Article Preview
Top1. Introduction
Virtual Learning Environments (VLE) have different models for online interaction, mainly synchronous and asynchronous tools. E-Learning courses allow personal compass and educational resources and services to a large number of students, but they have to comprise a high number of students’ dropouts (Queiroga et al., 2017). Discussion forums are among the most common interaction tools offered by VLE, mainly in Massive Online Open Course (MOOCs). The popularity of MOOCs includes accessibility to every person who has internet, scalability to handle any number of students with a wide diversity of needs and expectations, and flexibility they provide to learners to study according to their routine. However, students who seek to clarify concepts in online platforms may not get the attention they need, which may favor abandonment (Capuano & Caballé, 2020).
According to Márquez-Vera et al. (2016), dropping out of school comes from a long-term disengagement from school and classes. The student’s Motivational Profile is based on detecting characteristics that signal students’ demotivation and dropout possibility alerts teachers and tutors to the need for special attention. Therefore, being able to predict this behavior earlier could improve the students’ performances, as well as minimize their failures and disengagement (Neves et al., 2021). The students’ performances can improve given to the specific opportunities and guidance that they will receive in their trajectory, so educational data is needed to be extracted and analyzed.
Many researchers deal with the combination of intelligent techniques and data extraction from big datasets to develop adaptive e-Learning systems. In this context, Sentiment Analysis (SA) can assist in extracting human thoughts and perceptions from a large amount of data (Hemmatian & Sohrabi, 2019). “It is considered one of the research fields in text mining” (Alatrash et al., 2021). Machine Learning, Lexical, or Hybrid approaches are used in the SA context.
The Sentiment Analysis approach can be influenced by the data type, cost, speed, and method accuracy. According to Hemmatian & Sohrabi (2019), machine learning algorithms are fast and tend to have high accuracy; however, they need human involvement and have high costs due to the large training sets and the algorithms’ training time. On the other hand, lexical methods do not require human involvement, are dependent on the lexicon size, and have less accuracy but a low cost of implementation. Because of that, this work focuses on the lexical approach.
One of the biggest challenges of lexical approaches is to deal with semi or unstructured data, which requires Natural Language Processing (NLP) techniques (Hemmatian & Sohrabi, 2019). Furthermore, the dynamism of the language with the increased use of jargon, slang, symbols, images, and emoticons also makes it challenging to identify polarities and the generation of new lexicons. One way to improve the quality of lexical approaches is by using Computational Linguistics solutions, such as FrameNet Brazil (https://www.ufjf.br/framenetbr/), which deals with jargon and contexts.
Using only the students’ text sentiment is not enough to indicate if they are at risk of school dropout. It is also necessary to analyze the students’ Motivational Profile (MP), composed of their engagement, participation, assiduity, dedication, etc. In this way, the student´s text sentiment and his Motivational Profile define the student's Emotional State (ES), which can point to the possibility of dropout. Also, knowing the student's Emotional State (is helpful to expand the possibilities of adaptive e-learning (Alatrash et al., 2021; Faria et al., 2017). The solution proposed in this paper is the SASys (Sentiment Analyzes System) architecture, based on a polarized frame network to identify the student’s sentiment in texts, with the goal of increasing the assertiveness of detecting his Emotional State. Thus, the main research question is: Can an architecture, based on a lexical approach, using a polarized frame network, for Sentiment Analysis together with data from the student profile and access to a Virtual Learning Environment, identify the student Emotional State and the risk of school dropout?