Understanding emotions in text is crucial for applications in various domains, including mental health, microeconomics, social media analysis, and human-computer interaction. However, most systems focus only on classifying emotions, overlooking their underlying causes and the intensity with which they are felt. We introduce EmoCIE, a 3-step framework that performs emotion classification, cause extraction, and intensity estimation. The model extends a transformer-based architecture with optimized thresholds and integrates sentence embeddings for semantic reasoning, as well as a psychological lexicon (VAD) for modeling emotion strength. Across three benchmark baselines, including GoEmotions, HatEmoTweet, and NeuEmot, our approach outperforms state-of-the-art models in both high-resource and low-resource emotion categories. Our findings reveal that combining cause and intensity signals enhances accuracy and interpretability. Additional analysis shows alignment between emotion strength and model confidence. This work provides (1) a comprehensive model linking emotional labels to their causes and strength, (2) a psychologically informed method for estimating emotional intensity, (3) a novel interpretability framework that connects confidence with emotional reasoning, and (4) empirical results validating the model’s improved performance and transparency over prior approaches. Our code is publicly available at https://github.com/ademola1802/EmoCIE .