One of the cool things about the last half decade is the influx of people and technology from acquisitions. You’ll recall that this team’s original incarnation included two PeopleSoft people, Paul (@ppedrazzi) and Rich (@rmanalan).
I’ve found over the years that one cool byproduct of acquisitions is hidden technology gems. You know, projects or products that you didn’t know a company was doing or ones that didn’t seem obvious, e.g. Sun SPOTs, Cloud Office, Endeca’s Information Discovery capabilities, or RightNow’s Natural Language Processing (NLP) engine, Intent Guide.
Maybe you knew, but I didn’t. And if so, kudos for being so well informed.
Anyway, these gems, hidden or otherwise, can help existing applications by injecting new technology into them.
NLP may not grab as many headlines as something like the latest iPhone, iPad or Google Glass, but it can and does help make devices and their software easier for users to use. Many of us have changed our approach to search to suit the tool, e.g. the number of keywords in the average Google search has been increasing for years as users adapt their searches to fit Google’s engine. If you’re like me, you translate your mental question into a keyword string that fits what you think Google will understand best.
These mental gymnastics are common for frequent information seekers like me, but not for the average user, which is where NLP can help. Its goal is to understand a conversational question and provide answers without forcing iterations. If you’ve used Android’s voice features, Siri or have ever crossed horns with an automated telephone system (“I think you said spay my pill. Is that right?”), you’re familiar with NLP and with its decidedly mixed results.
These experiences run the gamut for users, which is why I’m interested in NLP, both good and bad.
I first heard about RightNow’s Intent Guide in a project introduction right after I joined Applications User Experience. After mentally high-fiving myself, I made a note to remember that because a) good NLP is hard to find and b) Intent Guide has RESTful APIs, making it perfect for the type of work we like to do.
So recently, Misha (@mishavaughan) went on a trip to Amsterdam and met some nice people who came over in the RightNow acquisition. She, correctly, assumed that NLP would be nerd-candy for me and was kind enough to connect me with one of those people, Margaret Salome.
Intent Guide uses its NLP engine to help online customers seeking assistance get the most relevant answers possible, quickly. Makes sense, right, but easier said (pun!) than done. Intent Guide has to account for many languages and both long and short queries, some real sentence questions, others Google-style keyword phrases. Oh, and then, there are the inevitable typos too.
But wait, there’s more. There are also industry-specific terms to consider, e.g. cashing a check or drawing money have different meanings in context, and also customer-specific brand names that may overlap with language definitions.
All of this has to happen quickly to give the best experience. Users want answers, immediately. So, relevancy and speed are what they want.
I would go on, but there’s a whitepaper that describes all the inner workings and coolness. However, I’m having one those exact moments with the Google, and I can’t seem to find it.
Update: Turns out, the whitepaper I have isn’t publicly available, which explains why I couldn’t find it. Never fear, there are several whitepapers and data sheets over on the Intent Guide resources page that will give you all the specs and goodness you crave.
Anyway, turns out Intent Guide is used by KLM Royal Dutch Airlines, so I had the opportunity to test it out live.
When you hit KLM’s home page, you’ll see a search field with an Ask button, which is where Intent Guide lives. The first question that came to mind was “what operating system does the in-flight entertainment system run?” suitably nerdy, and driven by a conversation Jeremy and I had last week, so top of mind.
Intent Guide results pop over the home page, and it did quite well interpreting my rather dumb question as a query about the in-flight entertainment. Nice. I then asked it a more germane question, “can I carry on a stroller?” which it perfectly identified in its results. It also adjusted for my colloquialism, matching “carriage” and “pram” to my Americanism, exactly like you want an NLP engine to do.
I played around a bit and discovered that KLM has Android and iOS apps; discovery is one of the things you want from NLP. I also tested its tolerance for typos, another must have. It performed well, e.g. “can haz android?” did return mobile-related results, until I tried too many typos, e.g. “do yo hav android” failed. Probably a good thing.
Anyway, after some light testing, I’m impressed and looking forward to finding a use for it in one of my percolating projects. NLP definitely helps bridge the gap between information and seeker, and when applied to datasets and user profiles that are relatively predictable, it can save a ton of time and frustration.
Imagine applying NLP to all the data in an enterprise. Then compare that to a trillion or so indexed web properties, or the Facebook status updates of a billion users. This is another one of those areas where enterprise users can benefit before consumer users.
Find the comments.