The first report of the Omicron variant triggered a lot of emotions towards vaccination in South Africa. These emotions are mostly expressed on social media such as Twitter. The result could weaken the confidence level of users even before they are vaccinated. Identifying these emotions and how they change before and during the Omicron variant can help, in understanding the dynamics in citizens' behaviour, towards vaccination for health policy-making. In this study, 23,000 vaccine-related Twitter posts were collected in South Africa, from 1 October 2021 to 15 January 2022 using Natural Language Processing techniques. The emotional classification of Twitter posts and their associated intensities were achieved using the Text2emotion pre-trained model. The results were validated using Naive Bayes with an accuracy of 76%, Logistic Regression (91%), Support Vector Machines (84%), Decision Tree (82%), and K-Nearest Neighbours (72%). The number of tweets significantly positively correlated with the increase in vaccination across all South African provinces (Corr<=0.532, P<=0.003) except Northern Cape province. The emotional intensities for vaccine-related posts showed a strong association with the increase in vaccination during Omicron (P<0.04) in Eastern Cape, Gauteng, Limpopo, and North West provinces than other provinces. The comparison of the intensities of the emotional classes differed across provinces before and during Omicron. The result of this research showed that social media data can be used to complement existing data, in understanding and predicting the dynamics in citizens' emotional behaviour towards vaccination, during a new COVID-19 variant or future outbreaks. The result could also inform health policy in planning, control, and management of provinces identified with vaccine hesitancy. It can also serve as a template or reference for related future academic research.