Chan, JohnnySundaram, DavidPeko, GabrielleLi, Yuming2024-12-162024-12-162024https://hdl.handle.net/2292/70852The surge in the global influence of social media and the complexities of digital communication have underscored the significance of sentiment and emotion analysis within the information systems field. This thesis addresses the nuanced challenges of extending sentiments and emotions across the dynamic landscape of social media, where traditional methods falter against the backdrop of growing data volume and complexity. It identifies five primary research challenges: conducting sentiment and emotion analysis in multilingual contexts; identifying finer-grained and mixed emotions; ensuring analysis efficiency in low-resource settings; personalising analysis for different user groups; and explaining outcomes from deep learning-based models. In pursuit of addressing these challenges, the thesis unfolds across five published papers. The first paper, published in PACIS (2021), introduces a multilingual sentiment analysis framework that leverages shared multilingual word embeddings and pre-trained language models, aiming to standardise sentiment analysis across languages. The second paper, published in Data & Knowledge Engineering (2023), proposes an advanced approach for fine-grained and mixed emotion analysis using a graph-structured storage for improved visualisation and categorisation. The third paper, presented at HICSS (2024), introduces a framework for hateful emotion recognition using knowledge distillation and data augmentation tailored for low-resource scenarios. The fourth paper, presented at AMCIS (2022), develops a system for personalised analysis of harmful emotions based on user profiles. The fifth paper, published in Decision Support Systems (2024), addresses the challenge of interpreting the outputs of deep learning models in sentiment and emotion analysis, offering an explainability framework grounded in emotion theory. In terms of its overarching methodological approach, this thesis is firmly anchored in the design science research methodology. Through the iterative cycles of design, implementation, and evaluation intrinsic to design science research, this thesis not only bridges theoretical gaps but also produces empirically validated artefacts that enhance the capabilities of sentiment and emotion analysis in handling multilingual texts, fine-grained emotions, resource constraints, personalisation, and explainability. The advancements contribute significantly to the academic realm and offer cutting-edge solutions for industry practices, including empathetic AI systems and enriched decision-making processes, marking a crucial evolution towards globally integrated, human-centric social media analysis.https://researchspace.auckland.ac.nz/docs/uoa-docs/rights.htmThesisCopyright: The author