Leveraging NLP and Machine Learning to Explore Student Experiences During and After the COVID-19 Pandemic

Abstract

The pervasive role of education within societal frameworks renders its disruption a matter of profound consequence for individuals. This thesis investigates the impact of the COVID-19 pandemic on university students, focusing on well-being, remote learning, and the use of social media. The research was motivated by the need to understand not only the immediate effects of the pandemic but also its longer-term consequences for students, which have received less attention in existing literature. Previous studies have tended to emphasise either quantitative surveys or social media data in isolation, and few have systematically compared computational approaches for analysing open-ended student feedback.
The study aimed to address these gaps by pursuing three objectives: (i) to examine lasting effects of the pandemic on students’ well-being and remote learning experiences across demographic groups; (ii) to compare traditional and transformer-based topic modelling methods for analysing qualitative feedback; and (iii) to explore the nature and themes of UK-based educational discussions on Twitter during the pandemic.
A multi-method approach was employed. The first study analysed large-scale survey data from higher education students to assess stress levels, coping strategies, learning satisfaction, and the influence of demographic variables. The second study used unsupervised topic modelling specifically Latent Dirichlet Allocation, Latent Semantic Analysis, Non-negative Matrix Factorisation, and the BERT-based BERTopic model to extract themes from open-ended survey responses about remote learning and mental health. The third study applied classification and topic modelling techniques to over one million UK-based tweets relating to education, filtering for relevant content and identifying key themes in public discourse.
The quantitative survey revealed that the pandemic had enduring effects on student well-being and learning, with notable differences by age, gender, and access to re-sources. Many students reported increased stress, ongoing challenges in adapting to post-pandemic learning environments, and mixed satisfaction with remote learning provision. Coping strategies varied, with social media playing a dual role as both a supportive outlet and a potential source of additional stress.
The topic modelling of open-ended feedback demonstrated that transformer-based approaches, particularly BERTopic, produced more coherent and nuanced themes than traditional models. These methods captured detailed issues relating to academic work-load, feelings of isolation, digital inequality, and preferred modes of study.
The analysis of educational tweets identified recurring themes including digital access, mental health concerns, policy changes, and community support. Comparisons between social media data and survey feedback highlighted overlaps in concerns about equity, resource provision, and the balance between in-person and online learning.
In conclusion, the findings underscore the lasting educational and psychological impacts of the pandemic, shaped by demographic and contextual factors. Social media emerges as a valuable source of real-time insight into student experiences, complementing formal feedback channels. Methodologically, the research evidences the benefits of advanced NLP models for analysing large volumes of unstructured educational data. The study offers practical implications for higher education policy and practice, including the need to strengthen digital infrastructure, enhance student support systems, and integrate diverse data sources into crisis preparedness planning.

Date of Award	13 Mar 2026
Original language	English
Awarding Institution	University of Portsmouth
Supervisor	Ella Haig (Supervisor) & Alaa Mohasseb (Supervisor)