dspm-rs

Maps graphs in a graph database $$DG$$ onto a multidimensional space $$MG$$ under a structural dimension $$M$$ using a mapping function $$φ()$$. The DS-preserved mapping preserves two things: distance and structure.

Paper: Leveraging Graph Dimensions in Online Graph Search

[Link] Authors: Zhu, Yu, Qin

Propose a distance- and structure-preserving (DS-preserved) mapping that automatically selects a small set of informative subgraphs as dimensions, enabling fast top‑k similarity search over large graph databases without expensive NP-hard operations.

The paper shows that mapping graphs into a learned structural dimension lets you answer similarity queries with vector operations while still respecting underlying graph structure. ArrowSpace follows the same pattern for high‑dimensional embeddings: its spectral index builds a structure‑aware space where λτ scores play the role of DS‑preserving coordinates, enabling fast, topology‑sensitive search over large vector corpora.

Core conceptual link

The paper learns a graph dimension: a small set of subgraphs that become coordinates of a multidimensional space, so that distances in that space approximate graph edit/MCS distances while preserving query‑time structure.

arrowspace builds a spectral index over an embedding matrix, using Laplacian‑based structure to define a space where similarity scores (λτ) reflect both content and dataset topology, not just raw cosine geometry.