Explainability

Why an item was recommended, made visible to the visitor and auditable by the team.

A recommendation that arrives with no reason is hard to trust, and in a Holocaust-related context trust and accountability matter deeply. The engine should not only deliver relevant content but make clear why each item is suggested.

Embeddings are latent and hard to read directly, so the system explains itself through signals a person can understand: similarity to something the visitor already viewed, alignment with a recognisable theme, or a link to a relevant entity such as a shared place or event. The aim is not one fixed form of explanation but a set of contextual cues that help visitors understand the reasoning, while preserving their agency over where they go next.

Explanations draw on the symbolic side of the system, which is inherently legible:

• Similarity cues: "because you viewed X" from the nearest-neighbour relationship.
• Thematic cues: alignment with a recognisable theme or topic cluster.
• Entity cues: a Knowledge Graph link, for example "recommended due to shared location and event."

flowchart LR
    rec["recommendation"] --> kg["KG path / entity"]
    rec --> theme["topic / tag match"]
    rec --> sim["similar to viewed item"]
    kg & theme & sim --> cue["human-readable cue"]

These mechanisms tie directly to the ethical framework of WP7. Human-readable signals support transparency, foster trust, and ensure accountability when adaptive systems are applied to sensitive content. Relevant metadata and filtering criteria are also logged for audit, keeping the decision process open to review and aligned with best-practice guidelines for respectful digital engagement.

In the implemented engine, every recommendation already carries a per-scorer breakdown on each ScoredCandidate, which is the raw material for these cues; see Ranking and the Recommendation model.