🧠 Prediction Filter Model for Turkish Stock Tweets
This is a fine-tuned BERT model that classifies whether a tweet contains a financial stock prediction. It’s used as the first step in a pipeline that filters and analyzes tweets from finance influencers.
Model Details
Model Description
- Developed by: damlakonur
- Model type: BERT-based binary classifier
- Language(s): Turkish
- License: MIT
- Finetuned from:
bert-base-cased
The model is trained to distinguish between general tweets and those that contain actionable predictions (e.g., "AKBNK will hit 55 next week").
Model Sources
- Repository: GitHub (private/local)
- Demo: Not hosted yet
Uses
Direct Use
This model is used in an NLP pipeline that:
- Collects tweets from Twitter influencers
- Filters tweets that contain market predictions
- Passes those tweets to downstream models for NER and sentiment analysis
Out-of-Scope Use
This model should not be used to infer stock prices or financial advice. It only filters text based on prediction intent, not accuracy or legitimacy.
Bias, Risks, and Limitations
- The model is trained on a relatively small dataset (~2,000 Turkish tweets)
- May not generalize to informal slang, sarcasm, or misspellings
- Overfits to formats commonly used in Turkish finance Twitter
Recommendations
Users should validate predictions with financial data sources. This model should be one component in a broader information system.
Training Details
Training Data
The dataset consists of Turkish tweets labeled as:
LABEL_1: General tweet LABEL_0: Contains stock prediction Data was collected and labeled manually from Twitter.
Citation
@misc{damlakonur2025predictionfilter, title={Prediction Filter Model for Turkish Finance Tweets}, author={Konur, Damla}, year={2025}, url={https://huggingface.co/your-username/prediction-filter-bist30} }
How to Get Started with the Model
from transformers import pipeline
classifier = pipeline("text-classification", model="your-username/prediction-filter-bist30")
classifier("AKBNK will reach 55 TL next week!")