An essential endeavor in natural language processing, sentiment analysis entails determining the sentiment expressed in a text. Transformer-based models like BERT have attained state-of-the-art performance in sentiment analysis tasks. However, these algorithms may have difficulty distinguishing sentiment-laden words. In response, we proposed combining lexicon-based sentiment analysis and transformer-based models. This study investigates the effect of lexicon-based sentiment analysis, particularly SentiStrength, on BERT's efficacy in sentiment analysis tasks. Experimental evaluations reveal that incorporating sentiment lexicons enhances the accuracy and F1-score of classical sentiment analysis compared to the baseline BERT model. Our findings demonstrate the value of incorporating external knowledge sources into transformer-based sentiment analysis models. © 2023 IEEE.