Memory · Software component
Multi-Signal Relevance Ranker
Software componentMemoryCognition & Memoryarc:MultiSignalRelevanceRanker
A reranker that orders retrieved memory candidates by a weighted combination of semantic similarity, temporal recency decay, learned importance and other signals such as frequency or source reputation.
Responsibility. Ranks retrieved memory candidates by a weighted multi-signal score.
Also known as: Recency-relevance-importance scoring, Weighted retrieval scoring
Variant of Reranker abstract
Relationships
is configured by structural
is invoked by dependency
reads dependency
Design guidance
- SHOULD tune signal weights per domain (e.g., weight recency highly where situations change rapidly, similarity where root causes are stable).
- SHOULD distinguish recency from frequency so novel spikes and stable recurring patterns trigger different actions.
- SHOULD incorporate knowledge recency and source reputation scores so higher-quality current sources surface ahead of slightly more similar low-quality ones.
Quantitative guidance
As stated by the sources; verify before use.
- Example weights alpha=0.7, beta=0.2, gamma=0.1; beta=0.5 strongly favours recent episodes (Ch5.7).
- Frequency signal example: 15 episodes with similarity >0.85 in past 7 days (Ch5.7).
Classification
- Patterns
- retrieval_score = alpha*similarity + beta*recency + gamma*importanceExponential recency decay exp(-lambda*days)Separate recency and frequency scoringSource-reputation weighting
- Quality attributes
- Functional suitability: correctness and validity (ISO/IEC 25010 | NIST AI RMF: valid)Maintainability (ISO/IEC 25010)
- Risks mitigated
- Surfacing obsolete highly similar memoriesMissing emerging recent patterns