1 University of Minho (PORTUGAL)
2 Universidade Federal do ParĂ¡ (BRAZIL)
About this paper:
Appears in: ICERI2020 Proceedings
Publication year: 2020
Pages: 9041-9048
ISBN: 978-84-09-24232-0
ISSN: 2340-1095
doi: 10.21125/iceri.2020.2010
Conference name: 13th annual International Conference of Education, Research and Innovation
Dates: 9-10 November, 2020
Location: Online Conference
Students' opinions are pivotal instruments in improving teaching and learning processes. Their analysis, and the sentiments expressed in them allows for obtaining rich and varied information elements that help in understanding the viability of computational tools and their real contribution to the improvement of the teaching process and learning, in a concrete way. However, the analysis of opinions expressed in natural language is not a simple process. In order to make it possible, we need to apply several techniques and models of natural language processing, syntactic and semantic analysis of texts, artificial intelligence, and computational linguistics. The process of analysing opinions is further complicated when we intend to extract the sentiments expressed in the opinion texts. There are several approaches for extracting and analysing sentiments from a text written in natural language. For example, in order to recognize a given sentiment we need to understand the meaning of the different terms that one or more phrases contain according to a certain application context, in specific, or to the common reality, in general. In this paper, we present and demonstrate an opinion and sentiment analysis module that we developed for a specific computational platform for assessing student knowledge in specific teaching and learning domains. When developing this module we wanted to collect and analyse the students' opinions, expressed in natural language texts, with the objective of improving the performance and quality of the referred evaluation system, as well as adjusting its functionalities to the operational requirements expressed by students. Additionally, through the analysis of the expressed sentiments, we were also able to evaluate effectively several other aspects (operationally, clarity, interface, difficulty, content, etc.) that are directly related to student satisfaction, regarding the evaluation of the system. Thus, we will be able to know, with some certainty and foundation, if students are in fact using the system in a useful way.

The opinion and sentiment analysis process was implemented in 5 distinct stages, namely:
1) pre-processing, in which we remove textual tags, separate words, convert formats, and correct syntactic errors;
2) tokenization, in which we create a structure with words related to opinions and sentiments;
3) removal of stop words, in which we remove words considered irrelevant;
4) analysis of sentiments, in which we identify the polarity of words; and
5) result validation, where we analyse and validate the sentiments extracted in the texts.

This module is supported by a document store (NoSQL document database), where are stored all the texts that contain students' opinions, organized by time and by the type of analysis we pretend to apply. The texts are collected previously through a set of forms specifically designed for this purpose, which can be filled by students when using the system. We finish this paper by analysing the quality and usefulness of the results obtained, especially in terms of the sentiments expressed and their usefulness for understanding the type of use that students actually make of the system. In addition to the sentiment analysis module, a sentiment visualization dashboard was developed for allowing the system manager to view and track the level of user satisfaction.
Educational Environments, Evaluation Processes Customization, Artificial Intelligence, Natural Language Processing, Opinion and Sentiment Analysis.