Kingland continues to push the technology envelope with enhancements to its Text Analytics Platform Suite, which we touched on in last week’s catch up on the company with president Tony Brownlee. The enhanced solution includes new administration menu options with deeper analysis and metrics for unstructured data sources and events, and improvements to Named Entity Recognition (NER) training.
The improvements to NER training increase accuracy in identifying organisations and people that are intuitively highlighted by text analytics within documents. In this case, targeted retraining of the NER models has increased the accuracy of stock models from 60% to nearly 90% in aggregate over numerous source document sets. This saves hours of reading and analysis time for teams manually reviewing documents for specific entities and related events.
By way of example, Kingland says that while the average person can read and comprehend a 100+ page document in several hours, text analytics solutions can read and process the same document in minutes.
Matt Good, chief technology evangelist at Kingland, explains these developments as a response to industry leaders wanting to efficiently solve data challenges around searching and extracting data from a variety of unstructured data sources. Use cases include onboarding, Know Your Customer (KYC), underwriting, compliance, risk monitoring, Anti-Money Laundering (AML), and sentiment analysis.
He adds: “Most enterprises have hundreds of thousands, if not millions of documents used by dozens or hundreds of processes. Organisations want speed, accuracy and the comfort of knowing that they are making business decisions based on extracted data that provides context with their counterparties, people, events and general entities of interest.”
Features of the Kingland Text Analytics Platform Suite include: unstructured source integration, which integrates directly with RSS feeds, crawls bot friendly websites, and supports imported documents in PDF, HTML and other source formats to allow business decisions based on more complete information; data identification and extraction, which combines trained and configurable language models to identify, tag and extract entities, people, events and other data attributes; and language modelling and training, which applies different models to different types of unstructured data sources and documents to support unique, fine-tuned analysis across documents including legal documents, financial documents and news articles.