I mentioned last month that Poblish has been using OpenAmplify‘s semantic/sentiment analysis service to give technology a shot at making sense of the vast sea of content that is the political blogosphere, in such a way as to help policymakers make better informed decisions. As I’ve said before:
Billions of individual thoughts and personal experiences have been written about, from all conceivable perspectives. No policy process will come up with ideas that have never been thought of before; so existing content represents a knowledge base that should not be ignored
In my piece at Left Foot Forward, earlier this week, I imagined a future in which such tools could take a source article and use this content to automatically, dynamically identify counter-arguments, hopefully before bad policy is made. Well, we have the content, we know that counter-arguments are out there, some of which may very well not yet have crossed the mainstream media’s horizon, and we hope – and believe – that technology can help us find them.
Only a very small percentage of Poblish’s articles have so far been semantically analysed (OpenAmplify are very kindly letting us evaluate their software for free, so the number of articles we process is limited), but all new articles are – and for those articles that have them, Poblish is now displaying the results in the page’s sidebar. Here are the results for the following article.
The way we display the results is simplistic at best, but essentially what we’re showing are the main topics from the article, divided into their relevant category, and coloured as follows:
- Blue: favourable references (or “polarity”). Dark blue for wholly positive (never negative), light blue for generally positive (but occasionally negative).
- Red: unfavourable references. Red for wholly positive (never negative), pink for generally positive (but occasionally negative).
- Grey: neutral references, or a mixture of positive and negative ones.
Clearly there are successes and failures in the above list. Sunny Hundal‘s name appears as a mere noun, rather than a human name (though I wonder if the fact that his surname was misspelled in the original article is relevant here) and some of the polarities seem a little random.
Bear in mind, though, that each set of results you see was the result of an analysis of one, single article, without any context. Give the tool 200,000, however, and we can be certain that insights will start to massively outweigh mistakes. Context is critical, and – just as we don’t judge people or texts on the basis of what we objectively see – semantic applications should not be regarded in isolation, but as part of a vast network of humans and machines, using different techniques to identify and weave links between pieces of information, gradually improving our understanding of them.
All in all, the questions I’m interested in are:
- Do we believe semantic analysis can work?
- Do we believe that it can reveal insights that it would be impractical for human beings to find?
- Do we believe that those insights might be just the ones we need?
- Is it worth us investing more in such solutions?
I’d offer a yes to each of those questions, and have had a lot of fun evaluating OpenAmplify, but: what do you think?
