SemJoin: Semantic Join Optimization

2026-06-28

AI Summary

This work addresses the prohibitive computational cost of large-scale semantic join, which typically requires invoking a large language model (LLM) for every tuple pair, and the limited adaptability of existing methods across diverse data and predicates. The authors propose a novel LLM-agent-based decision pipeline that introduces, for the first time, an LLM-consultant-driven dynamic routing mechanism to adaptively select between clustering-based and classifier-based strategies according to table-specific characteristics. By integrating unsupervised embedding clustering, sample filtering, and dynamic strategy scheduling, the approach achieves F1 score improvements of 20–33 points over ABJ on three benchmark datasets while substantially reducing LLM invocations—attaining superior performance to FDJ at token costs lower by one to two orders of magnitude.

Abstract

Integrating unstructured data into relational database systems is increasingly important as demand grows for natural language querying and analysis. A semantic join, joining two tables under a natural-language predicate, can be evaluated with a large language model (LLM), but comparing every pair of tuples requires O(M x N) LLM invocations and is cost-prohibitive at scale. Existing systems reduce this cost but typically commit to a single fixed strategy (e.g., embedding similarity or one batched scheme) regardless of the data or the join predicate. We propose an LLM-agent-based decision pipeline that optimizes semantic joins by matching the execution strategy to the characteristics of the underlying tables. An LLM advisor routes each join to one of two strategies: a Cluster Join, which prunes candidates via unsupervised embedding clustering and sample-based filtering, or a Classifier strategy for predicates that reduce to a shared discrete label set. Across three diverse datasets (IMDb reviews, email contradictions, and Stack Overflow tags), the advisor consistently identifies the optimal execution strategy for each workload. This dynamic routing proves decisive: it outperforms adaptive block join (ABJ) by 20-33 F1 points across all datasets while consuming fewer tokens on two of the three, and achieves higher F1 scores than featurized-decomposition join (FDJ) at one to two orders of magnitude lower token cost.

Problem

• semantic join
• large language model
• join optimization
• unstructured data
• relational database

Innovation

• Semantic Join
• LLM Agent
• Dynamic Strategy Selection
• Embedding Clustering
• Token Efficiency