Big data in the restaurant industry : a sentiment and quantitative analysis of online reviews to study the effect of restaurant attributes and context on customer satisfaction

(1)

Big data in the restaurant industry

A sentiment and quantitative analysis of online

reviews to study the effect of restaurant attributes

and context on customer satisfaction

Author Marlin van Drie Student number 11409428 Date of submission 22 June 2018

Version Final

Track MSc. in Business Administration - Digital Business University University of Amsterdam - Amsterdam Business School Supervisor Dr. Abhishek Nayak

(2)

2

STATEMENT OF ORIGINALITY

This document is written by Student Marlin van Drie who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

3

Acknowledgements

I would like to thank mr Nayak for all his help during the process of writing this master thesis. His feedback and our meetings have been very useful to me. Furthermore, I would like to thank my family and friends for distracting me from time to time and for supporting me unconditionally. I would also like to thank Ramon and Lisanne in particular, for all the time and effort they spend on rating the reviews.

(4)

4

Abstract

Nowadays, before making any purchase decision, customers tend to depend more on reviews from peers than on information from companies. Online reviews are very important for restaurants in particular, caused by the bulk of information reviews contain and by the highly competitive restaurant industry. It is important for restaurants to know their customer in order to be able to respond to their needs and preferences. In order to do so, it is important to know which restaurant attributes drive either positive or negative customer experiences and under which conditions.

This study focused on online reviews in the restaurant industry and makes a distinction between the restaurant experience at home (delivery) and the restaurant experience out of home (eating out). The paper uses online reviews from both Yelp and Grubhub (N=1988) in order to examine the effects of F&B quality, Customer Service, Value for Money, Overall Experience, Location and Delivery.

The results indicate that whether certain restaurant attributes are discussed and how they are discussed has a significant impact on the sentiment score of the review. The attribute sentiment has more power to explain the variance in sentiment score than the attribute discussion. The results show that the order of most discussed topics differ between the two context categories. Only the interaction effects between F&B quality and context and between overall experience and context on sentiment score are significant.

The study enriches the knowledge about restaurant reviews, shedding light on the influence of the most important restaurant attributes on customer’ satisfaction, and how this is influenced by context.

Keywords: Online reviews, Online Food Delivery, restaurant, restaurant attributes, sentiment analysis, electronic word of mouth

(5)

5

Overview of tables

Table 1: Operational definition and measurement method of each variable 29

Table 2: Star rating versus Sentiment Category 30

Table 3: Number of reviews with Attribute discussions per Context category 31

Table 4: Group Statistics of F&B 32

Table 5: Independent Samples T- test for Context and Sentiment Score 32

Table 6: Sentiment score per attribute 33

Table 7: Percentage of reviews with attribute discussions per sentiment category 33

Table 8: Group Statistics of F&B 35

Table 9: Independent Samples T- test for F&B and Sentiment Score 35 Table 10: ANOVA for F&B Sentiment and Sentiment Score 36 Table 11: Descriptives of F&B Sentiment and Sentiment Score 36 Table 12: Univariate analysis of interaction effect of F&B Sentiment

and context on Sentiment Score 37

Table 13: Group Statistics of Customer Service 38

Table 14: Independent Samples T- test for CS and Sentiment Score 38 Table 15: ANOVA for CS Sentiment and Sentiment Score 38 Table 16: Descriptives of CS Sentiment and Sentiment Score 38 Table 17: Univariate analysis of interaction effect of CS Sentiment

Table 18: Group Statistics of Value for Money 40

Table 19: Independent Samples T- test for VfM and Sentiment Score 40 Table 20: ANOVA for VfM Sentiment and Sentiment Score 40 Table 21: Descriptives of VfM Sentiment and Sentiment Score 41

(8)

8

Table 22: Univariate analysis of interaction effect of VfM Sentiment

Table 23:Group Statistics of Overall Experience 42

Table 24: Independent Samples T- test for Overall Experience and Sentiment Score 42 Table 25: ANOVA for Overall Experience Sentiment and Sentiment Score 43 Table 26: Descriptives of Overall Experience Sentiment and Sentiment Score 43 Table 27: Univariate analysis of interaction effect of Overall Experience Sentiment

and on Sentiment Score 44

Table 28: Group Statistics of Location 44

Table 29: Independent Samples T- test for Location and Sentiment Score 45 Table 30: ANOVA for Location Sentiment and Sentiment Score 45 Table 31: Descriptives of Location Sentiment and Sentiment Score 45 Table 32: Univariate analysis of interaction effect of Location Sentiment

Table 33: Group Statistics of Location 47

Table 34: Independent Samples T- test for Delivery and Sentiment Score 47 Table 35: ANOVA for Delivery Sentiment and Sentiment Score 47 Table 36: Descriptives of Delivery Sentiment and Sentiment Score 47

Table 37: Overview of the results 48

Table 38: The mean word count per Sentiment Category, per context 50 Table 39: Correlation between Word Count and Sentiment Score 51 Table 40: Mean sentiment score per Restaurant Category 52 Table 41: ANOVA for Restaurant Category and Sentiment Score 52 Table 42: The mean Sentiment Score per price category 53 Table 43: The mean Sentiment Score per price category 53

(9)

9

Table 44: ANOVA for Price Category and Sentiment Score 54 Table 45: Overview of the correlations between the covariates and sentiment score 54 Table 46: Hierarchical Regression Model of Sentiment Score (Attributes Y/N) 55 Table 47: Hierarchical Regression Model of Sentiment Score (Attributes’ Sentiment) 57

(10)

10

1. Introduction

Whether it is for business or for business scholars, big data is a hot topic. Big data are “the vast quantities of data that flow relentlessly from web sites, databases, information systems, mobile devices, social networks, and sensors” (Kimble & Milolidakis, 2015, P. 23). With the growing use of mobile devices and “the internet of things”, the availability of big data has increased. In the field of marketing research, big data is one of the key priority terms to focus on. The last few years many scholars researched the topic, but it remains one of the most poorly understood terms (Fosso Wamba, 2017; Frizzo-Barker, Chow-White, Mozafari & Ha, 2016). Côrte-real, Oliveira and Ruivo (2017) emphasize the importance of big data and see big data analytics as an effective factor in order to survive in competitive markets. The potential of big data is tremendous, but it appears that just a few companies know how to deal with it. Akter, Wamba, Gunasekaran, Dubey and Childe (2016) argue that big data analytics pays off for some companies, but not for others. In the end, big data can help to predict and ultimately help to make better informed decisions. This paradox of on the one hand the increasing importance and the potential, but on the other hand the lack of knowledge on how to deal, makes big data such an interesting topic for academic research. The emerge of web 2.0 in 2004 led to a new source of data: User Generated Content (UGC). While users only used to be able to read, web 2.0 made it possible for users to interact, which means that they became able to both read and write (Aghaei, Nematbakhsh & Farsani, 2012). There are many forms of UGC, but reviews are seen as the bulk of user-generated content (Ngo-Ye & Sinha, 2014). UGC is a form of unstructured data, like photos and text are. Due to a lack of practical tools to analyze UGC, UGC tends to be ignored in business research (Archak, Ghose & Ipeirotis, 2011; Deng et al., 2012 ; Liu, Singh & Srinivasan, 2016). However, in the form of online reviews, UGC plays a major role in the decision-making process for purchases. Niu and Fan (2018) mention that before making any

(11)

11

purchase decision, 78 percent of Americans between 18 and 64 years old consult online reviews. Surpassing that, Zhang, Zhao, Cheung and Lee (2014) argue that 90 percent of consumers read reviews when shopping online and 83 percent of them are influenced by these reviews.

Reviews can be both quantitative and qualitative. For quantitative reviews, customers give a rating, for example between 1 to 5 stars or a grade between 1 and 10. This form of review is easy to collect, store, process and analyze. Written reviews are seen as qualitative data. These reviews are harder to analyze, but consist of more valuable information (Gan, Ferns, Ye & Lin, 2017). The sentiment of a written review is a relevant determinant of consumer choices. When products or services are only reviewed by a numerical review rating, it is unjustified assumed that the products or services are one dimensional, while they are multi-dimensional. This means that customers are often only able to express their overall satisfaction regarding products or services, which is not specified for the different dimensions. At some online review platform, it is possible to rate different dimensions of the product or service, but these dimensions are preset by the platform. Another disadvantage of only analyzing numerical ratings is that the distribution of most reviews is bimodal, which means all products receive around the same average (Archak et al., 2011). Consequently, a numerical rating alone does not give the useful information the customer and the companies prefer.

Qualitative reviews on the other hand, give both the firms and the consumers a lot of information. Analyzing big data has the potential to gain important insights. However, data needs to be structured before it becomes information. With the help of employing and applying algorithms, companies can extract information from online reviews. This information helps companies to enhance their knowledge about customer behavior, sentiment, customer needs and (dis)satisfactions (Ngo-Ye & Sinha, 2014; Picazo-Vela, Chou, Melcher & Pearson, 2010). Analyzing the opinions and sentiments of online reviews helps to get an

(12)

12

idea of the general image of the public’s opinion. Besides that, written reviews can help to make predictions about customer behavior and to give explanations for variances in product demand (Archak et al., 2011). With the help of that knowledge, companies can adapt their manufacturing, distribution and marketing strategy accordingly (Dellarocas, Zhang & Award, 2007; Yu, Liu, Huang & An, 2012).

In order to evaluate online reviews, tools for sentiment analysis are often used to study opinions, emotions and sentiments. The satisfaction of customers is evaluated by assigning a sentiment score and indicating the review as either negative, neutral or positive. In the beginning, the results of sentiment analysis were not precise and reliable enough. However, the techniques for sentiment analysis improved thanks to the advances in machine learning and natural language processing. A recent development focused on the multidimensional reviews, where customers might like one aspect, but did not like another (Gan & Yang, 2015). Thanks to these improvements and to the increasing importance of online reviews, the use of sentiment analysis is becoming more popular to companies.

Online reviews are important to many business segments and the hospitality industry is one of them. Luca and Zervas (2016) found that with every extra star on Yelp, the revenue of that restaurant increases with 5-9 percent. Besides that, online reviews for restaurants offer rich metadata, consisting of the textual review, the numeric star rating as given by the customer, the price category, the type of food, the location and so on.

Eating out of home is an increasingly important part of the current lifestyle. The sales of the foodservice industry are 2.1 trillion dollars, which represent 10 percent of the Gross Domestic Product (Edwards, 2013). In America, the average spending on eating out of home is 44 percent of their total food budget. Compared with 2010, this is an increase of 10.4 percent . With other words, the restaurant sector is an interesting and a promising market.

(13)

13

Schuckert, Liu and Law (2015) reviewed articles published between 2004 and 2013 that are related to online reviews in tourism and hospitality. It appears that nine out of the 50 papers in this field were restaurant or food related, which is only 18 percent. The fact that the number of related research is limited is remarkable, given that the subject is becoming increasingly popular and important. In the highly competitive market of the hospitality industry it is especially important to know your customer, which makes it even more notable.

Past research on online reviews for restaurants, focused mainly on the restaurant visits and on food, service, ambience and price as the most important attributes for online reviews for these visits (Gan et al., 2017). However, with the evolution of digital technologies a new phenomenon for restaurants increased in importance: food delivery from restaurants. In 2017, the number of visits of a mobile app or website in order to order food from a restaurant in the USA was 1.9 billion. Compared to the previous year, this is an increase of 18 percent. Within 10 years, the number of restaurant meals eaten at home is expected to grow with 20 percent (McLynn, 2017). Online Food Delivery platforms (from now: OFD platforms) are gaining momentum and the revenues are expected to keep on rising.

Experience, online interaction, time saving, access convenience, prices and discounts are mentioned as important factors influencing attitude towards OFD (Yeo, Goh & Rezaei, 2017). Some research has been conducted about the differences for online shopping versus offline shopping (Forman, Ghose & Goldfarb, 2009). However, no research has been conducted about demand fluctuation for online versus offline shopping. Besides that, no academic papers focused on the differences for the experience at home (AH) versus out of home (OOH), specifically for restaurants. This experience is expected to be different, mainly caused by the differences in importance of a website, atmosphere, price and location. It is remarkable that there is hardly any the literature about OFD, while it is such an important phenomenon.

(14)

14

The fact that the market of OFD platforms is upcoming, makes it an interesting field to study. How do consumers experience the restaurant experience at home? What are the most important factors influencing consumers’ attitude towards their experience?

The main research question for this paper will be as followed:

Which restaurant attributes explain the variance in customer satisfaction as

expressed in online reviews, and to what extent is the influence of these attributes on customer satisfaction present for both the at home and the out of home restaurant experience?

As mentioned before, it is important to companies to analyze reviews in order to predict and adjust strategy and to stay ahead of their competitors. Analyzing reviews helps restaurants to predict demand and to gain more knowledge about their customers. The customer of nowadays is more evolved, better aware of value for money, and at the same time it asks for increasingly higher quality of food and service (Scozzafava, Contini, Romano & Casini, 2017). This makes the customers harder to satisfy. Therefore, it is important for restaurants to know their customers in order to adjust their strategy to their customers’ preferences. With the help of the insights of this paper, restaurants managers can make adjustments in order to improve customer satisfaction.

(15)

15

2. Literature review

2.1 Online reviews

Before making any purchase decisions, consumers consult information about the product or service. For years, word of mouth has been one of the most influential resources for obtaining information. However, traditional word of mouth has some limitations, because consumers are limited to their own social network and because the influence of WOM diminishes over time and distance (Duan, Gu &Whinston, 2008). However, digital technologies made the arrival of electronic word of mouth (eWOM) possible. EWOM means that customers place their evaluation of a product or service on a company’s or a third party’s website (Ngo-Ye & Sinha, 2014). Customers can post their opinions anonymous and they can reach many peers, even strangers. These characteristics make sure that consumers feel free to share their opinions. As a result, the volume of online reviews increases (Lee & Youn, 2009). Thanks to the increasing popularity of social media and online shopping, the amount of customers referring to online reviews before making any purchase decisions keeps on growing (Niu & Fan, 2018). This influence results, among other things, from the fact that customers prefer the information based on experiences of peers better than information that comes from the companies themselves. Online reviews are seen as more credible and trustworthy because the seller has no control over them (Lee & Youn, 2009; Chen & Xie, 2008). Besides that, writers of online reviews usually have no interest in selling the products or services, because their reasons for writing them have to do with a sense of belonging, reputation and the enjoyment of helping others (Ngo-Ye & Sinha, 2014).

The impact of online reviews is huge, for both customers and companies. Since online reviews gain importance, companies pay more attention on them. This is necessary, because online reviews can be seen as both an opportunity and a threat. However, the variance in

(16)

16

effects of online reviews depends mainly on category, location and other factors. (Huang, Chen, Yen & Tran, 2015).

Besides importance for businesses, online reviews are an interesting topic for researchers, which can be concluded from the fact that “product reviews” and “online reviews” together have almost 30 000 hits on Google Scholar (Huang et al, 2015).

2.1.1 Word Count

The more words a review consists of, the more useful the review is perceived by the readers. This results from the fact that readers perceive high word counts as a higher quality of the review (Cheng and Ho, 2015). However, Huang et al. (2015) found that this only holds true until a certain threshold. After a threshold of 144 words, the effect diminishes. However, they did not find any relationship between the length of a review and the level of satisfaction of the reviewer. They mention that the length of the review depends on the expertise and writing skills of the reviewer (Huang et al., 2015). Jurafsky, Chahuneau, Routledge and Smith (2014) describe how the length of the review is correlated with the price category of the restaurant. The more expensive the restaurant, the higher the word count of the review.

Vasa, Hoon, Mouzakis and Noguchi (2012) tried to find the relationship between star rating and review length of reviews for apps in the App Store. They found that especially dissatisfied customers leave long reviews while satisfied customers leave short reviews. Satisfied customers who left a 5-star rating often use a single word, for example when they only write ‘awesome’. However, as an exception, it turned out that reviews with a 2-star rating have a higher word count than reviews with a 1-star rating. The reviews with a 5-star rating were significantly shorter than for 2-star reviews, as the median for 5-star rated reviews was 54 characters and for 2-star rated reviews it is almost tripled with 144 characters (Vasa et al., 2012). Similarly, Racherla, Connolly and Christodoulidou (2013) found in a study in the

(17)

17

hospitality domain that the reviews with lower ratings have a higher word count than reviews with higher ratings. This means that the customers who are dissatisfied need more words to express their opinion than satisfied customers. They also found that word count is associated with price category. The higher the price rate, the greater the word count (Racherla et al., 2013).

2.2 Restaurant experience out of home

Jung, Sydnor, Lee and Almanza (2015) argue that good food quality is the most important factor when selecting a restaurant. Food quality is even considered as an essential precondition. It has been proven that when the food is not of good quality, the customers will not choose that specific restaurant even when a restaurant provides high service and low prices. Duarte Alonso, O’Neill, Liu and O’Shea (2013) agree with Jung et al. (2015) and point out that the food itself was central to the respondent’s decision-making process. They mention prior positive experience, a clean environment and hospitable service as other important factors. On the other hand, studies focused on dimensions for perceived disconfirmation. The six key dimensions found are facility aesthetics, lighting, layout, and service staff (Ryu & Han, 2011).

Gregory and Kim (2004) aimed to find whether the importance of the factors differ depending on the availability of information. Food quality, food type, value for money and atmosphere are ranked as the four most important factors for restaurant selection. This ranking was the same for both customers with information and for customers with a lack of information. Food quality was always ranked as highest, but for respondents with prior information, food quality turned out to be even more important than for respondents without.

However, Barnes, Meyer and Kinard (2016) argue that the quality of the food is a necessary, but not sufficient element for success. Service experience is expected to increase in

(18)

18

importance. They share their vision of the future wherein restaurants are not only expected to satisfy physical hunger, but emotional hunger as well. Supporting this view, Yu et al. (2017) analyzed online reviews and found that “friendly” was the most important and most frequently mentioned word for positive reviews, indicating that service is an important factor for positive experiences.

Gan et al. (2017) extended the knowledge about the important attributes for restaurants reviews. Past research mainly focused on food, service, ambience and price as most important attributes. However, Gan et al (2017) added ‘context’ to these attributes, as they argue that these are the five attributes in written reviews that explain the differences in star ratings. Moreover, they argue that context even belongs to the top 3 most important attributes, together with food and service.

Other factors influencing the experience of restaurants are for example category spanning (Kovács & Johnson, 2014), perceived authenticity (Kovács, Caroll & Lehman, 2014) and level of variety seeking (Ha & Jang, 2013). Variety seeking is triggered by the low quality of environmental dining factors, which means that restaurant with high quality are better equipped to retain existing customers (Ha & Jang, 2013).

Star ratings are important to restaurants, because many consumers refer to these when selecting a restaurant due to the easy access and high volumes of reviews. It appears that when restaurants are rated with 5 stars, consumers are willing to pay up to 20 percent extra (Kimble & Milolidakis, 2015). Restaurants should realize the importance of positive word of mouth and that commercial advertising alone will probably not be enough in order to generate good performance (Gregory & Kim, 2004).

2.2.1 Restaurant Category

For this study, the reviews from Grubhub and Yelp will be used. The type of cuisines as indicated by Yelp are used to indicate the restaurant category. However, no literature has

(19)

19

been found that explains the relationship between the restaurant categories as mentioned on Yelp and customer satisfaction. The restaurant categories used on the Yelp website are based on different types of cuisines or food types, like Vietnamese, Pizza, Sushi and many more. In previous literature, the restaurant categories are often somewhat vaguely described. For example, Johns and Pine (2002) argue that the order of factor importance varies between the “different styles” of restaurants. Gan et al. (2017) report that the perceived price depends on the “different types” of restaurants.

Brody and Elhadad (2010) showed a ranking of cuisine types from positive to negative, for three different aspects. It made clear that the ranking of cuisine types between food (general), mood and staff differed completely. This is in line with the findings of Ganu, Elhadad and Marian (2009), who also found that category distribution depends on the cuisine type. For example, the reviews about French and Italian restaurant mainly discussed service (20%), while the reviews of Chinese restaurants, Delis and Pizzeria’s mainly discussed the food.

2.2.2 Price category

As mentioned in the paragraph about word count; the lower the rating, the higher the word count. Besides that, it is described that the higher the price rate, the higher the word count. Similarly, it seems that a high price rate and a low rating are associated with each other. The price seems to positively correlate with negative reviews. According to Racherla et al. (2013) this is caused by the fact that the higher the price, the higher the expectations are. When expectations are not met, this leads to more extreme negative reviews.

On the other hand, a higher price is associated with online popularity. When the rating of quality, ambience and service are equal between restaurants, the restaurants with the higher price seem to be more popular than the restaurants with the lower price. As such, when a

(20)

20

restaurant strives to draw the attention of customers, a low price strategy is not effective (Zhang et al., 2010).

2.3 Restaurant experience at home

As mentioned in the section about the restaurant selection for the experience out of home, guests of restaurants tend to focus mainly on quality. Other mentioned important factors include for example service, price level, ambience, prior experience and food type. Are these factors as important when ordering online? It can be expected that it is not, because, for example, when ordering at OFD platforms, the atmosphere in a restaurant is expected to be less important or even not important at all.

The type of motivation of customer tends to explain variance in attitude towards OFD. Hedonic motivation has a moderate positive relationship with satisfaction. This means that consumers with hedonic motivations tend to have a more positive attitude towards OFD, which will ultimately increase the probability that those consumers will use OFD. Besides motivation, a person’s online purchase experience is mentioned as an important factor influencing satisfaction and purchase intention. Users of OFD platforms are more focused on saving effort and time. Especially experienced users prefer to put as less effort as possible when using OFD services (Yeo et al., 2017).

Pigatto, Machado, Negreti and Machado (2017) point out an important difference for the experience at home compared to the experience out of home. The process at home, OFD, is more focused on self-service, as the customer chooses the establishment, makes the order, choses the payment method and monitors the order until delivery all by himself. Restaurants should pay attention on the objectivity and clarity of the website. Understandability of the website is a great indicator for good results. The use of online ordering yields several benefits,

(21)

21

like the lower costs, workforce reduction, customization, convenience and control (Pigatto el al., 2017)

2.4 Hypotheses development

As mentioned in the previous sections, there are several differences in the restaurant experience at home versus out of home. This difference in context is expected to cause differences in the attributes mentioned in the corresponding reviews. These reviews are a reflection of customers’ attitude towards the experience. There are hardly any studies about attitude towards OFD (at home context), but several reviews have been found about the attitude towards the restaurant experience out of home. Overall, the most important restaurant attributes that have been focused on are food quality, service, price and ambience. After analyzing the findings of these studies and personally reading all the reviews of the dataset for this study, six factors are chosen to research. These factors cover the subjects mentioned in the reviews. These six factors are: F&B quality, Customer Service, Value for Money, Overall Experience, Location and Delivery.

While Johns and Pine (2002) argue the importance of attributes depends on the different outlets and different dining occasions, no literature has been found that study whether the importance of restaurant attributes differ between the two context categories of restaurant experience: AH or OOH. This makes it hard to predicts which impact the two different context categories will have on the importance of which restaurant attributes. For that reason, some general hypotheses have been developed.

2.7.1 F&B quality

Most studies researched the importance of food quality, rather than the quality of both food and beverage. However, for this research these terms will be used interchangeable. Food

(22)

22

is indicated as the most important factor for restaurants. It explains customer satisfaction, positive restaurant experience, return patronage and thus the success of a restaurant (Auty, 1992; Gan et al., 2017; Gregory & Kim, 2004). Quality is found to be the most important factor for consumers in their decision-making process for restaurants; it is even seen as a necessary condition. For that reason, great food quality maximizes success and can be an important weapon in the highly competitive restaurant industry (Auty, 1992; Gan et al., 2017). As researchers agree on the fact that quality and satisfaction are highly interrelated, it is expected to find similar results for this study.

H1: F&B quality has a significant effect on the attitude towards the

restaurant experience.

H1a: The effect of F&B quality on attitude towards the restaurant experience

is influenced by the context (AH/OOH).

2.7.2 Customer Service

Next to quality, service it often mentioned as one of the most important factors explaining consumer satisfaction. Gan et al. (2017) even describe how studies found that service is the most important factor for customer satisfaction and that service in full-service restaurants was even more important than food quality, physical design and price. In addition, they argue that service (humanic clues) plays a more important role than ambience (mechanic clues). Auty (1992) found that the quality of service leads to satisfaction, which has a positive impact on the intent of repurchasing. Overall, customer service is expected to have a significant effect on the attitude towards the restaurant experience as well.

(23)

23

H2: Customer service has a significant effect on the attitude towards the

restaurant experience.

H2a: The effect of customer service on attitude towards the restaurant

experience is influenced by the context (AH/OOH)

2.7.4 Overall Experience

The attribute “overall experience” embodies ambience, which is often mentioned as one of the important attributes. However, overall experience is a broader factor, as it covers several different aspects more, for example décor, music, temperature in the restaurant, crowdedness, acoustics, facilities like bathroom, concept of the restaurant and opening times. It is a well-recognized topic in hospitality research.

Crowdedness can be seen as something negative, as waiting times often become longer, and the noise becomes louder. Ha, Park and Park (2016) found previous studies describing how high levels of crowdedness resulted in negative consumer experiences. However, Ha et al. (2016) themselves found that some consumers value high levels of crowdedness, because they consider it as an indicator of high quality, good reputation and low food price However, when the restaurant is too empty, it works the other way around (Gregory & Kim, 2004; Ha et al., 2016). Besides crowdedness, cleanliness is an important factor for the overall experience.

However, Johns and Pine (2002) question whether the other experience factors have any impact on consumer satisfaction at all. They only point out quality and value for money as the most significant factors

H4: Overall experience has a significant effect on the attitude towards the

(24)

24

H4a: The effect of overall experience on attitude towards the restaurant

experience is influenced by the context (AH/OOH)

2.7.3 Value for Money

Although “value for money” received less research attention than the before mentioned two factors, it has been argued that is at least as important as food and service (Johns & Pine, 2002). A study of Gregory and Kim (2005) even found that, after food, value for money was the second most important factor. This is related to the fact that dining out becomes more common in modern lifestyle, which is the reason that consumers expect good food and service at a reasonable price. It is not surprising that Gan et al. (2017) found that price is one of the most mentioned topics in online reviews for restaurant experience. However, the role of price on preference is complex, since it can be considered as both a constraint and a quality que (Scozzafava et al., 2017).

H3: Value for Money/ Price has a significant effect on the attitude towards

the restaurant experience

H3a: The effect of Value for Money/ Price on attitude towards the restaurant

experience is moderated by the context (AH/OOH)

2.7.5 Location

In most studies food, price, service and ambience are indicated as the most important attributes that have an impact on consumer attitude towards a restaurant. However, Gan and Yu (2015) and Johns and Pine (2002) argue that location is at least as important as quality and service, or even one of the most important factors leading to the success of a restaurant. Gan

(25)

25

and Yu (2015) found that location is especially important when respondents do not have any information, compared to respondent who have.

H5: Location has a significant effect on the attitude towards the restaurant

experience

H5a: The effect of Location on attitude towards the restaurant experience is

moderated by the context (AH/OOH)

2.7.6 Delivery

No studies have been found which analyze online reviews in order to find the relationship between delivery and customer satisfaction. However, it is not expected that context will have any effect on the effect of delivery on the attitude towards the restaurant experience. This is caused by the assumption that only reviews in the at home context will discuss delivery. Therefore, there is no hypothesis about the influence of the context.

H6: Delivery has a significant impact on the attitude towards the restaurant

(26)

26

3. Methodology

For this research, a total of 1988 reviews have been studied, of which 996 for the AH context and 992 for the OOH context. These reviews are scraped from the websites of Yelp and Grubhub. Grubhub is an OFD Platform where customers can order food online in order to get it delivered to them. These customers are also able to leave a review about their experience. The reviews derived from Grubhub represent the AH experience. Yelp is a review platform for online reviews about restaurants, nightlife and home services. This source has been used to get data about the OOH experience.

First of all, reviews had to be in English since the language of this thesis in in English and since the English language is compatible for all used software tools during this study. The chosen location of the restaurants is New York. In New York many restaurants of different categories are located. Besides that, the official language in the USA, and thus New York, is English. Therefore, the chance of enough English reviews is plausible.

All restaurants on Yelp needed to meet the criterion of having at least 1000 reviews. The total number of restaurants in New York on Yelp are around 30 000, but only 105 of those meet this criterion. Subsequently, for those 105 restaurants it was checked whether they have at least a total of 100 reviews on Grubhub. From the 105 restaurant, 10 met the criterion ad were kept. These criteria have also been applied the other way around and in the end, 56 restaurants remained. By using these criteria to select the restaurants, the selection is not influenced by the order in which Yelp and Grubhub show the restaurants.

From the restaurants that were left, all reviews of the past two years (1 March 2016- 28 February 2018) were scraped with the help of the visual web scraping software WebHarvy. Reviews containing other languages than English and the reviews that were incomplete (only a star rating and no written text), were deleted. A time period of two years has been chosen in order to control for too many changes in the restaurants’ situation, like new chefs, new

(27)

27

owners, renovations and so on. Although it is short enough to control for major differences, the time period is long enough to gather enough reviews of a restaurant.

7131 reviews of 56 restaurants were collected, however, due to time constraints this amount was too big considering the time needed to analyze all of these manually. Subsequently, all restaurants of which the difference in number of reviews on Yelp versus Grubhub was too big, were deleted. For example, for one of the restaurants the number of reviews on Yelp was 650 and on Grubhub 14. These restaurants and its reviews have been deleted from the dataset. This criterion automatically helped getting rid of all restaurants with a number of reviews below 10 and above 100. In the end, the reviews of 30 restaurants were kept.

From the websites of Yelp and Gruhub, the name of the restaurants, the restaurant category, the price category, the date of the reviews, the star ratings and the textual reviews were collected. For the analysis, additional information was added. In the dataset file, the word count of the textual review was added. Besides that, two independent raters have reviewed all reviews, focusing on the six restaurant attributes. The raters received the description below in order to help them to judge the reviews:

1. F&B quality= quality of the food and/or drinks, think of the temperature, presentation, freshness, taste and so on.

2. Customer service= service in the restaurant, think of the behavior of the waiters, waiting time, correctness of the order when in restaurant, complaint handling in the restaurant or on the phone in case of delivery.

3. Value for money= price point, portion size, value for money, good deals. 4. Overall experience= ambience, décor, music, temperature in the restaurant,

crowdedness, acoustics, facilities like bathroom, concept of the restaurant and opening times. Besides these, some reviews do not specify what they like or dislike, so when, for example, only ‘it was great’ is mentioned, this is seen as the overall experience. 5. Location= physical location, think of the neighborhood, distance from the customer’s

(28)

28

6. Delivery= waiting time when ordered for delivery, behavior of the delivery person, correctness of the delivery order, online ordering system and so on.

The raters were free to interpret the above description in their own way. For all reviews, the two raters had to judge independently which attributes were discussed and thereafter how it is discussed. In order to express how the factor was discussed, the raters had to give a number between 0 and 3, with 0 meaning that the factor is not discussed at all, 1 meaning that it was negatively discussed, 2 meaning that it was neutrally discussed and 3 meaning that is was positively discussed.

Taking into consideration that two raters measured a variable on categorical scale, it was important to determine to what extent the two raters agreed. In this case, Cohen’s Kappa was used as the measure of inter-rater agreement for categorical scales. There was an almost perfect agreement between the two raters, κ= .954, p< .001.

Although the agreement was almost perfect, some disagreements still existed. Therefore, the two raters met with each other in order to try to agree upon all judgements. The reviews for which they could not come upon an agreement were deleted. In the end, eighteen reviews were deleted, while the other reviews were kept because the raters came to an agreement for those.

The sentiment analysis tool Lexalytics Semantria helped to determine the sentiment of the textual reviews. The output given was a sentiment score, which is a score between -2 and 2, and a sentiment category, being either negative, neutral or positive. Semantria failed to process four of the reviews, which had either too many punctuation marks relative to the amount of text, or had too many strange characters. As a quality measure of Semantria, 80% of the content must be alphanumeric. One example of the reviews that were deleted because it could not be processed is: “Super delicious! I wish I ordered more. ***(=^_^=)***”.

(29)

29

In the end, 1988 reviews of 30 restaurants were kept, of which 996 reviews for the AH context and 992 for the OOH context. Table 1 gives an overview of all used constructs for the analyses, with all variables and their operational definition.

Construct Variable Operational definition and measurement

method

Satisfaction Star Rating (dependent

variable)

Each review is rated by the customers with either 1, 2, 3, 4 or 5 stars.

Sentiment Score (dependent variable)

Continuous variable between -2 and 2 expressing the sentiment of the review Sentiment Category

(dependent variable)

Based on the sentiment score, the reviews are classified as either negative, neutral or positive

Factors Food Sentiment

(independent variable)

Either 0 for no mentioning, 1 for negative, 2 for neutral or 3 for positive

Customer Service Sentiment (independent variable)

Value for Money Sentiment (independent variable)

Overall Experience Sentiment (independent variable)

Location Sentiment (independent variable)

Delivery Sentiment (independent variable)

Context AH/OOH (independent

variable)

0 for AH (source: Grubhub), 1 for OOH (source: Yelp)

Covariates Word Count (independent

variable)

Total number of words in the review Restaurant Category

Category of the restaurant as indicated by Yelp: American, Asian fusion, Burgers, Chicken Shop, Chinese, Cuban, Delis, Diners, Hot Pot, Indian, Italian, Japanese, Kosher, Mediterranean, Pizza, Poke, Ramen, Sushi Bars, Thai, Turkish or Vietnamese

Price category

Each review is classified by Yelp into one of the 4 price categories: 1 (inexpensive), 2 (moderate), 3 (pricey) or 4 (Ultra High-End) dollar signs

(30)

30 Star Rating N of valid cases Mean Sentiment Score: Sentiment category

% Positive % Neutral % Negative

1 star 335 -0,407 0,00% 7,16% 92,84% 2 stars 199 -0,245 3,02% 24,12% 72,86% 3 stars 215 0,059 24,19% 50,70% 25,12% 4 stars 408 0,218 85,36% 15,44% 2,21% 5 stars 831 0,614 96,87% 2,89% 0,24% OVERALL 1988 0,261 60,31% 13,48% 26,21%

Table 2: Star rating versus Sentiment Category

As shown in table 2, most reviews have a positive sentiment (60%), which is also indicated by the number of star ratings with 4 and 5 stars (together 62%). The table also shows that the sentiment category matches the star rating of the review, this is obvious for example for the review ratings with 1 stars, for which only 0% is indicated as a positive review and 93% is indicated as having a negative sentiment.

It is still possible that reviews have been indicated with a negative sentiment score, while the reviewer was actually satisfied. An example of such a review is “Homemade ice tea sucks!!”. This review received a 4-star rating, but a sentiment score of -1,021. It is possible that the reviewer was very satisfied with everything, except the homemade ice tea, which was the only thing he mentioned anything about. Therefore, the sentiment score is not necessary representative of the reviewers’ satisfaction. This limitation should be taken into consideration when analyzing the results.

Sentiment analysis only indicates whether a text is positive, neutral or negative and Archak et al. (2011) believe that this is a disadvantage. They argue that the output does not provide enough useful information about the most important factors leading to the specific sentiment. During this study this disadvantage will be avoided by analyzing the sentiments of reviews for each specific attribute. This way it will be clear which attributes mainly lead to positive experiences and which ones to negative experiences.

(31)

31

4. Results

In this chapter, the results of the analysis are given and the hypotheses will be tested. To begin with, some findings of the overall dataset will be given. Subsequently, each attribute will be discussed in its own paragraph.

“Attribute discussion” means whether the attribute is discussed in a review, so yes or no. “Attribute sentiment” indicates which sentiment the two independent rates have assigned to the attribute. “Sentiment score” and “Sentiment category” are the outputs given of the Lexalytics Semantria tool.

AH Mentions as % of all AH reviews OOH Mentions as % of all OOH reviews Total Mentions Mentions as % of all reviews F&B quality 816 81,93% 924 93,15% 1740 87,53% Customer Service 41 4,12% 511 51,51% 552 27,77% Value for Money 127 12,75% 436 43,95% 563 28,32% Overall Experience 48 4,82% 331 33,37% 379 19,06% Location 3 0,30% 103 10,38% 106 5,33% Delivery 512 51,41% 0 0,00% 512 25,75% OVERALL 1547 1,55 2305 2,32 3852 1,94

Table 3: Number of reviews with attribute discussions per Context category

Table 3 shows the number of reviews that discuss the specific attribute, per context category (AH/OOH). Overall, F&B quality is the most discussed attribute in the reviews, namely in 88%. F&B quality is followed by value for money (28%), customer service (28%), delivery (26%) and overall experience (19%). Location is the least mentioned attribute, with 106 mentions 5% of all reviews discuss the location. The differences in number of discussions are high. The most mentioned attribute, F&B quality, is discussed at least three times more

(32)

32

often than the second most discussed attribute. The difference between the most and least mentioned attribute is even bigger; F&B is discussed at least fifteen times more often than location. It might be concluded that F&B is by far the most important attribute of all.

Almost all attributes have been discussed more in OOH reviews than AH reviews. However, this does not hold true for Delivery, which is only discussed in AH reviews and not in OOH reviews at all. Nonetheless, the highest mean of number of attributes discussed per review is for the OOH reviews. On average, the number of attributes discussed per review is 1.94, where AH reviews discuss 1.55 attributes, while OOH discuss 2.31 attributes. When delivery is not accounted for, the average number of attributes per AH review would only be 1.04. This would make the difference between AH and OOH bigger, with OOH mentioning on average twice as many attributes as AH reviews.

For both context categories, F&B quality is the most frequently discussed attribute. However, the rest of the order of importance of attributes differs between the two context categories. For AH reviews, F&B quality is followed by delivery, value for money, customer service and eventually location. For OOH reviews, F&B quality is followed by customer service, value for money, location and eventually delivery.

Context N Mean Std.

Deviation

AH 996 0,248 0,559

OOH 992 0,273 0,433

Table 4: Group Statistics of F&B quality

t df Sig. MD

Context -1,132 1986 0,258 -0,025

(33)

33

An independent-samples t-test was conducted to compare the sentiment scores for the two context categories; AH or OOH. There was no significant difference between the sentiment scores of reviews with the AH context (M = 0.27, SD = 0.43) and the OOH context (M = 0.25, SD = 0.56); t(1986) = -1.13, p = .26. The average sentiment score of reviews with the AH context is higher than for reviews with the OOH context. However, as mentioned, this difference is not statistically significant.

Mean Sentiment Score N Std. Deviation F&B 0,3089 1740 0,48 Customer Service 0,2167 552 0,46

Value for Money 0,2901 563 0,42

Overall Experience 0,3583 379 0,39

Location 0,3387 106 0,28

Delivery 0,1472 512 0,52

Table 6: Sentiment score per attribute

Table 6 shows the mean sentiment scores of the reviews discussing the specific attribute. On average, reviews discussing the overall experience had the highest sentiment scores. Customers agreed most in the reviews discussing location, as the standard deviation is the lowest of all (SD = 0.28). Consumers were most divided in reviews mentioning delivery, as the standard deviation for those reviews is the highest (SD= 0.52).

Sentiment Category N F&B Customer Service Value for Money Overall Experience Location Delivery Negative 521 73,51% 28,41% 21,31% 9,02% 1,73% 36,28% Neutral 268 86,19% 30,97% 32,46% 20,90% 5,97% 27,61% Positive 1199 93,91% 26,77% 30,44% 23,02% 6,76% 20,77% OVERALL 1988 87,53% 27,77% 28,32% 19,06% 5,33% 25,75%

(34)

34

Although already discussed, table 7 also shows that, overall, F&B is the most discussed attribute of all (88%), followed by value for money (28%), customer service (28%), delivery (26%) and overall experience (19%). Location is with 5% the least discussed attribute. When comparing the different sentiment categories, it is remarkable that none of the categories has the same order in most discussed attribute. However, F&B is for all different categories the most discussed and location is for all different sentiment categories the least discussed attribute. The other attributes differ in order per sentiment category.

Some conclusions of table 7 will also be discussed at each paragraph of the particular attribute.

4.1 Restaurant attributes

Each attribute will be discussed in its own paragraph. First, with the help of an independent t-test, it is tested whether there is a significant relation between the attribute discussion and the sentiment score. The independent t-test is used, because the dependent variable (Sentiment score) is numerical and the independent variable (attribute discussion) is categorical and only has two levels (is the attribute discussed in a review: yes or no).

Subsequently, it is tested whether there is a significant relationship between the factor’s sentiment on the sentiment score. This is tested with a one-way ANOVA test, because the dependent variable (Sentiment score) is numerical and the independent variable (attribute sentiment) is categorical and has more than two levels. The attribute sentiment can be either none, negative, neutral or positive.

Thereafter, the relation between the interaction of the attribute sentiment and the context, and the sentiment score has been tested by processing a two-way ANOVA test. The satisfaction is measured by the Sentiment Score. All hypotheses concerning the interaction effect have two independent variables (attribute sentiment and context) and one dependent

(35)

35

variable (sentiment score). Therefore, an independent factorial design, or a two-way ANOVA, has been performed.

Sentiment score= (b0 + b1 Context + b2 Attribute sentiment + b3 Interaction) + εi

4.1.1 F&B Quality

Overall, F&B quality has been mentioned in 1740 of the 1988 reviews, making it the most discussed attribute of all. Besides that, it is the only attribute for which the number of reviews that discuss it, is higher than the number of reviews that do not. Although it is the most discussed attribute for both context categories, it is even more important for the OOH experience. It has been discussed in 82% of all AH reviews and in 93% of all OOH reviews. Furthermore, it is remarkable that the percentage of reviews discussing F&B increases while the sentiment category increases. When looking at F&B discussions, the lowest percentage is for negative reviews (73.51%) and the highest percentage is for positive reviews (93.91%). This means that F&B is important for all customers leaving a review, but even more important when consumers have a positive attitude towards their experience.

Effect of F&B on Sentiment Score

Food N Mean SD

No 248 -0,077 0,536

Yes 1740 0,309 0,476

Table 8: Group Statistics of F&B quality

t df Sig. MD

F&B- Sentiment Score

-11,759 1986 0,000 -0,386

(36)

36

An independent-samples t-test was conducted to compare the sentiment scores for the difference in F&B discussion; whether it was mentioned or not. The results indicate a significant difference in the sentiment scores when F&B was discussed (M = 0.31, SD = 0.48) or not (M = -0.08, SD = 0.54); t(1986) = -11.76, p< .001. When F&B is discussed, the sentiment score is on average higher than when F&B is not discussed.

SS DF MS F Sig. F&B Sentiment 266,209 3,000 88,736 763,760 0,000 Error 230,508 1984,000 0,116 TOTAL 496,717 1987,000

Table 10: ANOVA for F&B Sentiment and Sentiment Score

F&B Sentiment N Mean SD None 248 -0,077 0,536 Negative 397 -0,329 0,293 Neutral 191 0,166 0,313 Positive 1152 0,552 0,305 TOTAL 1988 0,261 0,500

Table 11: Descriptives of F&B Sentiment and Sentiment Factor

An ANOVA analysis has been performed in order to examine the relationship between F&B sentiment and sentiment score. There were a statistically significant differences in the sentiment score for the different levels of F&B sentiment, F(3,1984) = 763.76, p< .001. The differences between all F&B sentiment groups were significant (p<. 001), with positive F&B sentiment having the highest sentiment scores, followed by neutral, no and eventually negative F&B Sentiment.

(37)

37

F&B and Context on Sentiment Score

SS DF MS F Sig. η² Context 0,386 1 0,386 3,375 0,066 0,002 F&B 259,155 3 86,385 755,612 0,000 0,534 Context*F&B 2,505 3 0,835 7,303 0,000 0,011 Error 226,362 1980 0,114 TOTAL 631,874 1988

Significant at the P< .05 level

Table 12: Univariate analysis of interaction effect of F&B Sentiment and context on Sentiment Score

The interaction effect of F&B quality and Context on Sentiment Score is significant, F(3,1980)= 7.30, p<.001. This effect, with η²= 0,011, is a low effect. In this model, the effect of F&B also has a main significant effect on sentiment score (F(3,1980) = 755.61, p<.001), but context does not (F(1,1980) = 3.38, p = .066). The effect size of F&B sentiment is high (η²=0.53).

4.1.2 Customer Service

Customer service has been mentioned in 552 reviews, which is 28% of all reviews. From these 552 mentions, 7% is in AH Reviews and 93% is in OOH reviews. There is no notable distribution when looking at the sentiment categories. The percentage of customer service discussions for the categories is between 27% and 31%.

From the 552 mentions, 325 mentions are indicated by this studies’ raters as positive mentions, which is 59% of all customer service mentions. 185 mentions were indicated as negative and 42 mentions were indicated as neutral. This implies that customer service is mainly something the consumers write about when they are either satisfied or dissatisfied. Looking at the AH context, none of the customer service mentions was rated as neutral.

(38)

38

Effect of Customer Service on Sentiment Score

Customer Service

N Mean SD

No 1436 0,278 0,515

Yes 552 0,217 0,456

Table 13: Group Statistics of Customer Service

t df Sig. MD

CS- Sentiment Score

2,437 1986 0,015 0,061

Table 14: Independent Samples T- test for CS and Sentiment Score

The results on the independent samples t-test indicate a significant difference in the sentiment scores when customer service was discussed (M = 0.22, SD = 0.46) or not (M = 0.28, SD = 0.52; t(1986) = 2.44, p = .02. When customer service is discussed, the sentiment score is on average lower than when customer service is not discussed.

Table 15: ANOVA for CS Sentiment and Sentiment Score

CS Sentiment N Mean SD None 1436 0,278 0,515 Negative 185 -0,256 0,364 Neutral 42 0,277 0,352 Postive 325 0,478 0,257 TOTAL 1988 0,261 0,500

Table 16: Descriptives of CS Sentiment and Sentiment Factor

SS DF MS F Sig. CS Sentiment 65,190 3 21,730 99,906 0,000 Error 431,527 1984 0,218 TOTAL 496,717 1987

(39)

39

The results of the ANOVA test indicate statistically significant differences in the sentiment score for the different levels of customer service sentiment, F(3,1984)= 99.91, p< .001. The differences between most CS sentiment groups were significant (p<. 05), with positive CS sentiment having the highest sentiment scores, followed by neutral, no and eventually negative customer service sentiment. However, the difference between no customer service sentiment and neutral customer service sentiment was not significant (p = 1.00)

Customer Service and Context on Sentiment Score

SS DF MS F Sig. η² Context 0,716 1 0,716 3,296 0,070 0,002 CS 36,078 3 12,026 55,336 0,000 0,077 Context * CS 0,642 2 0,321 1,477 0,229 0,001 Error 430,527 1981 0,217 TOTAL 631,874 1988

Table 17: Univariate analysis of interaction effect of CS Sentiment and context on Sentiment Score

The interaction effect of context and customer service sentiment on sentiment score is not significant (F(2,1981) = 1.48, p=.23). In this model, the effect of CS sentiment also has a main significant effect on sentiment Score (F(3,1981) = 55.34), p<.001), but context does not (F(1,1981)= 3.30, p=.070). The effect size of CS sentiment is moderate (η²=0.08).

4.1.3 Value for Money

Of all 1988 reviews, 563 mention value for money, which is 28%. From these 563 mentions, 23% is for AH reviews, while 77% is for OOH reviews. There is no notable distribution when looking at the sentiment categories. When looking at all sentiment

(40)

40

categories, the percentage of value for money discussions lies between 21% and 32%. It is least discussed in the negative reviews, compared to neutral and positive reviews.

Effect of Value for Money on Sentiment Score

Value for Money N Mean Std. Deviation No 1425 0,249 0,529 Yes 563 0,290 0,416

Table 18: Group Statistics of value for money

t df Sig. MD

VfM- Sentiment Score

-1,648 1986 0,099 -0,041

Table 19: Independent Samples T- test for VfM and Sentiment Score

The results on the independent samples t-test indicate that there is no significant difference in the sentiment scores when value for money was discussed (M = 0.29, SD = 0.42) or not (M = 0.25, SD = 0.53; t(1986) = -1.65, p = .10. When value for money is discussed, the sentiment score is on average higher than when it is not discussed. However, as mentioned, these differences are not significant.

SS DF MS F Sig. VfM Sentiment 39,540 3 13,180 57,197 0,000 Error 457,177 1984 0,230 TOTAL 496,717 1987

(41)

41 VfM Sentiment N Mean SD None 1426 0,249 0,529 Negative 168 -0,110 0,397 Neutral 108 0,422 0,263 Positive 286 0,478 0,291 TOTAL 1988 0,261 0,500

Table 21: Descriptives of VfM Sentiment and Sentiment Factor

The results of the ANOVA test indicate statistically significant differences in the sentiment score for the different levels of value for money sentiment, F(3,1984) = 57.20, p < .001. The differences between most VfM sentiment groups were significant (p <. 05), with positive VfM sentiment having the highest sentiment scores, followed by neutral, no and eventually negative VfM Sentiment. However, the difference between neutral value for money sentiment and positive value for money sentiment was not significant (p = .73)

Value for Money and Context on Sentiment Score

SS DF MS F Sig. η² Context 0,463 1 0,463 2,013 0,156 0,001 VfM 34,240 3 11,413 49,576 0,000 0,070 Context*VfM 1,318 3 0,439 1,908 0,126 0,003 Error 455,836 1980 0,230 TOTAL 631,874 1988

Table 22: Univariate analysis of interaction effect of VfM Sentiment and context on Sentiment Score

The interaction effect of value for money and context on sentiment score is not significant, (F(3,1980)= 1.91, p=.126). In this model, the effect of Value for Money has a main significant effect on sentiment score (F(3,1980) = 49.58), p < .001). but context does not (F(1,1980)= 2.01, p=.156). The effect size of value for money sentiment is moderate (η²=0.07).

(42)

42

4.1.4 Overall Experience

Of all 1988 reviews, 379 mention Price/ Value for Money, which is 19%. From these 379 mentions, 13% is for AH reviews, while 87% is for OOH reviews. The more positice the sentiment category, the higher the percentage that discuss the overall experience. When looking at overall experience discussion, the lowest percentage is for negative reviews (9%) and the highest percentage is for positive reviews (23%). This implies that overall experience is more important when consumers have a positive attitude towards their experience. This is in line with the earlier finding that reviews mentioning overall experience have the highest mean in Sentiment Score.

234 reviews have a positive overall experience sentiment. This means that if consumers mention overall experience, in 61% of the reviews the overall experience mention is positive. For the AH context, there was no review at all that mentioned the overall experience in a neutral way.

Effect of Overall Experience on Sentiment Score

Overall Experience N Mean Std. Deviation No 1609 0,238 0,520 Yes 379 0,358 0,387

Table 23:Group Statistics of Overall Experience

t df Sig. MD

EX- Sentiment Score

-4,242 1986 0,000 -0,121

Table 24: Independent Samples T- test for Overall Experience and Sentiment Score

The results on the independent samples t-test indicate a significant difference in the sentiment scores when overall experience was discussed (M = 0.36, SD = 0.39) or not (M =

Big data in the restaurant industry : a sentiment and quantitative analysis of online reviews to study the effect of restaurant attributes and context on customer satisfaction