Natural Language Processing: The Death of Unstructured Data
The Quick & Dirty
This week’s spotlight hits Natural Language Processing.
Our biggest challenge: endless amounts of Unstructured Data.
It’s like the dark matter of the Internet. It’s the black gold for almost everyone: medical researchers, lawyers, search engines, city planners, and yes, marketers.
A perfected Natural Language Processing platform paired with automated analytics and tomorrow’s chatbots could potentially scale in such a way that, quite possibly, a company could generate a new, hyper-informed, always-on cognitive marketer for every single customer and every single potential customer.
The Bigger Picture
Two of the biggest focuses for developers of cognitive computing are:
- Natural Language Processing
- Image Recognition.
Because of our biggest challenge: endless amounts of Unstructured Data. It’s like the dark matter of the Internet. It’s the black gold for almost everyone: medical researchers, lawyers, search engines, city planners, and yes, marketers.
As Alissa Lorentz wrote for Wired, the proliferation of data has created “a data mining gold rush that will soon have companies and organizations accruing Yottabytes (10²⁴) of data.”
If that’s hard to swallow, try chasing it with the following quote from Phil Fernandez, Chairman and CEO of Marketo, writing for the Huffington Post:
That wealth of data can bring us closer than ever before to our customers, and give us the ability to build real relationships with every customer based on what they’re saying — and not saying. It’s the Holy Grail for marketers.
Whoever can conquer unstructured data can spot patterns, create meaning, find solutions, and finally create solutions for some of the marketing world’s most significant hurdles, which includes analytics, social media monitoring, sentiment analysis, content creation, and personalization, so much of which depends on Natural Language Processing.
The growth of marketing technologies requires new ways of thinking, acting and strategizing. And to tame the complexities and capitalize on the potential of a digital world, marketers need to embrace software-smart management frameworks.
If you think this new wave of marketing technology is just hype, consider Brinker’s 2016 Marketing Technology Landscape Supergraphic of 3,874 company logos. Before you digest it, though, also consider that his 2011 Supergraphic comprised only 150.
Today we’ll focus on natural language processing (NLP), and next week, image recognition.
Natural language processing and its sister sector, natural language generation, is a field concerned with interactions between computers and natural language. (For our purposes, we’ll lump NLG in with NLP.)
A perfect natural language processing system would comprehend all the vagaries of every spoken and written language, regardless of dialect, slang, jargon, poor grammar, spelling mistakes, or accent. It would also give relevant, informative, and articulate responses or perform other useful tasks like providing analytics or text summaries, among other things.
Machine learning is the only way to get there. No amount of manpower could hand code every instance of linguistic ambiguity, slang, or accent.
A few fun examples I found on Quora illustrate such sentence-parsing problems.
A car flew off a bridge.
Not literally flying.
I liked your product as much as I like being poked in the eye with a stick.
Not a positive sentiment.
Or grammar that looks, at a glance, to be identical but in fact isn’t.
Time flies like an arrow.
Fruit flies like a banana.
THE MICKEY TEST
And as far as AI assistants go, it’ll be a long road to service the entire world.
The famous Turing Test posits that a genuine AI could talk with a human without the human knowing she’s talking to a computer. To do that, however, the computer must understand what everybody is saying.
I propose an intermediate test, The Mickey Test.
When a computer can successfully understand Brad Pitt’s character Mickey the Gypsy from the film Snatch (2000), Natural Language Processing will be on its way.
The extreme of cognitive computing in marketing is probably at McCann-Erikson Japan, which “employs” the world’s first AI creative director, AI-CD β.
AI-CD β was developed through McCann’s Creative Genome Project, part of a taskforce of Millennial employees, who built their own creative director by deconstructing, analyzing, and tagging a large swath of TV commercials. This was inspired by Netflix’s use of data to find what most viewers would be interested in watching as they launched their own content, which is how they decided to pull the trigger on House of Cards over anything else.
As of yet, no announcement of any AI-CD β creative triumphs has surfaced.
More conservatively ambitious marketing related tasks are being farmed out to cognitive-powered natural language processors, though.
For instance, targeting and perfecting personalization are among the hottest topics in marketing technology.
Facebook “has more recently created an internal platform to harness artificial intelligence so it can deliver exactly the content you want to see.” (Recent allegations don’t call into question the system itself so much as what can be done despite the system, such as skewing news and trending topics to a particular political affiliation.)
Lenovo used data mining, Natural Language Processing, and analytics software to learn more about what their audience wants, which led to the development of the first Yoga tablet.
Google-backed startup Signpost just released their cognitive-powered customer tracking software, Mia. As founder, Stuart Wall told TechRepublic,
Mia automates [the process of uploading customer information], keeps track of every customer, prospective customer, and taps into email and transaction data, so there’s a perfect customer record that’s cross-referenced and always up to date.
Marketers can expect to increase customer engagement across the entire lifecycle by getting to true one-to-one personalization at scale.
To get there, though, there’s a lot of sense-making to be done with all those unimaginably expansive reams of unstructured data.
Rocket Fuel is one of the many trying to get a full view of every customer.
In their own words, Rocket Fuel are “a full Programmatic Marketing Platform designed to go beyond 1:1 marketing by learning to predict what marketing actions to take with a particular person in a particular moment of time.”
Their ad-optimizing ideal, their founding principle, is called “omni-faceted optimization,” by which they mean to use all data at the service of solving a substantial swath of marketing challenges.
As written on SearchCio:
Today’s technology-connected customers care about wait times, ease of use and responsiveness. The deeper the technology is integrated into their lives, the more they’ll expect from the companies they do business with, [Rick] Davidson [president and CEO of the consultancy Cimphoni] warned. CIOs should begin to consider the machine learning and artificial intelligence (AI) investments they’ll likely have to make to meet evolving customer expectations
But as challenges mount, so do solutions.
A fully developed natural language processor and generator would represent a gargantuan leap in pretty much every facet of life and business, powering everything from smart homes and personal assistants to cutting-edge research that would’ve otherwise required thousands of human hands and millions of man hours.
The beauty of this for marketers will be what myriad insight could be gained from a deep analysis of every bit of language ever recorded.
A perfected Natural Language Processor paired with automated analytics and tomorrow’s chatbots could potentially scale in such a way that, quite possibly, a company could generate a new, hyper-informed, always-on cognitive marketer for every single customer and every single potential customer.
Join us next week as we dig into the other big hurdle into wrangling Unstructured Data: Image Recognition.
Originally published at persado.com on May 17, 2016.