Temporal Index Engine (ST-HNSW)
7. Temporal Index Engine (ST-HNSW)
Section titled “7. Temporal Index Engine (ST-HNSW)”7.1 Index Architecture
Section titled “7.1 Index Architecture”7.2 Search Algorithm
Section titled “7.2 Search Algorithm”7.3 Combined Distance Function
Section titled “7.3 Combined Distance Function”La distancia combinada es el corazón de ST-HNSW:
d_ST(query, candidate) = α · d_semantic(q.vector, c.vector) + (1 - α) · d_temporal(q.timestamp, c.timestamp) × decay(c.age)Donde:
α ∈ [0, 1]— peso relativo semántico vs temporal (configurable por query)d_semantic— coseno, L2, dot product o Poincaré según configuraciónd_temporal—|q.timestamp - c.timestamp| / temporal_scaledecay(age) = e^(-λ · age)— penaliza conexiones antiguas
7.4 Timestamp Graph (TANNS Integration)
Section titled “7.4 Timestamp Graph (TANNS Integration)”El HNT almacena los cambios en las listas de vecinos como un árbol binario de diffs, evitando duplicación de vecinos estables.