Interactive demos for the paper
Two SPARQL queries can look almost the same and return different answers, or look different and return the same answers. We execute a query log once, label query pairs by the overlap of their answers, and fine-tune encoders so that queries sit close when they return similar answers. At query time, only the query text is needed.
Why text similarity is the wrong signal, and a map of real SQuARE test queries: text-based embeddings next to results-aware ones.
Open demo 1 → Demo 2Browse real queries where text similarity retrieves the wrong query, and see how fine-tuning changes retrieval and result-cache decisions.
Open demo 2 →To our knowledge, the first SPARQL query embeddings trained on execution-derived answer overlap, with the SQuARE resource (DBpedia and Wikidata) for training and evaluation.