«Bachelor dissertation Alina Varfolomeeva Institute of Hospitality Management in Prague Department of Hospitality Management Major field of study: ...»
Companies that run analysis on a manual basis might gain a deeper understanding of customer feedback, but it tends to be inconsistent and timeconsuming when reviewing large quantities of data (Ziegler, Skubacz and Viermetz, 2008). At the same time, some of the companies that have adopted text mining techniques have failed to realize their expectations in using this method (Fenn and LeHong, 2012). Nevertheless, a proper use of text mining models has clear managerial benefits, including the availability of accurate and timely information, for better informed decision making.
The academic research on a subject of efficient use of text mining in hospitality sphere remains rather scarce, but at the same time – multidirectional.
The basics of application of text mining techniques in the hotel industry were discussed by Lau, Lee and Ho (2005). They propose text mining as a means of information management, and state its ability to analyse voluminous textual information that can be found in hotels internal database and on external websources. They also address the implementation cost of a text mining software and possible future technological advancement.
2.4.1 Sentiment Analysis Very close to the sphere of hotel service reviews are product reviews, which were discussed in the context of text mining by Hu and Liu (2004). They used methods of classification and summarization to mine product features, which were commented on by the customers. Then a sentiment analysis was used to identify opinion sentences and decide if each opinion sentence is positive or negative.
In the hospitality sphere a sentiment analysis was used in a research of Kasper and Vela (2011) to present a system which detects, collects and analyzes hotel customer comments, obtained from the Internet, in order to reveal classified and structured overviews of those comments. The system allows hotel managers to monitor what is published on the web about their establishments.
The majority of the studies which used a sentiment analysis have emphasized how both linguistic and nonlinguistic text mining methods (Taboada et al. 2011) contribute to better predicting the overall sentiment in a customer review.
Despite the importance of identifying sentiments, more specific information could be contained in textual customer feedback. With sentiment analysis, the program focuses solely on the final sentiment in the review, but with multiple emotions in the comment, this is not very useful, as part of the review, which might contain opposite sentiment, is neglected. However, if the focus of text mining analysis was connected to the different hotel departments and related customer interactions (e.g., with the staff, the bar, and the atmosphere), sentiment outcomes would have greater value for the future use (Ordenes et al. 2013).
2.4.2 Other Cases of Text Mining in Hospitality In 2006 Segall and Zhang present their research which applies the techniques of web mining to actual text of written comments for hotel customers using Megaputer PolyAnalyst®. Web mining functionalities utilized include clustering, link analysis, key word and phrase extraction, taxonomy, and dimension matrices.
Segall, Zhang and Cao (2007) discuss different features of two text-mining software tools: SAS® Text Miner and Megaputer PolyAnalyst®, specifically applied to hotel customer survey data. The focus is made mainly on technical side of the programs, rather than practical analysis of real hotel customer reviews.
Bjørkelund et al. (2012) study opinion mining on a base of travelers review websites. They analyze and visualize results of opinion mining studies using Google Maps in order to detect good and bad areas of tourism according to the amount of positive or negative reviews. An evaluation of the techniques presented, shows high accuracy in opinion mining, and that the prototype can help detect hotel features and possible reasons for changes in opinion as well as show "good" and "bad" geographical areas based on hotel reviews.
Tsujii at al. (2014) discuss a method to analyze hotel customer reviews in order to reveal common characteristics and examine ways to present beneficial information for the service improvement. In their paper, they extract information of evaluating points for foreign tourists from the hotels reviews posted by foreign tourists.
Lee, Singh, Chan (2011) reviewed service failures and ways to recover from them in the hotel industry with the help of keywords extraction.
Boiy and Moens (2008) present a machine learning approach to multilingual textual information and investigate on how to improve a success rate on drawing sentiments from multilingual set of textual documents.
Aciar (2009) addresses the problem of adapting natural language to the form which can be understood and utilized by computers for further identification of contextual information from consumer opinions and incorporating it in a system of recommendations. She uses classification text-mining techniques to create rules for obtaining contextual information, which were then used to sort through new reviews based on similarity to the previously analyzed documents.
Haruechaiyasak et al. (2010) discuss domain-dependence of opinion mining and design a framework for a feature-based opinion mining of Thai hotel customer comments. They construct a lexicon containing both domain-dependent and domain-independent types, and then the lexicon is used to build a set of syntactic rules based on the frequently occurred patterns. Based on this set of rules computer programs are able to extract more of the similar features from untagged textual documents.
Research team lead by Katerina Berezina from University of Florida conducted a study that demonstrates the application of a text mining technique for hotel operations. The main purpose of their 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 their study all reviews for one of the hotel in the South-Eastern U.S. were collected from the TripAdvisor.com.
2.5 Customers Feedback The nature of customer feedback can be classified as explicit or implicit depending on whether customers consciously or unconsciously provide a third party with information about their experiences (Ordenes et al. 2013). Companies collect explicit feedback through different channels (e.g., surveys, e-mail, online reviews), where they receive information directly from customers.
Customers give their implicit feedback through determined actions, without even knowing about it and without the hotel or any other company requesting the information (i.e., eye tracking, reading time, number of scrolling, number of clicks obtained in a webdocument) (Poblete and Baeza-Yates 2008). Both types of feedback contribute to a continuous learning process about customers.
Many establishments collect their feedback using quantitative methods because of how simple it is to analyze structured information. For example, questionnaires help companies analyze predetermined product/service quality attributes (Parasuraman, Zeithaml, and Berry 1988). Despite the importance of these data, using only predetermined attributes will inevitably result in only partial understanding of the customer experience (Macdonald et al. 2011; Vargo et al.
2007). Classifying reviews by predefined categories and using them to collect structured feedback provides companies with only the top of the iceberg – a small part of information about the entire customer e xperience (Caemmerer and Wilson 2010).
Nowadays to contrast these quantitative approaches to collection of structured data, technological advances offer customers new channels and social platforms through which they can share their feedback in a textual and unstructured format (Witell et al. 2011). Such form of feedback includes e-mails, online reviews, responses to open-ended questions, and social media conversations (Witell et al.
2011). In this case, the customer himself defines the process and timing of feedback and whether or not the context should be provided or it is enough only to give sores to the product. Arguably, this type of feedback is of greater relevance than structured approaches because its expression and content better reflect customer motivation (Belkahla and Triki 2011).
Taking into account that customers play active roles in generating relevant information on any product or service provides more valuable and complete sources of insight to companies (Belkahla and Triki 2011; Wirtz, Tambyah, and Mattila 2010). However, analysis of this information demands significant amount of time required to generate knowledge from large amounts of qualitative data (Janasik, Honkela, and Bruun 2009). Feedback from e-mails, customer reviews, short messages, and social media is constantly growing, leading organizations to develop more efficient approaches to measure and understand information (Zhan, Loh, and Liu 2010). Text mining methods are becoming more and more helpful as they offer a potential solution for dealing with great volumes of unstructured data in a textual form (Ur-Rahman and Harding 2011), whether such data are explicit or implicit and solicited or unsolicited.
Chapter 3. Methodology
3.1 Research Design With the aim of analyzing hotel customers comments using text mining software and comparing the results of manual and machine processing of the feedback, this study was carried out in Prague on the example of a four-star hotel Assenzio, to which the author of the thesis has an access as a front-office employee. The research was undertaken in the following steps: formulating of hypothesis, collecting reviews from TripAdvisor LLC, analyzing qualitative data, case study research.
In the previous chapter an extensive literature review was conducted to find out the direction of research, based on the already known information about text mining and gaps or controversies uncovered during the analysis of literature on the subject.
The hypothesis was developed based on literature review.
H1.: Text mining is a more precise and efficient method of analysis of customer feedback, than manual analysis.
H2.: Text mining can be used separately from other analysis techniques and present good results.
Having noticed the limited study on the efficiency of use of text mining in the hospitality sector, especially in Czech Republic, it was decided to follow the previous studies, made using TripAdvisor.com to mine customers’ feedback.
The purpose of this research is strictly exploratory, as it presents only possibilities of use of text mining in a real hotel establishment, and is limited to one particular hotel in Prague, Czech Republic.
The study utilizes inductive approach to gain new insights in the sphere of analysis of hotel customer comments, and uses qualitative methods to generate nonnumerical data.
3.2 Data Collection Procedure 3.2.1 The Object of Case Study The current research focuses on the analysis of guests’ reviews, made about hotel Assenzio****. Hotel Assenzio Prague is four-star boutique hotel, situated in a historical building in a Prague’s New Town; it is only 5 minute walk away from I.P.Pavlova public transport station, and about 15 minutes from Wenceslas Square. It is ranked #246 from 662 hotels in Prague on TripAdvisor on the moment of writing this study.
Hotel Assenzio**** is a medium sized lodging facility with the total of 69 rooms, including single, double and triple rooms, as well as junior suits and luxurious double mesonettes. Continental buffet breakfast is served in the hotel restaurant.
The hotel offers the standard hotel services, such as free wi-fi, lobby bar, laundry, concierge services, airport transfer, etc.
Hotel Assenzio**** was chosen for the case study, as the author of this thesis is a front-office employee of the hotel, which enables the research results to be compared to the real state of affairs. Moreover, the insight knowledge on the researched hotel can help with understanding of changes in reviews over time.
3.2.2 Collection and Preparation of Reviews As was stated above, for this thesis the case study research was chosen as a research strategy. Case study is one of the most commonly used qualitative methods of research. Rather than testing existing theory, it is used to develop a new insight on the research question. It might not be representative of the wider world it belongs to, but case study can be used as a pattern for a further research.
To start of the case study, reviews in English language were retrieved from TripAdvisor.com about hotel Assenzio****. The total of 44 English reviews from the period of 2012 – 2014 was collected for further analysis. The reviews were saved in a TXT format.
Some additional configurations were made with obtained reviews in order to ensure the best results of the case study. First, information not connected to reviews was removed from text files, such as excessive hyperlinks, repetitions of the beginnings of reviews and other text, which might make the case study research inaccurate. Then, customers’ feedback was printed out for read-through and manual analysis. After that, each review was saved as a separate file to be used in text mining analysis.
According to the findings of manual analysis most often guests are complaining
about the following (see Table 1):
Table 1 – Most common negative review groups
Air conditioning has appeared to receive the greatest number of negative comments; at least third of the collected reviews mention it. It was described as loud, poor, not working and non-existent. However, in a couple of other reviews stated that air conditioning was good. As both negative and positive reviews on this subject were made in the same time period, there is no correlation with any technical problem with it in the whole hotel. Further in the case study, showing the use of text mining, we would suggest possible solution on this question.
Nearly the same amount of negative reviews was received on the location of the hotel. Main concern for the guests is that hotel is situated next to a noisy street.
Other problem, mentioned by some of the guests is that hotel is located a bit far from the transport stop and too far from the center and historical part of the city.