A few weeks ago I received a pitch saying that Lexalytics, a B2B provider of sentiment and text analytics technology, powering brand and reputation management tools such as Radian6 and Scout Labs, had recently upgraded their Salience product by digesting all of Wikipedia to come up with their patent-pending technology, Concept Matrices. This product, they said, enables us to understand the complex relationships between words and their meaning, which provides valuable analysis to brands.
CBM: In layman’s terms what exactly is sentiment and text analytics technology?
Jeff Catlin: Pick some text, any text. And as if by magic, we’ll tell you who is being discussed, what the conversation is about, and if it’s positive or negative for those involved. This gets used by all kinds of different organizations – to see what people like, what they don’t, how well a product is working, if they need to put more beef on the hamburger, and, of course, by Big Brother to See If You Are Planning A Terrorist Attack.
On that note – don’t say anything on Facebook or Twitter or any other place on the Internet that you don’t want your grandmother and the government to read.
CBM: How did you get into this fascinating field?
JC: Having been involved in the search business before now; this is a natural next-step to search. Rather than searching for a word, you can start looking for the concepts you actually mean.
CBM: Salience product? Digesting Wikipedia? What’s going on?! How long did it take you to marinate on all of those web pages?
JC: Salience is our software that is integrated into various other systems. Much like man, we are not an island, and our software goes into other systems, systems with lots of text where someone wants to understand what’s happening without reading all of it and taking notes.
Digesting Wikipedia, well that’s the future. Our next release is all about understanding what people are talking about, and what better source to gain that understanding than the world’s largest encyclopedia? We’ve swallowed all of Wikipedia, so that we can understand that a “cat” is closely related to a “lion” and that it’s only very loosely related to a “walrus” (both animals). This sort of knowledge will help us truly understand what people are looking for when they ask questions.
CBM: Give us a simple man’s metaphor for Concept Matrices? Words, meanings, are we talking poetry here? When I say wave, will it know undertow? When I say valley, will it know mountain?
JC: You hit the nail on the head with your examples. When I think of oil, there are other concepts that go along with “oil” – “petroleum”, “gasoline” “natural gas”; and more tenuously related concepts like “solar power” and “wind”. Wikipedia has these relationships implicitly and explicitly stated, and we’re using these relationships to associate concepts like “cereal” to concepts like “rice” as well as concepts like “Frosted Flakes”.
CBM: So one day when I’m thinking of healthy food, will my networked refrigerator beep saying that my raw food breakfast is ready? What does the future of this technology look like in every day life? What are its greatest benefits?
JC: A true artificial intelligence requires a number of things – it requires the ability to understand sound, and know that it’s speech. It requires the ability to see lines and understand that they’re text. It requires the ability to take the words and the meaning of the words and make meaningful, useful connections between the concepts.
There are myriad companies working on each individual chunk of producing an artificial intelligence. Few of them have that as their explicit goal (as there are great intermediate uses for doing something like, say, turning speech into text; or text into meaning).
That’s the grand goal.
But, in the interim, being able to tell your fridge that you want breakfast food when it’s dinner time, have it understand what’s in there, and then suggest that you go out to Denny’s because all you have is ketchup and leftover Chinese food; that’s a very real possibility. In order to get there, your fridge has to understand what “breakfast” is, what food is generally associated with it, be able to cross match to potential recipes, and then look up on the map local restaurants that have food that could be classified as breakfast.
We’re not so far from that.