Kalles, Dimitris | Panagiotakopoulos, Christos T. | Kotsopoulos, Stavros | Vassilakopoulos, Michael | Kotsiantis, Sotiris | Sakkopoulos, Evangelos
Big Data | Learning Analytics | Machine Learning | Distance Learning | Sentiment Analysis
science-technology
At the present time, the technological boom has an emerging synergy with organizations, social and individual affairs. Information and Communication Technologies are triggering constantly developments in most fields of our society. By configuring new needs and services, huge amounts of information are accumulated in databases. The daily use of technological achievements and the enormous amount of recorded information, drive stakeholders to improve decision making with data analysis process. At a fast pace, big data techniques and intelligent systems are widely used to handle data and export value for policy making.
The domination of digital technology has expanded in the field of education and is inevitably linked to e-learning methods. The entry of ICT into education has changed conventional practices of learning and teaching around the world. Distance learning has widely adopted learning management systems as an immediate and alternative way for a variety of learners to gain access to educational resources and services. Global competitiveness and student expectations for personalized educational services are driving academic society to overcome the lack of physical abstention in distance learning by exploiting the plethora of students’ interactions in learning management systems.
Dropout rates are one of the main concerns in distance learning due to the impact on profits and the reputation of institutions. Financial constraints and daily time pressure intensify this phenomenon, as, the main profile of distance learning students is adults with professional and family responsibilities. Timely identification of students at risk has high practical value in effective students' retention services. Big data are applied to manipulate, analyze and predict students’ failure, supporting self-directed learning.
In order to support adaptive and personalized learning environments, the main effort of this dissertation is the research and development of a big data analytics framework. The goal is to identify trends and patterns in large datasets, providing an extremely detailed picture of student learning progress and identifying at-risk students on time. This environment activates tutors’ attention and improves their decisions on accurate feedbacks aiming at reducing dropouts and retaining students in the learning process.
This work uses datasets from four academic years and combines learning analytics methods, data mining techniques, sentiment analysis and NoSQL systems, as a large- data analytic framework. The assessment of specific learning analytics tools inside the Learning Management System indicates that tutors with a lack of learning analytics methodology can obtain a clear overview of students' engagement and progress in courses' contents, highlighting student’s weaknesses. The results of the sentiment analysis in forum texts underline that the emotion of students’ satisfaction was predominant, while the overall range of polarity and emotions were positive. This is even more evident for the most active students, where the overriding sentiment was positive.
Additionally, machine learning techniques reveal that sentiments’ indicators as models' variables improve model predictions on students' grades. The absence of tutors’ polarity and emotions' variables weakens the predictive ability of the models while the predictive error is further increased by the absence of students' emotions. Furthermore, it is noted that the models' forecasting capacity of the early weeks and periods of the academic year for students’ dropouts is satisfactory and an appropriate prediction tool was developed. It is also observed that the combination of a set of students’ actions in specific periods in the course and in the forum affects their progress. Moreover, the comparison of students’ actual grades and prediction grades, it can be an indicator of students' dishonesty.
A BIG DATA ANALYTICS FRAMEWORK TO SUPPORT ADAPTIVE AND PERSONALIZED LEARNING ENVIRONMENTS