The number of online reviews has increased exponentially with the growing use of the Web and social media platforms. To handle the large amount of reviews, machine learning methods are needed to analyze these opinions properly. Aspect-Based Sentiment Classification (ABSC) attempts to determine the sentiment of each aspect within a text, to give a more in-depth overview of the opinion. In this work, a new model called LCR-Rot-hop-Kfont++ is introduced. This model injects knowledge into the state-of-the-art LCR-Rot-hop++ model inspired by previous work on Kformer. We evaluate the model using the SemEval-2015 and SemEval-2016 datasets, for which knowledge is injected during training, testing, and both training and testing using a domain-specific ontology. We show that injecting knowledge improves the accuracy of ABSC. For smaller datasets, it is advised to inject knowledge at training time, while for larger datasets it is advised to inject knowledge at testing time.