Regresar

Sentiment Analysis of Peer Feedback with Deep Learning and Semantic Enrichment

Abstract:

The advancement of educational paradigms has driven the adoption of more integrated methods in the training and learning process, such as formative assessment. This form of assessment significantly contributes to the quality of learning that students obtain by giving and receiving feedback, and the immediate access that teachers can have to the progress of the class. Currently, higher education institutions seek to obtain the analysis of these unstructured texts through artificial intelligence. The objective of this study is to determine whether semantic enrichment through deep learning algorithms improves the performance of the task of classifying sentiment of textual feedback in Spanish. An experimental methodology was employed, using text mining techniques and natural language processing, along with Long Short-Term Memory (LSTM) and Bidirectional LSTM (BiLSTM) algorithms combined with pre-Trained Word2Vec/GloVe models for vocabulary formation. The results reveal that representations with semantic enrichment improve the performance of sentiment classification. The BiLSTM model with GloVe representation (F-Measure of 0.963) outperformed the baseline Support Vector Machines (F-Measure of 0.879) model with 1-g, 2-g, and Term Frequency-Inverse Document Frequency representations. It is concluded that semantic enrichment facilitated the resolution of terminological ambiguities, vocabulary expansion, and improvement in negation detection, among other relevant functionalities. As a continuation of this study, it is proposed to expand the peer assessment scenario with reverse evaluation, where the evaluated individual assesses the quality of evaluation of their peer evaluators, and two rounds are conducted to allow students to correct their tasks and improve their academic performance in the second round.