Aspect2Labels: A novelistic decision support system for higher educational institutions by using multi-layer topic modelling approach. (15th December 2022)

• Record Type: Journal Article
• Authors: Hussain, Shabir; Ayoub, Muhammad; Jilani, Ghulam; Yu, Yang; Khan, Akmal; Wahid, Junaid Abdul; Butt, Muhammad Farhan Ali; Yang, Guangqin; Moller, Dietmar P.F.; Weiyan, Hou
• Is Part Of: Expert systems with applications. Volume 209(2022)
• Publication Date: 2022-12-15
• Subjects: Aspect-based sentiment analysis -- Machine learning applications in education -- Aspect extraction -- Topic modelling -- Opinion mining -- Situational awareness; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2022.118119
• Languages: English
• ISSNs: 0957-4174

Abstract

Abstract: Aspect-based sentiment analysis (ABSA) has gained a rising concentration recently. It aims to provide a set of aspect terms and sentiments from a piece of text. Educational Data Mining (EDM) is now an essential tool for analysing pedagogical data. In academic institutions, student feedback is an influential gauge to measure the quality of the teaching–learning process. It helps higher education institutions to reconsider and improve their policies for student recruitment and retention. This paper proposed a situation awareness multi-layer topic modelling and enhanced hybrid machine learning approach for evaluating students' textual feedback data in academic institutions. The proposed Aspect2Labels (A2L) approach is divided our system into three layers. To preserve semantic information, we extracted general aspects terms in the first layer known as high-level aspects. We pulled low-level aspects terms associated with high-level aspect terms in the second layer and the third layer used for sentiment orientation. We used zero-shot learning, LDA, and different variants of LDA for the aspect extraction process. We performed annotation on unlabelled students' comments using our proposed A2L approach, and we obtained 91.3% accuracy in this process. We developed and tested novel algorithms for aspect terms mapping to label each aspect term to corresponding feedback. Different machine learning algorithms have been used to classify sentiments according to extracted aspects. We have also proposed and used Variable Global Feature Selection Scheme (VGFSS) and Variable Stopwords Filtering (VSF) to improve the performance of classifiers. We have managed to get 97% and 93% accuracy on the test dataset using Support Vector Machine (SVM) and Artificial Neural Networks (ANN), respectively. We highly suggest that our novel approach of aspect-oriented sentiment analysis could provide adequate understanding to analyse students' feedback. Highlights: Educational data mining is performed for higher educational institutions. Aspect2Labels, a multi-layer topic modelling framework to perform annotation. Proposed three rule-based algorithms for aspect extraction, validation, and mapping. Two data filtering techniques proposed to resolve sparsity and high dimensionality.