Detecting emotions online has increasingly become a critical concern for NLP researchers, particularly due to the proliferation of emotional expressions on social media and Web 2.0 platforms. Developing effective detection systems for languages with limited resources, such as Bengali, presents significant challenges. This paper introduces a novel dataset tailored specifically for contextual emotion detection in Bengali texts. The dataset creation process involved extensive data collection, preprocessing, human and automatic labeling, and label verification. This resulted in 20,247 annotated texts categorized into 27 different emotional categories. The dataset achieved a high Cohen’s score of 0.89, indicating strong agreement among annotators. We define context with multiple components: ‘WHO’, ‘WHEN’, ‘WHERE’, and ‘HOW’. Utilizing these context elements, we approximate a cognitive understanding of the posts, which facilitates emotion detection. We conducted comprehensive experiments using ML, DL, and BERT-based models to assess the dataset’s efficacy. Our findings underscore the pivotal role of context in emotion detection. Particularly noteworthy was the performance of the BERT-based model XLM-R, which achieved an impressive F1 score of 0.88 and accuracy of 0.85 when context information was utilized. These results highlight how incorporating context significantly enhances the accuracy of emotion detection systems. This research contributes to advancing robust methodologies for identifying and understanding emotional content effectively.