Natural Language Processing (NLP) is an area of Artificial Intelligence (AI) that allows a computer to read, recognize and analyze speech or text. When you use Alexa, Ok Google, Siri, or Cortana, you’re using NLP.
When you use predictive text to complete a word/sentence in your Gmail or Outlook replies, you’re using NLP. When you call an airline or a major bank or use a chatbot on your favorite eCommerce site, you’re likely using NLP. NLP has been around since the 50s but it’s gotten far more advanced over the last few years.
How does Natural Language Processing work?
First, human speech or text is converted into a language the computer understands. (Natural Language Recognition.) Then, the algorithms process and analyze it. (Natural Language Understanding.) Finally, The Machine generates a response. (Natural Language Generation.)
Where does all this language data come from?
Chatbots. Social media posts. Inbound emails to Customer Service. Voice mails. Text messages. Documents. Pages on your website. Blog posts. Videos. Reviews. Testimonials. Q&As. Feedback forms. Help Desk Tickets. Every spoken/written word you can get your paws on.
How does The Machine really understand the language?
It doesn’t. Yet. NLP turns human words into numbers so that The Machine can analyze the words. The AI looks at the number of words, the language; the sequence (word order); how the words relate (read: are structured) to each other, and so on. The Machine doesn’t comprehend; it uses statistics and probability to classify and predict. (In the interest of fairness, you could debate this with examples of supposed sentience, but they’re not widespread enough to be considered, especially from a marketing perspective.)
Why is NLP so important to marketers?
Natural Language Processing helps us better understand our audience (customers and prospects.) Using oodles of past and real-time data, NLP enables you to understand how your audience speaks; what they ask/inquire about; how they feel about your products/brand; what they like/dislike about your company, and frankly, in general; what their motivations are (read: what their propensity to purchase/act is); and more.
Analytics (like Google Analytics) show you what’s happening on your site. When appropriately used, NLP tells you why. Over the past few years, NLP has dramatically improved when it comes to deciphering what a customer says/means. It’s solid at handling typos and spelling errors, slang, dialects, accents, language translation, colloquialisms, and some sentiment (identifying anger, for example.) More importantly, it’s gotten lightyears better at assessing intent.
How are Marketers using Natural Language Processing?
Content Generation and Optimization
Brand Monitoring/Sentiment Analysis (what people think of/say about your brand)
Search Engine Optimization (Keyword research, taxonomy, meta descriptions)
Chatbot and Response Optimization
Credit/Payment Options Scoring and Targeting
Lead Generation Ranking/Scoring
Survey Analysis (open-ended surveys)
Review and FAQ development
Evaluating and Understanding Audiences
…and lots more!
As a marketer, how do I start using NLP?
First, figure out what you want to know. (This sounds like a throwaway answer, but so many people skip this and live to regret it.)
Ask yourself: “what’s THE question that I need to be answered?” (ONE question. Not all the questions.)
Next, figure out what sources you will use to get your data.
Finally, mine the data and evaluate it using an outside package or something you’ve built internally.
Some consultants recommend that you develop a solid hypothesis before you start. Sometimes that works. Other times it doesn’t. It depends greatly on how attached you are to your hypothesis and whether that attachment is biasing you and/or the project. NLP (as we know it today) is still very young, so it often chases things you might not see if you’re fixated on a specific outcome. (You see this when doing sentiment projects or trying to prove something worked. Confirmation Bias x eleventy bazillion.)
If you’re the type of person who needs a hypothesis and, more importantly, an answer to whether what you did was right or wrong, be sure to build in some testing. Also, please make sure that what you’ve found is repeatable. If it’s not repeatable, there should be a rock-solid reason why.
If you’re the type who prefers to go in with a hazy vision and follow the most exciting rabbits down their holes, be sure to rank/organize your findings from most important to least. (You’ll likely have many more findings than the person with the hypothesis.) Also, be sure that the numbers support your findings (in other words, it’s statistical, not anecdotal) and that the numbers are repeatable. As an aside, many companies find it helpful to have both types of people on their team.
What NLP projects are easiest for Marketers to start with?
As tempting as it is to jump right into your plan for World Domination via NLP content generation, begin with a handful of more manageable tasks first. Things where your Question is crystal-clear, and you can learn how the process works. You don’t necessarily need to know how the sausage is made, but you should know the ingredients, their quality, and what you can/should add or subtract to get the best product.
Specifically, that would be things like processing inbound emails, chat logs, and voice calls to see what customers want/need more information about; SEO tasks (keyword research, meta tags, etc.); competitive research about ONE attribute (pricing, shipping, quality, distribution, etc.); and simple (hot-cold, not ranking) lead scoring.
Incidentally, the better you understand the language clues you get from NLP tools, the more powerful you can make them. (Language clue examples – context/intent of search queries, keyword combinations, positive/negative/neutral sentiment, and the like.)
Have questions about Natural Language Processing? Have a tip you’d like to share? Tweet @amyafrica or write firstname.lastname@example.org.