Yelp Restaurant Reviews Research Articles

A Structural Topic and Sentiment-Discourse Model for Text Analysis

We consider the common setting where one observes a large number of opinionated text documents and related covariates, such as the text of online reviews along with the date of the review and the author demographic information. In this setting it can be of interest to understand how the covariates determine the text composition, as well as the prevalence, sentiment, and/or discourse of various discussion themes. Yet, most topic modeling methods in the machine learning literature are designed to summarize the text for the purpose of exploratory analysis and not to perform this type of formal statistical inference. Further, topic modeling methods generally do not try to estimate the sentiment or discourse of discussion along separate topics that can be critical in business applications (e.g., for summarizing service or product quality). We develop a topic model called the structural topic and sentiment-discourse (STS) model that introduces a new document-level latent variable that captures the sentiment and/or discourse (termed as “sentiment-discourse”) for each topic, which modulates the word frequency within a topic. These latent topic sentiment-discourse variables are controlled by document-level covariates to allow for experimental control and regression analysis. We also introduce new computational methods to resolve scalability issues that have forced previous models to restrict to a small number of categorical covariates. We benchmark the STS model on three real-world data sets from surveys, blogs, and Yelp restaurant reviews around the COVID-19 pandemic. Our model recovers meaningful results including rich insights about how COVID-19 affects online reviews, demonstrating that the STS model can be useful for regression analysis with text data in addition to topic modeling’s traditional use of descriptive analysis. This paper was accepted by Anindya Ghose, information systems. Supplemental Material: The online appendices and data files are available at https://doi.org/10.1287/mnsc.2022.00261 . An updated version of the R package implementing the STS model is available at https://CRAN.R-project.org/package=sts .

The impacts of geographic and social influences on review helpfulness perceptions: A social contagion perspective

Identifying and presenting helpful reviews to customers can significantly affect their purchase decisions. Although review helpfulness has been extensively explored in tourism research, extant studies have not sufficiently emphasized the unique characteristics of tourism products and investigated review helpfulness perceptions from both geographic and social influence perspectives. In this study, drawing on social contagion theory, we developed a theoretical framework to examine the impact of social contagion, specifically geographic and social proximities, on perceived review helpfulness. Our empirical analyses of Yelp restaurant reviews indicated that geographic and social influences have varying impacts on review helpfulness perceptions. Additionally, social contagions significantly moderated the impacts of various review- and reviewer-related factors, and product characteristics further moderated the contagion effect on perceived review helpfulness. This study provides valuable theoretical and methodological contributions to research on review helpfulness, especially in tourism contexts, and lays out the practical implications for various stakeholders.

Topic Modeling and Sentiment Analysis of Yelp Restaurant Reviews

The use of online consumer reviews (OCRs) websites like Yelp or TripAdvisor has increasingly gained popularity over the years. However, many products or services now have a large number of online reviews, which makes it difficult for consumers to decide which reviews to pay attention to, or for businesses to identify which critical areas to improve on. The use of text mining and sentiment mining techniques is deemed important to automatically process the content of online reviews and help improve review usefulness. Applying a four-phase research model, our study demonstrated data extraction and cleaning, topic modeling using Latent Dirichlet allocation (LDA) to extract five topics (Price, Time, Food, Service, and Location), sentiment analysis using Python TextBlob to aggregate consumer sentiment per topic from a Yelp restaurant reviews dataset, and model performance evaluation. We proposed that the design of recommender systems for OCRs or business decision-making systems can be faster, simpler, and more useful by integrating automatic topics extraction and sentiment analysis.

Mining behavioural and sentiment-dependent linguistic patterns from restaurant reviews for fake review detection

Online reviews are increasingly recognized as a key source of information influencing consumer behavior. This in turn implies that competitive advantage can be achieved by manipulating users’ perceptions about restaurants. The hospitality industry is particularly susceptible to this issue because products and services in this industry can only be rated upon consumption. Therefore, many efforts have recently been dedicated to developing automatic methods for detecting fake reviews based on data intelligence in this sector. Recent studies suggest that both the semantic meaning of consumer reviews and the sentiment conveyed may be useful indicators of fake reviews. However, the semantic meaning may be context-sensitive and may also disregard sentiment information. Moreover, the content analysis approach should be integrated with the reviewer's behavior to reveal their true intentions. To address these problems, we propose a review representation model based on behavioural and sentiment-dependent linguistic features that effectively exploit the domain context. Using a large dataset of Yelp restaurant reviews, we demonstrate that the proposed review representation model is more effective than existing approaches in terms of detection accuracy. It furthermore accurately estimates the average rating assigned by legitimate reviewers, which has significant managerial implications for the hospitality industry.

Detecting fraudulent online Yelp reviews using K-L divergence and linguistic features

Nearly all businesses that sell a product or service directly to consumers will be reviewed online. The ubiquity and importance of these online reviews to business success has woefully produced numerous opportunities for fraud. In this paper, we introduce and combine two tactics to distinguish fake Yelp restaurant reviews from real ones. Using genuine Yelp reviews, we employ two tactics. First, we examine 12 features across 7 linguistic categories and exploit the difference between the two datasets, achieving an average precision of 0.854. Next, we draw upon the Zipf distribution of ranked terms in the real and fake reviews and employ the Kullback–Leibler (K-L) divergence technique to augment our ability to discriminate between real and fake reviews. We obtain a final average precision of 0.867, which is an improvement over other techniques on comparable datasets.

Food “taste” on Yelp: displays of cultural omnivorousness and authenticity in dining choices and online reviews

ABSTRACT Styles of consumption indicate social status. This study aims to understand the influence of external constraints on status displays via two consumption styles – cultural omnivorousness and authenticity. Using a large-scale dataset of Yelp restaurant reviews and the 2015 American Community Survey, we find evidence that the ease – travel distance, neighborhood wealth, and the availability of culturally unique and authentic restaurants – with which status can be obtained via a consumption strategy influences preference for that strategy. Though limited to observations at the neighborhood level, our findings provide corroboration for existing theories about 1) individuals’ motivations to display cultural omnivorousness and authenticity in online review sites, and 2) the ways that these status displays are motivated by availability. Findings also indicate that while new media increase the access to information about cultural products, status display appears to have adapted to emphasize new scarcities, particularly localized knowledge.

The disproportion of crowd wisdom: The impact of status seeking on Yelp reviews.

This study shows that while status seeking motivates people to participate in crowdsourcing platforms, it also negatively impacts the bedrock of crowdsourcing–wisdom of crowds. Using Yelp restaurant reviews in 6 cities, we found that motivations of status seeking lead people to review a greater variety of restaurants, and achieving status further encourages this variety seeking as well as the targeting of more expensive restaurants for review. The impact of this individual-level tendency is confirmed by our aggregate-level analysis which shows that restaurants with higher price levels, higher uniqueness levels, and a larger percentage of elite reviews tend to obtain enough reviews to generate wisdom of crowds sooner than other restaurants. This leads to a different kind of distortion to crowd wisdom: an over-representation of status-conferring products and an under-representation of products that are not status-worthy. The findings suggest the importance of studying sources of distortion that are endemic to crowdsourcing itself.

Yelp Restaurant Reviews as Star Rating Predictors

Yelp Restaurant Reviews as Star Rating Predictors

Sentiment Analysis on Social Media Reviews Datasets with Deep Learning Approach

Thanks to social media, people are now able to leave guiding comments quickly about their favorite restaurants, movies, etc. This has paved the way for the field of sentiment analysis, which brings together various disciplines. In this study, Yelp restaurant reviews and IMDB movie reviews dataset were used together with the data collected from Twitter. Word2Vec (W2V), Global Vector (GloVe) and Bidirectional Encoder Representation (BERT) word embedding methods, Term Frequency-Reverse Document Frequency (TF-IDF), and the Bag-of-Words (BOW) were used on these datasets. Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), Recurrent Neural Network (RNN), Support Vector Machine (SVM), and Naive Bayes (NB) were used in the sentiment analysis models. Accuracy, F-measure (F), Sensitivity (Sens), Precision (Pre), and Receiver Operating Characteristics (ROC) were used in the evaluation of the model performance. The Accuracy rates of the models created by the Machine Learning (ML) and Deep Learning (DL) methods using the IMDB dataset were in the range of 81%-90% and 84%-94%, respectively. These rates were in the range of 80%-86% and 81%-89% for the Yelp dataset, and in the range of 75%-79% and 85%-98% for the Twitter dataset. The models that incorporated the BERT word embedding method have the best performance, compared to the other models with ML and DL. Therefore, BERT method is recommended for this type of analysis in future studies.

Sentiment classification using hybrid feature selection and ensemble classifier

This paper presents a Hybrid Feature Selection Technique for Sentiment Classification. We have used a Genetic Algorithm and a combination of existing Feature Selection methods, namely: Information Gain (IG), CHI Square (CHI), and GINI Index (GINI). First, we have obtained features from three different selection approaches as mentioned above and then performed the UNION SET Operation to extract the reduced feature set. Then, Genetic Algorithm is applied to optimize the feature set further. This paper also presents an Ensemble Approach based on the error rate obtained different domain datasets. To test our proposed Hybrid Feature Selection and Ensemble Classification approach, we have considered four Support Vector Machine (SVM) classifier variants. We have used UCI ML Datasets of three domains namely: IMDB Movie Review, Amazon Product Review and Yelp Restaurant Reviews. The experimental results show that our proposed approach performed best in all three domain datasets. Further, we also presented T-Test for Statistical Significance between classifiers and comparison is also done based on Precision, Recall, F1-Score, AUC and model execution time.

What factors affect consumers’ dining sentiments and their ratings: Evidence from restaurant online review data

Online review data refers to customer opinions (e.g., text reviews, numerical ratings, and personal information) left on online retailing and review platforms. Here, we provide insight for food-related consumer sentiment study by using online review data. We present a case study using Yelp restaurant review data on what affects restaurant customers’ sentiment responses to dining out. In our case study, a sentiment-analysis method was used to extract and generate sentiment indices from 175,879 text-based Yelp restaurant reviews. We examined the relationships between our sentiment indices and ratings and showed that online review data can provide information about consumer dining sentiments. By using semi-automatic content analysis method, we first counted the most frequent 300 words in all text reviews and manually categorized those most frequent words into five topics: “food,” “service,” “expenditure,” “social,” and “miscellany”. We then computed the fraction of topic words in each category over the total topic words for each review. We found that (1) consumers used more positive sentiment words than negative sentiment words in their Yelp restaurant text reviews. (2) The proportion of positive sentiment words in reviews was positively related to ratings; the proportion of negative sentiment words in reviews was negatively related to ratings. (3) Relative to food, consumers used more sentiment words (both positive and negative) when they were discussing restaurant service. (4) Consumers used the least sentiment words when they were discussing social-related topics compared with the other topics (food, service and expenditure). (5) Consumers rated restaurant service higher than they rated food.

A Weakly Supervised WordNet-Guided Deep Learning Approach to Extracting Aspect Terms from Online Reviews

The unstructured nature of online reviews makes it inefficient and inconvenient for prospective consumers to research and use in support of purchase decision making. The aspects of products provide a fine-grained meaningful perspective for understanding and organizing review texts. Traditional aspect term extraction approaches rely on discrete language models that treat words in isolation. Despite that continuous-space language models have demonstrated promise in addressing a wide range of problems, their application in aspect term extraction faces significant challenges. For instance, existing continuous-space language models typically require large collections of labeled data, which remain difficult to obtain in many domains. More importantly, previous methods are largely data driven but overlook the role of human knowledge in guiding model development. To address these limitations, this study designs and develops weakly supervised WordNet-guided deep learning to aspect term extraction. The approach draws on deep-level semantic information from WordNet to guide not only the selection representative seed terms but also the pruning of aspect candidate terms. The weak supervision is provided by a very small set of labeled data. We conduct a comprehensive evaluation of the proposed method using both direct and indirect methods. The evaluation results with Yelp restaurant reviews demonstrate that our proposed method consistently outperforms all baseline methods including discrete models and the state-of-the-art continuous-space language models for aspect term extraction across both direct and indirect evaluations. The research findings have broad research, technical, and practical implications for various stakeholders of online reviews.

The Disproportion of Crowd Wisdom: The Impact of Status Seeking on Yelp Reviews

This study shows that while status seeking motivates people to participate in crowd-sourcing platforms, it also negatively impacts the bedrock of crowd-sourcing – wisdom of crowds. Using Yelp restaurant reviews in 6 cities, we found that motivations of status seeking lead people to review a greater variety of restaurants, and achieving status further encourages this variety seeking as well as the targeting of more expensive restaurants for review. The impact of this individual-level tendency is confirmed by our aggregate-level analysis which shows that restaurants with higher price levels, higher uniqueness levels, and a larger percentage of elite reviews tend to obtain enough reviews to generate wisdom of crowds sooner than other restaurants. This leads to a different kind of distortion to crowd wisdom: an over-representation of status- conferring products and an under-representation of products that are not status-worthy. The findings suggest the importance of studying sources of distortion that are endemic to crowd-sourcing itself.

Artificial Intelligence for Surveillance in Public Health.

Summary Objectives : To introduce and summarize current research in the field of Public Health and Epidemiology Informatics. Methods : The 2018 literature concerning public health and epidemiology informatics was searched in PubMed and Web of Science, and the returned references were reviewed by the two section editors to select 15 candidate best papers. These papers were then peer-reviewed by external reviewers to give the editorial team an enlightened selection of the best papers. Results : Among the 805 references retrieved from PubMed and Web of Science, three were finally selected as best papers. All three papers are about surveillance using digital tools. One study is about the surveillance of flu, another about emerging animal infectious diseases and the last one is about foodborne illness. The sources of information are Google news, Twitter, and Yelp restaurant reviews. Machine learning approaches are most often used to detect signals. Conclusions : Surveillance is a central topic in public health informatics with the growing use of machine learning approaches in regards of the size and complexity of data. The evaluation of the approaches developed remains a serious challenge.

Cross-Cultural Comparisons of Review Aspect Importance

Previous text mining studies identified key common aspects in online restaurant reviews. However, it is not clear how important these aspects are for consumers. In this exploratory study, we used Yelp restaurant reviews on an ethnic food item, ramen noodles, and assessed the importance of each aspect to both U.S. and Japanese consumers. The results show that food and atmosphere are far more important than the other common aspects in both the U.S. and Japan. However, we found noticeable differences between consumers in the two countries regarding how the food aspect plays a role on star ratings. Both implications and a future research agenda are discussed.

Discovering foodborne illness in online restaurant reviews.

We developed a system for the discovery of foodborne illness mentioned in online Yelp restaurant reviews using text classification. The system is used by the New York City Department of Health and Mental Hygiene (DOHMH) to monitor Yelp for foodborne illness complaints. We built classifiers for 2 tasks: (1) determining if a review indicated a person experiencing foodborne illness and (2) determining if a review indicated multiple people experiencing foodborne illness. We first developed a prototype classifier in 2012 for both tasks using a small labeled dataset. Over years of system deployment, DOHMH epidemiologists labeled 13 526 reviews selected by this classifier. We used these biased data and a sample of complementary reviews in a principled bias-adjusted training scheme to develop significantly improved classifiers. Finally, we performed an error analysis of the best resulting classifiers. We found that logistic regression trained with bias-adjusted augmented data performed best for both classification tasks, with F1-scores of 87% and 66% for tasks 1 and 2, respectively. Our error analysis revealed that the inability of our models to account for long phrases caused the most errors. Our bias-adjusted training scheme illustrates how to improve a classification system iteratively by exploiting available biased labeled data. Our system has been instrumental in the identification of 10 outbreaks and 8523 complaints of foodborne illness associated with New York City restaurants since July 2012. Our evaluation has identified strong classifiers for both tasks, whose deployment will allow DOHMH epidemiologists to more effectively monitor Yelp for foodborne illness investigations.

Crowd-Driven Competitive Intelligence: Understanding the Relationship Between Local Market Competition and Online Rating Distributions

Crowdsourced online mean ratings of local businesses are increasingly being used to infer the market power of a business. An important consideration in making this inference is whether two identically rated businesses (e.g., 4 stars) encounter the same local competitive dynamics if they face contrasting local market competition. To analyze this, the authors combine demographic, socioeconomic, and Yelp restaurant reviews for 372 isolated markets in the United States. The empirical estimates demonstrate that an increase in overall competition—measured as total number of businesses in a market—leads to a broader range and to a decrease in the average of a market’s mean rating distribution. This implies that two restaurants with equal mean ratings differ substantially in the way they are affected by local competition depending on whether they are located in a large or in a small market. For example, a larger market has proportionately more lower rated restaurants, whereas higher rated restaurants have relatively fewer comparable substitutes and face less competition in such a market. Therefore, restaurants with identical mean ratings should ideally be assessed differently if they operate across markets with dissimilar numbers of competitors.

How online reviews create social network value: the role of feedback versus individual motivation

Online reviews, like those compiled at Yelp, Amazon, Epinions and Tripadvisor, allow consumers to create social network value for other consumers, and to influence the success or failure of reviewed businesses. These reviews, produced without compensation, are a prominent example of how new social media have altered the challenges facing strategic marketing. Although the production of online reviews has exploded, little is known about what motivates ordinary consumers to produce reviews. We surveyed Yelp review writers, and then compiled an archive of Yelp restaurant reviews and the feedback received on those reviews. This data-set enabled a longitudinal analysis of the role of feedback as an external motivation for review production. Although receiving positive feedback increases the probability that a novice reviewer will continue to produce reviews, such feedback had no detectable effect on the productivity of established reviewers. Results point to an emerging segment of consumers who have a need to publish and reach a mass audience.

A Theory of Politeness and Rudeness in Cooperative Communication

It is common for one who has information to share it cooperatively with one who needs it. Perhaps surprisingly, this information is often not communicated in the simplest possible way. For example, Standard and Poor's assigns ratings of at least "B-" to 97% of corporate issues, and segments these issues into 16 categories (AAA, AA, AA-, etc.). A full 7 of these 16 categories are devoted to issues with nearly identical default probabilities, leaving only 9 categories to cover the wide variety of default probabilities found in speculative corporate debt. Equally concerning, Yelp restaurant reviews are predominantly positive, with an average of 3.8 stars out of 5. This limits the site's usefulness in distinguishing the highest quality fare. I show that the purpose of a reviewer generates the optimal distribution of reviews. If it is most important to separate great from good, then reviews will tend to be harsh, in the sense that most reviews will be below average. If it is most important to separate bad from worst, then reviews will tend to be polite, in the sense that most reviews will be above average. Importantly, politeness and harshness are emergent properties of the optimal messaging rule. Results are consistent with casual observation, and provide testable implications across a variety of settings, including credit reports, analyst ratings, credit ratings, wine ratings, referee reports, customer reviews, grade inflation, and letters of recommendation.