Sentiment analysis and opinion mining on educational data: A survey

Thanveer Shaik, Xiaohui Tao, Christopher Dann, Haoran Xie, Yan Li, Linda Galligan

Abstract

Sentiment analysis, also known as opinion mining, has emerged as a pivotal Natural Language Processing (NLP) application for identifying human intentions within reviews. In the education sector, these techniques are increasingly employed to capture student feedback and enhance pedagogical practices. This survey explores the role of emotional analysis across four levels—document, sentence, entity, and aspect—while evaluating the progression from lexicon-based approaches to advanced AI methodologies, including deep learning and transformers. By reviewing the impact of these tools on decision-making and evaluation, this paper identifies current challenges such as multi-polarity and opinion spam, while outlining future directions for the field.

Introduction

The digital transformation of education, accelerated by the proliferation of Massive Open Online Courses (MOOCs) and Learning Management Systems (LMS), has generated an unprecedented volume of qualitative data. While quantitative metrics like star ratings provide a surface-level view of student satisfaction, they often fail to capture the nuanced "why" behind the numbers. Sentiment Analysis (SA) offers a solution by automating the extraction of emotional orientation from text, allowing institutions to listen to the "voice of the student" at scale and transform raw feedback into actionable pedagogical insights.

Levels of Sentiment Analysis

Research in educational sentiment analysis is typically categorized into four hierarchical levels of granularity:

  1. Document Level: This approach evaluates the overall sentiment of an entire review or post, providing a general "vibe" of the student's experience.
  2. Sentence Level: This level determines the polarity of individual sentences, accounting for the fact that a single review may contain a mix of positive and negative sentiments.
  3. Entity Level: Analysis focuses on sentiment directed toward specific entities, such as a particular instructor, a specific software tool, or a textbook.
  4. Aspect-Based Sentiment Analysis (ABSA): Widely considered the "gold standard" for institutional improvement, ABSA identifies sentiment toward specific attributes, such as "course content," "lab facilities," or "teaching style." This allows administrators to pinpoint exactly which areas of a course require intervention.

Methodological Evolution

The technical landscape of SA in education has evolved through several distinct phases, moving from manual rule-based systems to sophisticated neural architectures.

Unsupervised and Lexicon-Based Approaches

Early methods relied on predefined dictionaries (e.g., SentiWordNet, VADER) to calculate sentiment scores based on word frequency. While computationally efficient, these methods often struggle with domain-specific jargon. For instance, a student describing a course as "challenging" may be offering praise in an advanced physics context but expressing frustration in an introductory elective.

Supervised Machine Learning

Traditional algorithms such as Support Vector Machines (SVM), Naive Bayes, and Random Forest remain reliable benchmarks. Studies indicate that SVM and Naive Bayes often achieve 72% to 81% accuracy for document-level classification. Notably, Random Forest has demonstrated exceptional performance in specific aspect identification tasks, with F1 scores reaching as high as 98%.

Deep Learning and Transformers

The advent of Deep Learning introduced Convolutional Neural Networks (CNNs) for feature extraction and Long Short-Term Memory (LSTM) networks for capturing the sequential nature of language. These models consistently outperform traditional SVMs in predicting student outcomes, such as the likelihood of course failure based on forum participation.

Currently, Transformers—specifically BERT (Bidirectional Encoder Representations from Transformers)—represent the state-of-the-art. By utilizing attention mechanisms to understand context, even "shallow" BERT-CNN architectures (using only 6 layers) have achieved 92.8% accuracy, balancing high performance with computational efficiency.

Data Sources and Linguistic Challenges

The survey highlights a diverse range of data sources, including MOOC platforms like Coursera and Udemy, education-related social media threads, and internal LMS logs from platforms like Moodle. Despite the wealth of data, several linguistic hurdles remain:

  • Negation and Polarity Shifts: Words like "not," "never," or "hardly" can completely invert the sentiment of a sentence. BiLSTM models have proven particularly effective at identifying these shifts.
  • Domain Specificity: General-purpose commercial tools (e.g., IBM Watson or Azure Text Analytics) often fall short in educational settings. Fine-tuning models on specialized educational corpora is essential for capturing the nuances of classroom discourse.
  • Opinion Spam: Similar to product reviews, educational data is susceptible to fake reviews or automated spam, necessitating the development of robust detection algorithms.

Future Directions

The field is moving toward more complex, multimodal analysis. Future research aims to integrate text with video (facial expressions during lectures) and audio (tone of voice in discussions) to provide a holistic view of the student experience. Furthermore, the integration of Reinforcement Learning (RL) suggests a path toward "pedagogical conversational agents" capable of adjusting their dialogue in real-time based on the student's emotional state.

Conclusion

Sentiment analysis is no longer merely a research interest; it is becoming a core component of educational quality assurance. By moving from simple polarity detection to nuanced, aspect-based insights, AI enables institutions to make data-driven decisions that directly improve the learning-teaching environment.

Summary updated

Abstract

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.
BibTeX
@article{shaik2023sentiment,
  author = {Thanveer Shaik and Xiaohui Tao and Christopher Dann and Haoran Xie and Yan Li and Linda Galligan},
  title = {{Sentiment analysis and opinion mining on educational data: A survey}},
  journal = {Nat. Lang. Process. J.},
  year = {2023},
  doi = {10.1016/j.nlp.2022.100003},
}
Figure