Independent research: GraphRAG makes AI agents 80% more truthful | Read the report

NODES 26 — November 12, 2026

Improving Text2Cypher with Test-Time Strategies

Session track: Modern Applications

Session time:

Session description:

Text2Cypher systems translate natural language questions into Cypher queries that are executed on graph databases to retrieve structured results. In practice, this process is not always reliable. Even when generated queries appear correct, they often fail during execution due to syntax errors or mismatches with the underlying graph schema. This work focuses on improving reliability at test time, without changing the training process, by combining confidence-based decoding with structured filtering based on grammar and schema constraints.

Speaker

photo of Makbule Gulcin Ozsoy

Makbule Gulcin Ozsoy

Senior Software Engineer, Neo4j

Makbule Gulcin Ozsoy is a Software Developer/Machine Learning Engineer, mainly working on recommender systems, ranking and information retrieval.