Guest Survey Sentiment Analysis and Topic Mining Made Simple

This entry is part 1 of 2 in the series Guest Survey Analysis and Topic Mining  

Guest surveys are an important part of any business that is dependent upon repeat business. And the best surveys are those that have open-ended questions that prompt guests to tell us a little more about their experience. Often times the most useful nuggets of information can be garnered from these open-ended questions. Whether they mention an employee that went above and beyond to assist them or perhaps alerting management to something that requires some attention.


Although guest comments are so important to operations they can be very difficult to process, analyze, and report upon. Especially at an enterprise level. Lets face it. There are many different ways to say that the temperature was too cold at a restaurant:

  • Temperature was cold…
  • A/C was too high…
  • Turn down the A/C…
  • It was frigid in there…

These are all important facts that management at all levels needs to be aware of but can easily be lost in the hundreds, thousands, or even millions of other non-topical words in guest comments.

Another very important hurdle to consider is that often guests will score their experience high and say lots of nice things in their comment. However, they will often add something negative or something that needs addressed. For example, here is a guest comment that gave a good score (4) and is mostly positive but there is a tidbit at the end that management would want to be alerted about:

“We always enjoy our meal at your restaurant. We try to come as often as possible, especially on 2 for $20 night. Service is always fantastic! I normally score it a 5 but this time I gave my visit a 4 since we have noticed that service was slow and the bathrooms are dirty.”

If management is only looking at the “lower” score guest comments (ie 1-3) for negative sentiments, the very pertinent topics of service slow and bathrooms dirty may never be seen and lost forever.

The purpose of this series is to show a method of “munging” through guest comments and extracting the important topics. Then we will work through how to re-evaluate the sentiment of the comment even if the guest tends to score high either to be polite or loyal. Here is a summary of the process. Follow the links below to see the details of each and how the comment above is transformed at each step:

Once we have a count of negative terms, we can use this value in determining the sentiment. In addition, we can take those bi-grams that were produced and count the number of occurrences. If, for example, in the course of a day 1000 surveys contain 30 instances of the bi-gram “cold food” or “slow service”, then our process has assisted in bringing these issues to the attention of management.

Series NavigationGuest Comment Processing Step 1 – Normalizing Phrases >>
Posted in

Leave a Reply

Your email address will not be published. Required fields are marked *