DICO-2 : An Improvement of the Contextualized Lexical Resource DICO for Educational Opinion Mining Using Machine Learning Classifiers

Opinion mining is currently considered one of the most active research fields. It aims to analyze people's feelings, which makes it an important element in the decision-making process. It is introduced in different fields, especially in the education one. The objective of this work is to improve a lexical resource called DICO, which we have presented in our previous works for educational opinion mining, through a recalculation of polarities, in order to improve the opinion detection process. This improvement is done by training machine learning classifiers with EDUCA, a corpus built from educational communities, including a set of manually annotated subjective comments. This approach is implemented using the WEKA platform which allows the generation of classification models. These models classify the terms of the lexical resource DICO and, the polarities attributed by these models are used for their recalculation to get a better lexical resource called DICO-2. This last is evaluated by comparing the classification of the EDUCA corpus using DICO with that one using DICO-2. The results obtained show a better classification of opinions with increase of the performance, especially the accuracy.