After earning a spot on the 2023 AIFintech100, in this blog our engineers unpack three AI-driven technologies that make Sigma’s platform an accurate and scalable risk management tool for our current and future clients. Ready to geek out? Keep reading!
Sigma’s AI processes 4.5 million+ articles monthly from 200,000+ publishers in total, filtering for adverse news involving corporations or individuals. We determine and classify risk types using event tags, and also surface relevant entities mentioned in the article. Finally, articles referring to the same real-world event are grouped together to increase efficiency during due diligence reviews.
Using the power of large language models (LLMs) and their ability to excel at various natural-language processing (NLP) tasks, Sigma delivers world class news and adverse news screening performance.
From 200,000+ news sources, Sigma’s Adverse News model only extracts news relating to risks that are important, as defined by the end user. This model was built on transfer learning from the DistilBERT LLM.
Next, relevant event tags for each article are determined by a supervised deep learning model built on the RoBERTa LLM. Sigma’s model was trained on a proprietary and growing corpus of expert tagged news articles, which include the following events:
Risk managers can optimize their teams efficiency by specifying only the event tags of interest using filter sets. It's important to note that Sigma’s model considers news events that may not be adverse, but could be important to risk managers. These tags include the following events:
Sigma’s Named Entity Recognition (NER) technology identifies and extracts specific entities mentioned in news articles, such as organizations, individuals, and locations. NER greatly improves the relevance of articles bound to the entity of interest, versus keyword matching which is an outdated approach found in many legacy systems. Sigma’s NER model leverages BERT (another LLM), empowered by self-attention among other approaches.
Noteworthy news events are often covered by several publishers, resulting in users having to sift through multiple articles on the same topic. Sigma combines articles covering the same event into a single event for a consolidated view that streamlines reviews. To identify which articles are similar, Sigma’s system embeds text from each article into a latent vector space via universal sentence encoding.
Within the unstructured text of any news article is a set of relationships between entities that represent a knowledge graph for the news event being discussed. For example, if an entity is mentioned in an article about a negative event, it doesn't always signal wrongdoing on that entity's part.
In the near future, Sigma’s solution will only show articles where the risky actors match the entities you’re interested in. Moreover, by extracting relevant entity relationships from news articles, Sigma will establish powerful links between entities, giving risk and compliance teams an extra edge in identifying network-based risk.
Not only is Sigma’s matching system on the leading edge when it comes to accuracy, it’s also 10 times faster through the use of Golang, an open source programming language developed by Google. The matching algorithm has been tuned using generative AI to emulate real-world text variants at a massive scale.
Sigma has developed a highly efficient proprietary matching algorithm and system that leverages generative AI to generate thousands of variations of entity names that are commonly used in different countries and languages, accounting for misspellings. Parameters were tuned using this massive test set and continue to be optimized through further additions.
Today, Sigma’s AI-tuned matching algorithm helps risk and compliance teams uncover and address scenarios that may not have been considered before. Future plans to incorporate AI and deep learning models directly into matching algorithms will further improve the accuracy of matching, particularly with challenging names.
Sophisticated global crime networks have made risk management a complex task. Traditional strategies that focus solely on individual transactions or client screening are no longer effective, a topic we recently explored on an expert panel about next generation risk management. In today’s risk landscape, network risk mitigation tactics are important if risk and compliance teams want to effectively mitigate financial crimes.
With the imminent launch of the Sigma360 Platform, users will be able to assess the connections between entities in a fast and scalable way. These analytics are derived from trusted data sources, while the Sigma360 technology platform can analyze risk across entity relationships at a point in time or on an ongoing basis, such as those between owners, founders, officers and subsidiaries.
This powerful new network graph enables users to explore beyond their initial search results and understand how an entity is connected to the broader world, while also accessing a singular stream of related risk intelligence and proprietary risk scoring spanning more than 150 risk sub-indicators.
Sigma’s AI-powered platform is used by leading financial institutions and high-risk businesses to identify, screen, monitor and review clients and their relationships, with the following benefits:
To learn how Sigma delivers modern risk mitigation workflows for risk and compliance teams, request a demo today.
This week, the ICIJ reported on a new study by a group of renowned academics that revealed that, despite past scandals, the industry continues to...
Hello Sigma network! As the year comes to a close, not only have we celebrated being named in the RegTech 100 and winning the 2020 Benzinga Awards...