2. AML #1 issue: False Positives
https://www.innovativesystems.com/blog/how-defuse-ticking-time-bomb-bad-data-aml-compliance
3. Technology Strategies to reduce False Positives
Compliance-specific Data Quality
Tuning of Machine Learning
Enhanced list database
Robotics Process Automation
Innovative AI / Machine Learning
4. NLP Core Solution on False Positives issue
Augment AML monitoring and investigation with Deep learning-
driven natural language processing (NLP) technologies
News and social media sentiment analysis, entity recognition,
relation extraction, entity linking and link analysis on different
data sources to provide additional evidence to human
investigators for final decision-making
5. NextGen AML summary
Harnessing heterogeneous open data in AML
compliance to make it more robust and updated
Using different levels of sentiment analysis to
identify negative evidence of a target entity
Using Name Entity Recognition (NER) and
relation extraction (RE) to build the knowledge
graph of a target entity, and analyse the hidden
and complex connections between the target
entity and existing suspicious entities from the
fraud bases
6. NextGen AML components
1. Transaction Monitoring
Suspicious transactions are
flagged. Name, location,
account details and other
relevant data are extracted
2. Name screening
Sanction list or ML
cases from bank
records and other fraud
bases
5. Knowledge Graph
A collection of information
gathered on the previous
stages
4. Sentiment Analysis
Geo-spatial time-series
sentiment analysis to
find clues of the target
entity
3. NER, ER and Record Linking
Collecting unstructured negative
news data to extract relational facts
in real time and build an entity-
specific KG regarding the target
entity
6. Confidence score
Generated along with
evidences from each
module
8. NER, ER and Record Linking
Named Entity Recognition (such as Organization, Person, or
Location, and predefined taxonomies), Entity Relations and
Record Linking is the ability to recognize that records across
different tables and data sources, using different schemas and
even in different languages are in fact referring to the same
entity (a person/ a company). Fuzzy Match and ER modules
removes ambiguity and reveals previously hidden relationships in
the data.
9. Sentiment Analysis
Get comprehensive data coverage of news,
blogs, discussions and reviews
Monitor dark web data and detect data
breaches in near real-time
Leverage over 100TB of structured historical
data as far back as 2008