Aspect-Based Sentiment Analysis (ABSA) goes beyond general sentiment classification by associating opinions with specific features or aspects of a subject. This paper presents “AspectSense,” a scalable, real-time ABSA framework that processes Twitter data to extract aspect-sentiment pairs with high accuracy and interpretability. Unlike traditional multi-class sentiment models that may introduce ambiguity, our approach improves clarity by using strict binary sentiment classification—labeling each aspect as either positive or negative. KeyBERT is used for unsupervised aspect extraction by identifying semantically relevant phrases using transformer-based embeddings, eliminating the need for predefined aspect lexicons. DistilBERT, a lightweight yet efficient variant of BERT, is used for sentiment classification, offering near state-of-the-art performance with reduced computational overhead. Trained on a 50,000 tweet subset of the Sentiment140 dataset, our model achieves 88.5% accuracy and generates over 145,000 accurate aspect-sentiment pairs. These structured outputs enable fine-grained insights essential for feedback analysis, brand monitoring, and opinion tracking. The modular architecture ensures domain adaptability and seamless integration into social media analytics pipelines. Evaluation metrics include accuracy, precision, recall, and F1-score, with performance exceeding LSTM and standard BERT baselines.