Cost-based selection of Semantic Bloom Filter realizations

Determine a cost-based method for selecting between unary Semantic Bloom Filters and pairwise semantic screens as physical realizations of a registered necessary condition in JEVDB.

Background

JEVDB supports two physical realizations for registered necessary conditions in semantic joins: unary Semantic Bloom Filters, which can reduce an entire target relation before candidate-pair materialization, and pairwise screens, which operate after pairs have been formed and can provide more selective pruning. The paper establishes the correctness conditions for lossless pruning but does not develop a cost model or optimizer rule for choosing between these alternatives.

The unresolved problem is to determine when the broader relation-level reduction of a unary filter outweighs the greater selectivity of pairwise screening, accounting for factors such as candidate cardinality, extraction cost, data distributions, and model-inference cost.

References

JEVDB supports both unary SBFs and pairwise screens as physical realizations of a registered necessary condition. Detailed cost-based selection among these alternatives is left to future work.

— Prune First, Decide Fast: Scalable Semantic Query Processing with JEVDB  (2610.02046 - Wang et al., 1 Oct 2026) in Section 3.2.1, subsection “Correctness and cost”