Hotels operate in a competitive, rapidly-changing environment. Increasing commoditization of the hotel product has amplified the competitive nature of the business. It has become very important to understand what makes customers return, or not to return; what makes them recommend a hotel or discourage friends and relatives from visiting; what items, services, and features create value for customers; and its important to create and communicate a clear brand image.
In analyzing hotel consumer behavior, I often refer to research study findings that investigate the impact of technology on hotel guest satisfaction and their likelihood to return. Although hotels ask guests for their opinions in an effort to understand satisfaction and predict future behavior, very often guests leave a property without providing their input. It’s not uncommon, however, for their opinions to later turn up on online review sites.
With the development of social media, online review websites have become providers of rich information about hotel properties. These websites are not only useful to travelers, but also serve as hotel performance indicators to property owners and managers. However, it can be challenging and time consuming to work with these large volumes of information in an attempt to identify particular patterns and trends.
What’s in a Keyword
Enter text mining, a data mining technique that explores textual data to establish meaningful patterns and rules hidden in the data. Text mining may be particularly useful in predictive analytics, and in particular in the analysis of online reviews. A team of university researchers* conducted a study that demonstrates the application of a text mining technique for hotel operations. The main purpose of the study was to analyze online hotel reviews and identify what makes satisfied customers happy, and what can lead to unhappy customers.
For the purpose of this study all reviews for one of the hotel in the South-Eastern U.S. were collected from the Tripadvisor.com. The information was collected using PASW Modeler, an online robot developed for the purposes of this research. A total of 2,511 reviews were recorded. The file contained all standard data fields that appear on tripadvisor.com: hotel name, guest name, trip type, comment, and rating scores for value, rooms, location, cleanliness, service, and sleep quality. Next, all reviews were divided into two categories based on the guest’s indicated intention to recommend or not recommend this hotel to others.
The first analysis that was applied to the document is word categorization. The main purpose of this type of analysis is to compare and contrast the main topics in the reviews of happy and unhappy customers. The top most frequently used word categories in positive reviews were: place of business (in this case, it was the word “hotel”), rooms and staff. The top three categories discussed in negative reviews were place of business (“hotel”), rooms, and furniture.
It’s logical for hotel reviews to contain words referring to the hotel itself and its rooms as the core lodging product. However, we already start seeing differences at the third category. Happy customers switch to intangible components (e.g. staff), while unhappy customers keep talking about the tangible aspects of hotel operations (e.g. furniture). Also, the results showed that the financial category was more dominant in negative reviews. This category included such words as cost, discount, and credit card. This suggests that unhappy customers more are concerned about financial issues (such as the price they paid for their perceived poor experience) than are happy customers (indicating that guests will be less concerned with price if they were satisfied).
Word Pairs Reveal Problem Areas
The study also included text link analysis, which looks at pairs of words frequently used together. Here, the word chosen for evaluation was “no”. The results found the following pairs of words in positive reviews: no wait, no minutes, no complaint, no noise, no questions, no issue, etc. Despite the use of the word “no,” these combinations actually have a positive meaning. On the other hand, the negative reviews for “no” often contained the following word pairs: no room, no balcony, no breakfast, no refund, no care, no return, etc. These word pairs, found as an outcome of the text mining analysis, identified several categories of guest complaints that might require
Text mining may be a useful tool for hotel managers to keep an eye on the reviews of their own, and even competing, properties. Even though technology cannot perform the entire analysis for managers, it can provide a fairly accurate synopsis of what’s going on in a hotel.
* Research team lead by Katerina Berezina from University of Florida, along with researchers from University of South Florida Sarasota-Manatee and University of Central Florida.