[Metadata Improvement]: GPT topicCategory hallucinations

[gtsueng]

Issue Name

GPT topicCategory hallucinations

Issue Description

Records with topicCategory hallucinations | hallucinated topicCategory
https://data-staging.niaid.nih.gov/resources?id=zenodo_8112069 | Human biology
https://data-staging.niaid.nih.gov/resources?id=zenodo_7574524 | Human biology
https://data-staging.niaid.nih.gov/resources?id=zenodo_5001210 | Human biology
https://data-staging.niaid.nih.gov/resources?id=zenodo_7104970 | Human biology
https://data-staging.niaid.nih.gov/resources?id=zenodo_6139140 | Human biology
https://data-staging.niaid.nih.gov/resources?id=zenodo_6787653 | Human biology
https://data-staging.niaid.nih.gov/resources?id=zenodo_8009444 | Human biology
https://data-staging.niaid.nih.gov/resources?id=zenodo_2629148 | Human biology
https://data-staging.niaid.nih.gov/resources?id=zenodo_7226308 | Human biology
https://data-staging.niaid.nih.gov/resources?id=mendeley_89d2d85g9s | Human biology
https://data-staging.niaid.nih.gov/resources?id=dataverse_10.7910_dvn_meudnp | Human biology
https://data-staging.niaid.nih.gov/resources?id=dataverse_10.7910_dvn_mh3osr | Human biology
https://data-staging.niaid.nih.gov/resources?id=dataverse_10.7910_dvn_qupdyi | Human biology
https://data-staging.niaid.nih.gov/resources?id=mendeley_gbypg9zky3 | Human biology

It appears that GPT is overly prescribing the Human biology topicCategory to any documents mentioning humans, including human impact on the environment.

To do:

• Identify other frequently hallucinated terms
• Develop heuristics for filtering out excess hallucinated terms
  • The heuristic maybe length-based (name and description provided to GPT)
  • Can be based on the other topicCategories GPT found (e.g. - remove Human biology, 'Ecology' and 'Biodiversity' also topicCategories)
  • Can be restricted to repositories (only apply the heuristics to Mendeley and Zenodo)

Related issues

NIAID-Data-Ecosystem/nde-metadata-corrections#20

Issue Discussion

No response

Please select the type of metadata improvement

• Standardization (normalizing free text to an ontology)
• Augmentation (adding values for metadata fields missing values)
• Clean up (addressing redundancy or messy metadata)
• Structure (changing the structuring of the metadata to support front end UI features)

Meta URL

No response

Related WBS task

https://github.com/NIAID-Data-Ecosystem/nde-roadmap/issues/4

For internal use only. Assignee, please select the status of this issue

• Not yet started
• In progress
• Blocked
• Will not address

Status Description

No response

Request status check list

• This metadata improvement has yet to be discussed between NIAID, Scripps, Leidos
• This metadata improvement does not need to be discussed between NIAID, Scripps, Leidos
• This metadata improvement has been discussed/reported between NIAID, Scripps, Leidos
• This metadata improvement has been implemented locally to generate data for review
• This metadata improvement has been implemented on Dev
• This metadata improvement has been implemented on Dev and the results have been reviewed and approved for staging
• This metadata improvement has been implemented on Staging
• This page/documentation/change has been approved for Production
• This page/documentation/change has been implemented on Production

[gtsueng]

Results of investigation into Human biology: https://docs.google.com/spreadsheets/d/12NAPPzntfVXzDSLMlKgkbNMl6l_6kL-4mAryX2vmMS0/edit?gid=0#gid=0

[gtsueng]

At least for Human biology, the we can use the following inclusion criteria to reduce the bias:

Include 'Human biology' only if one of these other topicCategory values is included:

• Anatomy
• Transcriptomics
• Developmental biology
• Oncology
• Physiology