Google Research

DAG-structured Clustering by Nearest-Neighbors

International Conference on Artificial Intelligence and Statistics (2021)


Hierarchical clusterings compactly encode multiple granularities of clusters within a tree structure. Hierarchies, by definition, fail to capture different flat partitions that are not subsumed in one another. In this paper, we advocate for an alternative structure for representing multiple alternative clusterings, a directed acyclic graph (DAG). By allowing nodes to have multiple parents, DAG structure is not only more flexible than a tree but also allows for points to be members of multiple clusters. We describe a large-scale, map-reduce-based algorithm to infer these DAGs. Our algorithm works by simply merging nearest neighbor substructures to form a DAG structure. Our algorithm is supported with theoretical guarantees showing its representational capacity over tree-based algorithms. Further, we provide comprehensive empirical experiments on large-scale clustering benchmarks and entity resolution datasets. Our results show that our method is as scalable as and more accurate than state-of-the-art tree-based techniques.

Research Areas

Learn more about how we do research

We maintain a portfolio of research projects, providing individuals and teams the freedom to emphasize specific types of work