X-AI Insight: AI solution for AI tweet analysis
Challenge
Our client is an independent analytical center that monitors English-language public opinion on the war in Ukraine via X (formerly Twitter). Their goal is to identify shifts in sentiment around topics like invasion, refugees, sanctions, and NATO membership in real time.
Previously, this analysis was manual—slow, resource-intensive, and prone to subjective bias. Given the scale of data, it became impossible to maintain consistent quality or speed. Additionally, traditional keyword-based systems failed to filter bot-generated content or understand the nuanced context of social media language. The challenge was to develop an AI-powered solution that could automatically collect, filter, classify, and visualize sentiment data from English tweets in real time—without violating GDPR and while ensuring high precision, adaptability, and scalability.
Our client is an independent analytical center that monitors English-language public opinion on the war in Ukraine via X (formerly Twitter). Their goal is to identify shifts in sentiment around topics like invasion, refugees, sanctions, and NATO membership in real time.
Previously, this analysis was manual—slow, resource-intensive, and prone to subjective bias. Given the scale of data, it became impossible to maintain consistent quality or speed. Additionally, traditional keyword-based systems failed to filter bot-generated content or understand the nuanced context of social media language. The challenge was to develop an AI-powered solution that could automatically collect, filter, classify, and visualize sentiment data from English tweets in real time—without violating GDPR and while ensuring high precision, adaptability, and scalability.
Solution
WEZOM built an end-to-end AI platform for tweet sentiment and topic classification using NLP and transformer-based models. The backend architecture was built in Python, integrating the Twitter API via Tweepy for scalable tweet collection and preprocessing. We cleaned the data using heuristics and Pandas, removing bots, spam, duplicates, non-English tweets, and external content.
For classification, we fine-tuned two transformer models: BART-large-MNLI for multi-topic classification and DeBERTa-v3 for sentiment analysis. The models were trained on over 3,000 manually labeled tweets using relevant war-related hashtags (#ukrainewar, #donbas, #nato, etc.). The analytical pipeline was built using HuggingFace Transformers, Scikit-learn for validation and metrics, and Matplotlib for offline visualization.
To ensure GDPR compliance, we stored only tweet texts and hashed user IDs. The result is a web-based dashboard for real-time visualization of sentiment and topic trends, fully integrated with the client’s internal systems. This platform allows analysts to track the dynamics of public opinion with a few clicks—saving time, reducing manual labor, and improving insight quality.
WEZOM built an end-to-end AI platform for tweet sentiment and topic classification using NLP and transformer-based models. The backend architecture was built in Python, integrating the Twitter API via Tweepy for scalable tweet collection and preprocessing. We cleaned the data using heuristics and Pandas, removing bots, spam, duplicates, non-English tweets, and external content.
For classification, we fine-tuned two transformer models: BART-large-MNLI for multi-topic classification and DeBERTa-v3 for sentiment analysis. The models were trained on over 3,000 manually labeled tweets using relevant war-related hashtags (#ukrainewar, #donbas, #nato, etc.). The analytical pipeline was built using HuggingFace Transformers, Scikit-learn for validation and metrics, and Matplotlib for offline visualization.
To ensure GDPR compliance, we stored only tweet texts and hashed user IDs. The result is a web-based dashboard for real-time visualization of sentiment and topic trends, fully integrated with the client’s internal systems. This platform allows analysts to track the dynamics of public opinion with a few clicks—saving time, reducing manual labor, and improving insight quality.
Results
The platform has transformed the client’s workflow. They now process tens of thousands of English-language tweets daily, automatically classifying them by tone and topic. The dashboard provides real-time insights into sentiment shifts and emerging trends, helping the team respond faster and with greater clarity.
This system replaces manual work and provides a more objective, scalable view of public opinion during wartime. Its modular architecture allows easy adaptation to new platforms (like Reddit or Telegram), and advanced filtering ensures resilience against bot-generated content.
The solution is already in regular use by the client’s analysts and can serve as a model for similar geopolitical monitoring tools. With high performance across key NLP metrics (Precision, Recall, F1-score), and full GDPR compliance, this project showcases the power of AI in complex, sensitive contexts.
The platform has transformed the client’s workflow. They now process tens of thousands of English-language tweets daily, automatically classifying them by tone and topic. The dashboard provides real-time insights into sentiment shifts and emerging trends, helping the team respond faster and with greater clarity.
This system replaces manual work and provides a more objective, scalable view of public opinion during wartime. Its modular architecture allows easy adaptation to new platforms (like Reddit or Telegram), and advanced filtering ensures resilience against bot-generated content.
The solution is already in regular use by the client’s analysts and can serve as a model for similar geopolitical monitoring tools. With high performance across key NLP metrics (Precision, Recall, F1-score), and full GDPR compliance, this project showcases the power of AI in complex, sensitive contexts.
