Parallax Background

National Inter-jurisdictional data sharing system for person data

State government agencies are able to provide services to citizens within that state. However, services and outcomes for citizens can be improved when information is shared between states. Factil was tasked with developing this inter-jurisdictional data sharing system. There are considerable barriers to inter-agency information sharing, which has serious consequences for the safety of vulnerable citizens nationally.

Factil developed Kalinda to solve this challenge. Kalinda is an aboriginal word that means “to look out” or “to see”. Kalinda’s design is based on creating a hosted central system, housed in a secure data centre, that receives a “thin-layer” of data from the state-based systems of participating agencies and provides a web interface for government officials to search nationally for relevant information.

The central system consists of:

  • Database Server holding person, location and relationship information, user data, and public data, that has been indexed for fast partial-match retrieval.
  • An Integration Server that receives batch and incremental data from participating agencies and loads the data into the database. The integration server also uses a machine-learning algorithm to pre-compute confidently-matched records which are also loaded into the database.
  • A Web Server which handles user login, search and related functions for government officials, and a range of data administration functions for data stewards and systems administrators.

Kalinda supports two search modes:

  • A general query mode that searches for records that match a name, personal characteristics and/or relationships with people or locations.
  • A record matching mode that finds records from other agencies that match the given search record.

For each search, Kalinda returns a set of results to the users. These results are ranked in order of importance.

To implement the Kalinda search process, Factil has developed an advanced machine learning algorithm to pre-compute confidently matched records and powerful database indexing techniques to perform both exact and partial matches. We believe that both algorithms are amongst the most advanced of their kind in the world. The Kalinda search process has been specifically designed to handle a wide range of spelling variations, common name alternatives, typographical errors, and missing or inconsistent data.

Kalinda has been implemented with intuitive, easy-to-use web interface that is similar to Google search and can be used on mobile, tablet and computer devices. Factil has installed an instance of Kalinda in an Australian Signals Directorate ISM-certified PROTECTED Government Cloud environment. The system was loaded with large-scale data sets of the three participating agencies. Factil has taken a statistically rigorous approach to developing and evaluating the record matching algorithm, taking into consideration the proportion of false positives and false negatives in the record matching results. The F2 score is a combined measure of result accuracy. We were very happy with the record matching F2 score of 93% for the Proof of Concept. We are optimistic that with further effort we will be able to get the accuracy into the high 90’s. We did indeed achieve this goal on a separate project.