• Member: Efe Celik
• Member: Ulas Ucrak
• Member: Eray Acikgoz

In this study, we aimed to analyze the sentiment and topic distribution of Bitcoin-related tweets using both synthetic and real datasets. Our goal was to:

• Predict tweet sentiments (bearish, bullish, neutral). Hugging Face App
• Assess the relationship between tweet sentiment and Bitcoin price movements.
• Evaluate the influence of users based on follower count and verification status.
• Compare topic similarity between synthetic and real datasets.
• Determine if influential users make more accurate predictions.

• Synthetic Tweet Dataset: Hugging Face Dataset
• Real Tweet Dataset: Kaggle Dataset
• Bitcoin Price Dataset: Kaggle Dataset

• Synthetic Dataset: Text cleaning, tokenization, stop-word removal, and lemmatization.
• Real Dataset: Applied similar preprocessing steps for consistency.

• Word2Vec was used for text embeddings.
• Trained Models:
  • Logistic Regression
  • Neural Network
  • Long Short-Term Memory (LSTM)
• Model Outputs: Saved for further analysis and prediction tasks.

• Merged Data: Real tweets and Bitcoin price data aligned using timestamps.
• Prediction Models: Logistic Regression and Neural Network were used to label real tweets.
• Sentiment Labels: Bearish, Bullish, Neutral.

• Metrics: Follower count, retweets, verification status.
• Accuracy Analysis: Compared average influence scores for correct and incorrect predictions.
• Impact Study: Analyzed if influential users made more accurate predictions.

• Analyzed how sentiment trends influenced Bitcoin price changes.
• Compared Logistic Regression and Neural Network results.

• Users were labeled as 'Trust' or 'Don't Trust' based on prediction accuracy and influence scores.

• Applied LDA (Latent Dirichlet Allocation) and DTM (Document-Term Matrix).
• Compared topic distributions between synthetic and real datasets.

• Developed an interactive sentiment prediction application using the Logistic Regression model.
• Input: Text (e.g., tweets or comments).
• Output: Sentiment prediction (bearish, bullish, neutral).

This study demonstrated the effectiveness of using synthetic datasets for training sentiment analysis models and validated their performance on real-world data.

• Explored the connection between sentiment trends, price changes, and user influence.
• Despite limitations in measuring market impact, the developed application serves as a practical tool for real-time sentiment analysis.

🚀 Future Work: Further refine influence measurement techniques and enhance market impact analysis.

🔗 Explore the Project
Stay tuned for updates and feel free to contribute! 🚀