NeutrinoBI: Natural Language Search Meets Information Discovery
UK-based Neutrino Concepts is taking search in a new direction -- delivering answers to questions, not a ranked list of results -- and helping users start data discovery in a whole new way.
- By Stephen Swoyer
- April 16, 2013
If you think that business intelligence (BI) vendors have licked the problem of search, you haven't spoken with the folks at UK-based Neutrino Concepts Ltd.
Neutrino markets NeutrinoBI, a natural language data discovery technology.
NeutrinoBI's model isn't so much a Google-like search but search like Wolfram-Alpha does it.
A more familiar example might be search à la Siri, the spoken-word natural language search (NLS) facility for Apple Inc.'s iOS platform. In any case, Neutrino officials say, users interact with NeutrinoBI the same way they'd interact with either of these technologies: by posing natural-language questions.
"[Neutrino] started out in 2007. Our founders came from a very technical background [where they] were working with large-scale data warehousing projects. They came up with the single tenet of merging search engine technology with traditional BI tools," explains Jonathan Woodward, Neutrino's CEO.
Neutrino wasn't conceived as a conventional BI search tool, Woodward maintains. For example, he says, it isn't designed to return full or partial matches in response to specific search terms, nor is it intended to function as a knowledge or content management tool that makes it easier to find information in reports, scorecards, or other business information artifacts. The difference between conventional BI search and NLS is that the latter attempts to return answers to questions instead of a (ranked) list of results.
"The focus was on natural language search, [with the idea] that you could discover [information] without having to understand the structure of a database -- without any coding or without any knowledge of SQL," he explains. "We don't convert to SQL; with [natural language search], you're not limited to set-based theory of SQL."
On the other hand, anyone who's used Wolfram Alpha -- to say nothing of Siri -- knows that this doesn't always work as advertised. Ask Wolfram Alpha to rank the five largest countries by land area, for example, and it'll (quickly) give you exactly what you asked for; ask it to return the average selling price (in dollars, Euros, or any other currency) of a quart of milk and you'll have a lot less luck. For this reason, officials position NeutrinoBI as a starting point for discovery. Ask it for the answer "to the ultimate question of life, the universe, and everything" and it won't (like Wolfram Alpha or Siri) respond with "42."
Instead, it generates interactive visualizations. The emphasis, says Toni McAlindon, operations director with Neutrino, is on self-service exploration, analysis, and synthesis: users can drill-down into and combine visualizations, in the process creating new mash-up views.
"When you type in a natural language search -- for example, 'show me the sales of fruit in the UK' -- what [NeutrinoBI] will do is go to our semantic layer, which has been built from talking to different databases, building an index, and doing all of the different combinations of visualizations that could exist," she explains. Neutrino's semantic layer is, in effect, a federation or virtualization layer. "It builds all of this into the index, then when you [perform] the search, it will pick out all of the [approximate matches] and render the data."
The results of an NLS -- including both analysis and mash-ups -- can be persisted back into NeutrinoBI, McAlindon explains. "You can, with your fingers, take two different data sources, two different charts, ... [and] generate on the fly a new chart that's fully managed, [which] fully reflects [any] changes [that you've made]," she explains.
Right now, Neutrino positions itself as a complement to (and not as a replacement for) the data warehouse -- much like other discovery-oriented vendors. Data in the DW can serve as a primary source for discovery -- especially when it's blended with data from other sources.
"There's just no way people are going to get it all into the data warehouse anymore. That's now gone. What you need to be able to do is you need to be able to pick and choose from different data sources based on your different use cases," she says.