Classification_for_StudentFeedback

Label to predict:

0 : Negative

1: Neutral

2: Positive

Fine-tuning PhoBERT for Vietnamese Student Feedback Analysis

In the realm of Natural Language Processing (NLP), the Vietnamese language poses its own set of challenges and intricacies. Fine-tuning language models tailored to Vietnamese, such as PhoBERT, has emerged as a pivotal endeavor in advancing NLP applications within the Vietnamese-speaking community. Here, we introduce a model fine-tuned on Vietnamese Student Feedback data, an essential domain in educational assessment and improvement efforts.

Model Overview PhoBERT: PhoBERT, short for "Pre-trained Pho Vietnamese BERT," is a transformer-based language model specifically pre-trained for the Vietnamese language. Leveraging the BERT architecture, PhoBERT captures contextual information and semantic nuances within Vietnamese text, enabling it to understand and generate Vietnamese content effectively. Specifically, I used sup-SimCSE-VietNamese-phobert-base (https://huggingface.co/VoVanPhuc/sup-SimCSE-VietNamese-phobert-base) to train the model and I have a good result

Training Details (https://huggingface.co/datasets/uitnlp/vietnamese_students_feedback) Dataset: The model is fine-tuned on a dataset consisting of Vietnamese student feedback, a corpus rich in linguistic diversity and educational insights. This dataset provides valuable feedback on various aspects of educational experiences, including teaching quality, course content, and overall satisfaction. The Dataset splits 3 main parts: Train(11.4k rows), valid (1.56k rows), test (3.17k rows) The Dataset includes 3 features: Sentences, Sentiment and Topic. I used 2 column names: Sentence and Sentiment to classify sentimnet.

Training Duration: The fine-tuning process spans 15 epochs, with each epoch iterating over the entire dataset. Despite the considerable depth of training, the model demonstrates efficiency, achieving promising results within a reasonable timeframe.

Hyperparameters:

Learning Rate: Set to 2e-5, the learning rate governs the step size in the optimization process during fine-tuning. A carefully chosen learning rate facilitates effective weight updates while preventing overshooting or stagnation. Batch Size: With a batch size of 64, the model processes 64 data samples in each training iteration. This batch size strikes a balance between computational efficiency and model stability, facilitating smooth convergence during training. Performance Loss: The fine-tuned model exhibits an impressive loss metric, averaging around 0.002 throughout the training process. This minimal loss signifies the model's ability to accurately predict student feedback sentiments and insights with high precision. Impact and Applications

Student Feedback Analysis: By accurately analyzing student feedback, educational institutions can identify areas of improvement, enhance teaching methodologies, and foster a more conducive learning environment. Educational Assessment: The model aids in automating the assessment of educational quality and effectiveness, providing educators and administrators with actionable insights to optimize educational practices. Natural Language Understanding: With its nuanced understanding of the Vietnamese language, the model serves as a cornerstone for developing advanced NLP applications catering to Vietnamese speakers, including chatbots, summarization tools, and sentiment analysis systems. In summary, the fine-tuned PhoBERT model represents a significant stride in leveraging advanced NLP techniques for educational enhancement and linguistic analysis within the Vietnamese-speaking community. With its robust performance and versatility, this model promises to revolutionize the landscape of educational assessment and linguistic research in Vietnam and beyond.