cirad-astre / topomatch Goto Github PK

While the local-alignment algorithm currently used seems to perform better than alternatives such as fuzzy-matching or distance-based methods in most cases, sometimes it fails to match pretty obvious terms.

Implement other competing methods to be used either as an alternative or simultaneously. If simultaneously, we can display the best guesses from each method to be presented as alternatives to the user for the manual fixes.

I'm currently performing transliteration from UTF-8 to ASCII for standardising toponyms.
However, in Windows systems with some extended ASCII encoding (latin-1, ISO-8859-1, whatever) this have given some problems.

Test this and implement a more robust approach that takes into account the current locale.

Given a topomatch object, write a function that walks the user interactively through all the unmatched toponyms, presenting the candidate terms ordered by score and letting the user pick the right choice (or NA for no match).

Return a named vector with fixes suitable to be included as the fixes argument of transcribe().

Use case: need to match both province and district names for two sources.

1. Match province names
2. Match district names, within provinces. (this functionality is missing)

Idea: additional argument to topomatch() that performs matches by corresponding groups given two vectors of corresponding province names.
This is a bit fragile. Think about it.