Text Analytics in Hospitality
By Jeff Catlin, CEO - Lexalytics
As true as the idea is, I find this anthem frustrating: what does "listen better" really mean?
- Who is talking about my services?
- What are they discussing?
- How are they feeling?
These questions — the who, what, and how — represent a more comprehensive understanding of the elements of customer satisfaction, and answering them in full will reveal a more complete picture. But listening to each individual customer is hard when you have thousands of reviews and comments to sort through. Manual analysis is time-consuming and carries many challenges and drawbacks.
To accurately hear your customers' voices, you need the modern marvel that is automated text analytics. Today, I'll explain why.
Online Reviews are a Big Deal
Let's begin by agreeing that the Internet — and the reviews on it — are enormously influential. In 2013, Travel Weekly reported that TripAdvisor and Yelp claimed over 200 million and 39 million unique visitors each month, respectively — and in 2014, TripAdvisor's TripBarometer survey found that 95% of United States travelers say that reviews influence their choice of hotels. An independent study commissioned by TripAdvisor a year earlier went even further, finding that 80% of travelers read at least 6-12 reviews before they choose a hotel; and still another survey indicated that for 29% of consumers, positive online reviews are the most important factor in their booking decision.
Any business serving the hospitality industry needs to constantly be listening to what their customers are saying. Not long ago, this meant analyzing surveys and questionnaires delivered to the customer at a "moment of truth" — after they completed a purchase, read a number of articles, or something similar. But surveys are inherently limited in their scope: questions are closed and answers are typically chosen from a list of pre-written responses. Surveys and questionnaires are quick to create, simple to distribute, and easy to interpret, but they offer a black-and-white picture of customer opinion that shows none of the greyscale that makes up the reality of day-to-day life.
The Internet offers consumers the chance to write far more nuanced opinions, above and beyond the simple options that pervade mass surveys. Online reviews allow for a critical phrase: "yes, but". Consider this sentence, a common part of any review you might find on TripAdvisor:
"The room service staff was friendly, but the food took a while to get here".
Simple survey questions don't allow for multifaceted responses like the above. That one-line review offers magnitudes more actionable insight than a simple question-and-answer like you might find in a survey:
Q: "Room service was attentive and agreeable." A (choose one): "Strongly Disagree /Disagree /Neutral/Agree/Strongly Agree".
Which response do you want guiding your customer service?
Manual Analysis is the Past
The early days of sentiment analysis utilized manual systems: real humans reading through individual reviews, flipping between travel websites, checking news outlets for mention of a brand. But manual analysis is tedious and inefficient. Combing through thousands of reviews and articles by hand takes enormous amounts of time, and brings no guarantee that you gather all of your mentions or score them reliably (human productivity is notorious for being influenced by coffee). Today more than ever, the resources required for manual analysis aren't worth the investment. Accuracy of manual analysis is high (humans average 80% accuracy in sentiment analysis), but recall (how many mentions you find out of every existing mention) is low and the opportunity costs are staggering.
Let a Computer Do the Work for You
Automated text mining tools built for businesses erupted onto the tech scene ten years ago and immediately revolutionized the way businesses listen to their customers. Computer analysis is the single best way to sift through online reviews, article mentions, Tweets, comments, and any other text you want. You're no longer limited to close-questioned survey responses; now, those informational treasure-troves of open-ended responses and long consumer reviews are within easy reach. Text analytics engines process hundreds of thousands of documents per day, as accurate or better than human analysts. The kicker: computers don't get bored or frustrated, don't take sick days, and never forget their morning coffee. Automated text mining is quick, effective, and reliable, and contains none of the drawbacks inherent in manual analysis.
What is Text Mining?
There are a few technical terms to understand: Text mining, otherwise known as text analytics, refers to computer processes that turn unstructured text (reviews, Tweets, comments, etc.) into structured data. Text analytics involves breaking text down into components (entities, themes, concepts) that offer insights into what the original writers were discussing.
Natural Language Processing — interpreting text that doesn't conform to traditional rules of grammar — is the method by which modern text analytics engines interpret all of the vagaries and colloquialisms of modern human conversation. Language is constantly evolving, and so are text mining tools. NLP is a complicated discipline deeply intertwined with machine learning and the study of computer systems.
Sentiment analysis is the key functionality of text analytics for hospitality. Automated sentiment analysis saves you time and money.
There are many facets to the text mining industry, but none is more important for hospitality professionals than the sentiment analysis of text. Precisely how sentiment is analyzed varies on your chosen analytics solution: text analysis engines will tell you the overall sentiment of a group of documents, imparting a sense of the general tone being conveyed by those particular reviews, articles, or tweets. Refined text mining tools are capable of taking huge quantities — hundreds of thousands of unstructured documents — and analyzing sentiment directed at individual entities and themes.
For an example of how this works, take this hotel review:
Hotel offers free wifi which was great , I think its free wifi in the lobby hence why there is so many people in there. Only negative was the temperature in the room stayed around 24 / 25 degrees celsius the whole time we were there (small room on 12th floor).
Looking at sentiment associated with entities, "wifi" will return a positive result and "12th floor" will return a negative result. After you've analyzed a set of such reviews and seen similar recurring results, you'll want to look deeper — analysis software guides you through the system, showcasing the themes and concepts associated with each individual entity. In this case, you find the "12th floor" entity associated with the theme "climate" through the mention of temperature. You send a maintenance worker to investigate the air conditioning on the 12th floor, and he reports that there is a small defect in an AC unit that's preventing guests from adjusting the temperature.
A small amount of money and a short amount of time later, the air conditioning is functioning normally and you can turn your attention to other matters. But before you move on, you write up a quick apology in reply to the reviewer who tipped you off and make it clear that the problem is now fixed — after all, TripAdvisor found that 87% of travelers had an improved opinion of a hotel after reading an appropriate management response to a bad review. You've moved proactively to avoid more negative feedback, and your prompt response impresses potential customers reading your reviews. All thanks to automated text analytics.
A Wide Lens Can Be Misleading
Sentiment analysis is a powerful tool, but will cost you if misused. If you're processing large quantities of documents, don't be afraid to look beyond the initial layer of analysis. Some words contain many entirely different meanings: say you've analyzed a large batch of reviews, and found that the entity "bat" is mentioned numerous times and with strongly negative sentiment. This could mean that a group of bats are flying around your hotel at night and disturbing your guests — but it could also mean that your guests are having a bad experience at the batting cage down the road. The wide lens of entity sentiment could reference either option, or something else entirely. Narrow the lens, focus on the problem, and you'll distinguish between flying mammals and inanimate wooden sticks.
It's not just about the short-term.
Sentiment analysis of text is influential beyond the short-term. In fact, many of the most meaningful insights come from long-term analyses. By maintaining tabs on long-term sentiment, you can:
- Watch public response to new initiatives
- Identify and monitor recurring issues within certain departments (housekeeping, dining, etc.)
- Even track public opinion of a competitor that just opened its doors across town.
- The applications of automated text analytics are virtually limitless.
- Go analyze some text.
The ability to analyze the sentiment of text is perhaps the greatest tool that modern technology has gifted the hospitality industry. In a world where every customer review is available to every other customer, where each guest knows what each other guest liked and disliked about your service, it is more important than ever that you are aware of and acting on these conversations. Don't spend money on unwanted services and upgrades: with modern text analytics, you'll minimize your costs and improve customer satisfaction by identifying and acting on what matters most to your guests. So get out there, choose a text mining solution, and discover what your customers are really saying about you.
Jeff CatlinMore from Jeff Catlin