This paper presents a novel approach to improve test case prioritisation in continuous integration environments by integrating semantic features derived from test case descriptions. The approach uses a Convolutional Neural Network (CNN) model that considers both structured and semantic features. The target variable for prioritisation is defined using a threshold on calculated priority values, helping to identify the most critical test cases. We further engineer interaction and polynomial terms with the semantic feature to capture complex relationships. These polynomial terms allows the model to capture non-linear relationships. For example, a small change in the semantic feature might not matter much in the middle range, but a large implicit_prob (e.g., above 0.9) might have a disproportionately strong effect on the priority. Through a fivefold cross-validation, our results demonstrate that while the semantic features do not statistically significantly improve overall F1-score, they yield a statistically significant enhancement in Average Percentage of Fault Detection (APFD) (p-value = 0.0055). This indicates that incorporating semantic understanding enables the model to effectively reorder test cases, leading to earlier detection of critical faults and increased efficiency in testing efforts within live CI/CD environments.