NSFW Content Classifier

Model Description

This model is a fine-tuned text classification model for detecting NSFW (Not Safe For Work) content in text messages. It is based on a transformer architecture and has been trained to distinguish between explicit sexual content (NSFW) and safe content (SFW).

Developed by: MyPlaygirl AI Team
Model type: Text Classification
Language: English
License: MIT

Labels

• 0: NSFW (Not Safe For Work)
• 1: SFW (Safe For Work)

Performance Metrics

Evaluated on 100 test samples with 90% confidence threshold:

Overall Performance

• Accuracy: 1.0000
• Average Confidence: 0.9924
• High Confidence Predictions (≥90%): 98/100 (98.0%)
• High Confidence Accuracy: 1.0000

Detailed Metrics

Confusion Matrix

                 Predicted
              NSFW    SFW
Actual NSFW     50      0
       SFW       0     50

Usage

Basic Usage with Confidence Scores

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch
import torch.nn.functional as F

# Load model
model_name = "surya120/myplaygirl-nsfw-classifier"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(model_name)

# Prepare text
text = "Your text here"
inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True, max_length=512)

# Predict with confidence scores
with torch.no_grad():
    outputs = model(**inputs)
    logits = outputs.logits
    
    # Get probabilities
    probs = F.softmax(logits, dim=-1)
    confidence, prediction = torch.max(probs, dim=-1)
    
    prediction = prediction.item()
    confidence = confidence.item()

# Interpret result
label = "NSFW" if prediction == 0 else "SFW"
print(f"Prediction: {label}")
print(f"Confidence: {confidence:.2%}")

# Apply confidence threshold (recommended: 90%)
CONFIDENCE_THRESHOLD = 0.90
if confidence >= CONFIDENCE_THRESHOLD:
    print(f"High confidence prediction: {label}")
else:
    print(f"Low confidence prediction: {label} (consider manual review)")

Training Data

The model was trained on a custom curated dataset containing:

• Training samples: Text messages with explicit sexual content (NSFW) and safe conversational content (SFW)
• Labels: Binary classification (0 = NSFW, 1 = SFW)
• Data source: Custom generated dataset focused on conversational text moderation

Training Details

• Base model: Transformer-based sequence classification model
• Fine-tuning: Supervised learning on labeled NSFW/SFW text data
• Objective: Binary text classification for content moderation

Testing & Evaluation

The model was evaluated on 100 balanced test samples (50 NSFW, 50 SFW):

Test Results Summary

• ✅ Perfect classification: All samples correctly classified
• ✅ High confidence: 98/100 predictions above 90% confidence
• ✅ Zero false positives and false negatives

Confidence Threshold Analysis

• Recommended threshold: 90%
• Average confidence: 99.24%
• High confidence accuracy: 100.00%

For predictions below the confidence threshold, manual review is recommended for production use.

Limitations

• The model is trained specifically on text-based content
• Performance may vary on content types not well-represented in the training data
• Should be used as part of a comprehensive content moderation system

Ethical Considerations

This model is designed for content moderation purposes. Users should:

• Be aware of potential biases in classification
• Use the model responsibly and in compliance with applicable laws
• Consider the context and implement appropriate human review processes