Event Ontology Alignment Prompt Template

This prompt is designed for ontology engineers and knowledge graph architects who need to map event types extracted from unstructured text to a standard event taxonomy. Use it when you have an upstream extraction pipeline producing raw event labels and you need to normalize those labels into a controlled vocabulary such as CAMEO, IDEA, or a custom enterprise ontology. The prompt forces the model to produce an alignment decision, a confidence score, and an explicit gap flag for every input event type. It is designed for batch processing of event type strings, not for end-to-end extraction from raw documents.

The ideal workflow pairs this prompt downstream of an event extraction prompt. Raw event labels from the extraction step are fed into this alignment prompt in batches, and the output is validated before ingestion into a graph database or event store. The prompt requires a defined target ontology with clear event type definitions, hierarchical relationships, and synonym mappings. Without a well-documented target ontology, the model cannot produce reliable alignments and will default to low-confidence guesses or gap flags. The prompt also assumes that input event types are already extracted and normalized to strings; it does not handle document context, coreference, or event argument structures.

Do not use this prompt when you need to extract events from text—pair it with an event extraction prompt upstream. Do not use it when the target ontology is undefined, when you require real-time sub-50ms responses, or when the input contains ambiguous event labels without sufficient context for disambiguation. For high-stakes domains such as intelligence analysis or legal proceedings, always route low-confidence alignments and gap-flagged events to human ontology curators for review before ingestion. The prompt is a normalization tool, not a replacement for ontology governance.

This template is designed to be pasted directly into your prompt layer. It accepts a list of extracted event types and a target ontology, then returns an alignment mapping. Every variable is enclosed in square brackets—replace them with your actual data before sending the request. The prompt instructs the model to classify each input event, provide a confidence score for the mapping, and explicitly flag any events that cannot be aligned to the target ontology, preventing silent misclassification.

textCopy
You are an ontology alignment specialist. Your task is to map a list of extracted event types to a standard event taxonomy.

INPUT EVENT TYPES:
[EXTRACTED_EVENT_TYPES]

TARGET ONTOLOGY:
[TARGET_ONTOLOGY_DEFINITION]

CONSTRAINTS:
[CONSTRAINTS]

OUTPUT SCHEMA:
{
  "mappings": [
    {
      "input_event": "string",
      "mapped_ontology_class": "string | null",
      "confidence": "high | medium | low",
      "rationale": "string",
      "is_gap": "boolean"
    }
  ],
  "unmappable_events": ["string"],
  "ontology_gap_notes": ["string"]
}

INSTRUCTIONS:
1. For each event in [EXTRACTED_EVENT_TYPES], find the single best match in [TARGET_ONTOLOGY_DEFINITION].
2. If no suitable class exists, set mapped_ontology_class to null and is_gap to true. Add the event to unmappable_events.
3. Provide a concise rationale for every mapping decision, especially for low-confidence matches.
4. If you identify a recurring pattern of unmappable events, add a note to ontology_gap_notes suggesting a new class.
5. Do not force a match. A null mapping with a gap flag is preferred over an incorrect classification.
6. Adhere strictly to [CONSTRAINTS].

After pasting this template, replace [EXTRACTED_EVENT_TYPES] with a JSON array of event type strings from your extraction pipeline. Populate [TARGET_ONTOLOGY_DEFINITION] with the full hierarchy, class definitions, and synonyms of your target taxonomy. Use [CONSTRAINTS] to enforce domain-specific rules, such as 'Do not map financial events to healthcare classes' or 'Prefer the most specific subclass.' The output is a structured JSON object ready for programmatic ingestion. Always validate the response against the provided schema before writing to your ontology store, and route any mapping with a confidence of 'low' or an is_gap flag of true to a human review queue.

This prompt is not a standalone tool; it is a mapping stage inside an event extraction pipeline. The typical flow is: raw text → event extraction prompt → event structures → this alignment prompt → aligned event records → downstream ingestion. The alignment prompt expects already-extracted event types as input, not raw documents. You must supply the target ontology as structured context—either inline in [ONTOLOGY_DEFINITION] or via a retrieval step that pulls the relevant ontology branch before calling the model. The prompt produces aligned classifications, mapping confidence scores, and ontology gap flags. Your application code should treat the gap flags as first-class signals, not errors: they indicate where your ontology needs to grow.

Validation and retry logic: Parse the model output against a strict schema that requires aligned_type, confidence, mapping_rationale, and gap_flag fields for every input event. If the output fails schema validation, retry once with the validation error injected into [CONSTRAINTS]. If confidence falls below your threshold (we recommend 0.7 as a starting point for automated ingestion, 0.5 for review-queue routing), route the event to a human review queue rather than silently ingesting a low-confidence mapping. Log every mapping decision with the input event type, the aligned ontology class, the confidence score, and the gap flag. This log becomes your ontology drift detector over time.

Model choice and tool use: This prompt works well with GPT-4o, Claude 3.5 Sonnet, and comparable models. Avoid smaller or older models for this task—ontology alignment requires reasoning about taxonomic fit, not pattern matching. If your ontology is large (hundreds of classes), do not paste the entire ontology into the prompt. Instead, use a retrieval step: embed the input event type description, retrieve the top-k candidate ontology classes, and supply only those candidates in [ONTOLOGY_DEFINITION]. This keeps the prompt within token budgets and reduces the risk of the model selecting a superficially similar but incorrect class from a distant branch. For high-stakes domains like legal or clinical event taxonomies, always require human review of gap-flagged and low-confidence mappings before they enter your knowledge graph or analytics pipeline.

What to avoid: Do not use this prompt to extract events from raw text—that is a separate extraction stage. Do not treat the alignment output as ground truth without confidence checks. Do not ignore repeated gap flags for the same event type across multiple documents; those are signals that your ontology needs a new class or a broader parent. Do not run this prompt without versioning both the prompt template and the ontology snapshot used; when the ontology changes, you need to know which version produced which mappings. Wire the prompt version and ontology version into your mapping log for auditability.