Blocking

Abstract

As we have seen in Chapter 2.1, grouping entity descriptions in blocks before comparing them for matching is an important pre-processing step for pruning the quadratic number of comparisons required to resolve a collection of entity descriptions. The main objective of algorithms for entity blocking, formally defined in Section 3.1, is to achieve a reasonable compromise between the number of comparisons suggested and the number of missed entity matches. In Section 3.2, we briefly present traditional blocking algorithms proposed for relational records and explain why they cannot be used in the Web of data. Then, in Section 3.3, we detail a family of algorithms that relies on a simple inverted index of entity descriptions extracted from the tokens of their attribute values. Hence, two descriptions are placed into the same block if they share at least a common token. As we will see in Section 3.4, a more precise similarity (e.g., Jaccard) comparison of two entity descriptions can be achieved by post-processing the blocks of the inverted index and thus further reduce the number of entity pairs that need to be compared.