Active learning, accelerated data labeling, new knowledge models and auto-generation of extraction rules drives scalability and efficiency of natural language AI
Expert.ai today announced new features have been added to its natural language (NL) platform, enabling applications to be put into production faster, with the highest accuracy possible and at scale. New capabilities include enhanced active learning to further improve data labeling; pre-trained knowledge models to both accelerate the delivery of NL applications as well as proactively select and quickly validate the desired outputs; and expanded extraction functionalities based on auto-generation rules for extraction.
Companies want to scale their use of AI but struggle to measure its value and fill the gap between technical promise and business reality. Typically, within NL, there are challenges attributed to the lack of suitable or sufficient documents for training a model. With a machine learning-only approach, tasks such as annotation and data labeling are onerous. Annotation is a human-led activity dependent on subject matter experts that determines the overall project scalability and success. The problem is that the manual selection and labeling of the right business documents is tedious as it requires extensive resources and is very time-consuming.
“Putting AI in production, scaling it across the business and measuring its true ROI remains a challenge for many organizations,” said Luca Scagliarini, chief product officer at expert.ai. “When it comes to natural language applications and content, the solution lies in embedding technology and knowledge that complements language intensive processes that typically require a human to read and comprehend content. These expert.ai Platform enhancements prioritize the need organizations have to develop AI that can work quickly and efficiently in real-world enterprise scenarios.”
The most effective AI-based NL platform
Recently named a Strong Performer for text analytics platforms by Forrester, expert.ai provides the most effective mix of NL tools and techniques for seamlessly scale business use cases. This is because the platform is purpose-built to handle the complexity of unstructured language data. Today, the company unveiled additional Platform enhancements including the following features:
- Active learning to accelerate and improve the efficiency of data labeling. The output of annotation activities can be enhanced by learning a concept with fewer examples, both propagating annotations across multiple documents with the click of a button. It also automatically suggests which documents to annotate for the greatest impact on overall model accuracy..
- Original document view to improve annotation and extraction confidence by seeing, and making, annotations with the context of the original document file format.
- Auto-generation of extraction rules leverage machine learning to enable users to automatically generate rule-based models for entity or concept extraction use cases.
- New „smarter from the start“ knowledge models deliver natural language applications to production faster with higher levels of domain-specific business accuracy. Users can access customizable pre-built rules-based models to classify documents and extract entities, insights and relationships specific to a vertical domain or use case. New knowledge models include banking email classification, temporal information extraction, a finance knowledge graph extension (providing full understanding of events, organizations, markets, tickers and indices), hate speech detection, corporate crime identification for intelligence and operational risk mitigation.
- Enhanced taxonomy management increases effectiveness and productivity when building reusable enterprise-wide content taxonomies. As the platform is purpose-built for low code natural language workflow, it can be leveraged as a single, end-to-end solution for document classification, terminology-driven taxonomy indexing, metadata enrichment and knowledge base creation and maintenance.
- Expanded language support to create expert.ai projects in Dutch, Russian and Portuguese.
The enhanced platform release further includes a new document data extraction engine (beta version) that requires less post-extraction processing while ensuring a higher quality. This converts native PDFs and returning text, titles, headers, footers and tables with relevant metadata such as bounding box, reading order and document information.