GRU-Powered Emotion Detection: Breaking New Ground in Artificial Intelligence
Anirudh Jindal, Kanwarpartap Singh Gill, Sonal Malhotra, Swati Devliyal · 2024
This study looks at how to understand emotions in language using a type of technology called bidirectional gated recurrent units (GRUs). Old models that only look at data in one direction are not as good as bidirectional GRUs because they can’t understand relationships in data from both before and after. For jobs like understanding emotions, where the situation matters a lot, being able to consider things from both sides is really important. A step-by-step plan is used to create the proposed model. An embedding layer is the first step that changes word numbers into dense vectors, which are needed to work with text data. A dropout layer with a rate of 0. 5 is added to avoid overfitting. Next, there are two GRU layers that can process information in both directions. They are set up to return sequences, which helps to understand the timing and order of events better. A batch normalization layer helps make training more stable and faster. An extra bidirectional GRU layer helps the model better understand complex relationships from both sides. It has six parts and uses a softmax function to sort feelings into six categories. The last layer is fully connected. The Adam optimizer is added to the model, and it uses sparse categorical cross-entropy to measure how well the model is performing. Accuracy is a way to see how well something performs. This design shows that bidirectional GRUs are good for mood analysis, working accurately $94 \%$ of the time. It does a better job at guessing feelings because the model can see complicated connections in ordered information, as the results show. The model summary shows the different parts and options for the settings that help make the results really good.