CoreSemDB
Benchmark for hybrid semantic-relational query processing over text-rich databases.
CoreSemDB is a benchmark for evaluating hybrid semantic-relational query processing over databases with rich textual content.
Problem
Text-rich databases require queries that combine relational predicates with semantic interpretation over long-form or semi-structured text. Existing database benchmarks are strong at structured query processing, but they do not directly test whether systems can jointly handle symbolic constraints, text-grounded semantics, and query-level evaluation.
Core idea
CoreSemDB turns semantic-relational query processing into a benchmarkable workload. It targets the same broad problem space as SEMA, but from the evaluation side: instead of proposing an execution engine, it defines workloads for comparing systems that combine relational query processing with semantic reasoning over text.
My role
Benchmark framing and research positioning for semantic-relational query processing over text-rich databases.
Evidence
- CoreSemDB, COLM 2026, to appear
- Focuses on hybrid semantic-relational query processing.
- Targets databases whose records contain substantial textual content.
Impact
CoreSemDB complements system work such as SEMA by providing an evaluation substrate for semantic-relational database workloads. The benchmark helps separate system capability from prompt engineering or ad-hoc task construction.