Measure quality on several organizations

[EgorBu]

To have a better understanding of the quality of the approach let's measure performance in near-real life scenario:

• collect a list of organization with different scale
• apply eee-identity-matching to collect identities based on the current approach
• evaluate it using ground truth data from ghtorrent

[warenlg]

I dealt with this issue while solving #30, and here are the resulting visuals, for 20 open source stacks.

The bubble size being here proportional to the number of unique identities output by the identity matching algorithm.
I removed IBM and intel from the chart because we performed badly on their codebase (~60% precision and recall) and it squashed the rest of the chart. This is under investigation.
These results does not include the current work on bot detection.

[vmarkovtsev]

I consider this task as done. @warenlg can you please create a follow-up with the investigation of IBM and Intel.