1 of 16

Vocabularies and use of machine learning in the new IMOS Australian Ocean Data Network (AODN) Portal.

​

Natalia Atkins and Yuxuan Hu

2 of 16

  • Introduction to the AODN and IMOS
  • AODN Vocabularies – concentrating on parameters and platforms
  • GCMD mapping to parameter categories
  • Use of Artificial Intelligence (AI) and Machine Learning (ML) to improve discoverability of all AODN hosted and harvested metadata records.

​

​

​

Overview

3 of 16

4 of 16

AODN published vocabularies on Research Vocabularies Australia

Platform Category

10 concepts

Version 1.2

Discovery parameter

217 concepts

Version 1.7

Many re-use term from BODC P01 or map to P07

Platform

524 concepts

Version 6.5

Re-uses terms from BODC L06 and C17

Parameter Category

67 concepts

Version 3.0

Some mapped to GCMD in vocabulary

5 of 16

AODN Portal (https://portal.aodn.org.au)�

6 of 16

​

AODN Partners utilising parameter and/or platforms

  • Australian Institute of Marine Science (AIMS) (~2600* records)
  • Institute for Marine and Antarctic Studies (IMAS) (~1300*)
  • CSIRO Marine National Facility (~4600*)

(Note – all hosted AODN metadata records do where applicable)

​

AODN Partners not

  • Australian Antarctic Division (AAD) (~1900)
  • Geoscience Australia (~2000)
  • National Computing Infrastructure (NCI) (~100)
  • eAtlas (~500)

​

* not all

​

​

​

​

​

​

​

AODN usage of AODN vocabularies

7 of 16

​

Usage

  • AAD (no AODN parameter usage)
  • Picks up “other” IMOS, AODN, IMAS, CSIRO records

​

Mapping exercise

  • Audited the metadata for GCMD terms
  • Generated a list of granular terms
    • Earth Science | Oceans | Bathymetry/Seafloor Topography | Water Depth

​

    • Mapped these to existing AODN Parameter Category (IMOS focused)
    • Identified new Parameter Category terms required
    • Left un-mapped if not regarded as a “parameter”

​

New AODN Portal

- https://portal-beta.aodn.org.au/

AODN usage of GCMD keywords

8 of 16

​

But what about organisations and metadata records that don’t utilise either the AODN Parameter Vocabularies or GCMD keywords?

​

  • eAtlas
  • Geoscience Australia
  • National Computing Infrastructure (NCI)
  • And all “other” records from other organisations

​

​

​

​

​

​

​

​

AODN usage of vocabularies

9 of 16

AI/ML does not replace AODN vocabularies or human metadata curation. It depends on them. The vocabularies provide the structure, and AI/ML helps more records benefit from that structure.

AODN usage of AI/ML for vocabulary-supported discovery

10 of 16

    • Vocabularies work best when records are connected to the right concepts. However, some records have useful titles and abstracts while their vocabulary metadata is incomplete.
    • In addition, users may not know the official vocabulary term to search for.

The discovery gap

11 of 16

ML supports discovery in two directions: it helps connect records to vocabulary concepts, and it helps connect users' language to vocabulary concepts.

Machine learning as vocabulary support

AODN vocabularies

Record-side support

Missing curated vocab

ML vocabulary classifier

AI-predicted concept

Portal filters

User-side support

User's everyday language

Semantic vocab search in back-end

Suggested vocab concept in front-end

Portal search suggestions

12 of 16

From missing vocabulary to discoverable record

Use case 1

13 of 16

From everyday language to a vocabulary concept

Use case 2

14 of 16

    • Vocabularies provide the controlled concepts.
    • Human expertise provides the authority.
    • ML helps build the bridge to discovery.

​

Responsible AI-supported discovery

15 of 16

​

​

Questions??

Natalia.atkins@utas.edu.au

Yuxuan.hu@utas.edu.au

16 of 16