Composing responses to negative hotel reviews: A genre analysis

This study probes into the issue of how top hotels respond to customers’ complaints in public space by exploring their responses to negative reviews (RNRs) through the lens of discourse genre analysts with the aim to make concrete contributions to both English language teaching and service industries. This study examined a corpus of top UK hotels’ RNRs posted on TripAdvisor to address three research objectives: (1) to identify moves and steps of RNRs and their occurrence frequency, (2) to identify typicality of each move, and (3) to generate recurrent word combinations (RWCs) of each move and step. Covering all the communicative functions appeared in the corpus, six distinctive moves were identified along with several steps. As for the typicality, moves were categorized into three types: Obligatory, Conventional, and Optional moves. The lists of RWCs generated confirmed the identity of each move and step as each has its own distinctive list of combinations creating clear cuts between moves and steps. The brief guidelines for composing RNRs were then developed based on the findings of the three components. Subjects: English for Specific Purposes; Discourse Analysis; English Language Teaching


PUBLIC INTEREST STATEMENT
The emergence of online platforms and usergenerated review features have changed the way businesses interact with customers and how customers make purchasing decisions. Online customers' complaints can no longer be ignored as they spread virally and damage businesses' image. In response to the potential threats, dealing with online complaints is now becoming a challenging task for hotel managements, making a call for ways to effectively handle online complaints. Responses to negative reviews (RNRs) has persistently received attention from scholars not only in the field of business, but also language and linguistics. While business researchers tend to focus on the potential impacts of RNRs, discourse genre analysts are interested in the linguistic side of the RNR genre. By gaining insight into "how" RNRs are constructed in the way they are, this study will shed some light on how professionals handle customers' complaints in public and ways to compose RNRs.

Introduction
There is no denying the fast-growing importance of customer-posted reviews in online business societies. The review feature is very much expected these days, as it provides prospective customers with honest feedback on a particular product or service. Generated by customers themselves, the reviews are perceived to be more trustful than words from businesses, which most of the time, report only the bright side of the story (Floyd, Freling, Alhoqail, Cho, & Freling, 2014). A large body of literature has confirmed that such aspect of online business has influenced customers purchasing decision and corporate image (Bambauer-Sachse & Mangold, 2011). The intangible nature of hospitality services has made the review section the first stop before proceeding to the next stage of decision-making. PhoCusWright (2014) discovered that over 70% of customers checked out others' reviews before booking a hotel. While positive reviews are found to be beneficial to businesses in boosting sales, negative reviews made themselves an interesting topic for research. Vermeulen and Seegers (2009) assert that negative reviews have received tremendous attention from scholars as they have impacts on customers' purchasing decision, leading to a call for extensive investigation on ways to effectively manage negative reviews.
Like many modern online platforms, TripAdvisor allows managements to directly interact with their clients in its review section: the business representatives can post responses to customers' reviews which can be seen by any visitors. As suggested by previous studies, responses to negative reviews (RNRs) have the potential to achieve service recovery, which is an ideal goal for hospitality service providers when suffering from service failure and/or customer dissatisfaction (e.g., Ho, 2017;Sparks, So, & Bradley, 2016;Thumvichit & Gampper, 2018). Thus, RNRs are now much more than just a casual interaction between the two parties, rather they play a leading role in handling customers who are not satisfied with services. Despite countless articles published on social media regarding how to effectively respond to negative reviews and strategies suggested by experts, research and empirical evidence are still far from a wellcrafted conclusion. The impacts of RNRs have been investigated by scholars in the field of business (e.g., Davidow, 2000;Sparks et al., 2016); on the other hand, the language dimension of RNRs interests linguistic analysts (e.g., Ho, 2017;Panseeta & Todd, 2014;Thumvichit & Gampper, 2018;Zhang & Vásquez, 2014) who explored RNRs with Swales (1990) ESP-based genre analysis, leading to the establishment of RNRs as a genre and sub-genres. Both monospecific and cross-sectional analyses on RNRs have been done to gain insight into linguistic phenomena in RNRs. While the former focuses on the generic structure and linguistic features of RNRs generated by a single hotel group, the latter concerns similarities and differences between RNRs generated by two groups of hotels. However, the practical applications of such analyses remain disputable. Although the textual organization and some salient linguistic choices were unveiled, further investigation is still required in order to maximize the benefits of the findings.
Given the increasing importance of RNRs in recovering clients and hotels' image (Bambauer-Sachse & Mangold, 2011;Vermeulen & Seegers, 2009), this study extends the existing genre framework of RNRs to the development of a ready-to-use instructional manual which is beyond the scope of the previous studies. The objective of this study is threefold; (1) to identify moves and steps of RNRs and their occurrence frequency, (2) to identify typicality of each move, and (3) to generate recurrent word combinations (RWCs) of each move and step. This study's findings will be beneficial to business related English for Specific Purposes (ESP) instruction and professionals in service industries.

ESP-based genre analysis
In general, the term genre is known as a type of text. Defined in the field of applied linguistics; however, "a genre comprises a class of communicative events, the members of which share some set of communicative purposes" (Swales, 1990, p. 58). Swales' definition and analytical framework have changed the way researchers and practitioners process ESP texts and revolutionized ESP research and practice (Cheng, 2015). The trend of ESP-based genre analysis has been around in the field for over 25 years, being famous for the capability to uncover communicative functions, sequences of generic structures (if any), and linguistic features of a particular text type. Such elements distinguish the text from one another, creating genres and sub-genres within the same community. Despite being eminent for different components in research articles from various fields, later in 2000s, Swales' ESPbased genre analysis was made even more popular by ESP researchers whose studies focus on professional documents such as business letter of negotiation (Pinto Dos Santos, 2002), replies to customer inquiries (van Mulken & van der Meer, 2005), text computation letters (Flowerdew & Wan, 2010), and purpose statement in graduate school applications (Samraj & Monk, 2008).
The main implication of the findings from analyzing professional genres has been that the ESP researchers and practitioners are expected to put them into practices by, for example, raising EFL/ ESL learners and professionals' awareness of the existence of the conventional structure of a particular text type. Although Bhatia (1993) has demonstrated the application of a genre analysis's findings in the pre-pedagogic stage, the findings of genre analysis studies are still underused to develop instructional materials for language learners and professionals. That is, the pedagogic application of genre analysis studies is not parallel with the contribution that genre analysis offers in adding understanding of a discourse community and linguistic phenomena. In reality, the role of ESP-based genre analysis is much more than just addressing questions of how and why the text is constructed the way it is, but rather, it plays a crucial role in bringing authenticity into ESP classrooms and offering practitioners or potential practitioners resources that can help them achieve their shared communicative purposes. With the current study, we commit to promote the utilization of a genre analysis's findings in responses to the original goal which is to make practical contribution to both language education and professional communities.

RNRs as a genre
Due to an increasing interest in RNRs as a genre, several generic structures of RNRs have been proposed. Zhang and Vásquez (2014) are the first to explore RNRs through the lens of discourse genre analysts. By analyzing RNRs produced by four-and five-star hotels in China, they proposed a generic structure comprising ten moves which was arranged based on the frequency. Panseeta and Todd (2014) investigated RNRs of five-star hotels in Thailand and then assigned five moves containing some steps within each. Panseeta and Todd (2014) generic structure was organized by the sequences and typical locations of moves and steps. Ho's (2017) study focuses on RNRs produced by hotels in various cities in Asia, and another set of communicative functions was discovered. A variety of the generic structures proposed by the previous study implies a variation of the RNR genre, leading to a call for future research on RNRs generated by service providers located in other regions. Nevertheless, there are room for replication. In some studies, the existing generic structure was adapted in alignment with the emergence of new communicative functions appearing in a certain corpus. Napolitano (2018), for example, employed Zhang and Vásquez (2014) move structure to analyze how restaurant managements in UK and Italy reply to negative comments on TripAdvisor. Thumvichit and Gampper (2018) employed Panseeta and Todd (2014) move structure as a starting point to analyze RNRs produced by local hotels and top hotels. Since some new communicative functions emerged while coding was proceeding, the adopted move structure was adjusted. Thumvichit and Gampper (2018) cross-sectional study, however, expanded the traditional analytical framework of RNRs. Apart from the move-step identification and frequency analysis, their work draws on the discourse community members for improved understanding of why RNRs are written in the way they are. They concluded that the managements strive for either service recovery or image protection.
Despite a growing body of genre analysis research on RNRs, the concrete application of the findings has not yet been emphasized. Thus, the current study will take another step closer to the real-world application of genre analysis. This study extends the framework of RNR genre by utilizing genre analysis to develop a guideline manual for composing an effective RNR. The following research questions navigate this study to the development of the instructional manual: (1) What is the frequency of occurrence of each move and step?
(2) What is the typicality of each move ?
(3) What are the most frequently occurred recurrent word combinations of each step?

Corpus compilation
All the RNRs were drawn from the review section of UK hotels' homepages on TripAdvisor. Initially, the hotels were identified by inserting the search codes of Luxury and 5-star. This type of hotels is known to provide top-end services for their clients. We then browsed through each hotel's homepage to locate an RNR. To ensure the negativity of the reviews that RNRs had to deal with, RNRs in the Terrible section was targeted. It is important to note that not every top hotel has a negative review nor an RNR in this section. Thus, without an RNR in this section, some hotel were excluded from this research. In the case that there are more than one RNR in the Terrible section, the most recent one was taken from each hotel as it reflects how present-day RNRs are composed. After visiting all the hotels' homepages, we identified 72 RNRs produced by 72 different hotels, making an RNR corpus of 72 RNRs of 8,037 tokens.

Analyzing moves
To identify the frequency of each move and step of RNRs, this study employed Panseeta and Todd (2014) move structure of RNRs as an initial coding scheme. The rationale behind this decision is that the framework provides detailed descriptions for moves and steps. The adopted framework comprises five moves and several steps representing distinctive communicative units. Following Swales's (1990) notion of move analysis, each communicative unit was labeled in accordance with the function it serves in the message. Coding proceeded as we read through the corpus multiple times to manually label each expression and ensure clear-cuts between coded units.
Due to the subjective nature of the manual coding involving researcher's own judgments, the reliability of coding has to be evaluated (Biber, Connor, & Upton, 2007). Like many previous topdown move-based studies (e.g., Henry & Roseberry, 2001;Kanoksilapatham, 2005), the current study used inter-coder reliability check to measure the degree to which coders agree on the coding descriptions and the identification of communicative units. Cohen's kappa (k) was used as a statistical method to determine the achievement of coding. Another individual, who holds a strong background in business discourse and tremendous experience in English language teaching, was invited to serve as an external coder. Initially, a short training session was given to familiarize the other coder with the coding scheme and ensure full comprehension of moves and steps' boundaries, and the coding rubric was provided. 25% (18 RNRs, 116 communicative units) of the total number of RNRs were randomly selected and assigned to the other coder for individual coding. The inter-coder reliability examination yields and the k value of .926 1 which is considered excellent 2 (Fleiss, 1981). Despite the subjectivity in the coding, such figure implies strong inter-coder reliability in identifying communicative units between coders.

Analyzing typicality of moves
Typicality of communicative units refers to the degree of conformity with the norm each move holds in a particular communicative event. Typicality analysis is considered one of the integral components of genre analysis studies because of its practical implication in textual construction. Addressing the typicality of moves allows prospective members of a discourse community to realize the degree of conformity of each communicative function. In general, moves can be categorized into three different kinds based on their typicality in the text-obligatory, conventional, and optional moves. The typicality of each move can be identified through frequency analysis, which provides the information regarding how frequently each move occurs in the text. However, using frequency data alone is not sufficient to justify each type of move. Thus, we divided moves of RNRs into three different types based on their frequency of occurrences and previously proposed definitions, which together define the terms obligatory, conventional, and optional. The cut-off points were decided after the frequency analysis and discussion of relevant literature.

Examination of recurrent word combinations
Following the typicality analysis, recurrent word combinations (RWC) were examined. We chose RWCs as a target linguistic area not only because they play a leading role in characterizing RNR genre as they reflect the language choices that fulfill communicative purposes, but they also inform the prospective members of alternative expressions they can use to compose RNRs. AntConc 3.4.4w, a corpus analysis toolkit for concordancing and text analysis, was used to generate RWCs of each move and step. We followed Altenberg's (1998) definition, taking those strings occurring more than once in the identical form into consideration. That is, any combination that occurs more than once would be qualified. Prior to the examination, sub-corpora of steps were compiled, so we could generate RWCs of each step. Each sub-corpus was then examined independently. To ensure that the RWCs generated here sound communicative and are not repeatedly presented in the lists, the AntConc 3.4.4w was configured to identify the range of 2-to 4-word combinations. The ideal is to identify the lengthiest combinations of every step, and thus, we had to bring down the gram size as some steps might not contain the lengthiest combination. The RWCs generated here were then recruited to develop the guidelines for writing RNRs.

Moves of RNRs
Analyzing the corpus of top hotels' RNRs leads to the identification of six moves and 18 steps, as contrasted with five moves and 16 steps proposed by Panseeta and Todd (2014). As coding was proceeding, a new move called brand positioning emerged. This move operates independently with its two steps-stating hotel's commitment and confirming hotel's standard. Some steps were adjusted for the current corpus in term of names and functions. Presented in Table 1, moves and steps were organized based on their usual locations, since they might appear in more than one location. For example, Salutation was placed in the first position of the structure because it is typical for the writer to start the message by addressing the reviewer. The sequence of RNRs' moves was absent, contradicting with Panseeta and Todd (2014) findings.

Salutation (M1)
Salutation is used to establish the relationship between the responder and the customer who posted a negative review on the website. It can only be found at the very beginning of the messages. Salutation in RNRs is governed by the same principle of business letters and emails. The function of this move is simply to address reviewers. Salutation confers the situations where the writer addresses the reviewer. To perform this move, the writer has to mention the name of reviewer, so they can address and show respect to the reviewer. (1) is the example of salutation.
(1) Dear + [reviewer's name] Dear was found to be frequently used across the corpus. In fact, this is the same one that is commonly used in formal and semi-formal business writings where an appropriate salutation is necessary. It implies that the RNRs can be considered a formal business text type even if they are publicly accessible.

Acknowledging feedback (M2)
Acknowledging Feedback (M2) is used to signify to customers that their comments are acknowledged and to make references to the customers' reviews. Hayes (2017) comments on this function that it is very important to stay engaged with clients and let them know that the feedback is well taken care of. This move can serve three distinctive functions in RNRs-expressing gratitude, valuing feedback, and expressing regret/concern/apology. These steps were used as strategies by the writer to acknowledge customers' complaints.
4.1.2.1. Expressing gratitude (M2Sa). One of the strategies the writer frequently uses to explicitly show the acceptance of customers' complaints is Expressing gratitude. Despite the complaints filed by customers, hotels have to express their gratitude. Simple gratitude expression can be seen in (2-3).
(2) Thank you for your review.
(3) Thank you for taking the time to share your experience with others regarding you stays at + [hotel's name] 4.1.2.2. Valuing feedback (M2Sb). To acknowledge customers' feedback, the writer may reassure the reviewer that his/her complaint is very important and it will definitely be taken seriously. Hotels raise the value of customers' feedback to show their willingness to solve the problem and improve. (4) is an example of valuing feedback.
(4) All feedback we receive is of utmost importance to us and only through your words we can learn and improve. 4.1.2.3. Expressing regret/concern/apology (M2Sc). Another way to fulfill the second move is to express regret, concern, or apology. The writer uses expressions of regret, concern, or apology to assure the reviewer that the complaint filed by a customer really concern the hotel management. The examples are shown in (5-6).
(5) We were disappointed to learn that on this occasion we failed to meet your expectations (6) … .so I can only apologise that it did not live up to your expectations on this occasion.

Brand positioning (M3)
Brand positioning is one of the integral components in marketing. Brand positioning refers to a systematic strategy to build and maintain one's positive image (van Riel & van Bruggen, 2002). Normally, brand positioning strategies are commonly known for their effectiveness in establishing competitive advantages; however, they are also beneficial in obtaining the purposes of the organization. It is commonly known as the conceptual place a business wants to take over in its target customer's mind: it shapes how customers perceive a business. Even in the case of RNRs, such strategy is used to build a positive mental image about the hotel. This move, in RNRs, plays a vital role in reputation management since it spreads a positive message about a hotel across the community. There were two steps realized in this move-stating hotel's commitment and confirming hotel's standard. Dealing with complaints constitutes the key part of RNRs as it plays a leading role in both image protection and service recovery, which are perceived as communicative purposes of RNRs (Thumvichit & Gampper, 2018). This move is employed to handle customers' complaints. As the analysis was proceeding, it was found that hotels involve a variety of sub-functions to fulfill this move. The result of the analysis shows three different steps the writers use to handle customers' complaints including explaining causes of the incident, reporting action taken, and admitting mistakes. 4.1.4.1. Explaining causes of the incident (M4Sa). This step can be seen when the writer makes an attempt to explain what caused the incident. That is, the writer brings up information and/or evidences to elucidate the incident. This can be as simple as stating what happened. The writer may also depict another side of the story to inform the customer that there is a reason behind the incident. See (9-10) for the examples.
(9) Over the last 25 years there have been a number of changes at the hotel as you correctly say.
(10) Our hotel has a "joiner free" policy but the maximum occupancy must not exceed the room limitation However, this step can also be viewed in the negative way. The reviewers and the public may perceive it as excuses hotels try to make to cover up their faults.
4.1.4.2. Reporting action taken (M4Sb). Reporting action taken/investigation result is used when hotel managements report what they have done to address the incident. Once the action has been taken, the hotel informs the reviewer of the update of the situation and/or what they have done. The example can be seen in (11).
(11) We have taken your comments regarding the noise from the corridor very seriously and we are currently working on solutions to resolve this noise issue.
In some scenarios, the writer decided to respond to the reviews before taking any action. The writer may inform the review of what he/she is going to do to address the incident. (12) is the example.
(12) … . we will discuss your comments with our senior management. 4.1.4.3. Admitting mistakes (M4Sc). In Admitting mistakes, the writer admits that it is the hotel's fault and apologizes to the reviewer. According to the research conducted by Nottingham School of Economics, University of Nottingham (2009), unhappy clients tend to be more willing to forgive the company that apologizes rather than the company that monetarily compensates. As such, a genuine apology can increase the chance to recover services. Expressions of apology may be employed to inform the customer that the hotel is taking responsibility for the situation. See (13)(14) for the examples.
(13) I'm sorry your complaint was not properly addressed at the time.
(14) I am truly sorry that we were mistaken on this occasion and that this detracted from what should have been a relaxing break.
It is important to note that this step is different from expressing regret/concern/apology. Although expressions of apology can be identified in both steps, by restating of the incident, only admitting mistakes is considered as the step that provides a genuine apology.

Concluding remarks (M5)
Concluding Remarks has to do with how the writer finalizes the message. In general, the final paragraph was devoted to this move, which characterizes the ending part of RNRs. In business correspondence, it is often seen that the write leaves the reader with a positive impression of both the writer and the message. In this regard, ending RNRs with a positive impression promotes service recovery and image protection, which are the main communicative purposes of RNR genre, and thus ending RNRs with this move allows the writer to completely fulfill the communicative purpose. The following steps were detected in the corpus. 4.1.5.1. Expressing gratitude (2) (M5Sa). Expressing gratitude (2) is similar to expressing gratitude (M2Sa) discussed earlier. Before closing the message, the writer may express his/her second gratitude in order to create the last positive impression. The example of this step can be seen in (15).
(15) Once again we appreciate to your kind comment and feedback to our service … .
The second-time gratitude looks very much like the first-time gratitude except for the fact that it has been expressed for the second time. This step can simply be identified through second-time markers such as once again. 4.1.5.2. Expressing regret/concern/apology (2) (M5Sb). As is the case with the divergence between expressing gratitude and expressing gratitude (2), expressing regret/concern/apology (2) is the counterpart of expressing regret/concern/apology. After expressing regret/concern/apology is employed early in the message, the writer concludes the message with an identical expression. (16) is the example.
(16) … .and once again I apologise that these experiences were below your, and our, expectations.
It can be seen that this case is similar to the case of expressing gratitude (2) where the secondtime marker may be used to indicate that the expression has been made for the second time.
4.1.5.3. Asking for a return visit (M5Sc). To wrap up the message, the writer may try to convince the customer to give his/her hotel another chance after the applications of previous moves which altogether strive to recover services. The writer may encourage a return visit by showing the desire to welcome the customer again. See (17) for the example.
4.1.5.4. Soliciting direct contact (M5Sd). Soliciting direct contact is realized in the conclusion part of the message. This step firmly serves the two main communicative purposes-image protection and service recovery. As for the image protection, hotels try to take the issue out of public view by offering a private channel of communication. This step also creates an opportunity to recover the relationship between hotels and customers. That is, the hotel may insist on taking responsibility for the incident. The writer simply asks the customer to contact him/her directly through telephone or email, and thus a phone number and/or an email address are provided. The direct contact given to the customer is the only concrete evidence which can confirm that the hotel is willing to take responsibility for what has happened. See (18)  This step may co-occur with asking for a return visit. For example, the writer combines asking for a return visit with promising to improve services to reassure that the customer will receive a better experience if he/she returns to the hotel. In this case, promising to improve services adds a strong sense of willingness to bring customers back, and that makes the statement even more convincing. The combination can be seen in (20).
(20) I do hope you will give us another opportunity to welcome you back at which time we will spare no effort in providing you with the perfect visit.

Closing (M6)
As mentioned, RNRs to some extent took on the style of formal business correspondence, especially when it comes to opening and closing. The following are the steps that are realized in this move.
4.1.6.1. Sign off (M6Sa). As with traditional business correspondence, sign off were found throughout the corpora. Traditionally, closing the message with a sign off shows the writer's respect and/or appreciation for the reader. Although there are plenty of sign offs out there ranging from informal to formal ones, the writers of RNRs tend to use semi-formal and formal sign offs to flow with the convention of business correspondence. Closing the letter or email with the writer's name is very common in the world of business. Although the writer responds on behalf of his/her hotel, his/her personal name are often included in the closing part as it can also be used as a reference in future interactions.
4.1.6.3. Job title (M6Sc). To provide more information about the writer, his/her job title is added. Apart from the writer's name, his/her position may be included for a future reference. This will make it easier to identify the writer. By analyzing this step, more information about the actual writers of RNRs was also discovered. Hotels often have someone in a management position handled customers' complains. This implies that customers' complains are taken seriously by hotels.
4.1.6.4. Contact information (M6Sd). The writer may leave contact information in the closing part. This step is different from soliciting direct contact, where the contact information is always included.
4.1.6.5. Affiliation (M6Se). The last step of this move is Affiliation (M6Se). The writer may include his/her corporate affiliation in the closing part. In Pinto Dos Santos's (2002) study of business letters, an affiliation included in the final part of the message was realized as company credentials (step 4 in move 4).

Typicality of moves
Once the move structure was established, we then browsed through the corpus to identify the frequency of each communicative unit. The frequency analysis reveals a total of 567 communicative units identified. Before deciding the typicality of each move, frequency of each move and step was identified (see Figure 1 and Table 2). The typicality was decided based on the frequency of moves and previous literature. To put it differently, this study draws on relevant literature and the frequency of moves to define the typicality of moves and create the cut-off points between moves, which together inform the degree of conformity of each move.

Obligatory
According to Xiao and Cao (2013) study of moves in research article abstracts composed by native and non-native speakers of English, obligatory communicative units were those that appear at 90% or above in the corpus. Cheung (2008) created a stronger criterion with the minimum of 95.6% as it was the highest frequency that could be identified in the corpus. Apart from the frequency of moves, no additional information was added to explain their decisions. Tessuto (2015) describes an obligatory move as a communicative unit that appears in every text within a particular dataset. This means that the cut-off point for the obligatory move was set at 100%: lower than 100% will result in disqualification.
However, due to the nature of public and online business correspondence, RNRs are considered an open-ended text where a distinctive convention has not yet been fully developed. The pattern of RNRs is fairly flexible. For instance, some writers applied the convention of business letters or emails as they aim to achieve similar purposes, while others do not follow such convention strictly. As a result, there is no move that occurs in every single text in the corpus. If we followed Tessuto's (2015) definition, there would be no move that is qualified for the term obligatory. Nevertheless, there are two moves that occur in almost every text in the datasets. Acknowledging feedback, concluding remarks, and dealing with complaints occurred in 97.2%, 97.2%, and 90.2% of the total RNRs respectively. The cut-off point of 100% for obligatory, as suggested by Tessuto (2015), had to be slightly adjusted in conformity with the nature of online correspondence. Thus, the cut-off point of 90% was applied.

Conventional
Conventional moves, also known as typical or quasi-obligatory moves, refer to moves that frequently occur but are not as frequent as obligatory moves. They are commonly employed by community members to achieve their communicative goals. In other words, it is typical to see them being realized in a particular dataset. As with the case of obligatory moves, there is no absolute cut-off point for conventional moves. Cheung (2008) sets the cut-off point at 59.3% as it corresponds to her dataset of sales letters. Tessuto (2015) uses the term quasi-obligatory moves to describe the moves that occur between 83% and 88%. The cut-off points seem to vary widely from one dataset to another. In some cases, conventional moves were not even presented. In Yang's (2015) study, for example, obligatory and conventional moves are integrated under the term obligatory moves, which in this case, refers to the move that occurs more than 60% in the dataset. It implies that conventional moves are important, but, unlike obligatory moves, not compulsory for meeting the communicative purpose. As such, we define conventional moves as the ones that frequently occur, and the cut-off point was set at 70%. Thus, the only move that is qualified for this term is closing.

Optional
Optional moves, as its name suggests, are the moves that are available to be chosen. They do not occur as frequently as obligatory and conventional moves. According to Henry and Roseberry (1998), optional moves are extra communicative units added by writers or speakers to fulfill communicative purposes more effectively. As with obligatory and optional moves, the operationalization of optional moves varies from one study to another. In Shi and Wannaruk (2014) (2) Expressing regret/concern/apology (2) Asking for a return visit  based study of research articles in agricultural science, optional moves refer to the move that occurs less than 60%. What seems to be obvious is that optional moves are the least frequent compared with obligatory and conventional, and the term optional is meant to cover the rest of moves that do not meet the frequency requirement of either obligatory or conventional moves. Since the current study set the cut-off point for conventional moves at 70%, optional moves would be the move that occurs less than 70%. Therefore, brand positioning and salutation are the only optional moves. Table 3 shows the categorization of moves in RNRs proposed in by this study.

RWCs in RNRs
Apart from moves and their typicality, this study explores RWCs occurred in RNRs. To do so, the sub-corpora presenting individual steps were compiled and used as the source to provide RWCs of each step. Table 4 presents the size of the sub-corpora.
Once the sub-corpora were complied, AntConc 3.4.4w was used to generate RWCs of each step (see Table 5). It was configured for 2-to 4-word combinations. The findings show that each step seems to have its own distinctive set of RWCs. RWCs were completely absent in some steps simply because they involved personal information such as a name, job title, and contact information. Apart from move analysis, which earlier distinguished the steps based on the functions they perform in the   messages, the steps were further distinguished by their own linguistic properties, which confirm their identities in RNR genre. To put it differently, the combination of move and linguistic feature analyses makes each communicative function distinctive not only in the way it functions but also the way it is composed. However, the identification of RWCs presented here is still ambiguous when it comes to pedagogical implication. As can be seen, many of the RWCs identified here look incomplete even with content words. Although generating RWCs of each step is a great way to capture its distinct linguistic features, the pedagogical implication of such combinations is questionable. Chen and Baker (2010) compared the uses of RWCs in native-speaker and non-native speaker academic writing, and they suggest that it is necessary to support quantitative analysis with qualitative analyses which expands the analysis by looking into concordance lines. Thus, the RWCs identified were qualitatively analyzed. At this time, the surrounding elements (e.g., collocates, concordances) were taken into consideration. To this end, after identifying the most frequently used RWC in each step, we read through the concordance lines where the most frequently used RWCs were discovered. For instance, the lead expression thank you for (M2Sa) was captured and analyzed for optional collocates. See (23-26) for examples.  (24) Thank you for taking the time to post … .
As it can be seen, there is a variation of collocates of the lead expression thank you for. Thus, the recurrent collocates such as taking the time to, sharing, and reviewing were extracted and used to develop the guidelines. The recurrent collocates are meant to provide potential writers with options. Another great example is the lead expression I am sorry (M2Sc). See (27)(28)(29) for examples.
(28) I am sorry to read … .
The collocates hear, read, and that were included in the guidelines as they repeatedly used by the writers. For those steps in which a RWC was absent, the frequently used words identified in the previous section were used to support the manual development. Guidelines for writing RNRs were presented in Appendix.

Conclusion
This study responds to the emergence of computer-mediated interaction between hotels and customers in social media platforms and the rapidly expanding interest in response to customers' complaints as a genre. Employing a top-down approach, this study was navigated from macro to micro level. Panseeta and Todd (2014) move structure of RNRs was used as a starting point. As the data were being coded and recoded, some new communicative units emerged; and the final move structure proposed by this study includes six moves, each of which performs a distinctive function.
Representing the strategies writers employed to fulfill the communicative purpose, steps, were identified in each move. The frequency of each move was then used to identify the typicality of each move. Thus, moves were categorized into three different types-obligatory, conventional, and optional-based on the frequency of occurrence and definitions from previous studies. Obligatory moves are the moves that occur most frequently followed by conventional moves which are preferred by most writers of RNRs. Those moves occurring less than 70% of the corpus were labeled optional.
To verify the identity of each step and develop guidelines for composing RNRs, RWCs were examined. Sub-corpora representing each step were examined individually to explore their linguistic properties. The findings indicate that there was a clear cut between steps, since the RWCs generated by the corpus tool were different from one sub-corpus to another. The identification of RWCs made great contributions to the development of guidelines for RNR construction. The guidelines were developed based on the empirical evidence of frequently used words and recurrent word combinations.

Implications
This study was initiated with the desire to gain insight into RNR genre as it plays vital roles in the hospitality industry and probably other service industries and is one of the tasks English as a Second Language (ESL)/English as a Foreign Language (EFL) students are expected to fulfill in business-related ESP courses. Due to the rapid growth of business interactions on online public platforms, business-related ESP courses should prepare students for professional communication in online platforms. Analyzing moves shows several distinctive communicative functions, which learners need to understand in order to achieve communicative purposes. It is also beneficial for learners to be aware of the typicality of each move in that obligatory and conventional moves should be prioritized as they are used frequently by professionals. Learners can also benefit from the findings of the linguistic feature analyses as they reflect the authentic choices of language used by professionals. Since a practical teaching implication is very much of concern in this study, the guidelines for RNR construction were developed to support not only English language learners, but also practitioners. The guidelines can be used as they are further developed for specific classroom instructions and professional trainings. However, the guidelines only provide learners with commonly used expressions. That is, relying solely on the guidelines is not the intention of the developer. They are still required to consider other factors such as situations they have to deal with and institutional policies. Even with the guidelines, learners need to be exposed to a wide variety of situations, which require different strategies.

Recommendations for future research
Due to the limited scope of this study, like any research, this study makes room for further investigations of RNRs. The scope of this study is limited to the investigation of top UK hotels' RNRs; thus, the findings presented here can only represent the products of top UK hotels. The discovery of new moves and steps suggests that the variations of RNR genre do exist. Future research is encouraged to analyze RNRs of other groups of hotels to add in-depth understandings of RNRs.
As far as pedagogy is concerned, this study has some practical implications for language classrooms. Although genre analysis research has been widely recognized for its practical applications in language teaching, further investigation on such applications is still insufficiently carried out. Future research should include, but not be limited to, the effects of the materials developed from the findings of this study on writing performance, for instance, and learners' and instructors' attitudes towards the materials.
Since the genre will undergo changes over time, this study must be considered only a snapshot. The last but not the least recommendation is to examine how the genre may have changed, and what factors led to the change.

Move
Step Expression Example Salutation (M1) * -Dear + reviewer's name "Dear … … .," Acknowledging feedback (M2) *** Expressing gratitude (M2Sa) Thank you for (taking the time to write/share) your feedback/comment/ review/experience "Thank you for taking the time to share your experience with us" Valuing feedback (M2Sb) All guest feedback is important/valuable "All feedback we receive is of utmost importance to us and only through your words we can learn to improve." Expressing regret/concern/apology (M2Sc) I am sorry/disappointed to hear/ read that [Expressing regret/concern] + customer's overall experience "I am sorry to hear that you didn't enjoy your last stay." I would like to apologize for [Expressing an apology] "I would like to apologise for any inconvenience" Positioning hotel brands (M3) * Stating hotel's commitment (M3Sa) We aim to + hotel's commitment "We strive to provide excellent an exclusive service experience for every guest" We strive to/for Confirming hotel's standard (M3Sb) The level of service you experienced + defamiliarize the incident " … .the level of service you experienced during your recent stay is not indicative of high standards at our hotel … ." I can assure you that your experience (Continued)

Move
Step