Nothing Special   »   [go: up one dir, main page]
Next Article in Journal
The Differential Effects of Augmented Reality and Product Presentation Strategies on Brand Recall: An Embodied Cognition Perspective
Previous Article in Journal
Signaling Effects in AI Streamers: Optimal Separation Strategy Under Different Market Conditions
You seem to have javascript disabled. Please note that many of the page functionalities won't work as expected without javascript enabled.
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
JTAER
Volume 19
Issue 4
10.3390/jtaer19040145
jtaer-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Consumer Sentiment and Hotel Aspect Preferences Across Trip Modes and Purposes

by
Osnat Mokryn
Department of Information Systems, University of Haifa, Hanamal St. 65, Haifa 3303220, Israel
J. Theor. Appl. Electron. Commer. Res. 2024, 19(4), 3017-3034; https://doi.org/10.3390/jtaer19040145
Submission received: 16 September 2024 / Revised: 13 October 2024 / Accepted: 29 October 2024 / Published: 4 November 2024
Browse Figures
Figure 1
<p>Study 2 Phase 1 Flowchart: Evaluation of unsupervised clustering methods.</p> "> Figure 2
<p>Study 2 Phase 2 Flowchart: Evaluation of the prominence of each aspect per each trip mode and purpose group.</p> "> Figure 3
<p>Prominence of hotel aspects as calculated by the average frequency of terms per hotel aspect as expressed by each trip mode group. (<b>a</b>) Prominence of Location; (<b>b</b>) Prominence of Service; (<b>c</b>) Prominence of Food; (<b>d</b>) Prominence of Room.</p> ">
Versions Notes

Abstract

:
Travelers’ perceptions of hotels and their aspects have been the focus of much research and are often studied by analyzing consumers’ online reviews. Yet, little attention has been given to the effect of the trip mode, i.e., whether the person travels alone or with others, on travelers’ preferences as sentiment. Here, we study the influence of the trip mode and purpose using a mixed-methods approach. We conducted a user study to evaluate the perceptions of reviews across trip modes and found that star ratings do not consistently capture the sentiment in text reviews; on average, solo travelers’ text reviews are perceived as more negative than the star ratings they assigned, whether they travel for business or pleasure. We then analyzed over 137,000 reviews from TripAdvisor and Venere and found that a co-occurrence network approach naturally divides the text of reviews into hotel aspects. We used this result to measure the importance of hotel aspects across various traveler modes and purposes and identified significant differences in their preferences. These findings underscore the need for personalized marketing and services, highlighting the role of trip mode in shaping online review sentiment and traveler satisfaction.

1. Introduction

Consumer’s online opinions have been the focus of much research [1,2,3,4,5].
Online opinions or reviews have become a significant source of information for consumers looking to make informed decisions about purchasing products or services [6,7,8], with “77% of shoppers specifically seek out websites with ratings and reviews” [9]. The increasing use of review platforms has resulted in a vast amount of user-generated content, including hotel reviews. Hotel reviews can offer valuable insights into guests’ experiences, preferences, and complaints. Online opinions are critical in shaping consumer perceptions of hotels and other accommodations in the hospitality industry [5,10,11,12]. Eliciting insights and understanding the content of reviews has been the focus of much interest [13,14,15,16,17]. Uncovering the information hidden in reviews is critical for understating consumer needs, improving recommender systems, and promoting insights for businesses [18,19,20,21]. However, the influence of the trip mode on the sentiment expressed in reviews and travelers’ preferences has received limited attention [22,23,24,25].
Trips can have various purposes, such as leisure or business, and travelers’ accommodation expectations often differ depending on the purpose of their trip [26,27,28]. The term mode refers to the composition of the travel group, whether it be single travelers or different types of groups with various trip purposes, i.e., family vacations, romantic getaways, or work [24].
Here, we investigate the influence of the trip’s mode and purpose using reviews. To that end, we extract over 137,000 reviews from two major review sites, namely, TripAdvisor and Venere. Our analysis offers two contributions:
A user study that evaluates human perceptions of review texts written by travelers across different travel modes and purposes.
Data-driven experiments that assess travelers’ preferences for various hotel aspects based on their trip mode and purpose.
For the first contribution, we conduct a user study to examine perceptions of reviews and determine whether the review text provides additional insights beyond the star ratings, and whether these insights vary across different trip modes and purposes. Online opinions contain ratings that customers give to the hotel. These ratings are often used for sentiment analysis, which involves identifying the overall sentiment expressed in the review (positive, negative, or neutral). It is well known that the sentiment expressed in reviews correlates with customers’ satisfaction [29]. However, the relation between the text of online opinions and the star rating in these reviews is complex and hard to infer. In this experiment, we seek to understand whether the star rating is predictive of the perceived sentiment of the review text and whether it holds across reviews written by travelers with different travel modes and trip intents. Building on this finding, we continue with our second contribution.
The second study evaluates the prominence of various hotel aspects for the different mode and purpose groups by extracting hotel aspects from the reviews and devising a metric to identify the prominence of each aspect for each mode group.
Here, we are looking to identify the hotel aspects from the text of the reviews. The importance of extracting hotel aspects from online opinions lies in the insights that can be gained about customer perceptions and preferences. By analyzing the aspects customers mention in their reviews, researchers can identify which hotel features are most important to customers and which may need improvement. Aspect-related information can be used in hotel recommender systems, in the development of targeted marketing campaigns, for enhancing customer experiences, and in informing product development [18,30,31,32,33,34]. Much research exists on methods to automatically extract hotel aspects from online opinions [31,35,36,37,38,39].
Extracting textual components from online opinions can be challenging. Topic models, as well as prominent clustering methods, rely on the user specifying the number of aspects as a parameter [37,40,41]. These aspects are often determined according to common wisdom, as used by review sites [40]. However, assuming that aspects and their number are known in advance is problematic, as additional, latent aspects might exist in the text [42,43]. Yet, the number of hotel aspects need not be predetermined. How can we divide the text to clusters without predetermining their number?
To address this question, we evaluate several clustering methods using different representations of reviews to identify the most effective approach for mining hotel aspects and their associated terms. We conduct a comparative analysis of several methods, considering various representations of the reviews’ text and evaluating them against a gold standard created by two annotators. Figure 1 depicts this first phase of study 2, in which we evaluate various clustering algorithms against a baseline. The best-performing clustering method that matches best the gold standard, as found in Figure 1 step 3, creates clusters that correspond to the hotel aspects. Each cluster contains the terms relevant to that cluster.
In the second phase of study 2, we use the results of the first phase and combine the terms for each aspect in a separate list. Due to the unsupervised manner of the method, each aspect cluster contains all terms associated with that aspect. Thus, we have lists of terms, each corresponding to a different hotel aspect.
Figure 2 depicts this second phase of study 2, in which the prominence of each aspect per each travel mode or purpose is calculated using a metric we developed to assess the prominence of aspects for each mode and purpose group. The metric considers the occurrence of aspects’ terms in reviews of travelers with that purpose or travel model. In step 3, as depicted in Figure 2, the occurrences are aggregated and normalized and thus convey the differences in aspects’ prominence across the mode/purpose groups.
The two studies build on each other. In the first, we examine the perceptions of reviews to determine whether the review text provides additional insights beyond the star ratings and whether these insights vary across different trip modes and purposes. Building on this finding, we use the text of reviews to determine hotel aspect preferences of travelers in different modes and purposes.
The work provides valuable insights into travelers preferences and shows that different travelers modes and purposes alter travelers preferences.
By understanding these preferences, businesses in the hospitality industry can tailor their services and marketing strategies to better meet the diverse needs of travelers, thus enhancing customer satisfaction and loyalty. The findings highlight the importance of considering trip mode and purpose when analyzing online reviews and interpreting consumer feedback to improve hotel offerings.

2. Materials and Methods

2.1. Background and Related Works

Hotel guests vary in their expectations, and these often rely on the purpose of their trip, may it be leisure or business [23,27]. Moreover, their expectations of the hotel, the importance of its different aspects, and how they are appreciated, are expected to vary based on the purpose of the trip [23,27]. Glaesser et al. [11] found that a majority of trips are motivated by either leisure or business. Rajaguru and Hassanli [23] examined the interaction between the trip’s purpose and the hotel’s star rating on guests’ satisfaction as extracted from their reviews. The study considered the review star ratings on TripAdvisor as a user satisfaction indicator and found that business travelers differ in their expectations from leisure travelers. Like this work, the study considered “couple’s trip, a trip with family, a trip with friends and a solo trip” as leisure [23]. Recent studies, however, relate to these various compositions additionally as the trip mode [15,24,28]. Here, we will refer to the purpose of the trip in the case of business, and to the mode of the trip in cases of leisure trips.
Further, we conduct an experiment to see whether the review’s star rating correlates with the perceived sentiment of the review text and whether reviews written by people with different trip purposes and modes differ in how their text is perceived compared with their review rating.
Online reviews act as a form of electronic word of mouth in which consumers share their experiences [44]. Reviews have a major impact on consumer decisions as well as on firm profits [9,45]. In the tourism industry, electronic word of mouth has emerged as a key source of information and a powerful catalyst for decision-making [10,18,46,47,48,49]. Understanding reviews and their sentiment has thus become a key interest [2,20]. However, understanding the sentiment expressed in reviews is often considered a “suitcase research problem” that consists of many parts [50].

Determining Sentiment and Aspect Extraction

Within this suitcase research problem [50], extracting aspects is a fundamental task. Understanding the various aspects contained in the text, which is achieved through aspect extraction and sentiment polarity, has attracted extensive attention [51,52]. Wang et al. [53] studied the preferences of different purpose groups by analyzing the ratings given by the various groups to various hotel aspects and looking for distinctive terms in the text of their reviews. However, they used the aspects that were determined by the review platform. The work in [24] is the closest to ours, as it researches the sentiment expressed in reviews written by travelers across various trip modes. It uses unsupervised methods to identify prominent terms and use these terms as aspects. Yet, often in reviews, different words may be used to refer to the same general aspect of a hotel. For example, words like ’area’, ’street’, and ’metro’ may all refer to aspects of a hotel’s location. Hence, we identify clusters by looking for groups of words about a specific aspect.
Fully supervised methods are very prevalent in recent years [2,54], yet they require that the aspects are identified in advance. Among the semi-supervised methods, topic modeling, mainly Latent Dirichlet Allocation [55], is a prominent method for extracting aspects from reviews [51,56,57,58,59,60,61,62,63]. In Latent Dirichlet Allocation, each review or set of reviews is viewed as a random set of words over a distribution of latent topics, while each topic is a multinomial distribution of words. The method is semi-supervised in that the number of topics is predetermined, requiring one to decide in advance on the number of possible aspects in the reviews.
Unsupervised methods partition the text into clusters or communities of similar content to detect the aspects [18,64,65]. Clustering methods can be used to group the text of hotel reviews into coherent clusters, followed by extracting aspects discussed within each cluster [16,18,39,66].
Unsupervised clustering methods utilize machine learning algorithms to identify patterns and relationships between words and phrases used in the reviews, providing a more efficient and scalable way to analyze the content of hotel reviews. The effectiveness of clustering algorithms depends on the dataset and the application [39,67]; knowledge from the application domain is used implicitly or explicitly in every clustering algorithm [68]. Choosing the similarity measure, selecting the pattern representation scheme, and selecting a grouping scheme are examples of implicit knowledge [68,69].
In this study, we evaluate several unsupervised clustering methods and find that when the text of reviews is represented as a graph of co-occurring terms, it can be naturally divided to reveal the various aspects discussed in the reviews. We then use this approach to identify which aspects are most prominent for trip mode and purpose.

2.2. Data

The dataset used in this study was extracted from two well-known travel search engines, namely, TripAdvisor.com and Venere.com. We selected TripAdvisor and Venere as our primary sources for hotel reviews because their extensive and diverse user bases provide rich data essential for understanding consumer preferences and needs across various trip modes and purposes. These platforms offer a wide range of reviews from different traveler compositions—including solo travelers, families, and business groups—and purposes such as leisure and business trips.
The data were collected for four European cities: Munich and Berlin in Germany and Milan and Rome in Italy. For each hotel, the data contain general information about the hotel (e.g., name, address, average rating, stars, price, etc.) and a set of reviews written by hotel guests. The reviews include the travel purpose or mode of the reviewer, nationality, rating, review text, and additional metadata. The data include reviews that were written before January 2011. We collected 84,968 reviews for 1930 hotels from TripAdvisor and 52,266 reviews for 1845 hotels from Venere. Table 1 summarizes our data.
The ratings extracted from TripAdvisor reviews range from 1 to 5, whereas in Venere, the review ratings range from 1 to 10. Also, Venere allows those who write reviews to indicate their purpose or mode when staying at a hotel. They have many fine-grained mode/purpose categories; thus, we aggregated them into our own travel mode or purpose groups: [Couple, Family, Group, Single traveler, Business, Not Specified]. We will refer to them in the following text as travel mode to save repetition; however, they include the purpose “business”. Table 2 shows the number of reviews for each travel mode, with five mode categories. For each mode group, we obtained thousands of reviews, from 6541 reviews written by people on business trips to 60,113 reviews written by couples. Overall, we collected information for 3775 hotels corresponding to 137,234 reviews.

2.3. Experiment One Design: Comparing Human Perceptions of Textual Evaluations and Star Ratings from Travelers with Different Trip Modes

In this experiment, we provide evidence that the review text adds information over mere ratings and should be analyzed according to the trip’s mode.
The relationship between a user’s rating and the text of their review is tricky for the following reasons. Some cultures tend to rate high, while others rate low [70]. This distinction in behavior can also happen across individuals. For example, a “3” rating may not mean the same thing to different people. So, two individuals could rate a hotel as “3”, yet one may use many positive adjectives in their review, while the second could contain mostly negative adjectives. We found that more than 80 % of the ratings from the two datasets (with the Venere data normalized to that scale) are between 3 and 5, with little use for low ratings. This trend of a relatively narrow distribution of user ratings is not specific to our dataset and has also been reported for other datasets like the Netflix and MovieLens datasets. This first hints at why ratings alone are inadequate to help users distinguish between hotels and the importance of mining the review text for insights.
We investigate the perceived sentiment of the review text as follows. First, we examine how the review text differs from ratings by examining the overall perception of the reviews by Mechanical Turk (MT) workers. Second, we examine the experiment results and ask whether the different travel mode groups rank differently, in which case the rating they give might be biased.
To evaluate the correlation between the numerical ratings and the text of reviews in our datasets, we designed the following online experiment using Mechanical Turk workers. The idea was to let the Turkers read the reviews and estimate which star rating the original reviewer gave based solely on the text. A Turker’s estimate thus captures the review’s sentiment perception based on their text, and we can compare this to the actual star rating.
We recruited 760 Mechanical Turk workers. Each worker was given the text of five different hotel reviews, and their task was to estimate the star rating. We used reviews from 50 different hotels taken from Venere.com, where ratings are on a scale of [1–10] (we used the original ratings, un-normalized to the 1–5 scale). Based on the text, we then asked the workers to estimate the rating the reviewer gave. The primary objective of this study is to analyze the discrepancies between estimated and actual customer review ratings within the hospitality industry. By quantifying the extent and direction of these differences, the study aims to identify whether the text of reviews may hold a different sentiment than the one perceived in the star rating and whether reviews written by people with various trip modes or purposes are perceived differently.
We calculated the average completion time across all Turkers and did not include the estimations of 17 workers whose completion time was less than the average time to complete—two standard deviations. We obtained 5–7 perceptions of the star rating for each review, yielding 3715 estimates.
Unlike studies that prioritize inter-rater agreement to assess consistency among evaluators, this analysis was centered on understanding the nature and extent of estimation discrepancies relative to actual ratings. By quantifying how estimations deviated from actual ratings, the study provides a macro-level perspective on the value of star rating to reflect the actual sentiment in the text of a review.

2.4. Experiment Two Design: Identifying Travelers’ Hotel Preferences by Trip Mode Through Unsupervised Aspect Extraction

Here, we identify aspects by clustering the aggregated texts of the reviews to identify the terms associated with the different hotel aspects. The results of the clustering algorithms are evaluated against a gold standard.
Clustering is a popular unsupervised learning technique in various fields, including natural language processing, computer vision, and social network analysis. There are many approaches to clustering [60,71,72,73], including hierarchical clustering and partition or flat clustering. Hierarchical clustering aims to produce a nested series of partitions, while partitional clustering produces only a single partition of the collection of items into clusters [68,74]. Examples of partitional clustering algorithms include k-means [75,76], DBSCAN [77], spectral clustering [78,79], and affinity propagation [80]. Graph-theoretic clustering methods are also related to partitional-based clustering. Examples of hierarchical clustering algorithms include the minimum-variance method (ward) [81] and complete-link [82]. The number of clusters, k, is usually either an input parameter or found by the clustering procedure itself [83].
In natural language processing, the choice of similarity measure, pattern representation scheme, and grouping scheme can significantly impact the performance of clustering algorithms [68]. For example, some studies have found that the K-means algorithm may converge to a locally optimal solution, resulting in suboptimal clusters [84]. Similarly, DBSCAN may not be effective for high-dimensional data such as text [67]. On the other hand, spectral clustering algorithms have been reported to be quite effective for high-dimensional data such as text [67].
Another approach to text partitioning is to use graph-based community detection algorithms [85]. Frequent features pertaining to different aspects are often found in sentences together. For example: The Location was great, and the room was very clean. This might be a barrier to spectral methods but may prove less problematic if the text of all reviews is represented as a graph. We experimented with three models: spin glass [86], Fastgreedy [87], and Walktrap [88]. The spin glass algorithm seeks to find the maximal modularity by minimizing the energy of the feature network graph. The Hamiltonian (Equation (1)) is used to accomplish this, where the existing internal edges and non-existing external links minimize the Hamiltonian while existing external and non-existing internal links increase it. The spin glass model aims to find a partition that minimizes the Hamiltonian by assigning the community indices as spin states.
H ( { σ } ) = − ∑ i ≠ j a i j A i j δ ( σ i , σ j ) ︸ internal links + ∑ i ≠ j b i j ( 1 − A i j ) δ ( σ i , σ j ) ︸ internal non − links + ∑ i ≠ j a i j A i j ( 1 − δ ( σ i , σ j ) ︸ external links − ∑ i ≠ j b i j ( 1 − A i j ) δ ( σ i , σ j ) ︸ external non − links
where A i j is a boolean adjacency matrix, σ i ∈ 1 , 2 , … q denotes the indices of the communities, with q the number of maximal communities. Ref. [89] showed that the division does not depend on q for large initial q values.

Data Representation for Community Detection

Community detection algorithms divide graphs into clusters of highly connected elements. To create the graph, we build a network graph in which each node corresponds to a feature (a term). We then continue to choose how to assign the links. Several works consider the probability of two graph nodes being connected as high, given their adjacency in the origin data, which here corresponds to the probability of two terms (features) appearing together in a sentence [18,89].
Two metrics were used for the link weight: one is WordNet distance; this metric measures the distance from the least common ancestor node in WordNet of two features. The other is the PMI-Pointwise mutual information weight measures the information overlapping between two random variables [90,91], described in Equation ((2)):
P M I i j = log ( p ( i ∧ j ) p ( i ) · p ( j ) )
where p ( i ) is the probability that the feature i appears in a sentence.

3. Results

3.1. Experiment One: Perceptions of Reviews Compared with Star Ratings

We checked whether the workers’ rating estimates were similar to those the person writing the text gave, by evaluating the correlation between the numerical ratings and the Mechanical Turkers’ perception (estimate) of the text of reviews in our datasets. We computed the difference between the estimated and actual ratings for each estimate. We computed one average across all of the overestimates (when the estimate exceeded the actual rating) and a second average across all of the underestimates (when the estimates were lower than the actual ratings). The results are shown in Table 3 where the percent difference captures the average difference in ratings. There is a clear skew between the estimated rate and the actual rate. The skew is more significant when the workers perceive reviews as negative. Namely, the difference between the actual ratings and the ratings one might expect from the text is about 16%. This means that reviewers may rate generously with the star system but write more negative commentary in their reviews. Two conclusions are thus possible. In one scenario, the ratings reflect the true opinion of the user even though the text contains negative commentary. In the other scenario, the text is closer to the actual user opinion, and the star rating is overly generous. In either case, these results indicate that star ratings do not consistently capture the sentiment in text reviews, and thus the predictive relationship between text and reviews is weak.
We further examined whether mode groups differ by the star ratings they give. That is, whether people with different trip modes rate hotels differently.
Table 4 summarizes our findings for the difference in ratings per different travel modes. Indeed, we found a distinction in the average rate given by different mode groups. While the highest average rating given by people with the same mode is 4.15 (families), the lowest is 3.6 (business people). Single travelers, on average, rated hotels lower than groups. However, the results were not significant.

Human Perception of Review Sentiment Across Different Trip Mode Groups

We showed that the text of different mode groups differs in the importance it gives to the different hotel aspects. Here, we examine if the review text of different mode groups is also perceived differently.
We return to our Mechanical Turk experiment and ask whether the skew varies across the mode groups. Recall that the skew is captured by the difference between the actual rating and the estimated rating of the Turker based on the review text. Table 5 summarizes the results for the difference in ranking per different mode groups. We can see that the average difference between the perceived and real rates is very similar to the one in the previous experiment. In 60% of cases, reviews are rated as more negative than the actual rating. When reviews are perceived as more negative, the estimation is, on average, 1.67 points lower than the actual rating. When the text is perceived as more positive than the actual star rating, the difference is only 0.94 on average.
There is a clear difference in how the review text of various mode groups is perceived, and the skew is not the same across mode groups. For example, the percent difference for underestimates is 13.2 % for families but reaches 19.3 % for single travelers. When the estimated rating is higher than the actual rating, the difference is 10.9 % for couples but only 5.6 % for families. Interestingly, the group that experiences the biggest disconnect between the perception of their reviews’ sentiment and their ratings are the Single Travelers. The average difference between the perceived and real ratings was 19.3% when estimates were lower than the real ratings, which was the case in about 55% of the reviews. This means that the review text of single travelers is perceived as far more negative than the corresponding star rating they gave.
To check the correlation between the star ratings and the perceived sentiment, we used the Wilcoxon matched-pairs signed-ranks test. We found a high correlation with a p value of 0.0002 (when p ≤ 0.05 two groups are significantly correlated).
We held an additional experiment to validate our findings in the Mechanical Turk. We asked the workers to estimate reviews’ ratings when they knew the trip’s mode. The results were very similar, with no significant difference. Thus, the additional information of knowing the writer’s trip mode did not affect the estimations.
Next, we continue to understand the aspect preferences of various mode groups.

3.2. Experiment Two: Preference of Mode Groups for Hotel Aspects

In this experiment, we first find the hotel aspects and the terms associated with each aspect. First, we identify the most suitable clustering algorithm by evaluating several clustering algorithms.
For the partitional approach, we used k-means [75], DBSCAN [77], affinity propagation [80], and spectral clustering [78,79]. We further evaluated the Ward hierarchical clustering [81] algorithm. This type of algorithm is usually more accurate than k-means but suffers from efficiency problems [84]. For a graph-based algorithm, which can also be considered to be partition based, we evaluated spin glass [86], Fastgreedy [87], and Walktrap [88].

3.2.1. Clustering Evaluation

To evaluate the performance of the clustering algorithms, they are compared against a gold standard created by two annotators. The annotators were given all the review sentences and a list of the terms extracted from them (448 features). They were then asked to create clusters such that items within a cluster pertain to a specific hotel aspect. There were debates between the annotators about two of the features, and both annotators accepted the final clusters. One of the debates was on the feature “WiFi”. One annotator claimed it should belong to the Room aspect, while the other thought it should belong to the aspect Service; because “WiFi” is a service that the hotel provides and not part of the room, the agreement was that this feature would be included in the Service aspect. The second debate was regarding the feature “noise” that could be assigned to either Location or Room aspects; After examination of reviews mentioning noise, the annotators agreed to include it in the aspect Location, because reviewers were more likely to refer to the location of the hotel when they were discussing the noise.
The clustering algorithms k-means and DBSCAN did not perform well. Some of the created clusters contained a single term, with one large cluster containing the vast majority of the terms. Surprisingly, Ward hierarchical clustering produces the same poor results. On the other hand, the performances of spectral clustering, affinity propagation (AP), spin glass model, Fastgreedy, and Walktrap algorithms are good. Hence, we present a comparison of their results.
For the link weights, we found that using graphs with the PMI metric significantly outperformed the WordNet distance metric for all the clustering algorithms in the experiment (the best F1 score with WordNet distance is 0.29 ). Hence, in our results, we report on graph-based clustering using PMI weights.
Table 6 shows the results for the competitive algorithms. We used the Rand index to evaluate each clustering against the ground truth (reference clustering) [92]. The Rand index calculates the percentage of data points assigned to the same cluster in algorithm-generated and reference clustering. It ranges from 0 to 1, with 1 indicating perfect agreement between the two clustering measures and 0 indicating no agreement beyond what would be expected by chance. We further evaluated the precision, recall, and F1 score, standard metrics for external evaluation of cluster results.
The results show that the spin glass model outperforms the other clusters. Over the corpus of reviews, the PMI-pointwise improvement of the spinning glass community detection algorithm produced six clusters of different sizes (note that the number of clusters is unsupervised). Each cluster and the set of features it contains can be intuitively thought of as a hotel aspect and its associated terms. Each feature can only end up in one cluster. Cluster names, determined by the terms associated with each cluster, are Location, Service, Food, Room, Value for money and Facilities. The “Value for Money” cluster name was selected to describe terms surrounding quality, value, experience, and cost, as these elements collectively define the concept of value for money [27].

3.2.2. Identifying Mode Groups’ Preferences for Hotel Aspects

We then use the above-found division of aspect terms to examine preferences for hotel aspects according to mode groups. Hotel review platforms require that the reviewers specify their mode group when writing a hotel review. We use this information to extract the various mode group’s preferences for hotel aspects. To do this, we introduce the prominence index metric, which measures the prominence by which travelers with the same trip mode write about specific hotel aspects compared to others. The prominence index  P I of a hotel aspect a for a trip mode t i is the average deviation of all the features that belong to an aspect; P I is described as follows:
PI a t i = ∑ f ∈ a dev f t i N f a
where N f a is the number of features f that belong to the hotel aspect a.
Let t i denote the trip mode and freq f t i denote the frequency of feature f for trip mode t i . The frequency of a feature per trip mode is the relative number of occurrences of feature f in sentences appearing in reviews that belong to t i . Similarly, the avg f is the average frequency of feature f for all the trip modes, and stdv f is its standard deviation. Then, the following is true:
dev f t i = freq f t i − avg f
Figure 3 shows the prominence index for four aspects of the hotel for each trip mode. We can observe many differences in Figure 3a across the mode groups. For example, couples tend to write and care more about the hotel’s location than others, business travelers focus more on the hotel service than others, single travelers are significantly less interested in the hotel food than other groups, and so on. These trends are also consistent across the other two hotel aspects (price, X). Note that we could consider the 6-tuple ( P I f o o d t i , P I l o c a t i o n t i , . . . , P I s e r v i c e t i ) as a sort of profile for each mode group. It is clear from these pictures that profiles differ significantly across the mode groups.

4. Discussion

The paper contributes to the research and understanding of the importance of using the mode and purpose of the trip in assessing the sentiment and aspect sentiment of the text of hotel reviews, as well as in understanding the needs and preferences of various mode and purpose groups.
Our first contribution explores the relationship between the rating given in online opinions, the sentiment expressed in the text, and the role of the trip’s mode or purpose. Reviews often reference various aspects, and the given star rating is not necessarily an exact sum of its various parts [51,93]. Here, we add the complexity of considering the various trip modes and their impact on the experience and expressed online opinions.
We conducted an online experiment asking the participants to estimate the star rating given to the hotel by the review’s author. Our Mechanical Turk experiment reveals a significant difference between how the review text is perceived, regardless of the mode of the trip, and the star rating the reviewer gave the hotel. More than 55% of reviews were perceived as more negative (i.e., the estimated star rating was lower) than their corresponding star rating. The effect was more pronounced for single travelers, either business people or backpackers, as the text of their reviews was perceived as significantly more negative than the ratings they gave.
Interestingly, our findings show that different mode groups express their experiences uniquely. For example, we observed that single travelers tend to give similar ratings as other travelers, but their textual reviews were perceived as much worse by Mechanical Turk workers. On average, there was a difference of 1.93 between the perceived rating and the actual rating for single traveler reviews, compared to an average of 1.54 lower estimations for reviews written by people traveling in groups. This finding contrasts with recent findings, which indicate that people traveling alone tend to give higher ratings to hotels than those traveling with families [94]. We also conducted an additional experiment where workers were given reviews while being told the context of the trip’s mode, but this additional information did not impact the estimations.
This result has managerial and practical implications. To better understand the feedback in reviews, it is best to mine the text separately for different mode groups. This would also allow us to differentiate the experiences of various mode groups and understand which hotel each group prefers. Similarly, Wang et al. [53] found that the proposed groups differ in the key factors they consider when choosing a hotel, and Levi et al. [18] showed that considering the mode of the trip results in better hotel recommendations.
Taking a second look at the finding that single travelers experience the largest disconnect between the perceived sentiment of their reviews and their ratings, we identify a potential engagement issue. A possible follow-up research question is whether people are more tolerant when engaging in social interaction than alone. Consider the following example: Imagine a user waiting for a large file to download. How would the user perceive and describe the download time if they are alone, compared to when a colleague initiates an online chat during the download? If engagement positively influences our perception of experiences, a subsequent research question is whether sites that facilitate engagement provide users with a better experience than those that do not. Other studies have also identified that travelers who journey with companions report higher levels of happiness than those traveling alone [95,96]. Their analysis attributed happiness to their relationships with their companions. We also propose considering the role of engagement, suggesting that it could have significant implications for social networks and e-commerce sites.
In our second contribution to e-commerce, we build on our finding that review text provides additional insights beyond the star ratings and that these vary across different trip modes and purposes. We introduce a framework to determine which aspects are more prominent for each mode and purpose group using review texts.
For trip purposes, our findings show that, interestingly, business travelers discuss the aspects of service and room more. Similar to previous results [53], we find that aspects of service and room are very important to business travelers. We further find that the food aspect is also prominent in their review, but they do not discuss the location. Interestingly, single travelers who are not business people and travel for leisure primarily discuss the room aspect and rarely consider the food aspect.
Thus, identifying only the mode of the trip is insufficient, and both the mode and purpose of the trip should be analyzed.
For trip mode, we further find that location is a prominent topic for couples and that the room aspect is seldom mentioned in reviews by groups and families. These findings contrast with those reported in [24], which found that “the top 6 aspects soliciting positive sentiments are the same for all trip modes.” This discrepancy may be attributed to methodological differences. While Yadav and Roychoudhury [24] treat each top term as an individual aspect, we cluster semantically similar terms together before calculating the prominence of each aspect. Consequently, our approach accounts for multiple references to each aspect, revealing how the importance of various aspects varies among travelers with different trip modes.
These findings demonstrate the importance of considering trip mode and purpose when analyzing textual reviews, especially in domains where more user data may be needed. By understanding how different modes and purpose groups perceive and express their experiences, more accurate and personalized recommendations can be developed, and a better understanding of the factors influencing customer satisfaction can be gained.
From an academic perspective, our research enhances the understanding of how different trip modes and purposes influence how travelers express their opinions online. Our methodology provides a foundation for future studies on hospitality and e-commerce consumer behavior. Our results demonstrated that over the corpus of reviews, the PMI-pointwise link weight metric and the spinning glass community detection algorithm outperformed other clustering algorithms and produced a good representation of the hotel aspects and the terms used to describe them by guests. By discussing unsupervised methods for aspect extraction from online opinions, we contribute to the development of new techniques for analyzing customer feedback. Using unsupervised methods allows for a more flexible and scalable approach to aspect extraction, which can be applied across various domains beyond the hospitality industry. Rule-based and clustering methods were also used to uncover insights from online opinions [18,65,97,98,99], yet the effectiveness of the algorithms depends on the dataset and the application [18,98]. We show that the choice of similarity measure, pattern representation scheme, and grouping scheme can significantly impact the performance of the various clustering algorithms.
The six aspects identified by this unsupervised learning algorithm (Location, Service, Food, Room, Value for money, and Facilities) align well with those identified by professional review sites such as Booking.com and TripAdvisor. These companies have developed their expertise in identifying the most important aspects of hotel reviews based on their experience in the industry. By comparing the algorithm findings to those of those professional sites, we can gain confidence in the accuracy of recommending using unsupervised learning algorithms to identify the key aspects that are most important to users. This finding contributes to a deeper understanding of the factors influencing consumer satisfaction in the hotel industry and has practical applications for businesses looking to improve their offerings and overall customer experience.
This work has several managerial applications. Our first study found that a significant number of reviews (60.3%) underestimated the actual ratings compared to 39.7%, which overestimated them. This suggests that e-commerce platforms can improve their ability to understand customer sentiment directly from review texts, especially for businesses targeting solo travelers, whether traveling for leisure or business. Additionally, since reviews from group travelers tend to be more positive on average, businesses catering to solo travelers should consider offering socializing opportunities for their guests.
There are also practical applications for the hospitality industry. By identifying which hotel aspects are most important to customers, hoteliers can focus on improving these features to enhance customer satisfaction. Our work highlights the importance of aspect extraction in gaining insights into customer preferences and perceptions and the challenges of using ratings for sentiment analysis and as an evaluating metric. In addition, by understanding the limitations of using ratings, hotels can develop more accurate and effective strategies for managing their online reputation and tailor it to their target mode groups.
There are several limitations to this study. We have used data from two major online platforms, TripAdvisor and Venere. While these platforms provide a robust dataset, it is important to acknowledge that the reviews were collected in 2011. Although the fundamental aspects of traveler behavior and preferences are likely still relevant, some changes in the hospitality industry or consumer expectations over time may not be fully captured. Additionally, they might not represent diverse opinions from different regions and demographics [70,100]. Future research could benefit from integrating more recent and diverse data sources and exploring alternative analytical approaches to validate and extend these findings. Additionally, this study is limited by its quantitative focus, as it did not explore the underlying reasons behind the observed estimation discrepancies. Without conducting a qualitative analysis or gathering detailed feedback from the raters, it is impossible to uncover the specific factors contributing to the over- or underestimations. Future research incorporating qualitative methods, such as interviews or focus groups with raters, would provide a deeper understanding of the motivations and challenges influencing their evaluations.
Overall, this paper contributes to a better understanding of how online opinions can be used in industry and research and the importance of considering the mode of the trip when evaluating them.

5. Conclusions

In this work, we demonstrated the importance of considering the mode and purpose of the trip in evaluating electronic word of mouth and showed that consumers’ preferences vary depending on their trip mode and purpose.
Our online study examining the human perception of reviews finds evidence that review text adds information above and beyond the rating data. The study also looked at the influence of mode on the skew between estimated and actual ratings. The results indicated that star ratings do not consistently capture the sentiment in text reviews, and the predictive relationship between text and reviews is weak, more so for the various trip mode groups.
We further showed that review text uncovers additional aspects that cannot be determined from star ratings and suggested an unsupervised method for extracting knowledge, such as hotel aspects, from it. Our results show that over the corpus of hotel reviews, the PMI-pointwise improvement of the spinning glass community detection algorithm outperforms other algorithms, creating a natural division of the text to the different hotel aspects.
Our research offers significant theoretical contributions by advancing the understanding of how trip modes and purposes influence the prioritization of hotel aspects in online reviews. Our methodology allows for identifying underlying patterns in traveler preferences, enriching the existing literature on sentiment analysis and consumer behavior in e-commerce. From a practical and managerial perspective, our findings have several valuable applications for e-commerce businesses in the hospitality sector. Extracting and analyzing specific hotel aspects from online opinions enables businesses to develop targeted marketing campaigns tailored to the distinct preferences of travelers based on their trip modes and purposes. For instance, by identifying that single travelers predominantly emphasize room quality while couples focus on location, companies can customize their promotional materials to highlight these aspects, enhancing the relevance and appeal of their offerings to different customer segments.
Furthermore, our results suggest that relying solely on star ratings may provide an incomplete picture of customer satisfaction and preferences. By integrating the mode of trip and the qualitative content of reviews into their evaluation processes, businesses can gain deeper insights into consumer feedback. This comprehensive approach allows for more informed decision-making, enabling companies to address specific areas of improvement and better align their services with the expectations of diverse traveler groups. Consequently, businesses can enhance customer satisfaction and loyalty by offering personalized experiences that resonate with their target audiences’ unique needs and preferences.

Funding

This research received no external funding.

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

Data will be made available upon publication.

Conflicts of Interest

The author declares no conflicts of interest.

References

  1. Pang, B.; Lee, L. Opinion mining and sentiment analysis. Found. Trends Inf. Retr. 2008, 2, 1–135. [Google Scholar] [CrossRef]
  2. Jain, P.K.; Pamula, R.; Srivastava, G. A systematic literature review on machine learning applications for consumer sentiment analysis using online reviews. Comput. Sci. Rev. 2021, 41, 100413. [Google Scholar] [CrossRef]
  3. Mokryn, O. The opinions of a few: A cross-platform study quantifying usefulness of reviews. Online Soc. Netw. Media 2020, 18, 100080. [Google Scholar] [CrossRef]
  4. Zheng, X.; Huang, J.; Wu, J.; Sun, S.; Wang, S. Emerging trends in online reviews research in hospitality and tourism: A scientometric update (2000–2020). Tour. Manag. Perspect. 2023, 47, 101105. [Google Scholar] [CrossRef]
  5. Madzík, P.; Falát, L.; Copuš, L.; Valeri, M. Digital transformation in tourism: Bibliometric literature review based on machine learning approach. Eur. J. Innov. Manag. 2023, 26, 177–205. [Google Scholar] [CrossRef]
  6. Wan, Y.; Ma, B.; Pan, Y. Opinion evolution of online consumer reviews in the e-commerce environment. Electron. Commer. Res. 2018, 18, 291–311. [Google Scholar] [CrossRef]
  7. Hu, X.; Yang, Y. What makes online reviews helpful in tourism and hospitality? A bare-bones meta-analysis. J. Hosp. Mark. Manag. 2021, 30, 139–158. [Google Scholar] [CrossRef]
  8. Chen, T.; Samaranayake, P.; Cen, X.; Qi, M.; Lan, Y.C. The impact of online reviews on consumers’ purchasing decisions: Evidence from an eye-tracking study. Front. Psychol. 2022, 13, 865702. [Google Scholar] [CrossRef]
  9. PowerReviews. Survey: The Ever-Growing Power of Reviews, 2023 ed. 2023. Available online: https://www.powerreviews.com/power-of-reviews-2023/ (accessed on 15 September 2024).
  10. Schuckert, M.; Liu, X.; Law, R. Hospitality and tourism online reviews: Recent trends and future directions. J. Travel Tour. Mark. 2015, 32, 608–621. [Google Scholar] [CrossRef]
  11. Glaesser, D.; Kester, J.; Paulose, H.; Alizadeh, A.; Valentin, B. Global travel patterns: An overview. J. Travel Med. 2017, 24, tax007. [Google Scholar] [CrossRef]
  12. Armutcu, B.; Tan, A.; Amponsah, M.; Parida, S.; Ramkissoon, H. Tourist behaviour: The role of digital marketing and social media. Acta Psychol. 2023, 240, 104025. [Google Scholar] [CrossRef] [PubMed]
  13. Xiang, Z.; Du, Q.; Ma, Y.; Fan, W. A comparative analysis of major online review platforms: Implications for social media analytics in hospitality and tourism. Tour. Manag. 2017, 58, 51–65. [Google Scholar] [CrossRef]
  14. Martin-Fuentes, E.; Fernandez, C.; Mateu, C.; Marine-Roig, E. Modelling a grading scheme for peer-to-peer accommodation: Stars for Airbnb. Int. J. Hosp. Manag. 2018, 69, 75–83. [Google Scholar] [CrossRef]
  15. Xu, X. Does traveler satisfaction differ in various travel group compositions? Evidence from online reviews. Int. J. Contemp. Hosp. Manag. 2018, 30, 1663–1685. [Google Scholar] [CrossRef]
  16. Hassani, H.; Beneki, C.; Unger, S.; Mazinani, M.T.; Yeganegi, M.R. Text mining in big data analytics. Big Data Cogn. Comput. 2020, 4, 1. [Google Scholar] [CrossRef]
  17. Meneses, R.; Brito, C.; Lopes, B.; Correia, R. Satisfaction and dissatisfaction in wine tourism: A user-generated content analysis. Tour. Hosp. Res. 2023, 14673584231191989. [Google Scholar] [CrossRef]
  18. Levi, A.; Mokryn, O.; Diot, C.; Taft, N. Finding a needle in a haystack of reviews: Cold start context-based hotel recommender system. In Proceedings of the Sixth ACM conference on Recommender Systems, Dublin, Ireland, 9–13 September 2012; pp. 115–122. [Google Scholar]
  19. Chen, L.; Chen, G.; Wang, F. Recommender systems based on user reviews: The state of the art. User Model.-User-Adapt. Interact. 2015, 25, 99–154. [Google Scholar] [CrossRef]
  20. Li, J.; Xu, L.; Tang, L.; Wang, S.; Li, L. Big data in tourism research: A literature review. Tour. Manag. 2018, 68, 301–323. [Google Scholar] [CrossRef]
  21. Zarezadeh, Z.Z.; Rastegar, R.; Xiang, Z. Big data analytics and hotel guest experience: A critical analysis of the literature. Int. J. Contemp. Hosp. Manag. 2022, 34, 2320–2336. [Google Scholar] [CrossRef]
  22. Oh, H.; Kim, K. Customer satisfaction, service quality, and customer value: Years 2000–2015. Int. J. Contemp. Hosp. Manag. 2017, 29, 2–29. [Google Scholar] [CrossRef]
  23. Rajaguru, R.; Hassanli, N. The role of trip purpose and hotel star rating on guests’ satisfaction and WOM. Int. J. Contemp. Hosp. Manag. 2018, 30, 2268–2286. [Google Scholar] [CrossRef]
  24. Yadav, M.L.; Roychoudhury, B. Effect of trip mode on opinion about hotel aspects: A social media analysis approach. Int. J. Hosp. Manag. 2019, 80, 155–165. [Google Scholar] [CrossRef]
  25. Moise, M.S.; Gil-Saura, I.; Ruiz-Molina, M.E. Implications of value co-creation in green hotels: The moderating effect of trip purpose and generational cohort. Sustainability 2020, 12, 9866. [Google Scholar] [CrossRef]
  26. Pizam, A.; Mansfeld, Y. Consumer Behavior in Travel and Tourism; Routledge: Abingdon-on-Thames, UK, 1999. [Google Scholar]
  27. Radder, L.; Wang, Y. Dimensions of guest house service: Managers’ perceptions and business travellers’ expectations. Int. J. Contemp. Hosp. Manag. 2006, 18, 554–562. [Google Scholar] [CrossRef]
  28. Banerjee, S.; Chua, A.Y. In search of patterns among travellers’ hotel ratings in TripAdvisor. Tour. Manag. 2016, 53, 125–131. [Google Scholar] [CrossRef]
  29. Luo, J.M.; Vu, H.Q.; Li, G.; Law, R. Understanding service attributes of robot hotels: A sentiment analysis of customer online reviews. Int. J. Hosp. Manag. 2021, 98, 103032. [Google Scholar] [CrossRef]
  30. Ong, B.S. The perceived influence of user reviews in the hospitality industry. J. Hosp. Mark. Manag. 2012, 21, 463–485. [Google Scholar] [CrossRef]
  31. Hernández-Rubio, M.; Cantador, I.; Bellogín, A. A comparative analysis of recommender systems based on item aspect opinions extracted from user reviews. User Model.-User-Adapt. Interact. 2019, 29, 381–441. [Google Scholar] [CrossRef]
  32. Kanje, P.; Charles, G.; Tumsifu, E.; Mossberg, L.; Andersson, T. Customer engagement and eWOM in tourism. J. Hosp. Tour. Insights 2020, 3, 273–289. [Google Scholar] [CrossRef]
  33. Chu, S.C.; Deng, T.; Cheng, H. The role of social media advertising in hospitality, tourism and travel: A literature review and research agenda. Int. J. Contemp. Hosp. Manag. 2020, 32, 3419–3438. [Google Scholar] [CrossRef]
  34. Reyes-Menendez, A.; Correia, M.B.; Matos, N.; Adap, C. Understanding online consumer behavior and eWOM strategies for sustainable business management in the tourism industry. Sustainability 2020, 12, 8972. [Google Scholar] [CrossRef]
  35. Zhang, L.; Liu, B. Aspect and entity extraction for opinion mining. In Data Mining and Knowledge Discovery for Big Data: Methodologies, Challenge and Opportunities; Springer: Berlin/Heidelberg, Germany, 2014; pp. 1–40. [Google Scholar]
  36. Chen, H.; Yin, H.; Li, X.; Wang, M.; Chen, W.; Chen, T. People opinion topic model: Opinion based user clustering in social networks. In Proceedings of the 26th International Conference on World Wide Web Companion, Perth, Australia, 3–7 April 2017; pp. 1353–1359. [Google Scholar]
  37. Xue, W.; Li, T.; Rishe, N. Aspect identification and ratings inference for hotel reviews. World Wide Web 2017, 20, 23–37. [Google Scholar] [CrossRef]
  38. Wan, C.; Peng, Y.; Xiao, K.; Liu, X.; Jiang, T.; Liu, D. An association-constrained LDA model for joint extraction of product aspects and opinions. Inf. Sci. 2020, 519, 243–259. [Google Scholar] [CrossRef]
  39. Shu, Z.; González, R.A.C.; García-Miguel, J.P.; Sánchez-Montañés, M. Clustering using ordered weighted averaging operator and 2-tuple linguistic model for hotel segmentation: The case of TripAdvisor. Expert Syst. Appl. 2023, 213, 118922. [Google Scholar] [CrossRef]
  40. Tubishat, M.; Idris, N.; Abushariah, M.A. Implicit aspect extraction in sentiment analysis: Review, taxonomy, oppportunities, and open challenges. Inf. Process. Manag. 2018, 54, 545–563. [Google Scholar] [CrossRef]
  41. Ray, B.; Garain, A.; Sarkar, R. An ensemble-based hotel recommender system using sentiment analysis and aspect categorization of hotel reviews. Appl. Soft Comput. 2021, 98, 106935. [Google Scholar] [CrossRef]
  42. Zhang, J.; Lu, X.; Liu, D. Deriving customer preferences for hotels based on aspect-level sentiment analysis of online reviews. Electron. Commer. Res. Appl. 2021, 49, 101094. [Google Scholar] [CrossRef]
  43. Zhao, M.; Liu, M.; Xu, C.; Zhang, C. Classifying travellers’ requirements from online reviews: An improved Kano model. Int. J. Contemp. Hosp. Manag. 2024, 36, 91–112. [Google Scholar] [CrossRef]
  44. Mokryn, O.; Bodoff, D.; Bader, N.; Albo, Y.; Lanir, J. Sharing emotions: Determining films’ evoked emotional experience from their online reviews. Inf. Retr. J. 2020, 23, 475–501. [Google Scholar] [CrossRef]
  45. Mishra, A.; Satish, S. eWOM: Extant research review and future research avenues. Vikalpa 2016, 41, 222–233. [Google Scholar] [CrossRef]
  46. Litvin, S.W.; Goldsmith, R.E.; Pan, B. Electronic word-of-mouth in hospitality and tourism management. Tour. Manag. 2008, 29, 458–468. [Google Scholar] [CrossRef]
  47. Filieri, R.; McLeay, F. E-WOM and accommodation an analysis of the factors that influence travelers’ adoption of information from online reviews. J. Travel Res. 2014, 53, 44–57. [Google Scholar] [CrossRef]
  48. Strandberg, C.; Nath, A.; Hemmatdar, H.; Jahwash, M. Tourism research in the new millennium: A bibliometric review of literature in Tourism and Hospitality Research. Tour. Hosp. Res. 2018, 18, 269–285. [Google Scholar] [CrossRef]
  49. Wen, J.; Lin, Z.; Liu, X.; Xiao, S.H.; Li, Y. The interaction effects of online reviews, brand, and price on consumer hotel booking decision making. J. Travel Res. 2021, 60, 846–859. [Google Scholar] [CrossRef]
  50. Cambria, E.; Poria, S.; Gelbukh, A.; Thelwall, M. Sentiment analysis is a big suitcase. IEEE Intell. Syst. 2017, 32, 74–80. [Google Scholar] [CrossRef]
  51. McAuley, J.; Leskovec, J. Hidden factors and hidden topics: Understanding rating dimensions with review text. In Proceedings of the 7th ACM Conference on Recommender Systems, Hong Kong, China, 12–16 October 2013; pp. 165–172. [Google Scholar]
  52. Ma, Y.; Peng, H.; Cambria, E. Targeted aspect-based sentiment analysis via embedding commonsense knowledge into an attentive LSTM. In Proceedings of the AAAI Conference on Artificial Intelligence, New Orleans, LO, USA, 2–7 February 2018; Volume 32. [Google Scholar]
  53. Wang, L.; Wang, X.k.; Peng, J.j.; Wang, J.q. The differences in hotel selection among various types of travellers: A comparative analysis with a useful bounded rationality behavioural decision support model. Tour. Manag. 2020, 76, 103961. [Google Scholar] [CrossRef]
  54. Liang, B.; Su, H.; Gui, L.; Cambria, E.; Xu, R. Aspect-based sentiment analysis via affective knowledge enhanced graph convolutional networks. Knowl.-Based Syst. 2022, 235, 107643. [Google Scholar] [CrossRef]
  55. Blei, D.M.; Ng, A.Y.; Jordan, M.I. Latent dirichlet allocation. J. Mach. Learn. Res. 2003, 3, 993–1022. [Google Scholar]
  56. Moghaddam, S.; Ester, M. ILDA: Interdependent LDA model for learning latent aspects and their ratings from online product reviews. In Proceedings of the 34th International ACM SIGIR Conference on Research and Development in Information Retrieval, Beijing, China, 24–28 July 2011; pp. 665–674. [Google Scholar]
  57. Guo, Y.; Barnes, S.J.; Jia, Q. Mining meaning from online ratings and reviews: Tourist satisfaction analysis using latent dirichlet allocation. Tour. Manag. 2017, 59, 467–483. [Google Scholar] [CrossRef]
  58. Hu, N.; Zhang, T.; Gao, B.; Bose, I. What do hotel customers complain about? Text analysis using structural topic model. Tour. Manag. 2019, 72, 417–426. [Google Scholar] [CrossRef]
  59. Sutherland, I.; Kiatkawsin, K. Determinants of guest experience in Airbnb: A topic modeling approach using LDA. Sustainability 2020, 12, 3402. [Google Scholar] [CrossRef]
  60. Jia, S.S. Motivation and satisfaction of Chinese and US tourists in restaurants: A cross-cultural text mining of online reviews. Tour. Manag. 2020, 78, 104071. [Google Scholar] [CrossRef]
  61. Kalnaovakul, K.; Promsivapallop, P. Hotel service quality dimensions and attributes: An analysis of online hotel customer reviews. Tour. Hosp. Res. 2023, 23, 420–440. [Google Scholar] [CrossRef]
  62. Sann, R.; Lai, P.C. Topic modeling of the quality of guest’s experience using latent Dirichlet allocation: Western versus eastern perspectives. Consum. Behav. Tour. Hosp. 2023, 18, 17–34. [Google Scholar] [CrossRef]
  63. Kumar, A.; Chakraborty, S.; Bala, P.K. Text mining approach to explore determinants of grocery mobile app satisfaction using online customer reviews. J. Retail. Consum. Serv. 2023, 73, 103363. [Google Scholar] [CrossRef]
  64. Chen, L.; Martineau, J.; Cheng, D.; Sheth, A. Clustering for simultaneous extraction of aspects and features from reviews. In Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, San Diego, CA, USA, 12–17 June 2016; pp. 789–799. [Google Scholar]
  65. Nazir, A.; Rao, Y.; Wu, L.; Sun, L. Issues and challenges of aspect-based sentiment analysis: A comprehensive survey. IEEE Trans. Affect. Comput. 2020, 13, 845–863. [Google Scholar] [CrossRef]
  66. Irfan, R.; King, C.K.; Grages, D.; Ewen, S.; Khan, S.U.; Madani, S.A.; Kolodziej, J.; Wang, L.; Chen, D.; Rayes, A.; et al. A survey on text mining in social networks. Knowl. Eng. Rev. 2015, 30, 157–170. [Google Scholar] [CrossRef]
  67. Zeimpekis, D.; Gallopoulos, E. Principal direction divisive partitioning with kernels and k-means steering. In Survey of Text Mining II; Springer: Berlin/Heidelberg, Germany, 2008; pp. 45–64. [Google Scholar]
  68. Jain, A.; Murty, M.; Flynn, P. Data clustering: A review. ACM Comput. Surv. 1999, 31, 264–323. [Google Scholar] [CrossRef]
  69. Shalileh, S.; Mirkin, B. Summable and nonsummable data-driven models for community detection in feature-rich networks. Soc. Netw. Anal. Min. 2021, 11, 1–23. [Google Scholar] [CrossRef]
  70. Chen, M.M.; Seach, K.; Inversini, A.; Williams, N. Different cultures review hotels differently. Tour. Hosp. Res. 2023, 14673584231198410. [Google Scholar] [CrossRef]
  71. Kaufman, L.; Rousseeuw, P.J. Finding Groups in Data: An Introduction to Cluster Analysis; Wiley: Hoboken, NJ, USA, 2009; Volume 344. [Google Scholar]
  72. Ezugwu, A.E.; Ikotun, A.M.; Oyelade, O.O.; Abualigah, L.; Agushaka, J.O.; Eke, C.I.; Akinyelu, A.A. A comprehensive survey of clustering algorithms: State-of-the-art machine learning applications, taxonomy, challenges, and future research prospects. Eng. Appl. Artif. Intell. 2022, 110, 104743. [Google Scholar] [CrossRef]
  73. Curiskis, S.A.; Drake, B.; Osborn, T.R.; Kennedy, P.J. An evaluation of document clustering and topic modelling in two online social networks: Twitter and Reddit. Inf. Process. Manag. 2020, 57, 102034. [Google Scholar] [CrossRef]
  74. Grira, N.; Crucianu, M.; Boujemaa, N. Unsupervised and semi-supervised clustering: A brief survey. In A Review of Machine Learning Techniques for Processing Multimedia Content, Report of the MUSCLE European Network of Excellence (FP6); Ercim: Cannes, France, 2004. [Google Scholar]
  75. MacQueen, J. Some methods for classification and analysis of multivariate observations. In Proceedings of the Fifth Berkeley Symposium on Mathematical Statistics and Probability, Berkeley, CA, USA, 27 December 1965–7 January 1966; Volume 1, p. 14. [Google Scholar]
  76. Ikotun, A.M.; Ezugwu, A.E.; Abualigah, L.; Abuhaija, B.; Heming, J. K-means clustering algorithms: A comprehensive review, variants analysis, and advances in the era of big data. Inf. Sci. 2023, 622, 178–210. [Google Scholar] [CrossRef]
  77. Ester, M.; Kriegel, H.P.; Sander, J.; Xu, X. A density-based algorithm for discovering clusters in large spatial databases with noise. In Proceedings of the KDD, Portland, OR, USA, 2–4 August 1996; Volume 96, pp. 226–231. [Google Scholar]
  78. Ng, A.Y.; Jordan, M.I.; Weiss, Y. On spectral clustering: Analysis and an algorithm. Adv. Neural Inf. Process. Syst. 2002, 2, 849–856. [Google Scholar]
  79. Von Luxburg, U. A tutorial on spectral clustering. Stat. Comput. 2007, 17, 395–416. [Google Scholar] [CrossRef]
  80. Frey, B.J.; Dueck, D. Clustering by passing messages between data points. Science 2007, 315, 972–976. [Google Scholar] [CrossRef]
  81. Ward, J.H., Jr. Hierarchical grouping to optimize an objective function. J. Am. Stat. Assoc. 1963, 58, 236–244. [Google Scholar] [CrossRef]
  82. King, B. Step-wise clustering procedures. J. Am. Stat. Assoc. 1967, 62, 86–101. [Google Scholar] [CrossRef]
  83. Fortunato, S. Community detection in graphs. Phys. Rep. 2010, 486, 75–174. [Google Scholar] [CrossRef]
  84. Manning, C.D.; Raghavan, P.; Schütze, H. Introduction to Information Retrieval; Cambridge University Press: Cambridge, MA, USA, 2008; Volume 1. [Google Scholar]
  85. Mikhina, E.K.; Trifalenkov, V.I. Text clustering as graph community detection. Procedia Comput. Sci. 2018, 123, 271–277. [Google Scholar] [CrossRef]
  86. Reichardt, J.; Bornholdt, S. Statistical mechanics of community detection. Phys. Rev. E 2006, 74, 016110. [Google Scholar] [CrossRef] [PubMed]
  87. Clauset, A.; Newman, M.; Moore, C. Finding community structure in very large networks. Phys. Rev. E 2004, 70, 066111. [Google Scholar] [CrossRef] [PubMed]
  88. Pons, P.; Latapy, M. Computing communities in large networks using random walks. In Proceedings of the Computer and Information Sciences-ISCIS 2005: 20th International Symposium, Istanbul, Turkey, 26–28 October 2005; pp. 284–293. [Google Scholar]
  89. Reichardt, J.; Bornholdt, S. Detecting fuzzy community structures in complex networks with a Potts model. Phys. Rev. Lett. 2004, 93, 218701. [Google Scholar] [CrossRef] [PubMed]
  90. Church, K.; Hanks, P. Word association norms, mutual information, and lexicography. Comput. Linguist. 1990, 16, 22–29. [Google Scholar]
  91. Church, K.; Gale, W.; Hanks, P.; Kindle, D. 6. Using Statistics in Lexical Analysis. In Lexical Acquisition: Exploiting On-Line Resources to Build a Lexicon; Psychology Press: East Sussex, UK, 1991; p. 115. [Google Scholar]
  92. Rand, W.M. Objective criteria for the evaluation of clustering methods. J. Am. Stat. Assoc. 1971, 66, 846–850. [Google Scholar] [CrossRef]
  93. Sadiq, S.; Umer, M.; Ullah, S.; Mirjalili, S.; Rupapara, V.; Nappi, M. Discrepancy detection between actual user reviews and numeric ratings of Google App store using deep learning. Expert Syst. Appl. 2021, 181, 115111. [Google Scholar] [CrossRef]
  94. Radojevic, T.; Stanisic, N.; Stanic, N. Solo travellers assign higher ratings than families: Examining customer satisfaction by demographic group. Tour. Manag. Perspect. 2015, 16, 247–258. [Google Scholar] [CrossRef]
  95. Radojevic, T.; Stanisic, N.; Stanic, N. Inside the rating scores: A multilevel analysis of the factors influencing customer satisfaction in the hotel industry. Cornell Hosp. Q. 2017, 58, 134–164. [Google Scholar] [CrossRef]
  96. Yang, R.; Tung, V.W.S. How does family influence the travel constraints of solo travelers? Construct specification and scale development. J. Travel Tour. Mark. 2018, 35, 507–516. [Google Scholar] [CrossRef]
  97. Liu, Q.; Gao, Z.; Liu, B.; Zhang, Y. Automated rule selection for aspect extraction in opinion mining. In Proceedings of the Twenty-Fourth International Joint Conference on Artificial Intelligence, Buenos Aires, Argentina, 25–31 July 2015. [Google Scholar]
  98. Sasmita, D.H.; Wicaksono, A.F.; Louvan, S.; Adriani, M. Unsupervised aspect-based sentiment analysis on Indonesian restaurant reviews. In Proceedings of the 2017 International Conference on Asian Language Processing (IALP), Singapore, 5–7 December 2017; pp. 383–386. [Google Scholar]
  99. Setiowati, Y.; Djunaidy, A.; Siahaan, D.O. Aspect-based extraction of implicit opinions using opinion co-occurrence algorithm. In Proceedings of the 2022 5th International Seminar on Research of Information Technology and Intelligent Systems (ISRITI), Yogyakarta, Indonesia, 8–9 December 2022; pp. 781–786. [Google Scholar]
  100. Mariani, M.M.; Borghi, M.; Okumus, F. Unravelling the effects of cultural differences in the online appraisal of hospitality and tourism services. Int. J. Hosp. Manag. 2020, 90, 102606. [Google Scholar] [CrossRef]
Jtaer 19 00145 g001
Figure 1. Study 2 Phase 1 Flowchart: Evaluation of unsupervised clustering methods.
Figure 1. Study 2 Phase 1 Flowchart: Evaluation of unsupervised clustering methods.
Jtaer 19 00145 g001
Jtaer 19 00145 g002
Figure 2. Study 2 Phase 2 Flowchart: Evaluation of the prominence of each aspect per each trip mode and purpose group.
Figure 2. Study 2 Phase 2 Flowchart: Evaluation of the prominence of each aspect per each trip mode and purpose group.
Jtaer 19 00145 g002
Jtaer 19 00145 g003
Figure 3. Prominence of hotel aspects as calculated by the average frequency of terms per hotel aspect as expressed by each trip mode group. (a) Prominence of Location; (b) Prominence of Service; (c) Prominence of Food; (d) Prominence of Room.
Figure 3. Prominence of hotel aspects as calculated by the average frequency of terms per hotel aspect as expressed by each trip mode group. (a) Prominence of Location; (b) Prominence of Service; (c) Prominence of Food; (d) Prominence of Room.
Jtaer 19 00145 g003
Table 1. Summary of hotel review data.
Table 1. Summary of hotel review data.
Web SiteHotelsReviews%Reviews
TripAdvisor.com193084,96861.9%
Venere.com184552,26638.1%
Total3775137,234
Table 2. Reviews per travel mode.
Table 2. Reviews per travel mode.
Travel ModeReviews%Reviews
Couple60,11343.8%
Family17,55712.8%
Group12,3069.0%
Single Traveler11,1248.1%
Business65414.8%
Not Specified29,59321.5%
Summing137,234
Table 3. Overall Mechanical Turk Results.
Table 3. Overall Mechanical Turk Results.
Type% DifferenceReviews
Estimation > Rate 9.4 % 1474
Estimation < Rate 16.7 % 2241
Total 13.8 % 3715
Table 4. Average rating per mode group on TripAdvisor.
Table 4. Average rating per mode group on TripAdvisor.
Mode GroupAverage Rating (Stdv)
Family4.15 (0.8)
Couple4.08 (0.8)
Group4.02 (0.8)
Single3.8 (0.8)
Business3.6 (1.0)
Table 5. Mechanical Turk results by trip mode on the Venere dataset.
Table 5. Mechanical Turk results by trip mode on the Venere dataset.
Mode GroupTypeAverage Difference%DifferenceReviews
CoupleEstimation > Rate 1.09 10.9 % 685
CoupleEstimation < Rate 1.67 16.7 % 954
FamilyEstimation > Rate 0.56 5.6 % 241
FamilyEstimation < Rate 1.32 13.2 % 404
GroupEstimation > Rate 1.02 10.2 % 89
GroupEstimation < Rate 1.54 15.4 % 158
Single TravelerEstimation > Rate 0.83 8.3 % 426
Single TravelerEstimation < Rate 1.93 19.3 % 659
Not SpecifiedEstimation > Rate 1.41 14.1 % 33
Not SpecifiedEstimation < Rate 1.51 15.1 % 66
TotalEstimation > Rate 0.94 9.4 % 1474
TotalEstimation < Rate 1.67 16.7 % 2241
Table 6. Feature clustering evaluation.
Table 6. Feature clustering evaluation.
AlgorithmRand-IPrecisionRecallF1
Spin glass0.89340.76750.71930.7426
Fastgreedy 0.8453 0.6430 0.6211 0.6319
Walktrap 0.7037 0.3722 0.5621 0.4479
Spectral 0.7586 0.4513 0.5989 0.5147
AP 0.7905 0.5752 0.0775 0.1366
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Mokryn, O. Consumer Sentiment and Hotel Aspect Preferences Across Trip Modes and Purposes. J. Theor. Appl. Electron. Commer. Res. 2024, 19, 3017-3034. https://doi.org/10.3390/jtaer19040145

AMA Style

Mokryn O. Consumer Sentiment and Hotel Aspect Preferences Across Trip Modes and Purposes. Journal of Theoretical and Applied Electronic Commerce Research. 2024; 19(4):3017-3034. https://doi.org/10.3390/jtaer19040145

Chicago/Turabian Style

Mokryn, Osnat. 2024. "Consumer Sentiment and Hotel Aspect Preferences Across Trip Modes and Purposes" Journal of Theoretical and Applied Electronic Commerce Research 19, no. 4: 3017-3034. https://doi.org/10.3390/jtaer19040145

APA Style

Mokryn, O. (2024). Consumer Sentiment and Hotel Aspect Preferences Across Trip Modes and Purposes. Journal of Theoretical and Applied Electronic Commerce Research, 19(4), 3017-3034. https://doi.org/10.3390/jtaer19040145

Article Metrics

Back to TopTop