X-AI Insight: Real-Time AI Analysis of War-Related

Our client is an independent analytical center monitoring public sentiment on the war in Ukraine via X (formerly Twitter). To improve speed and accuracy, WEZOM was tasked with building a fully automated AI solution for classifying English-language tweets by topic and sentiment in real time.

Objective
Develop a scalable platform that collects tweets via API, filters bots and irrelevant content, performs emotional and thematic classification using NLP, and visualizes results through a GDPR-compliant dashboard—enhancing decision-making and reducing manual analyst workload.

Solution & Technology

• Model architecture: We used BART-large-MNLI for topic classification and DeBERTa-v3 for sentiment analysis, fine-tuned on 3,000+ manually labeled tweets across hashtags like #ukrainewar, #nato, #refugees.

• Pipeline: Built in Python using HuggingFace, Pandas, and Scikit-learn, with validation metrics including Precision, Recall, and F1-score.

• Filtering: Tweets were cleaned of spam, bots, and non-English content; only text and hashed user IDs were stored to meet GDPR standards.

• Dashboard: A real-time interface lets analysts track sentiment trends and topic distributions across thousands of daily tweets.

Results
The client now performs instant analysis of tens of thousands of tweets, improving the accuracy and transparency of insights on global public perception. The system is already used by the client’s analytics department and has proven scalable and adaptable to other platforms like Reddit or Telegram. Its bot filtering and war-topic tuning make it a powerful tool for media and policy analysts alike.