Airbnb Customer Satisfaction Analysis

In response to the growing importance of data-driven insights in the Airbnb marketplace, this project delves into the complex relationship between listing features, geographic location, and customer satisfaction in New York City. By employing techniques such as principal component analysis, clustering analysis, regression models, text mining, and sentiment analysis, the study seeks to uncover the specific factors that most significantly influence guest experiences and review scores. The findings aim to provide Airbnb hosts with strategic recommendations for optimizing their listings, ultimately improving guest satisfaction and aligning their offerings more closely with traveler preferences.

Abstraction

The continued rise in popularity of alternative accommodation platforms, Airbnb, presents the transformative shift in customer travel behavior. Travelers prioritize safety, trust, and quality of guest experience. Data-driven insights are essential for hosts to refine their listings and ensure they consistently exceed guest expectations. This study explores muti-factors that influence customer satisfaction on Airbnb, with a specific focus on the bustling metropolis of New York. We aim to discover the impact of different determinants. Specifically, how the geographic locations and overall listing performance affect the customer’s experience. Through a blend of principal component analysis, clustering analysis, regression models, text mining, and sentiment analysis, this research aims to reveal strategic adjustments for hosts, seeking to dissect the interplay between location-specific attributes and quality of listings. Ultimately, our goal is to gain a full-scale understanding of customer satisfaction drivers, providing actionable recommendations for optimization and improving the alignment of guest preferences with host offerings for a better travel experience.

Literature Review

Amidst the expanding popularity of Airbnb, extensive research has been conducted on the preferences influencing customers’ accommodation choices. However, the combined influence of listing features and geographical factors on Airbnb guest satisfaction remains a notably under investigated area, highlighting a gap in understanding the comprehensive impact on consumer decisions.

In the realm of Airbnb, listing features such as price, amenities, location, and property type play a crucial role in shaping guest satisfaction. Studies have shown that these attributes significantly influence guest experiences and review scores (Guttentag et al., 2018). For example, the perceived value and quality of amenities can enhance guest satisfaction, while discrepancies in listing descriptions can lead to dissatisfaction (Xu & Li, 2016). Understanding the impact of these features is essential for hosts to optimize their listings and for researchers to develop models that accurately predict guest satisfaction.

Geographic location is another critical factor that influences guest satisfaction in Airbnb accommodations(Varma et al., 2016). Proximity to attractions, neighborhood characteristics, and regional differences play a significant role in shaping guest experiences (Kai Ding et al., 2021). Studies have found that guests prioritize locations that offer easy access to desired places and a favorable neighborhood environment (Sutherland & Kiatkawsin, 2020). Moreover, Kai Ding found that guests staying in lower-priced rooms are more concerned about the convenience of exploring the surrounding area, while those in higher-priced rooms prioritize the surrounding environment and interior facilities(Kai Ding et al., 2023). These insights contribute to a deeper understanding of the role of geographic factors in customer satisfaction and highlight the need for Airbnb hosts to consider these factors in their listings.

Research Question

Research Question 1: How do listing features influence review score ratings, and what areas could be targeted for improvement?

Null Hypothesis

• Listing features do not significantly influence review score ratings.

Alternative Hypothesis

• Listing features significantly influence review score ratings, indicating areas of potential improvement.

This research question aims to investigate the impact of various listing features, specifically numeric variables, on review score ratings in the context of Airbnb listings in New York City. The analysis will focus solely on numeric factors such as price, amenity_count, cleaning_fee, verification_count, and cancellation_policy. To address this question, a multifaceted approach will be employed, including the use of gradient boosting machine (GBM) modeling, which incorporates numeric predictors to predict review score ratings. Additionally, feature importance analysis will be conducted to identify the most influential numeric factors affecting review score ratings. By understanding these factors, hosts can identify areas for improvement and enhance the overall guest experience.

Research Question 2: How does the geographic location of Airbnb listings impact customer satisfaction?

Null Hypothesis

• The geographic location of Airbnb listings does not significantly impact customer satisfaction.

Alternative Hypothesis

• The geographic location significantly impacts customer satisfaction, with variations across different neighborhoods.

The second research question aims to explore how the geographic location of Airbnb listings affects guest satisfaction. This includes examining the neighborhood of the listings, observable features between different neighborhoods, and the overall appeal of the area to guests. To investigate this, principal component analysis (PCA) and clustering analysis techniques will be employed. PCA will help identify underlying patterns and reduce the dimensionality of the review scores data, while clustering analysis, specifically K-means clustering, will group listings into clusters based on their satisfaction scores. Additionally, multivariate analysis of variance (MANOVA) will be conducted to determine whether there are statistically significant differences in satisfaction scores among different neighborhoods. This comprehensive analysis will shed light on the relationship between listing location and customer preferences, offering valuable insights for hosts and platform administrators.

Raw Data Description

We selected datasets of Airbnb listings and customer reviews from Kaggle as our data sources for investigating our research problem. These datasets contain detailed information on each listing, including host details, property features, and the rating and comments provided by customers.

According to our research questions, property characteristics, host information, and geographical attributes are the most relevant information in analyzing each listing. Property characteristics include property type, room type, accommodation capacity, and amenities. Host-related information includes host identifiers, verifications, response metrics, and hosting history. Although our focus is on New York listings, geographic details such as neighborhood information and location coordinates are also included. These datasets enable us to assess the relevance of factors influencing customer satisfaction levels.

While the comments field in the raw review dataset may appear messy, it holds significant importance for our sentiment analysis. Customer survey data encompasses overall review scores and specific ratings for various aspects of their stay, such as cleanliness, description accuracy, and communication. These comprehensive datasets offer broad information about Airbnb listings and provide ample support for further in-depth analysis.

Data Cleaning Process

The data cleaning process involved several conscious steps to ensure the integrity, consistency, and reliability of the dataset. We treat this as one of the most important phrases as this will be the foundation for further analyses.

Initially, an overview of the review and listing datasets was conducted using essential functions like glimpse(), str(), head(), and dim(). This initial examination provided insights into the structure, dimensions, and variable types of the datasets, guiding subsequent cleaning procedures. In the review dataset, only the relevant columns (listing_id and comments) were retained, streamlining further cleaning processes.

It was observed that the comments column contained non-English text, necessitating a language detection process. Despite its time-intensive nature (approximately 80 to 120 minutes), this step was crucial for subsequent text mining analyses. Following language detection and filtering English-only comments, comments were grouped by listing_id and combined into a single string using summarize() and paste() functions. Simultaneously, stop words were removed from the combined comments to eliminate common language elements without significant semantic meaning, facilitated by unnest_tokens() and anti_join() functions.

Turning to the listing dataset, missing values were addressed using various strategies based on domain knowledge, missing percentage, and variable importance. Predictive modeling, replacement with predetermined values, and removal of rows with missing values were among the employed strategies. Data type transformations and formatting adjustments ensure consistency across the dataset, leveraging string manipulation functions like str_replace() and as.numeric(). Additionally, tokenization was performed on the amenities column to identify common feature offerings by hosts, followed by the creation of binary indicators(dummy variables) using mutate() and regular expression matching with str_detect().

Finally, the two datasets (review and listing) were merged into a single dataset named airbnb using the inner_join() function, linking listings based on their unique identifiers(id and listing_id). A comprehensive check for missing values was then conducted using the colSums() function on the final airbnb dataset, ensuring data structure and completeness.

For the fact that not all attributes may be utilized for subsequent data analysis, the missing values leftover were retained in their current state. This decision was made to preserve as many observations as possible. It is worth noting that the colSums() function will be applied before each subsequent analysis to determine whether additional cleaning processes are necessary, ensuring ongoing data quality assessment and refinement.

Techniques Chosen

R Studio will be our programming language because of its specialized features for data analysis. It provides powerful tools for manipulating diverse and unstructured data, along with a wide range of statistical packages for conducting detailed analysis. R Studio’s visualization capabilities allow us to create various graphs and charts, enabling clear and impactful presentation of our findings.

For our first research question, we will utilize predictive data modeling to quantify how listing features influence review score ratings. This approach is chosen to provide hosts with actionable insights to enhance guest satisfaction. Models like GBM are adept at capturing complex interactions between variables and can predict review scores even with non-linear relationships in the data. Additionally, these models offer feature importance analysis, aiding in the identification of key factors influencing guest satisfaction. Understanding these relationships will enable hosts to optimize their listings, potentially improving competitiveness in the marketplace.

For our second research question, we will utilize Principal Component Analysis (PCA) to condense the various review score metrics into key components. This method will help identify factors impacting customer satisfaction by revealing underlying patterns in the data. Additionally, PCA simplifies visualization and understanding of significant factors while reducing issues like multicollinearity among variables.

Moreover, on the basis of PCA results, we will employ clustering to segment the data into meaningful groups without prior labeling. This approach will assist in identifying natural groupings in the Airbnb dataset, such as types of listings or customer preferences based on satisfaction scores. Clustering will enable us to tailor improvement strategies for hosts and understand how different types of listings perform across various neighborhoods, aiding strategic decision-making for property management and marketing.

Additionally, we will employ text mining techniques to extract insights from textual data like reviews and amenities, to uncover trends and sentiments that influence satisfaction levels. By text mining from unstructured data, hosts can better understand guest sentiments, identify common issues, and gain qualitative insights into customer satisfaction.

Finally, sentiment analysis will be crucial in our study to analyze the emotional tone of reviews left by Airbnb guests. Techniques such as AFINN, Bing, and NRC will be used to extract sentiment scores/segments from customer comments, providing an overview of guest feedback. These methods will analyze natural language and identify whether sentiments expressed are positive, negative, or neutral. Insights gained from sentiment analysis will complement quantitative data, offering a holistic view of guest experiences.

Analytical Results

Text Mining Results

The analysis of the customer review score revealed a right-skewed distribution, indicating a trend where the majority of customers rated their experiences positively, with scores centered above 95.

We also investigated the customer sentiment using the Afinn sentiment analysis, which assigned scores to each review.

These scores were correlated with the review scores, revealing a moderate correlation coefficient of 0.4397068. This suggests that customers’ sentiment in text reviews is somewhat reflective of their numerical ratings.

R code: cor(airbnb$review_scores_rating, airbnb$afinn)

R Output: 0.4397068

In addition, we utilized text mining to analyze the distribution of sentiments categorized by Bing. The findings showed that the vast majority of reviews possess a positive sentiment.

A more detailed visualization of the NRC sentiment analysis results highlighted that reviews tend to evoke emotions such as trust, joy, anticipation, and surprise more frequently than negative emotions like sadness, fear, anger, and disgust. Therefore, in the visualization, we see a greater prominence of positive sentiments compared to negative ones, reflecting the overall positivity of the customer experiences

Our research also included an evaluation of customer satisfaction across different neighborhoods in New York. The metrics considered for the mean satisfaction score included description accuracy, check-in process, cleanliness, communication, location, overall rating, and value. The findings indicated that Staten Island stood out as the neighborhood with the highest level of overall satisfaction. In contrast, as a tourist and business hub, Manhattan earned the highest mean score for location, aligning with our expectations.

PCA & Clustering Analysis Results

Upon scaling the data and constructing a correlation matrix, we noted that the correlation between location and other ratings ranged from 0.3 to 0.41. Among all breakdown rating metrics, value scores displayed the strongest correlation with other ones.

PCA Analysis

Dive into PCA analysis, from the scree plot, we believe the first three principal components captured a substantial amount of the total variance in the data.

The PCA analysis with three components indicates that the first component explains 58.95% of the variance, the second component explains 11.17%, and the third component explains 10.87%, arriving a 80.99% cumulative variance explained.

Examination of variable contributions reveals that overall_rating, accuracy, and value contribute significantly to the first component, while cleanliness, location, check-in and communication contribute more to the second component. The third component could suggest that customers’ perceptions of Airbnb listings’ locations significantly impact their satisfaction levels. The location value not limited to the neighborhood level, may also related to the convenience or close to stores and transportation. This component reflecting variations in review scores associated with location differences. This analysis provides insights into the underlying structure of the review scores and the variables driving the observed patterns.

In conclusion, the substantial loading of the location variable in the third principal component of the PCA results (68.67) confirms its significant influence on the dataset’s variability. This aligns with moderate correlations (0.3 to 0.41) found between location and other rating metrics. Together, these findings highlight the pivotal role of location in shaping customer ratings, emphasizing its importance in the Airbnb experience.

Clustering Analysis

In clustering analysis, we applied k-means clustering to principal components derived from PCA to explore dimensionality’s impact on clustering outcomes. Clustering based on two principal components showed distinct, well-separated clusters, while clustering based on three components exhibited increased overlap, indicating reduced separability and greater complexity. This highlights the critical role of dimensionality selection in further analysis.

Additionally, our previous findings(in PCA) regarding the significant influence of location_review variable on dataset variability align with the observed overlap in clusters dominated by location_review feature. The multifaceted nature of location feature such as proximity to transportation, safety considerations, and cultural nuances, introduces complexity, potentially leading to similar patterns across different clusters and contributing to observed overlaps.

We then performed k-means clustering based on two principal components by PCA, establishing three clusters. Cluster 1 had the largest size with 32336 listings, followed by Cluster 3 with 9650, and Cluster 2 as the smallest with 736 listings.

Followed by a comprehensive analysis of each cluster based on various metrics. Clusters are characterized by average ratings across different aspects such as accuracy, cleanliness, check-in, communication, location, and value. Additionally, the summary includes the distribution of listings across different neighborhoods within each cluster. Observed that cluster 1 outperformed the others, showing the highest mean ratings across all evaluation elements. In comparison, cluster 2 fall behind and needs further improvements on customer experiences

It is worth noting that the average AFINN score across clusters aligns with the observed cluster characteristics, indicating the sentiment polarity of reviews within each cluster. This consistency suggests that the sentiment expressed in reviews corresponds to the different ratings metrics within each cluster. By considering sentiment analysis alongside other metrics, such as ratings and neighborhood distributions, we can gain a more comprehensive understanding of the factors driving customer satisfaction and preferences within each cluster.

In supplement of our analysis on the effects of geographical location on customer satisfaction level(RQ 2), we applied MANOVA tests. The result highlighted statistically significant differences in customer satisfaction based on neighborhoods with an extremely low p-value. These results proved the strong correlation between neighborhoods and customer satisfaction.

GBM Modeling Results

After running the gradient boosting machine (GBM) predictive modeling and extracting feature importance, it’s evident that features such as price and amenity_count play a significant role in predicting review scores ratings. The top features identified by the GBM model provide valuable insights into the factors influencing review scores. This information can guide hosts and property managers in making targeted improvements to enhance the guest experience and overall satisfaction.

Recommendations

Research Question 1

In addressing the question of how listing features influence review score ratings and identifying areas for improvement, our analysis reveals several actionable insights. Price and the count of amenities emerge as top influencers on review scores(from the feature importance chart above). Specifically, listings with more amenities tend to see an incremental increase in satisfaction, plateauing thereafter. Hosts should aim to offer a balanced array of amenities to positively impact their review scores. Intriguingly, our data suggests that an increased number of accommodations does not correlate with higher review scores. This indicates that guests may value quality over quantity. Therefore, it is advisable for hosts to optimize the number of guests accommodated to align with the perceived value. Moreover, we find that a flexible cancellation policy is significantly favored over stricter policies(5 is most strict). Hosts are encouraged to adopt a more lenient cancellation approach, which could result in higher guest satisfaction as indicated by our findings.

Research Question 2

Exploring the impact of geographic location on Airbnb guest satisfaction, our PCA and clustering analysis identify that Cluster 1, predominantly encompassing listings from Manhattan and Brooklyn, outperforms other areas with all aspects, especially accuracy, communication, and check-in experiences.

Conversely, listings belonging cluster 2 have a relatively low rating metrics, suggesting bad customer experiences and provide improvement recommendations areas for hosts. This conclusion is further supported by the sentiment analysis on cluster 2 listings.

In light of the MANOVA analysis, which significantly rejects the null hypothesis, it is clear that geographic location is a crucial factor in shaping guest satisfaction. Our recommendations are twofold: For hosts in high-performing areas like Manhattan and Brooklyn, continue to maintain the high standards of communication and check-in guest experiences as these are proven drivers of satisfaction. For emerging and underrepresented neighborhoods, Airbnb could implement targeted initiatives to elevate guest satisfaction. This includes host education programs focusing on check-in processes and guest interaction, and promotional campaigns to attract guests by highlighting unique local experiences.

Additional analysis - Beyond RQ

In the continuation of our investigation for RQ2, we delved deeper into the amenities offered by conducting text mining analysis of the top 20 amenities across different clusters. Our analysis indicates that Cluster 1 listings prominently feature amenities such as hair dryers, (pets)friendly, Wi-Fi, and smoke detectors, which are valued by guests.

These findings imply that equipping lower-rated Airbnb with popular amenities may strengthen its potential to improve customer satisfaction level. However, further investigations are required to test our assumptions.

Based on these insights, we recommend that hosts consider the strategic addition or promotion of these amenities to boost guest satisfaction. Specifically, from the chart below, we can observe that the average occurrence of top amenities has a clear difference across three clusters. Which further proposed the recommendations for hosts to take the top amenities into account, approaching to the desired cluster and attracting more customers, potentially, receiving higher review scores.

Lastly, our price prediction model using XGBoost highlights the potential overpriced listings, which may impact on booking rates and, subsequently, on review scores. Some listings showed a variance between the predicted and actual prices. We suggest that hosts calibrate their pricing strategies to avoid overestimating their listing value, which could deter potential guests and affect occupancy rates. A competitive and accurate pricing approach may facilitate quicker bookings and lead to improved guest satisfaction.

The datatable below shows the listings that are potentially overpriced, suggesting to reduce the listing prices to align with the competitive market.

Conclusion

This research has thoroughly investigated the factors influencing Airbnb guest satisfaction in New York City, focusing on listing features and geographic factors, revealing the impacts of these factors on Airbnb’s guest experience and review scores.

The result indicates that listing features do significantly influence the review score rating. Through the study, we can see that accommodation capacity, amenities, and cancellation policies play a greatly important role in guest satisfaction. From a regional perspective, there are significant differences in satisfaction among different neighborhoods. For example, high-performing areas like Manhattan and Brooklyn are benchmarks of excellence in guest services, particularly in accuracy, communication, and check-in experiences. In contrast, emerging neighborhoods present opportunities for strategic enhancements.

Advanced analytical techniques (like clustering, PCA, and sentiment analysis) provides insightful information, indicating the importance of location in guest satisfaction, and helps to comprehend consumer’s preferences and emotions. Through this study, the gap in existing research can be filled. Besides, it gives some helpful suggestions to hosts, which will help them to enhance their listings, utilize pricing strategies based on the market, and boost booking rates and guest satisfaction.

In summary, this study highlights that data-driven strategies can greatly influence the hospitality industry. Moreover, it suggests that Airbnb hosts can enhance competitiveness and align with guest expectations through this data. Based on this research, future research and continuous improvements in shared accommodation can achieve great development.