Анализ сентимента Telegram-каналов
A private client approached us with a hypothesis that Telegram channel publications may influence the behavior of some Moscow Exchange market participants. We built an ML module to analyze messages, detect company mentions, and determine publication sentiment.
Client
Private client
Period
1 week
Format
ML module
About the project
A private client approached us with a hypothesis that Telegram channel publications may influence the behavior of some participants in the Moscow Exchange market. To test this hypothesis, we needed to develop a module that analyzes Telegram messages, identifies company mentions, and determines the sentiment of publications. At the core of the project was the task of transforming an unstructured stream of messages into a clearer analytical signal suitable for further analysis.
The Challenge
We needed to quickly build a module for analyzing the sentiment of Telegram channel messages. The key requirement was to create a foundation that could collect and process messages from Telegram, identify company mentions, determine publication sentiment, and use the resulting data to test the hypothesis about the influence of content on market participant behavior.
Our Solution
- Developed an ML module for analyzing messages from Telegram channels
- Implemented logic for finding and normalizing company mentions in texts
- Built a data processing pipeline for message analysis and final signal generation
- Evaluated multiple sentiment analysis approaches: TF-IDF, BERT, CatBoost, and embedding models
- Designed a data structure linking companies, messages, and resulting analytical signals
- Delivered the full solution within tight deadlines for market hypothesis testing
Results
- ML sentiment analysis module delivered in 1 week
- Client received a tool for testing the hypothesis about Telegram publications influencing Moscow Exchange market participants
- Unstructured message stream transformed into a clearer analytical layer
- Foundation established for ongoing analysis of company mentions and publication sentiment
- Case demonstrates our ability to rapidly build applied ML solutions for research and analytical tasks
Similar projects
Voice AI Trainer
For RZD, we developed an educational platform where employees could study learning topics and take knowledge assessments with an AI assistant via voice interaction — a modern alternative to conventional tests and formal training.
Классификация новостей
For AI News, we built an ML microservice for a startup working with a stream of news from multiple sources. The product aggregated news from various channels, analyzed content, and delivered a processed feed without duplicating materials.
Let's buildsomething extraordinary.
Ready to start your next project? Reach out and let's discuss how we can help you achieve your goals.