bert_turkish_personality_analysis

bert_turkish_personality_analysis

This repository hosts a Turkish BERT model fine-tuned for multi-label personality trait classification. Built on top of dbmdz/bert-base-turkish-cased, this model predicts psychological and professional personality traits from Turkish text input.

🎯 Task: Multi-label Personality Trait Detection

Given a CV, personal statement, or written expression, the model assigns zero or more traits from the following set:

🏷️ Supported Labels

• özgüvenli – confident
• içe kapanık – introverted
• lider – leader
• takım oyuncusu – team player
• kararsız – indecisive
• abartılı – exaggerated
• profesyonel – professional
• deneyimli – experienced

The model supports multi-label classification using a sigmoid activation and thresholding logic.

🔧 Usage Example

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

# Load tokenizer and model
model_name = "MUR55/bert_turkish_personality_analysis"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(model_name)

# Sample text
text = "5 yıllık yöneticilik tecrübemle liderlik becerilerimi geliştirdim, aynı zamanda ekip çalışmalarına önem veririm."

# Tokenize and predict
inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True)
outputs = model(**inputs)
probs = torch.sigmoid(outputs.logits)

# Threshold to determine label presence
threshold = 0.5
labels = ["özgüvenli", "içe kapanık", "lider", "takım oyuncusu", "kararsız", "abartılı", "profesyonel", "deneyimli"]
predicted = [label for label, prob in zip(labels, probs[0]) if prob >= threshold]

print("Predicted traits:", predicted)

🧠 Model Details

• Base model: dbmdz/bert-base-turkish-cased
• Architecture: BERT with a linear classification head
• Task type: Multi-label classification
• Loss Function: Binary Cross Entropy with Logits
• Training Data: Custom Turkish dataset with personality trait annotations (e.g., CVs, social texts)

📈 Performance

Model was evaluated on a held-out portion of the dataset. Replace below with your real metrics:

🔍 Applications

• CV analysis and candidate profiling
• Smart recruiting and HR systems
• Social media or forum persona evaluation
• Turkish personality-aware recommendation systems

📁 Files Included

• pytorch_model.bin – fine-tuned model weights
• config.json – model configuration
• tokenizer_config.json, vocab.txt – tokenizer files

🤝 Acknowledgments

This project builds upon dbmdz/bert-base-turkish-cased. Thanks to the Turkish NLP community for contributions and datasets.

📬 Contact

If you have questions or suggestions, feel free to open an issue on the model page or contact the author.