Content Copyright © 2014 Bloor. All Rights Reserved.
This blog was originally posted under: The IM Blog
The best thing about being an analyst is meeting companies that are doing seriously cool things. There have actually been quite a lot of these recently but one that has really grabbed my attention is Arria.
Arria is based on research conducted at Aberdeen University in the area of natural language generation (NLG). Note that is different from natural language processing (NLP). As far as the latter goes, this is about interpreting input that is entered using natural language. NLG, on the other hand, is the reverse: it is about taking conventional data, analysing it and then producing a report in natural language. For example, The Met Office is a client of Arria’s and it uses the technology to generate its forecasts. Thus the forecast for my area for today is “outbreaks of rain or showers on and off throughout the day today. The rain could be heavy at times, but there will be drier spells too. Maximum Temperature 20C.” That’s generated by Arria and it can be re-generated as required and it’s done automatically rather than requiring someone to manually write the forecast. Which means that much more localised forecasting is possible and that forecasts have a consistent format.
Let’s take a different example. Another one of Arria’s customers is Shell, which is using Arria to produce reports for preventative maintenance/asset management on its oil rigs. Arria ingests sensor data from the relevant device (in the initial phase Arria is being used to monitor compressors, which average around 120 sensors) and whenever there is a fluctuation from an expected value (for example, too much vibration) on a sensor Arria can generate a report (and alert) for the engineer, not only giving him details of the out-of-range reading but also a history of relevant past readings (for example, has this particular component been on the blink for some time) along with recommended actions. Prior to the use of Arria, warning lights were used to warn of such variances but it typically needed an engineer to spend two to three hours of research before he could identify the problem and a likely resolution. Arria does this in moments.
Of course, it’s not trivial to set Arria up. Solutions are domain and client specific. According to Arria (which used to be known as Data2Text) it takes between 3 and 6 months to configure the rules (which are XML based) that underpin the system, depending on the complexity of the data analytics required. It can be deployed either in the cloud or on site and you can inject third party analytics if appropriate. Reports generated by Arria can include graphics and diagrams where that will be useful.
So, what Arria is doing is going a step beyond business intelligence (in its broadest sense) because it is not just presenting the data to you (lots of pretty charts) but is interpreting it, according to rules that you tell it. I think that’s really valuable.