Yesterday @royaheeee and I presented our paper, DAGE: DAG Query Answering via Relational Combinator with Logical Constraints, at #WWW2025. Existing query answering methods over knowledge graph embedding were limited to decomposable queries, called tree-form queries. Our method provides embeddings for relations that correspond to conjunctions of relations. This idea increases the expressive power of existing query embedding methods.

#KnowledgeGraphs #MachineLearning

https://dl.acm.org/doi/abs/10.1145/3696410.3714677

Institute researchers propose a new query embedding method with increased expressiveness of existing query embeddings for #KnowledgeGraphs. The paper will be presented at the Web Conference 2025 #WWW2025, in Sidney, Australia, from April 28 to May 2.

Query embedding methods are used to predict the answers to queries by embedding entities and queries. They decompose queries into subqueries, and perform complex reasoning since tasks are divided into subtasks. However, existing query embedding methods were limited to tree-form queries. In the paper "DAGE: DAG Query Answering via Relational Combinator with Logical Constraints," Yunjie He (@royaheeee), Bo Xiong, Daniel Hernández (@daniel), Yuquicheng Zhu (@yuqichengzhu.bsky.social), Evgeny Kharlamov, and Steffen Staab, developed a method to extend the capacity of existing query embedding methods for queries whose is structure can be seen as a DAG, and answer queries like "give all works x edited by an Oscar winner y and produced by an Oscar winner y." This query is not tree-form because it cannot be equivalent divided. Indeed, if we divide it into the queries "give all works x edited by an Oscar winner y" and "give all works produced by an Oscar winner y," we lose the requirement that y must refer to the same values in both subqueries.

See more on https://www.ki.uni-stuttgart.de/institute/news/New-query-embedding-method-with-increased-expressiveness-to-be-presented-at-the-Web-conference/