sentiment-thai-text-model

This model is a fine-tuned version of poom-sci/WangchanBERTa-finetuned-sentiment on an pythainlp/wisesight_sentiment.

Model description

This model is a fine-tuned version of poom-sci/WangchanBERTa-finetuned-sentiment, specifically tailored for sentiment analysis on Thai-language texts. The fine-tuning was performed to improve performance on a custom Thai dataset for sentiment classification. The model is based on WangchanBERTa, a powerful transformer-based language model developed for Thai by the National Electronics and Computer Technology Center (NECTEC) in Thailand.

Intended uses & limitations

This model is designed to perform sentiment analysis, categorizing input text into three classes: positive, neutral, and negative. It can be used in a variety of natural language processing (NLP) applications such as:

Social media sentiment analysis Product or service reviews sentiment classification Customer feedback processing

Limitations: Language: The model is specialized for Thai text and may not perform well with other languages. Generalization: The model's performance depends on the quality and diversity of the dataset used for fine-tuning. It may not generalize well to domains that differ significantly from the training data. Ambiguity: Handling of highly ambiguous or sarcastic sentences may still be challenging.

Training and evaluation data

The model was fine-tuned on a sentiment classification dataset composed of Thai-language text. The dataset includes sentences and texts from multiple domains, such as social media, product reviews, and general user feedback, labeled into three categories:

Positive: Indicates that the text expresses positive sentiment. Neutral: Indicates that the text is neutral or objective in sentiment. Negative: Indicates that the text expresses negative sentiment. More details on the dataset used can be provided upon request.

Training procedure

The model was trained using the following hyperparameters:

Learning rate: 2e-05 Batch size: 32 for both training and evaluation Seed: 42 (for reproducibility) Optimizer: Adam (with betas=(0.9, 0.999) and epsilon=1e-08) Scheduler: Linear learning rate scheduler Number of epochs: 5 The training used a combination of cross-entropy loss for multi-class classification and early stopping based on evaluation metrics.

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 5

Framework versions

• Transformers 4.44.2
• Pytorch 2.4.1+cu121
• Datasets 3.0.1
• Tokenizers 0.19.1