Keywords

Introduction

In quest of defining a successful restaurant, the existing literature focuses mostly on financial factors (Di Pietro et al. 2007; Harrison 2011; Susskind 2010). Another measure of success is customer loyalty and customer satisfaction (Han and Ryu 2009; Harrington et al. 2011). Nowadays, customer satisfaction is often expressed on online reviews and seems to influence potential customers to visit and dine in a restaurant. Restaurant owners should know that consumers’ driving force for sharing their positive opinion online is food quality rather than discussions about prices (Jeong and Jang 2011).

Apart from positive online reviews, ethnic restaurant success and customer attraction can be also achieved by offering authentic and high-quality products and services (Bryla 2015; Muller 1999; Namkung and Jang 2007; Sulek and Hensley 2004; Tsai and Lu 2012). If consumers perceive an ethnic restaurant as authentic they are more than happy to spread positive word of mouth (Lu et al. 2013). As such, consumers’ perceptions of authenticity need to be defined and especially how these are expressed nowadays in a digital global setting.

Marketers have defined authenticity as a social and commercial construction for differentiation and positioning (Becuţ 2011; Bryla 2015; Ebster and Guist 2005; Lu et al. 2015; Wood and Lego Muñoz 2007). In this study, authenticity is approached as a social projection which permits “various versions of authenticities regarding the same object” (Wang 1999, p. 352). This type of authenticity is not objectively defined but symbolically and personally constructed (Reisinger and Steiner 2006). Authenticity for the constructivists is a perception of cultures, which includes deeper meanings and different interpretations for every human (Lu et al. 2015). Personal experience and identity can additionally contribute to the characterisation of food as authentic or inauthentic (Chatzopoulou et al. 2019). Individuals are in an endless interaction with society, and so their personal experiences create the relations which may define authenticity. In our research, we explore the construction of authenticity meanings by the consumers of ethnic restaurants before vs. after visiting the country of origin of these restaurants. Moreover, we explore how consumers’ visit to the country of origin affects their online reviews and what can restaurant owners learn from these reviews to improve their ethnic restaurant businesses.

We first propose a methodology to extract when positive reviews are made for ethnic restaurants and also to depict authenticity meanings through graph representations. Then, sentiment analysis of consumers’ online reviews is outlined and so, the combination of those steps aids the exploration of ethnic authenticity perceptions and positive online reviews about it. As such, an innovative methodology is followed which integrated authenticity meanings extraction with a big data analysis.

Methodology

Modelling tourism data requires to take into account locations information, users’ properties and their interactions. Data are based on a TripAdvisor extraction of locations, users and their reviews. In the Neothentic database, we propose a graph data model and data operators dedicated to authenticity extraction and consumers’ reviews. Some studies focused on graphs to model trips with graphs (Brandes 2001; Sang-Hyun Lee et al. 2013; Shih 2006). Those analyzes focus on various centrality measurement methods on networks that are combined with maps. It proposes to identify interaction that can characterize tourism behaviors. We go one step beyond by characterizing authenticity paths in such graphs.

Our database is composed of geolocalized locations, restaurant reviews and users. Thus, a first filter is applied to locations in order to get only relevant ones. They are identified by type a cuisine type (a list of denominations such that [“Italian”, “Pizza”, “Sea Food”]), l a localization (lat, long) and rat a rating (rat ∈ ℝ ∧ rat ∈ [1.0, 5.0]).

To simplify localization, each location has been aligned with administrative areas (GADM). Each location is then linked to an area if its geolocalization (i.e., lat, long) is contained into the area’s shape (SpatialPolygon function SP), such that area = SP(l. lat, l. long). This area is composed of a country, a region, a department and a city: area (country, region, department, city). Thus, each location l is identified by: \( \boldsymbol{l}\in \mathbf{\mathcal{L}}\left(\boldsymbol{type},\boldsymbol{rat},\boldsymbol{area}\right) \).

A user u is identified by his nationality and age, \( \boldsymbol{u}\in \mathbf{\mathcal{U}}\left(\boldsymbol{country},\boldsymbol{age}\right) \).

A review is a note (n ∈ ℕ ∧ n ∈ [1, 5]) given by a user u on a location l at time t (t is in the discrete time domain \( \mathcal{T} \)). Each review is then defined by an event rt such that: rt = (l, u, n).

The stream of reviews \( \mathcal{S} \) is a time serie of \( {r}_{t_i} \) events: \( \mathbf{\mathcal{S}}=\left\{{\boldsymbol{r}}_{{\boldsymbol{t}}_{\mathbf{1}}},{\boldsymbol{r}}_{{\boldsymbol{t}}_{\mathbf{2}}},{\boldsymbol{r}}_{{\boldsymbol{t}}_{\mathbf{3}}},\dots \left.{\boldsymbol{r}}_{{\boldsymbol{t}}_{\boldsymbol{n}.}}\right\}\right. \)

Graph Data Model

In order to extract the authenticity experience of users in this time series, it is necessary to focus especially on users who have visited at least once the given destination (Italy) and have tested a “destination” restaurant in their country before and after the country of origin. To achieve this, we propose to model the time serie into a graph data model that represents the experience of each user corresponding to a given cuisine type (Italian).

Time Serie Specialization

Before producing a graph, we need to focus only on restaurants of a given cuisine type corresponding to the study. A filter σcuisine on “destination” restaurants keeps only those which corresponds to the cuisine type parameter:

$$ {\mathcal{S}}_{cuisine}={\sigma}_{cuisine}\left(\mathcal{S}\right)\iff {r}_{t_i}\in {\mathcal{S}}_{\mathrm{cuisine}}\Rightarrow \forall {r}_{t_i}\in \mathcal{S}\ cuisine\in {r}_{t_i}.l. type $$

We also need to keep specific localization of restaurants according to the protocol of our study. In fact, only restaurants located in the destination country, and those from the consumers’ country are to be kept. Thus, the destination operator δ produces a new time serie \( {\mathcal{S}}^{dest} \) that verifies users’ country or review destination:

$$ {\mathcal{S}}^{dest}={\updelta}_{dest}\left(\mathcal{S}\right)\iff {r}_{t_i}\in {\mathcal{S}}^{dest}\Rightarrow \forall {r}_{t_i}\in \mathcal{S} $$
$$ {r}_{t_i}.u. country={r}_{t_i}.l. area. country\vee {r}_{t_i}.l. area. country= dest $$

Finally, to produce the required time serie to produce the corresponding graph, we can combine both operators with the cuisine type and the destination. We can notice that the combination of operators can be permuted in order to optimize the process of extraction.

$$ {\mathcal{S}}_{cuisine}^{dest}={\updelta}_{dest}\left({\sigma}_{cuisine}\left(\mathcal{S}\right)\right) $$

For instance, \( {\mathcal{S}}_{Italian}^{Italy} \)denotes the serie of events where users reviewed Italian restaurants both in Italy and also in the consumers’ country.

Online Reviews Analyzes Framework

We can manipulate more easily the sequence of nodes for each user or a group of users according to the required study. A query language CypherFootnote 1 is available which allows manipulating the graph and to visualize how users behave on this graph.

We need to identify the experience before, during and after the user’s experience on a cuisine type. For this, we can execute queries on \( \mathcal{A} \) that extract the three sequences of circulation of users on the graph.

Table 1 proposes a query that extracts for each user the review sequence containing 3 paths in the graph. It specifies how sequences are extracted and filters that are applied on each of them. Three clauses are given: MATCH to give the pattern for paths, WHERE for the filters, RETURN to give the final result.

Table 1 Online reviews extraction in the Cypher query language
Full size table

Every edge is declared in the MATCH clause with a “-->” between nodes “()”. Stars between brackets say that we accept any length of the path (from 0 edges to n). The red path p1 corresponds to all the edges that occur before getting to Italy (first restriction in the WHERE clause) from nodes (b1) to (bn). Path p2 corresponds to the reviews in Italy (second restriction) from nodes (it1) to (itn), and path p3 after Italy (third one) from nodes (a1) to (an). Notice that those three paths are linked together by linked nodes it1 and a1 at the end of paths p1 and p2. Moreover, to specify that this long sequence of reviews is given by a single user, the fourth restriction in the WHERE clause says that all relationships are linked to user u (given at the end of path p1).

To finish with, the RETURN clause aggregates notes from reviews of each path in order to give the average rating before, during and after being in Italy. It will be called in the following the authenticity vector.

On top of that, we can refine queries by filtering the users’ country. For instance, we can add in the WHERE clause that u must come from the UK (u.user.country = “UK”). Consequently, we will obtain the authenticity vectors from British citizens. The set of all authenticity vectors can be visualized to show the distribution of ratings for each step.

This final step will produce the 3-step vectors that will be used to extract both ratings evolution and comments extraction for sentiment analysis. The aggregation of such vectors helps to have a global understanding of customers’ behavior on e-WoM.

Dataset

Global data were collected from TripAdvisor for the period 2010–2018 concerning reviews about 51,710 restaurants. The global study with the first filter (\( {\mathcal{S}}_{Italian}^{Italy} \)) collected data from 786,896 users who have put at least one review on an Italian restaurant. We applied this filter to an initial source of over 54,572,165 users. As such, we got 16,901,269 corresponding reviews for analysis from an initial source of over 300,084,943 reviews.

Findings

From the above described procedure, USA and UK reviewers’ comments have been collected from TripAdvisor platform concerning their experience in Italian restaurants. Their comments have been categorised in three different sections: before visiting Italy (step 0), during their visit to Italy (step 1) and after their visit to Italy (step 2). The purpose of doing so has been to explore whether food perceptions differ before the visit in the country of origin vs. after. The sentiment analysis was conducted with the use of NVIVO 12. First, we run a word frequency query for step 0, then for step 1 and finally for step 2. The most commonly words used per step are depicted on tree maps below. Secondly, a sentiment analysis of each and every word was conducted from consumers’ reviews in order to depict how these commonly used words are perceived by the consumers.

The sentiment analysis has shown that during their visit to the country of origin (Italy) the words pizza, pasta, Italian and friendly were missing from the 20 most frequent words of reviewers’ comments. The words appeared in the top 20 before they visited Italy. The words pizza, Italian and time appear only before the visit to Italy and again after their visit. The words: friendly, staff and excellent are missing from the top 20 most frequent words of reviewers’ comments after the visit to Italy even if they appeared before the visit in the country of origin. The words excellent and staff appear only in the top 20 of reviewers’ comments before and during the visit to Italy but, not after. As such, we may conclude that the country of origin affects perceptions of food quality and excellence making hard to give excellent reviews to an ethnic restaurant after visiting the country of origin. This is also evident by the less 5’s of reviews after the visit to Italy.

As can be seen in Fig. 1, the distribution of customers’ average ratings from authenticity vectors for UK and USA citizens is different from before to after visiting Italy. Their ratings are globally higher while dining in Italy. But we can notice that there are lower ratings after being in Italy as it tends to be more criticism between 4 and 5 average rating. As such, 57.14% of 4 s and more for USA users (resp. 43.48% for UK) before visiting Italy become 42.86% after their visit to Italy (resp. 39.13%). It means that US customers tend to be more criticism on ethnic authenticity after their visit than British citizen. Moreover, we can see that British customers leaved a better experience while being in Italy. Concerning the 5 s for UK users before vs. after visiting Italy were reduced by 7.2% while for USA users the number of 5 s witness a reduction of 5%.

Fig. 1
figure 1

Distribution of authenticity vectors from UK and USA citizens on Italian restaurants

Full size image

Based on the analysis of all reviews the current study answers the question ‘when are positive reviews made for ethnic restaurants’? In order to do so, we conducted a sentiment analysis of the reviews before visiting Italy vs. after the visit. As it can be seen on the tree maps of step 0 and 2, we may conclude that different perceptions exist about food aspects and the use of words good, great, pizza and restaurant. Concerning food, in step 0 there were no reviews about simplicity which is highlighted in steps 1 and 2 as a main characteristic of Italian cuisine. Moreover, only in step 2 is highlighted the value of real bread and not in step 0. Comments about nostalgia and how food reminded the country of origin can be seen only in step 2. Concerning the use of the word good, in step 0 the phrase good bread is missing whereas it has been pointed out in step 2. Moreover, the phrase “good flavoured dishes” is only used in step 2 but, not in step 0. Concerning the use of the word great, great location seems to concern consumers only in step 0 as it is missing from step 2. Consumers seem to care more about food aspects rather than the location. Concerning reviews about pizza, consumers seek for pizza variety only in step 0 but, not in step 2. They also tend to compare pizza with the country of origin only in step 2: “the best pizza outside Italia”. Finally, when it comes to the use of the word restaurant, consumers make comparisons with Italy only in step 2: “reminds me of being in Italy”, “I would call it more a good trattoria than a restaurant and don’t get me wrong, this is a compliment”.

Tree map analysis of step 0 (reviews on TripAdvisor before visiting Italy):

figure a

Tree map analysis of step 1 (reviews on TripAdvisor while visiting Italy):

figure b

Tree map analysis of step 2 (reviews on TripAdvisor after visiting Italy):

figure c

Based on the analysis of all reviews the current study explores when are positive reviews made for ethnic restaurants. To do so, we conducted a sentiment analysis of the reviews before visiting Italy vs. after the visit (see Table 2).

Table 2 Sentiment analysis – online reviews
Full size table

Concerning authenticity perceptions, a query was conducted in NVIVO 12 about the word authenticity and how this is used in the online reviews of consumers before (step 0) vs. during (step 1) vs. after their visit in the country of origin (step 2). A word cloud was created per step (Fig. 2).

Fig. 2
figure 2

Authenticity perceptions – word clouds before, during and after visiting the country of origin

Full size image

Authenticity perceptions are affected after visiting the country of origin and so the word clouds are different before, during and after the visit to Italy. For instance, pizza is perceived as part of the Italian food authenticity and so the word is included in the word cloud of step 0. However, after visiting Italy consumers realized that Italian cuisine does not necessarily include pizza but rather other food options. As such, the word pizza is not included in the word cloud of step 2. Gelato on the contrary is an important aspect for Italians and so it is included as part of authenticity in step 2 but, not before visiting Italy, step 0.

Finally, authenticity relies much on the actual food menu and authentic atmosphere which are depicted in step 2 but, not in step 0. Step 0 is more about the staff, the service and to feel closer to the Italian-like character whereas, after visiting Italy consumers perceive the core of authenticity to rely on great food, simplicity and the traditional menu choices.