SURROUND Ontology-powered Records Triage Robot
What does SORT do?
All organisations now rely on huge and massively-increasing volumes of electronic records that cannot be effectively assessed and classified via manual processes.
The SORT Robot effectively triages electronic records, aids record discoverability, and decreases operational expenditure through the ability to near-fully automate what is currently a manual task.
SORT has been designed to automate electronic record triaging processes smoothly, accurately, and accountably. SORT auto-classifies and sentences electronic records deemed to be of significance, using a powerful situational awareness capability for determining the context of the document, and therefore the appropriate actions required for retention or disposal. Based on its enhancements to electronic records, SORT supports improved record discoverability and faceted searching.
Auto-classification outcome accuracy is measurable, consistent, auditable, and comparable with manual sentencing practices. SORT provides a human-readable explanation for each assessment action.
Dashboards provide visibility of auto-classification events, which are aggregated to provide a wide range of KPIs connecting organisational strategy to business outcomes.
The SORT Robot uses a combination of advanced artificial intelligence, semantics, and machine learning techniques to:
- Characterise electronic records with Semantic metadata
Execute classification rules against the Records - the rules themselves are managed as reference data
- Different Record content and metadata analyses are performed to them test classification rules against
- Incorporate feedback on classification results to improve methods
- Both supervised and un-supervised Machien Learning takes place in the Robot
- Track provenance of all classifications and system executions to preserve decision transparency
How does it work?
Common APIs, Semantics + ML
The SORT Robot stores both data generated and reference data used in sorting as Semantic Web knowledge graphs. It uses the SURROUND Ontology Platform to manage the multi-part reference data knowledge graphs relevant to a sorting scenario.
System actions and generated data are have provenance recorded using the PROV Data Model standard. This ensure maximal operating transparency- used by the SORT Robot
The Robot accesses reference data and also presents interfaces for down-stream systems via a cutting-edge GraphQL API as well as other specialised APIs, such as SPARQL Endpoints. This means it can be meshed into multi-part systems quite easily.
The SORT Robot incorporates several machine learning scenarios operating at different conceptual levels that together allow a Robot deployment to improve its performance over time. The Robot can "learn" what better results look like from human feedback, it can "learn" what more efficient part-classifications of records look like and it by correlating reference data and results. It an also learn about drifiting result trends by re-examining its processing history which is preserved in a comprehensive provenance trace.
The SORT Robot's ML scenarios all read from and write back to RDF Knowledge Graphs and all learnign is tracked using the PROV Data Model. Using a very flexible and yet stadards-based data model for ML means we can adapt the learning schenarios per-application and present decisions made in easy to understand ways (i.e. not secret info!) if required to demonstrate ML traceability.
The SORT Robot can be adapted to classifying any kind of digital asset.
The particular record (asset) reading functiosn can be adapted for your dta types and the you can specify the classifications that you want to see records sorted according to. We can build up a collection of reference data as structured Knowledge Graphs that represent the knolwedge in your domain so that the SORT Robot can operate as your human expers do.
Whatever the deployment scenario: the SORT Robot will always track all decsions and learning made using standardised provenance so that you have auditable outcomes.