This study analyses data from the Sevilla and Malaga destinations in southern Spain. Both cities could be conceptualised as strongly leisure-orientated destinations.

On the one hand, Airbnb data are obtained from the Inside Airbnb website http://insideairbnb.com/ (available on 25 September 2021). InsideAirbnb is a third-party open-access site containing Airbnb listings' count data and is not associated with or endorsed by Airbnb or its competitors.

Our study preserves the amateur character of the host and selects only hosts with a single listing. Moreover, Airbnb listings are restricted to entire apartments to compare offers easily. The entire apartments represent 84–88% of the Airbnb offer in Seville and Malaga, respectively (AirDNA, 2022). In addition, Airbnb listings are considered outliers if their price and the number of beds fall outside the interval formed by the 5th and 95th percentiles. Our research removes listings with a price lower than 22 US dollars (not including cleaning fees or additional guests) and higher than 216 US dollars. Our study filters out all lodgings with more than six beds.

On the other hand, for strictly academic purposes, our study extracts (from TripAdvisor.com) the anonymous reviews of hotels, with initially more than 1000 reviews per hotel available on 25 September 2021, similar to Airbnb apartments in price. TripAdvisor reviews widely serve as a data source for previous studies (Ding et al., 2022; Gao et al., 2022; Liu et al., 2020, amongst others). The number of full-service hotels analysed here is 38 (13 and 25 hotels in Seville and Malaga, respectively). The reviews are obtained using a Python crawler.

The final dataset contains 34,917 (English) reviews from 2017 until 25 September 2021, precisely 23,259 Airbnb reviews and 11,658 hotel reviews. As a result, the volume of sentences is slightly unbalanced between the two types of accommodation.

Our text mining approach is divided into the following steps: (a) retrieval of documents and conversion from markup languages into (unstructured) plain text; (b) text preprocessing, i.e., segmentation into sentences and text normalisation; and (c) text analysis and extraction of knowledge from the corpus by identifying topics. Then:

• ■

  The text is split into logical constitutes (here, sentences) through the period (“.”), question mark (“?”), and exclamation mark (“!”) -implemented here with a customised set of rules (in Python language) that capture particular issues.

• ■

  Our approach becomes sentences in ASCII. It removes non-alphabets and any other kind of characters which might not be a part of the language. In addition, it discards extra whitespaces.

• ■

  It checks the spelling of sentences and fixes contractions and compound terms.

• ■

  It eliminates a list of common stop words to filter out overly common terms (not contributing to information content). Stop words are terms that are used frequently. Filtering them enables a reduction of the search space without losing the semantic meaning.

• ■

  Our study lemmatises the words so they can be analysed as a single item to reduce the variability of the language and avoid redundancy in the document vocabulary.

Our study proposes identifying guests' latent semantic structures (or topics) by analysing a bulk set of UGC through CTM-based data mining processing. In particular, our research applies a topic model with contextualised BERT embeddings to enhance the quality of topics; that is, overall coherence and interpretation (CombinedTM, CTM 2.2.0 in Python; cf.Bianchi et al., 2021a). Likewise, CTM fosters higher competitive topic diversity than classical topic modelling (e.g., Latent Dirichlet Allocation). As Bianchi et al. (2021a, p.760; cf.Bianchi et al., 2021b) conclude, “topic models [could] benefit from latent contextual information, which is missing in Bag-of-Word (BoW) representations”. CTM thus allows obtaining contextualised representations of documents. In this sense,

• ■

  The CTM here uses preprocessed sentences (derived from the cleansing process) for BoW. Our preprocessing reduces the variability of the language and avoids redundancy in the narrative vocabulary.

• ■

  BoW represents the input in an inherently incoherent manner, however. Accordingly, CTM combines BoW with contextualised embeddings from the not-preprocessed (only lemmatised) corpus to significantly increase topic coherence and competitive topic diversity. Here, the sentence encoding model is the pre-trained ‘all-mpnet-base-v2’ that maps sentences to a 768-dimensional dense vector space.

• ■

  Assuming there is no single correct path to select an optimal number of topics, our study tries different topics (10, 20, 30, 40 and 50). Our analysis of the coherence metric and a handy number of topics finally sets 30 topics (100 epochs and 50 samples). In addition, it estimates the document-topic distribution matrix to which a logarithmic transformation is applied due to its significant right skewness.

In short, the CTM is trained with narrative input representations that consider word order and contextual information and bypass one of the main limitations of BOW models (Bianchi et al., 2020).

In Table 1, our study displays the most relevant terms per topic. The topics are numbered from 0 to 29. The extracted topics are easily interpretable and offer a coherent impression.

Our study initially outlines the topics related to a general assessment of the service received. Topics 22 and 25 relate to guest dissatisfaction triggered by a negative service encounter that brings about complaints. That is to say, the service experience offered would not correspond with, for example, those reflected in the published photographs, consequently damaging the trust between the two sides involved. Therefore, service failures (and negative experiences) are determinants of the brand-hate users express through UGC. On the contrary, topics 11, 14 and 19 are associated with a positive overall experience (guests' satisfaction). Finally, topic 27 relates to behavioural intentions indicating whether guests remain in or defect from a hospitality firm.

Second, our study describes the topics related to the stay's features (or conditions) that could determine the levels of satisfaction or dissatisfaction. Below, our analysis summarises the semantic content of each topic:

• ˗

  Core services are associated with accommodation quality, i.e., topics 2 (accommodation size and type), 16 (functional space and decoration and cleanliness regarding core services), 21 (noise complaints derived from out-accommodation activities) and 28 (amenities such as water machine, refrigerator, drying machine, and bathrooms). Topic 20 also relates to feeling “at home” or hedonic comfort while staying in a hospitality accommodation. Finally, topic 7 refers to comfortable, branded accommodation with nice facilities. In other words, core services refer to the essential benefits that guests seek (e.g., a cosy place to live).

• ˗

  Relational drivers are inherent in hospitality services: responding to inquiries, keeping guests well informed, or simply making guests feel welcome. They are related to customer service performance (e.g., communication about directions, tips or advice, accommodation rules, Wi-Fi instructions, or check-in/out, amongst others), for instance, how staff deliver the service to make a hospitality experience satisfactory or dissatisfactory. Topics 9, 18, 24 and 26 focus on staff interactions (here, the term staff replaced the term host or the proper nouns used) and professionalism (e.g., friendliness and helpfulness, amongst others) associated with booking, checking guests in and out, or maintenance services. Topic 17 is related to additional services, e.g., seamless, effortless transaction and booking tasks.

• Relational topics are thus associated with a feeling of welcome or personal interaction with the staff, e.g., its reliability (proper performance), responsiveness (knowledge and courtesy) and empathy (ability to care).

• ˗

  Site-specific characteristics (convenience and access) are related to the location, i.e., easy accessibility to the hotel or other hotspots, attracting many visitors. Site-specific factors include distance to major attractions (topics 1 and 8) and pedestrian-orientated infrastructure encouraging people to walk quietly and peacefully from accommodation to hotspots (topic 29). Additionally, topic 12 focuses on public transportation hubs (trains or buses) and accessibility to tourist destinations by providing, for instance, shuttles. Finally, topic 15 concerns restaurants and groceries (neighbourhood amenities) that offer more high-quality experiences.

• ˗

  Property characteristics are related to services such as a terrace overlooking the city that enhances customers' delight (topic 6), a swimming pool for sunbathing (topic 13), or incidences such as long waiting times in crowded car parks (topic 3) -also related to location.

• ˗

  Topics 4 and 23 relate to staying with friends or family and its duration. Likewise, topic 0 is associated with an experiential value proposition for the different guest profiles- an evolving feeling of communion with family, friends, or other people- during hospitality experiences.

• ˗

  Topic 5 relates to value for money, i.e., allocating resources more efficiently. Value for money could be considered a relevant attribute for attracting guests (Nash et al., 2006), related to features such as comfort and stylish design.

• ˗

  Our analysis extracts topic 10 related to high-quality food, breakfast, or variety. However, as Airbnb rental apartments do not usually provide breakfast, our study does not consider this topic in subsequent approaches.

Our analysis employs the transformed sentence-topic distribution matrix (averaged by narrative) and focuses on necessary topics or critical conditions of (dis)satisfaction. Our study applies NCA (NCA 3.1.1 package in R, Dul, 2021) that identifies critical levels of these predictors (conditions) that must necessarily be present to achieve the desired result (Arenius et al., 2017; Dul, 2016). In other words, “performance will not improve by increasing the values of other determinants unless the bottleneck is taken away first (e.g., by increasing the value of the critical determinant)” (Dul, 2016, p. 15).

First, our analysis applies CR-FDH less sensitive to outliers. CR-FDH is the default technique for parametric data. Second, our research analyses the potential presence of a significant empty zone in the upper left-hand corner that would illustrate the constraint of the necessary condition. See Fig. 2. This means the necessity of X (or high level of X) for the existence of Y (or high level), i.e., there should be no narratives (or at least very few) with a low proportion of topic i and a high output value. Third, our study calculates the size of the necessity effect (d) or the relative size of the empty zone, i.e., to what extent the condition is necessary for the outcome. d ranges from 0 to 1. Our analysis considers only the conditions that are practically (here, d ≥ 0.1) and statistically significant (p-value ≤ 0.01) and, as a result, rejects the effect size being the result of random chance (cf.Dul, 2016 and Dul et al., 2010, amongst others). Fourth, our study sets c-accuracy ≥ 0.99, i.e., the percentage of observations on or below the ceiling. Finally, to be even tighter in determining the preconditions of necessity, the topics here need to be observed at no less than 30% (Bottleneck column in Table 2.) to reach an appropriate outcome level in practical settings (50%).

Fig. 2.

Upper-left-corner. Source: Dul (2021).

(0.21MB).

The top critical conditions of Airbnb dissatisfaction (in comparison to hotels) are the following:

• ■

  Topic 17 relates to additional services, such as reservations, waiting times, or the ease of making a reservation.

• ■

  Topic 9 is associated with staff recommendations, advice, and suggestions to enjoy life as a local or share local tips to give guests a unique experience.

• ■

  Topic 1 relates to proximity to the main heritage attractions.

• ■

  Topic 15 is composed of terms on facilities in the neighbourhood, for instance, restaurants and grocery markets that offer more high-quality experiences.

• ■

  Topic 24 is associated with staff interactions and professionalism, such as easy check-in/out.

In particular, to reach a 50% level of mention of dissatisfaction, topic 17 should occur at no less than 34.6%, topic 9 at no less than 30%, topic 1 at no less than 31.2%, topic 15 at no less than 35.6%, and topic 24 at no less than 31%. See the column' Bottlenecks-50%' in Table 2.

On the other hand, only three preconditions are revealed (topics 23, 20 and 26) as necessary conditions for the dissatisfaction of hotel guests during their stays. Topics 23 and 26 relate to the overall (positive or negative) experience with the family and trip length (topic 23) for both types of accommodation and, in addition, replies and scheduled messages to keep in touch with guests while staying in a hospitality accommodation (topic 26). Moreover, topic 20 relates to the perceived trustworthiness of the accommodation's photos. Without the 32.3% level of topic 20, there is no possibility of reaching the 50% level of output value (Dul, 2016). Hotels should thus provide an accurate description to shape guests' expectations.

Moreover, the main necessary conditions for satisfaction coincide in both types of accommodation, although the order of importance differs. For Airbnb guests, the top 3 topics are 17, 9 and 24. Regarding hotel guests, the top 3 topics are 23, 15 and 17. See also Table 2.

One essential contribution of this section is to answer the following research question: What key accommodation characteristics provide the most accurate and best separation of the two classes of data (i.e., Airbnb accommodations and hotels)? To answer this question, our analysis applies Extreme Gradient Boosting decision trees (from now on, XGBoost) on the transformed document-topic distribution matrix (averaged by narrative). XGBoost is an exciting version of the gradient-boost decision tree model for classification and regression issues (Chen & Guestrin, 2016). It includes a series of optimisations and regularisation techniques that help reduce overfitting. To improve the interpretability of the XGBoost model, our analysis also combines XGBoost with SHAP values (SHAPley Additive exPlanations) for plotting models' insights (cf.Lundberg & Lee, 2016). SHAP values are model-agnostic, present local accuracy, missingness, and consistency properties, and do not provide causality.

The topics' probability distributions per narrative (obtained from CTM), except satisfaction- and dissatisfaction-based topics, and topics 27 and 10 are here used as input features of the classification model. Our research also splits the data set into two subsets to ensure the effectiveness of our approach (Climent et al., 2019):

• ■

  80% of the data (train dataset) allows us to train the XGBoost model to find the best combination of parameters. Our research performs 10-fold cross-validation.

• ■

  The remaining 20% (validation dataset) validates the best-fitted model's performance and ensures that the model is generalisable.

The grid search of the scikit toolkit is used to select the parameters to achieve the best area under the curve (AUC) classification metric. The XGBoost classification model -or XGBClassifier- (after parameter tuning, i.e., colsample_bytree: 0.5, eta: 0.5, gamma: 0.2, max_depth: 4, min_child_weight: 9, n_estimators: 600, subsample: 0.4) obtains a prediction accuracy of 90% on the testing dataset. The ROC curves for the XGBClassifier are presented in Fig. 3. The ideal point is the upper left corner of the plot; that is, false positives are zero, and true positives are one. Our analysis also estimates the Matthews correlation coefficient (0.77 > 0.60). This coefficient generates a high score only if models can predict a high percentage of true positives and negatives.

Fig. 3.

ROC curves for XGBClassifier.

(0.11MB).

Once the tuned XGBoost model is fitted, the next step is to identify the importance of the features. In Fig. 4a the x-axis represents the average of the absolute SHAP value of each feature (i.e., magnitude) estimated from the testing data set. In Fig. 4b, our analysis plots the magnitude and directionality for Airbnb (class: 0) versus hotels (class: 1) to jointly visualise the features' importance and impact on the prediction. Fig. 4b shows how much each feature (or topic) contributes to the target variable (or accommodation type). Topics with high SHAP values (see x-axis) push thus towards class 1 (Hotels), while low SHAP values (see x-axis) push towards class 0 (Airbnb). In the following, our study presents the main results by accommodation type.

Fig. 4.

The importance of topics.

(0.26MB).

According to Airbnb stays the main features that predict the Airbnb narrative are the following.

• ■

  Topic 26 focuses on replies and scheduled messages to keep in touch with guests.

• ■

  Topic 4 is associated with staying with family, e.g., family as an essential component of relationships and interactions in hospitality staying.

• ■

  Topic 16 relates to being tidy, bright, cosy, or functional. Airbnb travellers frequently mention the cleanliness of the accommodation or its excellent design.

• ■

  Topic 9 relates to staff help, sharing local treats, and giving a unique experience.

The main features that show a positive impact (x-axis) on the model output -that are most likely to be mentioned in hotel narratives- are the following topics:

• •

  Topic 2 refers to accommodation size. Its long-tail trend reaches to the right (and not to the left), thus being highly predictive for some dataset instances but not for others.

• •

  Topic 24 is associated with ease of checking in/out and professionalism.

• •

  Topic 6 focuses on property characteristics related to services, such as a terrace overlooking the city that enhances customer delight.

• •

  Topic 12 focuses on public transportation hubs (e.g., trains or buses) to facilitate accessibility to tourist destinations or attractions. Hotel travellers mostly comment on transportation hubs or systems compared to Airbnb guests.

  Topic 6 affects some predictions by a significant amount, while topic 12 affects more predictions by a smaller amount.

• •

  Topic 0 is associated with an experiential value proposition to the different guest profiles; that is, an evolving feeling of communion with family, friends, or other people (strangers), during a hotel stay experience.

• •

  Topic 17 relates to additional services such as reservations or waiting times.

• •

  Topic 13 is associated with memorable experiences and spectacular views of the surroundings. Comments such as 'experiences associated with relax, sun, swim, enjoy, or sunset' are most likely to be mentioned in hotel narratives. The entire apartment could be in residential neighbourhoods without stunning views of the surroundings.

Our analysis identifies the (sufficient) antecedents of the overall guest stay evaluations. An additional contribution is thus to answer the following research question: Are the characteristics (detected and) related to hospitality services sufficient for the outcome to be produced in hospitality services?

To answer this question, four XGBoost regression models are generated for each hospitality class (Airbnb versus Hotels) and each of the two overall stay assessments (satisfaction versus dissatisfaction). Our analysis adopts 10-fold cross-validation in the training dataset (80% of the data) to select parameters. It chooses the model with minimum root-mean-square error through scikit-learn's parameter optimiser function (cf. scikit-learn 1.0.2 package in Python). The best (standard) hyperparameters for our four models and r-square values are displayed in Table 3. There is a slight gap between the train and testing results, most likely due to minimum overfitting. Although XGBoost does not assume normality, the residuals versus predicted values do not show significant dependency.

Our analysis employs SHAP values to gain a different view of each topic's importance in predicting the model's outcome. Annexes I-II show the importance of global features for our models. The global importance of each feature is taken to be the mean absolute value of that feature over all the given samples, and the blue bar indicates that the topic is positively correlated with the target variable.

Annexe I.

Simplified SHAP bar plot.

* Bars display mean(|SHAP value|) (average impact on model output magnitude). The blue colour means a topic is positively correlated with the target variable.

(0.31MB).

Annexe II.

Simplified SHAP bar plot.

* Bars display mean(|SHAP value|) (average impact on model output magnitude). The blue colour means a topic is positively correlated with the target variable.

(0.31MB).

Below, our analysis breaks down the results by dissatisfaction versus satisfaction and accommodation type.

Our models reach an r-squared of 0.724 (Airbnb model) and 0.705 (hotel model). The characteristics that determine dissatisfaction are similar between Airbnb and hotel guests, partly motivated by the choice of entire apartments compared to a hotel room (cf.Egresi et al., 2019). In this regard, the more significant the overlap of topics that promote dissatisfaction, the greater the potential threat of substitution between Airbnb and hotels.

According to the order of Airbnb guests, the most impactful attributes for both types of accommodation that are positively correlated with guest dissatisfaction are:

• ■

  Pedestrian-orientated infrastructure (topic 29).

• ■

  Hospitable and communicative staff (topic 18).

• ■

  An evolving feeling of community feeling with people (topic 0), i.e., family, friends, and strangers, during hospitality experiences.

• ■

  Property characteristics concerning location-based problems include long waiting times in crowded car parks (topic 3).

• ■

  Noise complaints from out-accommodation activities (topic 21), i.e., noise pollution or environmental noise.

• ■

  Value for money (convenient option), i.e., not paying more for a good or service than its quality or availability justify, and not necessarily regarding seeking the absolute cheapest accommodation (topic 5).

• ■

  In-accommodation amenities, e.g., microwave, refrigerator, air conditioner, and bath facilities (topic 28).

• ■

  Additional services such as reservations or waiting times (topic 17).

• ■

  Accommodation size and type in terms of room comfort (topic 2).

Although guests expect similarly, our topics suggest different features supporting the competitive advantage between the two types of accommodation. In the case of hotels, the most impactful features on dissatisfaction are as follows:

• ■

  Replies and schedules messages to keep in touch with guests (topic 26).

• ■

  Feeling “at home” or hedonic comfort while staying in a hospitality accommodation (topic 20).

Our satisfaction-based models reach an r-squared of 0.502 (Airbnb model) and 0.382 (hotel model). According to the order of the Airbnb guest sample, the set of main characteristics that are positively correlated with guest satisfaction are as follows:

• ■

  Topic 6 amenities regarding guests' delight (roof, terrace, pool, etc.).

• ■

  Topic 29 related to pedestrian-orientated infrastructure about people walking quietly and peacefully from accommodation to hotspots.

• ■

  Topic 23 related to the length of stay with a wife or husband.

• ■

  Topic 1 associated with proximity to the main cultural monuments.

• ■

  Topic 8 on distance to major attractions.

• ■

  Topic 4 regarding staying with a family; that is, family as an essential component of relationships and interactions in hospitality stay.

• ■

  Topic 18 focused on welcoming and communicative staff.

• ■

  Topic 7 regarding a whole and comfortable brand building with lovely facilities.

• ■

  Topic 9 associated with staff interactions, focused on recommendations, advice or tips, and suggestions to know how to enjoy yourself like a local.

Although guests expect similar features, our topics suggest different features supporting the competitive advantage between the two types of accommodation. The crucial satisfaction criteria differentiating Airbnb from the hotel are site-specific factors, e.g., distance to the old town (topic 8). In the case of hotels and compared to Airbnb, the main features relate to topics 24 and 26, that are associated with checking in/out, professionalism, fast and convenient replies, and scheduled messages to keep in touch with guests. The most relevant features that impact guest perceptions about hotel service quality are thus based on efficient and appropriate interactions between staff and guests. As Xu and Li (2016, p. 63) conclude, “staff performance seemed to be amongst the most influential factors in determining customer satisfaction […] that can strengthen the customer's relationship with the hotels”.

The study used a method that transformed guest narratives into a structured form to identify the underlying characteristics. Our approach enabled researchers to recall the aspects that mattered to guests and their evaluations of various aspects of hospitality. In particular, the study applied NLP and the CTM approach to increase topic coherence and competitive topic diversity. CTM uses pre-trained language representations (here, a specific versio of BERT) to support topic modelling. BERT is an ultimate enhancement in machine learning, such as sentiment classification, and topic modelling. Our fine-tuning approach fitted a small number of parameters. This approach therefore enabled: (1) overcoming the disadvantages of probabilistic generative topic modelling (i.e., LDA and PLSA could not fit into short texts appropriately due to severe data sparsity); (2) expanding previous studies based on structural scales; and (3) applying large-scale data sources following an exploratory approach.

Moreover, the use of transformers in our study of NLP is essential due to their effectiveness in processing and understanding language. For example, transformers allow us to learn contextual relationships between terms and phrases and to focus on relevant parts of the input sequence and have led to significant improvements in the accuracy of NLP. Transformers play a critical and further role in advancing NLP research and applications. In addition, although they have some challenges or limitations that hinder their applicability or scalability for NLP tasks, our research highlights the importance of using novel techniques to interpret hidden semantic structures in the data.

The study also applied NCA analysis to interpret the necessary conditions and answer the question of which critical characteristics of accommodation provide the most accurate and best separation of the two classes of accommodation (that is, Airbnb and hotels). Furthermore, the study considered the moderating effect of the accommodation type in differentiating the direction or strength of the relationship between hospitality characteristics and (dis)satisfaction strategies.

Therefore, the approach used in our study demonstrates the importance of data mining approaches in the academic domain. Using these techniques, researchers can identify the essential aspects of guests and their evaluations of different aspects of hospitality. As O'dell (2007) notes, "experiences are highly personal, subjectively perceived, intangible, ever fleeting and continuously on-going" (p. 15). Our approach even provides a more fine-grained understanding of hospitality services and can inform concerning the design of more effective policies and practices in the hospitality industry. However, using pre-trained language representations may lead to some loss of accuracy in topic modelling.

Beyond the above considerations, the study's interpretation of topics on guest (dis)satisfaction may face limitations due to the availability and quality of the data used for analysis. Firstly, it is essential to note that our study's sample size and generalisability may be limited because we collected data from a specific location. Thus, the results may not apply to other populations or regions. Therefore, research should study other destinations with different personalities and cultural backgrounds, seasonality patterns of tourists, or tourist attractions to generalise the findings.

Furthermore, guests' cultural scripts or demographic characteristics, such as age or income, affect customer narratives. Future studies should also focus on accommodations located downtown and in the surroundings of transport hubs such as airports, ports, and P2P accommodation with different categories (i.e., entire apartments, private rooms, or shared rooms). Finally, as Sainaghi and Baggio (2020) acknowledge, a disadvantage lies in differentiating business guest segments on the one hand and those of leisure guests on the other hand.

Moreover, the study's methodology may not capture the entire range of guest experiences, as the analysis focused on narratives and may not include all aspects of hospitality services that matter to guests. In addition, the study's approach may be biased towards positive experiences since the data collection relied on guests' voluntary participation, and guests who had negative experiences may have been less likely to participate.

Our study identifies differences and similarities between Airbnb and hotels by examining the impact of various features (or topics) on guest (dis)satisfaction and further revisiting intention. Firstly, our study reveals that the key necessary conditions for (dis)satisfaction in both types of accommodation are similar but differ in their order of importance.

• ■

  Regarding Airbnb, proximity to tourist attractions, staff recommendations and staff interactions and professionalism are the three most important themes for guest satisfaction.

• ■

  Additional services, the quality of area facilities, staff interactions, and professionalism are the most important themes regarding dissatisfaction with Airbnb.

• ■

  In hotels, the quality of facilities, additional services, staff interactions, and professionalism are the three most important themes for guest satisfaction.

• ■

  In hotels, the most important topics concerning dissatisfaction are the overall experience with the family and the length of the trip, scheduled responses and messages to keep in touch with guests, and perceived trust in the property's photos.

The higher the overlap (or dissimilarity) of the preferences or conditions that promote satisfaction with using Airbnb or hotels, the greater (or lower) the potential threat of substitution. In this regard, beyond singular aspects, our results suggest that Airbnb and hotels could be substitutive services concerning the conditions of need, as guests highlight experiences and prioritise similar topics (as necessary conditions) in their choice of accommodation. Although check-in/out by unprofessional hosts appears to underperform compared to traditional hotels (e.g., Huarng & Yu, 2019), the quality and professionalism of the staff (e.g., desk, pleasant, efficient, reception, professional, member, polite, attentive, amongst other terms) were here considered critical by guests staying in Airbnb housing and hotel guests.

Secondly, the classification analysis further supports this conclusion, as the essential topics associated with each accommodation type are distinct. For example, regarding the topics that classify Airbnb and hotels, the essential topics associated with Airbnb are scheduled responses and messages to stay in touch with guests, helpfulness of staff, and cleanliness and functional design. On the other hand, in hotels, the most relevant topics are properties with features that enhance the customer experience, spacious accommodation, professionalism and ease of staff check-in/out. In addition, family experience is an essential feature concerning Airbnb, while memorable experiences, spectacular views, and additional services, such as reservations and waiting times, are essential topics in hotels. These differences highlight the unique value propositions of each accommodation type, which can coexist and cater to different guest needs.

Thirdly, our study found that both Airbnb and hotels share a similar set of topics that contribute to guest dissatisfaction, with the overlap of topics suggesting (in this case) a potential threat of substitution between the two types of accommodation. Likewise, while Airbnb and hotels have common topics that affect guest experiences, the research findings also indicate that they may complement each other, given their unique features that appeal to guest preferences.

• ■

  The topics that most impact guest dissatisfaction with both types of accommodation include pedestrian-orientated infrastructure, hospitable and communicative staff, an evolving feeling of community, location-based problems, noise complaints, value for money, in-accommodation amenities, additional services, and accommodation size and type.

  Our results thus illustrate a power shift from hosts to guests. Farmaki and Kaniadakis (2020, p. 5) note that there emerges a “limboing between hotel-style and flat-sharing approaches” regarding the importance of “providing a safe, clean, and functional environment and being hospitable”. Hosts likely preserve a welcome feeling to compensate for being in residential neighbourhoods rather than a downtown tourism area and compete with the hotels offering comparable services. Airbnb has become a guest-orientated P2P platform, i.e., the sharing economy model also focuses on customers.

  On the other hand, hotels emphasise replies and scheduled messages to keep in touch with guests and feel at home or have hedonic comfort while staying in a hospitality accommodation. Tussyadiah and Zach (2015, p.9) conclude that “to appeal to consumers' social and environmental motivations, hotels could make guest experiences more personal (e.g., staff meaningfully interact with guests) and socially responsible”. By providing regular updates and responding promptly to guest inquiries, hotels can create a more personalised and attentive experience that helps guests feel comfortable and at home during their stay. This communication also helps to build trust and establish a sense of familiarity between guests and staff. Like Airbnb stays, hotels also develop evolving feelings of communion with friends and strangers (cf.Arnould & Price, 1993).

• ■

  Our study found that Airbnb and hotels have different features that support their competitive advantages. For example, crucial satisfaction criteria differentiating Airbnb from hotels are site-specific factors, such as proximity to cultural monuments and distance to major attractions. On the other hand, hotels prioritise efficient and appropriate interactions between staff and guests, with staff performance being amongst the most influential factors in determining customer satisfaction. This focus on service quality is vital for hotels, as they typically have to compete with other hotels in their immediate vicinity and must differentiate based on their service level.

  In addition, Airbnb often relates to providing unique and personalised experiences that allow guests to feel like locals and immerse themselves in the culture of a destination. However, beyond the opportunity to receive valuable local knowledge from the hosts, “the opportunity to interact with locals, and (..) the opportunity to have an authentic local experience” (Guttentag et al., 2018, p.14) is not extracted here as an essential characteristic - amongst the first ones- that impact guest satisfaction with the stay in an Airbnb apartment.

Consequently, accommodation providers should identify and prioritise the factors influencing guest satisfaction and dissatisfaction to stay competitive in the hospitality sector. The following discussion focuses on key aspects highlighted in the above-mentioned main results.

Our study identifies guest satisfaction and dissatisfaction conditions in hotels and Airbnb accommodations. Overall, one of the most significant differences between Airbnb and hotels is the role of hosts versus staff. As Guttentag (2015) notes, the traditional economy (e.g., hotels) could have a competitive opportunity in staff professionalism. Though the sharing economy rhetoric is associated with sustainability and local consumption, the role of the host is also crucial for P2P accommodation experiences (Lee & Kim, 2018; Tussyadiah & Zach, 2017).

In line with our results, Guttentag (2016) and Tussyadiah and Zach (2017) find that the feel welcome topics ("social interactions with hosts and hosts' efforts to welcome and accommodate guests", cf.Tussyadiah & Zach, 2017, p. 645) are associated with higher ratings. Previous studies have shown that the feeling of security and comfort from personal interaction with hosts is essential for the hedonic value of experiences and intentions to return (Lee & Kim, 2018; Lalicic & Weismayer, 2018). In particular, the quality of interaction between the hosts and guests is relevant in shaping the guests' experience (Tussyadiah & Zach, 2017). In addition, the hosts' behaviour and attitude, such as being friendly, attentive, and polite, can also personalise guests' perception of Airbnb and their likelihood to return or recommend it to others (Mody et al., 2019; Farmaki & Kaniadakis, 2020).

Similarly, staff performance is critical to customer satisfaction and loyalty in hotels. The professionalism and competence of the staff in providing quality services and addressing guests' needs impact guest satisfaction. As Xu and Li (2016, p. 63) propose, “staff performance seemed to be amongst the most influential factors in determining customer satisfaction […] that can strengthen the customer's relationship with the hotels”. Guests are more likely to choose hotels over Airbnb accommodations due to the quality and professionalism of the staff (Guttentag, 2015). In addition, friendly and polite interactions between staff and guests can improve their perception of the hotel and increase their likelihood of returning or recommending it to others.

In short, the efficient and appropriate interactions between accommodations managers and guests are here critical to guest satisfaction and dissatisfaction with Airbnb and hotels. Furthermore, hospitable and communicative staff are significant for satisfied and dissatisfied guests in both accommodations (Xu & Li, 2016). Therefore, hosts and hotel staff should focus on providing high-quality services and addressing guests' needs promptly and efficiently. For instance, hotels need to find ways to incorporate the social and interpersonal aspects of P2P accommodation, such as personal interaction with hosts, into their services to remain competitive. Digital technology could also improve guest communication, offer personalised services, or promote social and environmental responsibility to appeal to consumers' motivations. These strategies would help hotels compete with P2P accommodation platforms and meet guest expectations.

Our study finds that property characteristics significantly influence satisfaction in both types of accommodation, and property development and maintenance must be invested in to give guests a memorable and enjoyable experience to attract them and differentiate themselves from their competitors. Our study finds that property characteristics and value for money are critical factors in shaping guest experiences in both Airbnb and hotels. For instance, property characteristics, such as a terrace overlooking the city or room amenities, enhance the guests' delight and contribute to their overall satisfaction (Belarmino et al., 2019). In addition, traditional hotels have established reputations for providing safe and secure accommodations.

In this regard, an issue related to P2P accommodation platforms is the lack of consistency in service quality. On the one hand, hotels typically adhere to established standards and protocols to provide consistent service. On the other hand, Airbnb accommodations vary widely in quality, cleanliness, and amenities. This inconsistency can therefore concern guests who expect a certain level of comfort and quality during their stay.

Our study found that accessibility was a critical factor in customer satisfaction with Airbnb accommodations and hotels. From a (complimentary) sufficiency logic, people walking quietly and peacefully from the accommodation to hotspots is also one of the most impactful (sufficient) features for both types of accommodation on guest (dis)satisfaction. In particular, hotels cannot change their locations, and guests amplify (in their narratives) the advantage of the transportation hubs. “Location and accessibility (…) help customers find the hotel easily, provide a good view of the surroundings, and save time for customers seeking to visit nearby places of interest” (Xu & Li, 2016, p. 61; cf. also Sim et al., 2006). Based on the highest proportion of mentions in hotel reviews, hotel managers should highlight advantages related to site-specific characteristics, for instance, a strategic closeness to tourist hotspots or transportation hubs beyond core services.

In addition, there have been concerns about the impact of Airbnb on local communities, with accusations of contributing to gentrification and displacing long-term residents. Although these concerns can affect guests' perceptions, who are increasingly aware of their travel choices' social and environmental impacts, Airbnb has traditionally proven to be a popular option for travellers looking for unique and authentic experiences. Many guests appreciate staying in local neighbourhoods, interacting with hosts, and enjoying a more personalised travel experience. Surprisingly, the prospects of engaging with the local community and the potential to immerse oneself in an authentic local experience (Guttentag et al., 2018, p.14) have not been identified as crucial features - amongst the first ones- that significantly influence guest satisfaction with their stay in an Airbnb accommodation.

Furthermore, our results suggest that hotels should emphasise advantages related to location-specific characteristics, such as strategic proximity to tourist hotspots or transport hubs, beyond their core services to attract more customers. Highlighting location-specific features, such as proximity to tourist attractions or transportation hubs, could also attract more customers and provide a more holistic and satisfying experience.

Finally, it is vital to consider the controversial aspects of these P2P platforms from the guest's perspective. One of the primary concerns for guests is safety and security, which is often a top priority when choosing a place to stay. While traditional hotels have strict security measures in place, such as keycard access, CCTV cameras, and round-the-clock security personnel, Airbnb places the responsibility on hosts to ensure the safety and security of guests. As a result, this can concern guests, mainly when staying in unfamiliar locations where they may not know the area well.

Belarmino et al. (2019) conclude that room amenities emerge as a predominant topic for hotel guests, for instance, coffee, toiletries, or furniture. Amenities in the accommodation provide benefits associated with staying in a home and influence the comments amongst Airbnb hotel guests. Although enjoying the comfort of a home is not here highly mentioned amongst the main topics with a significant impact on (dis)satisfaction, noise pollution and in-accommodation amenities impact both types of accommodation. For instance, the noise would create unsatisfactory assessments during the Airbnb stay (cf.Gao et al., 2022). In particular, although Tussyadiah and Zach (2017, p. 647) reveal "no significant effect of the comfort of a home" (i.e., homey and peaceful atmosphere and the hosts' hospitality) on ratings, the noise would here create unsatisfactory assessments.

To sum up, our results show a change of power from hosts to guests. Farmaki and Kaniadakis (2020) point out that there is a limbo between the hotel and the apartment-sharing approach and the importance of security, a clean and functional environment and hospitality.

A more global concept that pure cost savings could remain a dominant motive amongst Airbnb travellers (Guttentag, 2015; Tussyadiah, 2015), value for money is a fundamental factor in booking Airbnb and hotel accommodation on dissatisfaction.

Value for money is a significant factor in booking both types of accommodations. Our study finds that guests are generally attracted to a combination of practical factors, such as cost savings and interactions with hosts or staff, rather than experiential drivers, i.e., a balance between affordability and personalised experiences to increase satisfaction. For instance, the growing sharing supply creates severe competition and forces (in conjunction with the reputational capital of the hosts; Ikkala & Lampinen, 2014, 2015) “hosts (..) set the standards high and adjust the prices” (Lalicic & Weismayer, 2018).

In sum, it suggests that both types of accommodation must balance their pricing strategies with providing high-quality service and personalised experiences to ensure guest satisfaction and revisit intention.

Finally, the pricing structure of Airbnb and hotels may also contribute to their differences. Airbnb accommodations are more affordable than traditional hotels, which may be attractive to budget-conscious travellers. On the other hand, hotels often provide higher service and amenities, which can justify their higher prices. Overall, these factors may contribute to the differences between Airbnb and hotels. However, further research is needed to fully understand the underlying factors contributing to these differences and how they may influence guest satisfaction and loyalty.

To sum up, based on the findings of our study, both Airbnb and hotels can promote their high-quality service and personalised experiences to ensure guest satisfaction and revisit intention by implementing the following recommendations:

• ■

  Develop and maintain a high level of professionalism amongst staff, including training in communication, problem-solving, and customer service skills.

• ■

  Emphasise the unique features and strengths of the accommodation, i.e., personalised interactions with hosts in the case of Airbnb and the quality and professionalism of staff in the case of hotels. On the one hand, focus on personalised interactions with guests by offering local recommendations or customised services, such as arranging a private city tour or a cooking class with a local chef. On the other hand, emphasise the quality and professionalism of staff by highlighting their credentials and expertise, such as certifications, industry experience, or awards. Also, showcase the hotel's unique features, such as a luxury spa, an exclusive dining experience, or an art collection.

• ■

  Invest in property development and maintenance to enhance the quality of accommodation, such as offering amenities and features necessary to guests, such as a terrace with a view or comfortable furniture. On the one hand, ensure the property is in excellent condition by investing in regular maintenance, repairs, and upgrades. This can include installing new appliances or furniture, repainting or redecorating, or adding new amenities like a hot tub or a home theatre system. On the other hand, it should be ensured that the hotel's infrastructure, facilities, and equipment are well-maintained and up-to-date. This can include investing in energy-efficient technologies, modernising the design and decor of the rooms, or upgrading the fitness or conference centre.

• ■

  Offer competitive pricing strategies that balance cost savings with high-quality service and personalised experiences. For example, on the one hand, offer discounted rates for more extended stays or repeat guests, or provide a referral programme that rewards guests for referring new customers. On the other hand, loyalty programmes can be provided that reward frequent guests with unique perks, such as room upgrades, late check-out, or free meals.

• ■

  Ensure precise and responsive communication with guests throughout their stay to foster a welcoming environment and establish a strong relationship. On the one hand, respond promptly to guest inquiries or concerns via email, phone, or messaging platforms. Provide clear and detailed instructions for check-in and check-out, and offer a 24/7 support line for emergencies. On the other hand, provide guests with a concierge service that can help them with transportation, dining reservations, or entertainment options. Also, provide regular updates on hotel events or promotions via email or social media, and offer a dedicated hotline for customer service.

By implementing these recommendations, Airbnb and hotels can attract more guests and differentiate themselves from competitors. Additionally, promoting high-quality service and personalised experiences can help to enhance guest satisfaction and revisit intention, which can lead to positive reviews and an increased reputation. By identifying key differences and features that impact guest (dis)satisfaction, the present study provides valuable insights into the potential substitution or complementarity between Airbnb and hotels in the tourist accommodation domain.