AI summaryⓘ
The authors study how to search through collections of small structured data records by turning them into text sequences for a computer to understand. They found that the order of fields in these sequences, often overlooked, greatly affects how well the search works after training the model. To fix this, they introduced a training method that mixes up the field order each time, so the model learns to recognize fields by their labels rather than position. They tested this approach on a large multilingual dataset of development statistics and showed it improves search accuracy, especially for less common languages. They also created a new benchmark of AI-generated queries to help evaluate retrieval methods and shared all their tools publicly.
structured metadatatext encoderfield orderfine-tuningpermutation invariancenDCG@10multilingual retrievalLLM-generated queriesbenchmarkDevDataBench
Abstract
We study retrieval over catalogs of structured metadata, where each record is a small schema whose fields answer different kinds of query. Embedding a record with a text encoder first serializes its fields into a string, which forces a choice of field order. We show this choice, usually treated as an implementation detail, silently controls retrieval quality once the encoder is fine-tuned. A standard fine-tune loses 7.4 nDCG@10 points when the index is rebuilt under a different field order, because it reads absolute position instead of the field labels. We propose permutation-invariant fine-tuning ($\textbf{PI-FT}$), which serializes each record under a freshly sampled field order with random field dropout, so meaning binds to the labels rather than to position. The change is about two lines in the data loader; it costs negligible in-distribution accuracy and cuts the order-change penalty to 0.2 points. We study this in the discovery of development statistics, a catalog of nearly 10,000 indicators that should be searchable in many languages by a model small enough to self-host. As AI assistants and agents increasingly mediate access to public data and statistics, this retrieval step decides whether an answer is grounded in the right indicator or series, making discoverability a precondition for disseminating data through AI. Because usage logs cannot provide training signal for indicators no one has searched, we generate the queries instead. $\textbf{DevDataBench}$ is a fully LLM-generated benchmark of grounded, facet-targeted queries across 15 languages, covering every indicator for both training and evaluation. A fine-tuned 118M-parameter CPU encoder outperforms every zero-shot baseline, including $\texttt{text-embedding-3-large}$ (0.707 vs.\ 0.556 nDCG@10), with the largest gains in low-resource languages. We release the benchmark, pipeline, models, and a reusable PI-FT framework.