Finetuned DistilBERT model card
This model is a fine-tuned version of distilbert-base-uncased post trained on a huge comments corpus for sentiment analysis.
Model description
The model is based on DistilBERT, a distilled version of BERT that is smaller, faster, and requires less computational resources while maintaining 97% of BERT's performance. This specific model has been fine-tuned to perform sentiment analysis on YouTube comments, classifying them into three categories: POSITIVE, NEUTRAL, and NEGATIVE.
The model is designed to understand the nuanced language used in social media comments, including slang, emojis, and informal speech patterns typically found in YouTube comment sections.
Model architecture
- Base model: DistilBERT (distilbert-base-uncased)
- Task type: Text Classification (Sentiment Analysis)
- Number of labels: 3 (POSITIVE, NEUTRAL, NEGATIVE)
- Label mapping: {"NEUTRAL": 0, "POSITIVE": 1, "NEGATIVE": 2}
Training and evaluation data
The model was trained on a custom dataset of YouTube comments with sentiment labels:
- Dataset: im-tsr/comments-sentiments
- Training set size: 7,20,977 samples
- Test set size: 36,558 samples
- Maximum sequence length: 128 tokens
Training procedure
The model was fine-tuned using the Hugging Face Transformers library with Pytorch.
Training hyperparameters
The following hyperparameters were used during training:
- Learning rate: 5e-05
- Train batch size: 64
- Eval batch size: 64
- Weight decay: 0.01
- Optimizer: AdamW
- Number of epochs: 3
- Maximum sequence length: 128
Evaluation results
The model was evaluated on a test set of YouTube comments. The evaluation metrics are:
- Accuracy: 0.730100
- F1 Score: 0.730025
Using Hugging Face Transformers
import torch
import transformers
from transformers import pipeline
def get_sentiment(text):
#Set to avoid warning messages.
transformers.logging.set_verbosity_error()
sentiment_classifier = pipeline(task="text-classification",
model="iam-tsr/distilbert-finetuned-sentiment-analysis",
tokenizer="iam-tsr/distilbert-finetuned-sentiment-analysis",)
return sentiment_classifier(text)
get_sentiment("I love programming!")
#[{'label': 'POSITIVE', 'score': 0.9759756326675415}]
Intended uses
This model is intended for analyzing sentiment in YouTube comments and similar social media text. It can be used for:
- Monitoring sentiment in YouTube comment sections
- Content moderation assistance
- Social media sentiment analysis
- User feedback analysis
- Tracking audience reaction to videos or content
Limitations
- The model is specifically trained on YouTube comments, which may have different characteristics from other text sources
- Performance may vary for comments in languages other than English
- The model may not handle sarcasm, irony, or cultural context effectively
- Limited to three sentiment categories which may not capture the full spectrum of emotions
- May not perform optimally on very short texts or texts with multiple conflicting sentiments
Links
Framework versions
- Transformers 4.53
- Pytorch 2.8.0
- Datasets 3.6.0