Reuters stories from 2014 have lots of weird NYT theme tags on

[rahulbot]

After tagging a big batch for Léo, he dug in and found a lot of Reuters stories had weird tags. For instance, story 636841788 is about the Syran conflict but is tagged with things like "fish and other marine life" and other random tags.

When I run the text of that story through the predict-new-labeller tool manually, all the tags > 0.2 make sense. So what happened? Not sure.

First step is to run this specific story as a test though the same side-band tagging harness and see what comes out.