, the learners have at their disposal a platform to interact both with the content and with other learners. Media metadata can be gathered through educational video websites, which in turn can provide important information about these videos. Today, the field of machine learning can provide important metadata techniques and methods to create new perspectives in the subject of the study. The present study attempts to delineate the framework of the use of machine learning methods in educational video lectures as well as to answer relevant questions concerning the metadata of the websites they host them. The two categories of metadata examined relate to the linguistic elements of the lexical content of the educational video lectures and the social elements of the interaction of the learners with the corresponding web site hosting them. The initial part of the study is devoted to their definition and their pre-processing, especially of the linguistic elements, analyzing the process of their conversion from unstructured to structured elements which can be later used algorithmically. The present research in the first part focuses on identifying the linguistic characteristics of speakers in popular educational video lectures. For this purpose an extensive analysis of the popularity of video lectures is made, the positive and the negative parameters that influence them are displayed, and the calculation formula is presented. With the use of machine learning techniques, the less and more popular lecturing videos are clustered, and the values of the linguistic features of the speakers are recorded in each case. Additionally, it has been found that it is useful to separate short and long in duration videos. This sets out the characteristics of the speakers for both the short and the long in duration popular videos and compares them. The next research section concerns the creation of an advanced framework for the search for educational video lectures using linguistic and social metadata. The aim is to present the user with a list of lectures based on the search he has done, taking, however, into account the views of the other learners who have already attended them, and expressed their opinion on them. At this point, machine learning is used by applying content similarity algorithms, and a new similarity methodology is introduced that adds a parameter called social weight, which is calculated based on the popularity of the educational video lecture. The confirmation of the method is made by conducting a user evaluation experiment, which reveals that better ranking is performed for the users during their search, thus saving them valuable time. The last part of the research is related to the automatic point of interest identification within an educational video lecture, combining two machine learning techniques. More specifically, the terminology and value of the video lecture is firstly extracted from users' comments. Secondly, the topics of video lectures are extracted from the corresponding transcripts. Each word in the topic is assigned a corresponding value in the first phase, and the sum of those values is defined as the topic score. The topic with the best score is selected as the most relevant video lecture. The words of the topic with the best score also define automatically the corresponding time points that include them (subtitles). Finally, a user evaluation experiment is conducted the results of which enhance the credibility of the proposed process. In this way, the users’ navigation through the video lecture is directed, helping the learner to find the point of interest. In conclusion, machine learning techniques and methodologies create new perspectives on the use of metadata in educational video lectures. The present research highlights how these methodologies can be applied, and answers relevant questions concerning the linguistic and social metadata of websites hosting educational video lectures.
Degree
thesis:*- Grantor dc:publisher
- Ionian University
- Year dc:date
- 2018
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Kravvaris, Dimitrios
- Κράββαρης, Δημήτριος
Subjects
dc:subject × 12Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/44572