Textual anomaly detection via Self-Supervision

Description

The textual anomaly detection solution aims to identify potential risks to human life and assets, including natural disasters, accidents, violence, terrorism and more. This process involves analyzing scraped content from diverse reporting sources and media platforms such as Twitter, Emergency RSS Feeds, Samdesk and news outlets. By leveraging ML techniques, these events are filtered, classified and subsequently published on a public warning platform. This enables customers to be alerted in near real-time and take proactive measures to safeguard their assets.

What We Achieved

1-Efficient filtering of incoming events

2-Automated classification

3-Cost reduction and easy distribution

Solution Features

1-Risk Assessment and Criticality: Our ML models enable us to filter events that pose an active risk to our customers: most events reported on news and social media are not critical in nature and traditional methods struggle to efficiently differentiate between levels of risk. For instance, distinguishing between a convenience store burglary and a major regional bank heist can be challenging. However, using textual anomaly detection, we can analyze hundreds of thousands of events and pinpoint the few that truly pose a danger to our customers’ assets.

2-Automation of Risk Classification:Using language models, we sanitize event data by removing irrelevant information. Additionally, we employ named entity recognition techniques to identify the event’s location and time. This enables us to swiftly notify customers with assets in proximity to the event. Furthermore, AI-based categorization methods, such as nearest centroids, are utilized to accurately identify the risk category associated with the event (e.g., fire, flooding, robbery, power outage, etc.). This ensures that customers receive precise and relevant information pertaining to their field or situation.

3-Active Learning & Continuous Adaption: Our system ensures that customers are promptly notified about emerging and unforeseen geopolitical situations, such as wars, civil unrest, and disease outbreaks (e.g., the war in Ukraine and Covid-19). By continuously adapting and learning from new data domains, we stay updated and equipped to protect customers’ assets effectively.