Highly heterogeneous data have boomed during the last decade, due to their largely distributed way of production: corporations of any size, individual users as well as automatic extraction tools have contributed a constantly increasing volume of heterogeneous and noisy information. Entity Resolution (ER) helps to reduce the corresponding entropy by identifying those pieces of information that refer to the same real-world objects.

Typically, blocking techniques are used to scale ER to large volumes of data. However, most of these techniques rely on schema information and are inapplicable to highly heterogeneous settings. Our work goes beyond existing blocking techniques, by introducing a novel methodology that is inherently crafted for voluminous, highly heterogeneous, and noisy data collections.

At the core of our approach lie three independent, but complementary steps: block-building (using redundant block assignments for effectiveness), meta-blocking (reducing the number of necessary blocks), and block processing (increasing efficiency of ER operations). Our experimental evaluation with three large-scale, real-world data sets demonstrates that our methodology can successfully handle very large and highly heterogeneous datasets, achieving an excellent balance between effectiveness and efficiency.