Automatic sentiment labelling and classification of multimodal data

In the digital era, sentiment classification plays a pivotal role in understanding public opinion, monitoring brand perception, and identifying emergent trends from vast amounts of user-generated content on different social media platforms. Accurate sentiment classification is difficult for businesses to evaluate customer feedback, for policymakers to gauge public sentiment, and for researchers to analyze emotional responses to significant events such as natural disasters, political developments, economic changes, and public health crises. However, the multimodal data shared in social media, which often includes text, images, and combinations, poses different challenges. These challenges include the complexity of obtaining sentiment labels, such as positive, negative, or neutral, for the vast datasets used in supervised machine learning and deep learning algorithms. Another challenge is to effectively extract features related to sentiment cues across diverse modalities and combine them for sentiment classification tasks. This research demonstrates the potential of automatic labelling for multimodal data as an efficient and useful alternative to human labelling. This research also examines the use of textual representations of images and multimodal data that combine image and text as an alternative approach for sentiment classification, showing improved performance when using existing sentiment classification machine learning models. This thesis makes a significant contribution to multimodal sentiment classification by providing deep insights into automatic labelling, textual and visual feature extraction, integration, and feature representations across diverse modalities for sentiment classification. This research demonstrates a new avenue for efficient, scalable, and useful solutions for sentiment classification in the era of big data for realworld needs.

Publisher

University of Galway

Rights

CC BY-NC-ND

cbn

Collections

University of Galway Theses (PhD Theses)