Fault Tolerant Trie (address matching)

Fault Tolerant Trie (address matching) is a data structure-based approach designed to improve the process of matching addresses, particularly when dealing with variations or errors in input data. The method leverages a trie, which organizes address tokens hierarchically, reflecting the geographic structure of real-world addresses. This structure allows for efficient representation and lookup of address data, encoding information such as whether a token marks the end of an address, the number of addresses reachable from a node, and whether a node is a leaf.

A key feature of this approach is its implementation of fault tolerance in address matching. The system introduces rules that allow for the skipping or ignoring of tokens in input addresses if doing so enables a better match within the trie. For example, tokens at the end of a messy address may be ignored if the resulting path in the trie leads to a node with sufficient matches, or tokens in the middle may be skipped if they do not correspond to any node at the current trie position. This flexibility is intended to handle real-world data issues such as extra, missing, or reordered address components, making the matching process more robust to input errors.

The approach is demonstrated interactively, allowing users to experiment with canonical and messy addresses and observe how the fault tolerant trie processes them. Additionally, the technique has been implemented as part of a DuckDB Community Extension called splink_udfs. Users can load this extension in DuckDB, create tables of address data, build tries by postcode group, and perform address lookups using the provided functions. There is also mention of ongoing work to incorporate the method into a package called uk_address_matcher for fast, accurate address matching.

This tool is relevant for those working with address data, such as data engineers or analysts involved in record linkage or data cleaning tasks.