Source-linked AI summary
Sentiment analysis and opinion mining on educational data: A survey
Thanveer Shaik, Xiaohui Tao, Christopher Dann, Haoran Xie, Yan Li, Linda Galligan
TL;DR
Educational sentiment analysis must capture student opinions from feedback while reducing reliance on manual annotation. This review synthesizes analysis levels, annotation techniques, AI methodologies, applications, challenges, and future directions, concluding that sentiment analysis is examined for decision-making, pedagogical concepts, and educational procedures while challenges remain.
Problem
Manual sentiment labelling of large volumes of student feedback is time-consuming and resource-intensive, creating a need for approaches that reduce manual annotation.
Method
The review covers document-, sentence-, entity-, and aspect-level analysis, lexicon- and corpus-based annotation, AI methodologies, applications, challenges, and future directions.
Results
The review presents sentiment analysis as relevant to educational decision-making, pedagogical concepts, and educational procedures, and reports reviewed applications and methods.
Takeaways & Limitations
Sentiment analysis can be used to process student feedback and support evaluation, decision-making, and pedagogical practices within education.
Takeaways & Limitations
Challenges in sentiment analysis remain and require attention to improve existing student feedback practices.
Abstract
from arXiv · showhide
Sentiment analysis AKA opinion mining is one of the most widely used NLP applications to identify human intentions from their reviews. In the education sector, opinion mining is used to listen to student opinions and enhance their learning-teaching practices pedagogically. With advancements in sentiment annotation techniques and AI methodologies, student comments can be labelled with their sentiment orientation without much human intervention. In this review article, (1) we consider the role of emotional analysis in education from four levels: document level, sentence level, entity level, and aspect level, (2) sentiment annotation techniques including lexicon-based and corpus-based approaches for unsupervised annotations are explored, (3) the role of AI in sentiment analysis with methodologies like machine learning, deep learning, and transformers are discussed, (4) the impact of sentiment analysis on educational procedures to enhance pedagogy, decision-making, and evaluation are presented. Educational institutions have been widely invested to build sentiment analysis tools and process their student feedback to draw their opinions and insights. Applications built on sentiment analysis of student feedback are reviewed in this study. Challenges in sentiment analysis like multi-polarity, polysemous, negation words, and opinion spam detection are explored and their trends in the research space are discussed. The future directions of sentiment analysis in education are discussed.
1. Introduction
The introduction frames sentiment analysis as a way to extract educational stakeholders’ opinions from student feedback and support learning, teaching, evaluation, decision-making, and pedagogy. It reviews analysis levels, annotation approaches, AI methodologies, applications, challenges, and future directions.
- Motivation: Student feedback combines qualitative comments with quantitative ratings and demographics, enabling analysis of courses, teaching, learning-management systems, and pedagogical procedures.Quantitative data provides statistical understanding, while qualitative analysis captures students’ intent and opinions.
- Sentiment analysis in education: Sentiment analysis labels comments as positive, negative, or neutral, with analysis ranging from document-level overviews to fine-grained sentence, entity, and aspect analysis.Aspect-level analysis can identify which parts of course delivery receive negative reviews for improvement.
- Annotation: Manual sentiment annotation is time-consuming, resource-intensive, and requires pedagogical understanding, motivating lexicon- and corpus-based unsupervised approaches.These approaches provide an initial understanding of student feedback and address manual labelling to a certain extent.
- AI methodologies: AI methodologies, including machine learning, deep learning, and transformers, process large volumes of student comments and classify or predict emotions in unlabelled text.Transformer-based methods use attention mechanisms to learn student opinions.
- Review scope: The review examines sentiment analysis across document, sentence, entity, and aspect levels, along with annotation trends, educational applications, adoption challenges, and future directions.Its research questions address the role of opinion mining, analysis levels, annotation trends, and challenges in education.
2. Sentiment Analysis in Education
Sentiment analysis in education extracts student opinions at document, sentence, entity, and aspect levels. These levels support applications ranging from course evaluation and engagement analysis to fine-grained opinions about tutors, courses, and assignments.
- Analysis levels: Four analysis levels are used: document, sentence, entity, and aspect.Document-level analysis determines overall course sentiment; sentence-level analysis aggregates positive and negative content; entity-level analysis links sentiment to educational entities; aspect-level analysis provides fine-grained category analysis.
- Entity level: Entity-level analysis combines entity extraction and sentiment analysis to identify opinions about teachers, courses, assignments, content, lessons, and curricula.Named Entity Recognition supports identifying entities such as people, companies, or locations, while SentiSW represents results as sentiment-entity tuples.
- Educational tools: Document- and sentence-level tools have been developed for educational forum posts and compared with commercial sentiment-analysis tools.In one course, educational-domain tools outperformed commercial tools at both document and sentence levels.
3. Sentiment Annotation techniques in Education
Educational sentiment annotation spans lexicon-based, corpus-based, machine-learning, deep-learning, and transformer approaches. These methods reduce reliance on manual labelling, while domain understanding remains a central challenge.
- Annotation foundations: Sentiment annotation labels documents, sentences, or phrases with semantic emotions such as positive, negative, or neutral.Educational feedback can also receive fine-grained emotional labels and polarity scores.
- Annotation foundations: Manual annotation becomes impractical for large student populations because it requires substantial time and resources.The review connects this constraint to the development of automated lexicon, corpus, machine-learning, deep-learning, and transformer tools.
- Lexicon-based approaches: Lexicon-based methods assign sentiment orientation to words or phrases using dictionaries, polarity values, sentiment classes, or fine-grained scales.They support unsupervised labelling but can ignore words outside the dictionary and struggle with domain or context understanding.
- Corpus-based approaches: Corpus-based methods use co-occurrence statistics, syntactic patterns, and semantic distance to adapt general sentiment lexicons to specific domains.The approach estimates a target word’s polarity from its relation to positive and negative words in text corpora.
4. Sentiment Analysis Impact on Education
Sentiment analysis is applied to educational feedback to examine engagement, pedagogy, infrastructure, course and teacher evaluation, and policy decision-making. Reviewed studies report predictive and classification results across these applications.
- Educational applications: Educational sentiment analysis is used to study student engagement, pedagogy, infrastructure limitations, course evaluation, and educational-policy decisions.Applications draw on course feedback and participation in online forums.
- Educational applications: Sentiment and opinion mining can support personalized learning by identifying students’ issues and aspect- or entity-level opinions.The review also describes sentiment extraction as a way to inform student-performance prediction and educational support.
- Evaluation and decision support: Ensembled classifiers for student comments achieved 90.32% accuracy, 93.80% F1-score, and 90.86% recall.The study combined individual classifier outputs for feedback analysis.
- Evaluation and decision support: Faculty-effectiveness prediction combined qualitative polarity scores with quantitative feedback, and random forest achieved 98.87% accuracy and 97.71% precision.The study used ten machine-learning algorithms to predict faculty effectiveness.
- MOOC analysis: In MOOCs, LSTM with GloVe word embeddings achieved the highest reported classification accuracy of 95.80% across evaluated models.The comparison included conventional supervised, ensemble, and deep-learning methods on 66,000 student reviews.
- Engagement analysis: Sentiment analysis has also been combined with topic modelling and similarity analysis to investigate learner discussions in online forums.Latent Dirichlet Allocation was used to extract topics from forum content.
- MOOC analysis: Aspect and sentiment analysis distinguishes knowledge-seeking from skill-seeking courses and identifies topic-specific positive and negative opinions.Reported positive aspects included learning, labs, job or career, and grades or tests; negative aspects included instructors, content, courses, fees, and teaching.
5. Challenges
Educational sentiment analysis faces challenges in handling negation, opinion spam, multi-polarity, and polysemous words. The review discusses annotation and AI-based approaches proposed to address these issues.
- 5.1. Negation Handling: Removing negation words such as “never” and “not” during preprocessing can invert the intended sentiment.A negative statement may therefore be processed as inverse sentiment when negation cues are discarded.
- 5.1. Negation Handling: Cue-and-scope modelling and recurrent neural networks were used to identify polarity shifts caused by negation.A BiLSTM approach outperformed SVM, Hidden Markov Model, and CRF baselines with an F-Score of 93.34%.
- 5.2. Opinion Spam Detection: Opinion spam detection is critical because students may submit fake feedback, while supervised approaches face manual labelling, imbalance, and computational costs.RF, SVM, and Gradient Boosting classifiers, together with SMOTE and ADASYN, were used to classify and rebalance spam datasets.
- 5.2. Opinion Spam Detection: Deep learning methods for opinion spam detection, including CNNs with GloVe embeddings, outperformed traditional machine learning methods.The comparison included SVM, KNN, Logistic Regression, Naive Bayes, MLP, and LSTM methods.
- 5.3. Multi-Polarity: Student feedback can contain positive and negative opinions about different course aspects, making sentiment annotation ambiguous or misleading.This multi-polarity challenge requires models that can represent contrasting emotions within the same feedback.
- 5.4. Polysemy: Polysemous words complicate sentiment analysis because their meanings vary by context, and estimating the number of meanings is difficult.Proposed solutions include adaptive cross-contextual embeddings, topic-based word vectors, LDA, WSD, and BERT embeddings.
6. Future Directions of SA in Education
Future research should develop education-based sentiment annotation and knowledge-oriented methods tailored to educational language and multimodal feedback. These directions aim to address ambiguity, labelling challenges, and educational data-mining problems.
- 6. Future Directions of SA in Education: Future work should focus on education-based sentiment annotation techniques as current research increasingly adopts AI methods.The review identifies this as a central future direction for sentiment analysis in education.
- 6. Future Directions of SA in Education: Education-based sentiment annotation is proposed to address semantic differences between student feedback and language used in general unsupervised techniques.The review links this direction to lexical ambiguity, data labelling challenges, and educational context.
- 6. Future Directions of SA in Education: Building an education-specific annotation corpus requires manually labelling educational documents in a dictionary-based approach.The resulting annotations could support reinforcement learning for goal-oriented learning tasks.
- 6. Future Directions of SA in Education: Educational knowledge bases could connect relationships among multimodal student-feedback data through topic ontologies.Knowledge-based systems are proposed to enhance educational data mining and address sarcasm, negation handling, and ambiguity.
- 6. Future Directions of SA in Education: Fine-tuning aspect-based and entity-level sentiment analysis requires annotating the educational entities and aspects that need emphasis.This direction targets more domain-specific analysis of educational feedback.
7. Conclusions
The review covers sentiment analysis across offline and online education, including fine-grained analysis, unsupervised annotation, and AI methods. It concludes that educational applications remain useful while challenges in opinion mining require further attention.
- 7. Conclusions: The review examines sentiment analysis for evaluating learning and teaching practices through student opinions.Its educational impacts include decision-making and pedagogical procedures.
- 7. Conclusions: Sentiment analysis ranges from detecting orientation toward course content or teaching to fine-grained topic-level analysis using topic modelling.The review distinguishes simple sentiment detection from more detailed analysis of student opinions.
- 7. Conclusions: Machine learning, deep learning, and transformer methods are used to analyse, predict, and classify opinions from large student groups.The review presents these AI methodologies as necessary for analysing large volumes of student opinions.
- 7. Conclusions: The review covers studies from both offline and online education, including aspect-level sentiment analysis and unsupervised dictionary- and corpus-based annotation.It surveys research across different educational modes and analysis granularity.
- 7. Conclusions: Challenges in educational opinion mining remain and require attention to improve existing student-feedback sentiment analysis.The conclusion identifies these challenges as an unresolved area for continued work.