Finfluencer-8B: Financial Tweet Analysis Model

Finfluencer-8B, developed by Baran Adanır & Ibrahim Enes Duran, is a specialized language model built upon Meta's Llama 3 8B architecture. Fine-tuned with Unsloth for efficiency, its primary function is to analyze financial and cryptocurrency tweets, converting raw text and optional visual context into structured JSON output.

Key Capabilities

• Sentiment Analysis: Determines if a tweet expresses Bullish, Bearish, or Neutral sentiment regarding financial assets.
• Entity Extraction: Identifies specific cryptocurrencies or stocks mentioned (e.g., $BTC, $ETH).
• Structured Output: Generates clean JSON data, making it suitable for database integration and automated processing.
• Visual Context Processing: Incorporates descriptions of attached images (like charts) into the overall analysis.
• Specialized Filtering: Includes strict protocols for filtering out noise, bragging, and non-actionable content, focusing on forward-looking trading setups.

Training Details

The model was trained using LoRA (Low-Rank Adaptation) with 4-bit quantization on a custom dataset of crypto-financial tweets. It is optimized for English, specifically financial and crypto terminology.

Intended Use Cases

Finfluencer-8B is ideal for developers and analysts needing to:

• Automate the analysis of financial social media data.
• Extract structured insights from cryptocurrency discussions.
• Integrate social sentiment into trading strategies.

Limitations

It is specialized for financial contexts and does not provide general conversational capabilities. The model offers analysis, not investment recommendations, and relies solely on provided input text without live internet browsing.