Are your analysts tired of manually searching for needles in the haystack of your data? Find out how we achieved 90% automation in the weekly review of news articles for our client with a custom solution powered by a domain-specific classification model.
Problem
Our client was looking for a way to improve the efficiency of their analysts by automating the review of tens of thousands of news articles every week. The articles had to be reviewed manually by specialists, who were responsible for scanning through articles to identify threats. Preliminary efforts to train a machine learning classifier produced noisy, confusing predictions.
Approach
We built our client a custom, automated news filtration system that reduces the number of articles for human review by 90%. The solution leverages a domain-specific language model that we bootstrapped for our client using a combination of open source software, pre-trained large language models (LLMs), hyperparameter tuning, and transfer learning.
Result
“It’s a slam dunk” says our client, who has since been acquired by BlueVoyant. Now the domain-specific model is part of their customer offering, automatically triaging incoming data on a regular basis and saving 20 hours of analyst time per week.
Reduction in Manual Review
Schedule your no-cost, risk-free consultation.