arXiv:2611.22485v1 Announce Type: new Abstract: Incremental entity resolution pipelines ask a large language model (LLM) whether a new record belongs to an existing cluster and show it the records already linked. Analyses of matching oracles assume that the answer depends only on the records compared. For the LLM matchers we test, it depends on cluster size, and the model lowers its bar for joining a cluster as the cluster grows. On French procurement suppliers from FOPPA, with SIREN identifiers as ground truth, we hold the new record fixed and vary only the number of verified members shown. Sixteen members instead of one raise the false match rate on similarly named companies by \dfmrMin--\dfmrMax{} percentage points in six open-weight models, while their area under the ROC curve rises by only \aucGainMin--\aucGainMax. In Claude Sonnet~5 and Haiku~5.5 the rise is \clSonD{} and \clHaiD{} points. The shift persists with street addresses, prompt variants and the BatchER and ComEM formats, and one false merge raises acceptance of the intruder's other records in every open model. It does not arise on Czech and Bulgarian TED notices, where one record with its full address already identifies the company. In online linking on confusable blocks, a single threshold set on labelled data raises recall by a factor of \fxRRmin--\fxRRmax{} for five models with no detectable loss of precision and no extra LLM calls.
Growing Clusters Lower the Bar: Profile Accretion in LLM-Based Incremental Entity Resolution
A new arXiv paper (2611.22485v1) reports that LLM-based incremental entity resolution matchers lower their bar for joining a cluster as the cluster grows: on French procurement suppliers from FOPPA, with SIREN identifiers as ground truth, showing 16 verified members instead of one raised the false match rate on similarly named companies by a measured margin across six open-weight models, while their area under the ROC curve rose only slightly. The effect persisted with street addresses, prompt variants and the BatchER and ComEM formats, and a single false merge raised acceptance of the intruder's other records in every open model, though it did not arise on Czech and Bulgarian TED notices where one record with its full address already identifies the company. In online linking on confusable blocks, a single threshold set on labelled data raised recall by a multiple for five models with no detectable loss of precision and no extra LLM calls.
Limit how many verified cluster members you show an LLM entity-resolution matcher, since 16 members instead of one raised false matches in six open-weight models, and tune a single threshold on labelled data to recover recall.
more in playbook: Evals and testing for LLM appsRun your AI side-project on zahid.host
EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.