Community tool proposal: GPT-based search bot built on BCIO ontology files #1212
Description
GitHub Issue — HumanBehaviourChangeProject/ontologies
Title
Community tool proposal: GPT-based search bot built on BCIO ontology files
Issue body
What this is
I have built an experimental GPT-powered search bot that uses the BCIO ontology files (bcio.owl + CSV hierarchies) as its sole knowledge source.
The bot is available here:
https://chatgpt.com/g/g-69a6ab4b5adc81919a5be72eecbd2f81-ontology-search-bot
How it works
The bot operates in two modes:
Mode 1 — Keyword search
User types search: [term] (in any language — EN/RU/DE).
The bot translates to English if needed, searches all BCIO files, and returns a table:
ID · Label · File · Hierarchy path
Mode 2 — Full situation analysis
User describes a behavioural situation in natural language.
The bot:
- Extracts slots: Behaviour · Population · Setting · Mode of delivery · Source · Style · MoA · BCTO
- Internally tags using BCW/COM-B/TDF as ranking indexes (not shown in output)
- Retrieves candidates using hybrid search (BM25 + semantic similarity)
- Scores using:
0.40 × semantic + 0.30 × keyword + 0.15 × hierarchy_bonus + 0.15 × policy_alignment - Outputs 1–2 bundles with ID/Label/file references — no invented terms
Hard rule: the bot only outputs terms that exist in the loaded ontology files. If no match — it states “no exact match” and suggests nearest nodes (parent/sibling/child).
Example dialogue
User: Students should listen in class, not play around. Grade 5, Germany. No restrictions.
Bot output (Bundle A):
audio presentation listening behaviour (BCIO:050427)instruct how to perform a behaviour BCT (BCIO:007058)provide feedback on behaviour BCT (BCIO:007023)attentional disposition (BCIO:050572)face to face mode of delivery (BCIO:011003)secondary school (BCIO:026026)
Attribution
All BCIO content is used under CC-BY-4.0.
Attribution: Human Behaviour Change Project, UCL; Michie et al.
Questions for the team
- Are there known limitations or versioning issues with the CSV hierarchy files I should be aware of?
- Would the team be open to listing this as a community tool on the BCIO website or repository README?
- Is there a preferred way to report cases where the bot finds no match — as potential gaps in the ontology?
Technical stack
- GPT-5.2 Thinking (ChatGPT Custom GPT)
- Knowledge files: bcio.owl.txt + 8 × BCIO-*.csv hierarchies
- Instructions: structured prompt with pipeline, scoring formula, and response format A–E
- Languages supported: EN / RU / DE (auto-detect)
Happy to share the full system prompt and instructions if useful for the team.