Victory from the jaws of “defeat”

Here’s an A.I. challenge: Find all news stories that have to do with Hillary Clinton’s Campaign but focus on her defeat.

If you try to do this with existing text analytic platforms you find that it’s not possible. You get lots of articles about Hillary Clinton’s campaign and how she will still win the election as highly relevant. In order to make this work, you need A.I. that recognizes when a piece of content is no longer relevant due to events that occurred after it was written, rendering the information useless. In our case, that event is “defeat”.

The reason why current A.I. systems fail this test is because the concept of defeat is not a named entity in the text. “Defeat” is described using lots of different words and concepts.

This was a challenge proposed by a large media company to my company, Intellogo. Intellogo offers unique solutions to this challenge. Let’s take a look:

Solution #1 — Give Intellogo a sample article as criteria

Intellogo uses the entirety of the article “Why Hillary Clinton lost the election: the economy, trust and a weak message” from the Guardian to find related content. This is a fast and easy method that gives great results because the source article is so much more than just a keyword. The results are in the middle in green.

Solution #1 Intellogo uses the entirety of the article “Why Hillary Clinton lost the election: the economy, trust and a weak message” from the Guardian to find related content. This is a fast and easy method that gives great results because the source article is so much more than just a keyword. The results are in the middle in green.

Solution #2 — Teach Intellogo the specific concept of “Hillary’s Defeat”

In this example, we teach Intellogo a new concept “AnalysisClintonLoss” (left column, top list item) using several different articles that analyze various reasons for her defeat. The benefit of this approach is that Intellogo learns to recognize stories about her loss that may take different view points. The results, in the center column are high quality using this method.

Solution #2 In this example, we taught Intellogo a new concept “AnalysisClintonLoss” (left column, top list item) using several different articles that analyzed her defeat. The benefit of this approach is that Intellogo learns to recognize stories about her loss that may take different view points. The results, in the center column are high quality using this method.

Solution #3 — Mix concepts together

Finally, we combine two concepts: Hillary Clinton and Defeat (middle column, bottom). The defeat concept is trained to recognize multiple kinds of defeat: sports, financial, love, etc. In this way it’s generalizable and able to be reused for other purposes. The other advantage of this method is that we can mix in more concepts if we want (e.g. the economy) to get even more refined results. You can see the results in the third column on the right. We think these are the best results.

Solution #3 In this example, we combine two concepts: Hillary Clinton and Defeat (middle column, bottom). The defeat concept was trained to recognize multiple kinds of defeat: sports, financial, love, etc. In this way it can be reused for other purposes to generally recognize defeat. The other advantage of this method is that we can mix in more concepts if we want (e.g. the economy). The results are in the third column on the right. We think these are the best results.

News, articles, books and conversations are more than just keywords as this challenge highlights. If we want to build systems that help us find the right content for the right person at the right time then we need to teach A.I. to recognize the more subtle features in our words. I think there are a ton of opportunities like this waiting to be snatched from the jaws of defeat.