Developing Multilingual Capabilities for All- Inclusive Learner Feedback

Initially Feedback is collected either quantitatively or qualitatively but failed to satisfy users who desire to provide feedback verbally .Many users prefer expressing opinions verbally, especially in multilingual settings, which can present challenges for traditional textbased sentiment analysis systems. To address this, our project integrates Gradio, a user-friendly interface tool, to allow users to record their audio feedback from both Telugu and English to English text directly on the webpage. This audio spoken input converted into text, making it ready for further analysis. In order to perform further analysis we have developed several models for the analysis of the feedback's sentiment. They include BERT with BERT tokens, BERT combined with Logistic Regression, and BERT with LSTM and multinomial Navie Bayes. We measured all the above models' performances in terms of accuracy and reliability. From all tested models, the model BERT with Multinomial Naive Bayes (Multinomial NB) achieved the highest accuracy rate, 98%. This model yielded the best performance in sentiment analysis , making it our most reliable option for the task.