In this paper, identifying emotions from text is one of the salient works in the Natural Language Processing (NLP) domain, such as customer service, social media analysis, and mental health monitoring. Here, we present a BERT-based multi-label classification model, which was previously trained on the GoEmotions dataset to predict 28 emotion categories, ranging from basic emotions (such as joy and sadness) to more nuanced “less common” ones (e.g., nervousness and admiration). It is suggested that the architecture of the system consists of an embedding module, a dual channel module, an emotion categorization module, and an explainability module. The textual features from the input sentences are extracted as embedding vectors with the help of embedding module that uses the pre-trained Bidirectional Encoder Representations from Shannon transformers (BERT) model. We optimize the architecture of our model for GPU acceleration and fine-tune it to improve the performance of emotion detection. In addition, we compare to related models (GPT, XLNet), describe the mechanisms of our system and dataset integration, and report the results of our evaluation. Textual data emotion detections are a high-speed developing field, from sentiment analysis to improving user-computer interfaces. In this paper, we propose a comprehensive method of emotion detection using Bidirectional Encoder Representations from Transformers (BERT) for the multi-label classification of emotions in text. To achieve this, we leverage the GoEmotions dataset, which includes 27 emotion categories and a neutral class, and develop a model to identify different emotions written or spoken in natural language accurately.