Human Emotion Ontology (HEO) · Schema version 1.0.0
Every node and every edge carries two related but distinct fields:
confidence — the degree of certainty that the construct (or relationship) is real and correctly specified:
| Level | Meaning | Node count | Edge count |
|---|---|---|---|
high | Multiple convergent lines of evidence; near-consensus construct. | 192 | 996 |
medium | Established in one or more frameworks; some disagreement or limited direct evidence. | 195 | 819 |
low | Single-source, speculative, or heavily contested. | 26 | 100 |
evidence_status — the epistemic category of the support:
| Status | Meaning | Examples |
|---|---|---|
empirically_established | Replicated empirical findings | basic-emotion expression studies; guilt→reparation; mixed emotions |
strongly_supported | Consistent support across studies/theories | shame/guilt distinction; envy/jealousy distinction; nostalgia as mixed |
plausible_theoretical | Theoretically motivated with partial evidence | many transitions_to edges; appraisal links |
disputed | Active disagreement in the literature | moral disgust; awe as blend; Plutchik dyads; autonomic specificity |
speculative | Little or no direct evidence | some culturally specific terms; tertiary dyads |
confidence and evidence_status are correlated but not identical: a high confidence node can still have a disputed evidence status for specific relationships (e.g., emotion.awe is a solid node, but its blend_of fear+surprise edge is disputed).
expression recognition (Ekman & Friesen 1971; Cordaro et al. 2018), evolutionary accounts (Nesse 1990), and neurobehavioral work (Panksepp 1998). The membership of the set is disputed (see registry §4.2), which is why some members are medium.
gratitude, compassion): high — each has a dedicated empirical literature.
— the distinctions are real in lexicons and norms (Warriner et al. 2013) but boundary placement is conventional.
soft.
Lutz 1988) but translations are approximate and the "distinct emotion vs. lexicalization" question is open (registry §4.14).
well-measured constructs (Smith & Ellsworth 1985; Lang et al. 1990); medium otherwise.
| Edge type | Typical confidence | Why |
|---|---|---|
subtype_of | high | structural (generated from node parents) |
opposite_of | high/medium | Plutchik pairs and circumplex contrasts are well established |
near_synonym_of, similar_to | medium | semantic; boundary judgments |
co_occurs_with | medium–high | mixed-emotion and comorbidity data where available; plausibility otherwise |
blend_of | high (primary dyads) → low (tertiary dyads) | Plutchik dyads well documented for primaries; idiosyncratic for secondaries/tertiaries |
transitions_to | medium (many) | some empirically supported (frustration→aggression; guilt→reparation); others plausible chains |
amplifies / inhibits | medium | supported where noted (e.g., fear inhibits curiosity — Gray 1982); weak where marked |
appraised_as, motivates, behavioral_expression_of, physiological_association | medium | component-process structure (Scherer 2001); physiology weakly specific (registry §4.11) |
culturally_related_to | medium/low | translation approximations |
Numeric dimensions values are prototypical estimates, explicitly marked confidence: low. They are anchored on Russell (1980) circumplex placements, Mehrabian & Russell (1974) PAD values, Warriner et al. (2013) ANEW-style norms, and Smith & Ellsworth (1985) appraisal locations. Use them for:
opposite_of pairs should have oppositevalence signs).
Do not use them as measured values for individuals, as ground truth for fine-grained inference, or as precise claims about any specific culture's emotion space.
strength ∈ {strong, moderate, weak, unknown} is a qualitative judgment. The ontology deliberately does not assign numeric strengths (the strength_value field exists in the schema but is empty throughout). Rationale: numeric similarity values are only defensible from specific norm datasets (e.g., semantic-embedding similarity over a fixed corpus), and inventing them would create fake precision. A future release can populate strength_value from, e.g., Warriner et al. (2013) distance norms or embedding-based similarity, with the source recorded.
Consumers can filter by confidence to get tiered views of the graph:
appraisals/physiology/behavior, well-studied complex emotions, and the strongest relationships.
and gradations.
specific nodes and contested blend/dyad edges — useful for hypothesis generation, not for claims of fact.
The evidence arrays on nodes and edges carry citation keys resolved to full references in data/ontology-meta.json → sources.