Text-based reviews are extremely valuable for companies and institutions, but also hard to process if there are many. Aspect-Based Sentiment Classification (ABSC) requires techniques for classifying sentiments in reviews towards different aspects. Neural network models turn out to be of great use in this domain. In this paper, we further explore the possibilities of knowledge injection into these models. Our research focuses on knowledge injection in the state-of-the-art LCR-Rot-hop-ont++ model. We explore the possibilities of injecting knowledge at training time, test time, and training and test times using various SemEval datasets. Based on statistically significant results, we conclude that knowledge injection during both the training and testing phases significantly enhances performance across the restaurant and laptop domains for larger datasets, surpassing the LCR-Rot-hop++ model. Additionally, our results demonstrate that injecting knowledge at the testing phase in the case of the laptop domain improves model performance even further.