E-mail/Text Tone Analyzer
Year:
2025
Overview
The Email Tone Analyzer is an AI-powered application that combines sentiment analysis and large language models (LLMs) to detect the tone and formality of written communication. It classifies emails and text messages as formal or informal while also capturing sentiment polarity. Designed to promote clarity and professionalism in digital communication, the tool helps users refine their writing tone across academic and business contexts.
Approach
The system integrates transformer-based architectures such as RoBERTa and Llama 3.2, along with weak supervision techniques to efficiently label a large dataset of emails. These automatically labeled samples were then used to fine-tune the Gemini LLM, improving its accuracy in recognizing subtle tone differences. The model was trained to provide both overall document-level and sentence-level tone classifications, ensuring fine-grained feedback for users.
Key Features
Tone Classification: Distinguishes between formal and informal writing styles.
Sentiment Analysis: Detects positive, negative, and neutral emotions in messages.
Weak Supervision Pipeline: Generates high-quality labeled data with minimal manual annotation.
Fine-tuned Gemini LLM: Achieves enhanced tone recognition through targeted model adaptation.
Sentence-Level Insights: Offers detailed tone breakdowns to guide message improvement.
