This article is more than 1 year old
Ask your data anything, says data viz biz Tableau as it flings 'patent-pending' tech at natural language tool
Plain language questions, automated data prep promised in first release of 2019
Natural-language loving data visualisation firm Tableau has emitted its latest release, which includes a plain language question tool, developed after its 2017 slurpage of startup ClearGraph.
The tool, named Ask Data, is bundled in as part of Tableau Server and Tableau Online in the firm's latest version, 2019.1, which is out today. Elsewhere in the release is a data management tool to schedules, automates and manages data preparation, a new mobile app, a Google Ads connector for marketeers to analyse website data more easily, and a PowerPoint export tool.
Tableau's sales pitch is that it wants to "democratise data" so anyone in a firm can have a go with it – and Ask Data is the latest such tool, The firm claimed people could type in plain language queries of their organisation's data into the tool and get accurate results, often in graph form, back in response.
The firm also reckoned Ask Data sidesteps some of the criticisms about natural language processing products – for instance that questions will be ambiguous or rely on context the machine doesn't have – with patent-pending technology and, of course, algorithms.
Tableau said Ask Data uses algorithms to profile and index data sources – although chief product officer Francois Ajenstat said that large datasets outside the Tableau stable won't all be indexed, as it would be too performance-intensive and would take too long.
But on those that it does index, the system will use this to offer up results people wants. So, if a person asks for sales data on "American furniture", Ask Data will know to filter the "product category" to furniture and the "country" field to the US.
"It combines statistical knowledge about a data source with contextual knowledge about real-world concepts: 'Furniture' is a common value for the 'Product Category' field and 'America' is a synonym of the country 'United States'," said the canned statement.
Similarly, it means the same results will be returned regardless of whether someone requests "sales" or "bookings".
And if a question could be asked in multiple ways or a person uses a colloquialism, Ask Data will use a parser to "cut through ambiguous language", and combine knowledge about the data source with past user activity and present a number of valid options.
For instance, in a demo for The Register, Ajenstat showed that the system had learnt that when a person types "internet", they might be seeking stats on internet usage or it might be a typo for "interest", so offers both options.
"Ask Data will learn how people ask questions and refine the model over time," he told us.
The tool has been in public beta since November 2018, and Ajenstat said there had been "several thousand customers" using it. The idea, he said, is for this to be used by people who aren't software experts – for instance, those who are analysts and people who are web-based users of governed data.
The tool will only work for English language queries, but Ajenstat said that the firm is working to support other languages.
The data visualisation firm is also considering how to support voice queries – Ajenstat said that setting it up so users can interact with the tool's search bar with voice would be "very straightforward", but that figuring out how to return results – which are often in the form of graphs or contain multiple data points – through smart speakers was "not as trivial".
Ask Data has been developed from ClearGraph, which the firm acquired in 2017 – at the time its third-ever buyout – and Ajenstat said there were more products and tools planned for the course of the year, all aimed at making interaction more conversational.
The firm earlier this month reported its earnings for the final quarter of 2018, ended December 31, posting revenues of $336.3m, a net income of $52.2m and annual revenue of $1.16bn, of which recurring revenue made up $841m.
However, despite the broadly positive picture it painted, the company's shares dropped about 5 per cent after the earnings were posted. The company's shift from one style of financial reporting to another –recognising revenue under a new accounting standard from 1 January onwards – saw the statement branded "overwhelming and confusing". ®