In the last decade, large numbers of comment texts have been generated on social media and websites. In the era of sentiment analysis, mining the role of emotional tendency in comments through deep learning technology is helpful for the timely classification of sentiment text as positive, negative, and neutral. Sentiment analysis is a task that predicts people's opinions on product reviews based on text data, and it's both a valuable and challenging task. This research study has utilized a novel deep learning based predictive framework, which is applied in analyzing the product reviews along with user opinion information. Firstly, the training set generates character vectors as input layers by using Bidirectional Encoder Representation of Transformers (BERT) and FLAIR embedding models, which are used to convert the product review into low-dimensional representation; and then uses this vector as input to a novel hybrid Bidirectional Long-Short-term memory model (Bi-LS TM) and Bidirectional Gated recurrent unit model (Bi-GRU), which are combined into a single architecture to predict the feature. Finally, the processed context information is classified using the softmax classifier. The resultant review shows the significant accuracy of our model.
BERT-BiLSTM-BiGRU-CRF: Ensemble Multi Models Learning for Product Review Sentiment Analysis
2022-12-01
673646 byte
Conference paper
Electronic Resource
English
Information Extraction from Contract Based on BERT-BiLSTM-CRF
TIBKAT | 2021
|BiGRU based online multi-modal driving maneuvers and trajectory prediction
SAGE Publications | 2021
|Springer Verlag | 2025
|