sentiment-analyzer-v2-final

Model Card for pauL9990/sentiment-analyzer-v2-final

Model Details

• Developer: Porjanya Paul Bordoloi (@pauL9990)
• Base Model: distilbert-base-uncased
• Architecture: Transformer (DistilBERT) + LoRA (Low-Rank Adaptation)
• Language: English
• License: Apache 2.0
• Framework: PyTorch + Hugging Face Transformers + PEFT
• Tags: sentiment-analysis, transformers, LoRA, NLP, HuggingFace, fine-tuning

---

Model Description

sentiment-analyzer-v2-final is a LoRA fine-tuned version of DistilBERT for binary sentiment classification.
This model is trained on the IMDB movie review dataset and enhanced with real-world short-form and mixed sentiment examples, optimized for:

• Gen-Z and informal expressions
• Sarcasm, ambiguity, and mixed opinions
• Lightweight, deployable inference (adapter-only)

---

Intended Use

Direct Use

• Review analysis in apps and websites
• Customer feedback classification
• Comment moderation (social platforms)
• Short-form content sentiment detection

Out-of-Scope

• Multi-lingual sentiment analysis
• Domain-specific tone modeling (e.g., legal, medical)

---

How to Use

from transformers import AutoTokenizer, AutoModelForSequenceClassification
from peft import PeftModel
import torch

tokenizer = AutoTokenizer.from_pretrained("pauL9990/sentiment-analyzer-v2-final")
base_model = AutoModelForSequenceClassification.from_pretrained("distilbert-base-uncased", num_labels=2)
model = PeftModel.from_pretrained(base_model, "pauL9990/sentiment-analyzer-v2-final")
model.eval()

def analyze_sentiment(text):
    inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True, max_length=256)
    with torch.no_grad():
        outputs = model(**inputs)
    probs = torch.softmax(outputs.logits, dim=1)
    label = "Positive" if torch.argmax(probs) == 1 else "Negative"
    confidence = round(probs.max().item() * 100, 2)
    return label, confidence

Citation

@misc{bordoloi2025sentimentanalyzer, title={Sentiment Analyzer V2 - LoRA Fine-Tuned DistilBERT}, author={Paul Bordoloi}, year={2025}, howpublished={\url{https://huggingface.co/pauL9990/sentiment-analyzer-v2-final}}, }