Integrated deep learning paradigm for document-based sentiment analysis

•The deep learning uses the BERT pre-trained vector representation model. The DBSA also used a multi-layer convolutional neural network with varying kernel sizes for feature extraction.•Therefore, the DBSA model adopts the versatility of bidirectional encoder representation from Transformers to improve accuracy. The model further utilized a multi-layer convolutional neural network with various kernel sizes to enhance feature extraction.•To further improve the accuracy of B-MLCNN, we extensively experimented on how max length, batch size, learning rate, and epoch size affected the proposed model.•We used IMDB, Movie Reviews (2002), Movie Reviews (2004), and Amazon reviews to compare the BERT, CNN, BERT-CNN, and B- MLCNN models.•We compared the model to other state-of-the-art models.