This model is a version of the cardiffnlp/twitter-xlm-roberta-base-sentiment-multilingual model that has been fine-tuned on a custom dataset of YouTube comments. The fine-tuning process was designed to improve performance on sentiment analysis for YouTube comments, which often differ in tone, slang, and structure from other social media platforms. After fine-tuning, the model achieved an accuracy of 80.17%.

Intended Use

The model is designed for sentiment analysis of YouTube comments. It accepts a list of text inputs (comments) and returns a sentiment label for each comment:

Positive
Neutral
Negative

This model can be used in applications such as video recommendation systems, content analysis dashboards, and other data analysis tasks where understanding audience sentiment is important.

How to Use

The model can be used via an API endpoint or loaded locally using the Hugging Face Transformers library. For example, using Python:

python

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

model_name = "AmaanP314/youtube-xlm-roberta-base-sentiment-multilingual"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(model_name)

# Example input
comments = [
    "This video aged like honey.", # Positive
    "This video aged like milk.", # Negative
    "It was just okay." # Neutral
]

inputs = tokenizer(comments, return_tensors="pt", padding=True, truncation=True)
with torch.no_grad():
    outputs = model(**inputs)
predictions = torch.argmax(outputs.logits, dim=1)
label_mapping = {0: "Negative", 1: "Neutral", 2: "Positive"}
sentiments = [label_mapping[p.item()] for p in predictions]
print(sentiments)

How It Was Trained

Base Model: cardiffnlp/twitter-xlm-roberta-base-sentiment-multilingual
Dataset: A custom dataset consisting of over 1 million YouTube comments, each annotated with one of three sentiment labels (Positive, Neutral, Negative).
Fine-Tuning Process: The model was fine-tuned using the following steps:
- Data Cleaning and Preprocessing
- Model Fine-Tuning
- Evaluation and Testing

Evaluation

The model was evaluated on a held-out test set of YouTube comments. It improved from a baseline accuracy of approximately 69.3% (when fine-tuned on Twitter data) to 80.17% on this dataset. This improvement demonstrates the benefit of domain-specific fine-tuning.

Citation

If you use this model in your research, please cite the original base model and this project:

@misc{cardiffnlp,
  title={Twitter-XLM-RoBERTa-Base-Sentiment-Multilingual},
  author={Cardiff NLP},
  year={2020},
  publisher={Hugging Face},
  howpublished={\url{https://huggingface.co/cardiffnlp/twitter-xlm-roberta-base-sentiment-multilingual}}
}

@misc{AmaanP314,
  title={Youtube-XLM-RoBERTa-Base-Sentiment-Multilingual},
  author={Amaan Poonawala},
  year={2025},
  howpublished={\url{https://huggingface.co/AmaanP314/youtube-xlm-roberta-base-sentiment-multilingual}}
}