Patentee duplicates

[cverluise]

Issue description

US patents before 1920 (format 1) listed each patentee twice, typically both as part of the header and the body of the text. We annotated both, we also collect both.

The side effect is that, for these formats, some metrics (e.g. size of the team) are deeply affected and could be misleading.

More

This can happen in 2 cases:

• US format 1 ~ 100%
• GB format 1 ~10% because of the provisional and final publication being both on the same document

Details

Version: 1.0.0rc1

[cverluise]

Deduplicate using relative Levenshtein distance on the name_text (iterat over all patentee couples).

• US format 1: 97%+ accuracy with threshold .43 (see doc)
• GB format 1: nothing yet

We created a new field is_duplicate which is True if we found a duplicate. Note that only 1 of the 2, the one bearing the less information, is marked duplicate.

Will be available as of v1.0.0rc2.

Leaving open in case we want to do something similar for GB