Beyond Star Ratings: How Online Reviews Can Forecast Restaurant Survival

An innovative study by Professor Hengyun Li and Mr Xiang (Kevin) Zheng of the School of Hotel and Tourism Management (SHTM) at The Hong Kong Polytechnic University (PolyU), working with co-authors, puts business survival prediction in the spotlight. Their research, focusing on the determinants of success in the restaurant trade, yields invaluable lessons for the hospitality and tourism industry as it adapts to economic uncertainty and transition.

 Protecting restaurants is a concern for governments, as the catering sector is among the largest sources of economic activity in areas dependent on tourism. Yet while crucial to economies’ health, “the tourism and hospitality sector is naturally vulnerable and bears the brunt of crises”, the researchers warn. Restaurants are no exception: with their limited resources and fragile finances, they are highly susceptible to failure when the business environment becomes more challenging. Buffering restaurants against economic storms to preserve the social and economic benefits they offer is thus paramount.

 Tourism and economics scholars have an important role to play here. “Amid global uncertainty and an ever-changing business environment”, the authors write, “researchers have increasingly striven to rationalise business failure and heighten firms’ likelihood of surviving”. In the restaurant context, scholars have identified various factors that affect the likelihood of long-term survival. Increasingly, these reflect trends in today’s social media age, in which online reviews can make or break a restaurant’s reputation.

 This has not gone unnoticed by businesses themselves, which are increasingly taking proactive steps to enhance their online word-of-mouth. “Major industry players (e.g., Marriott) have begun tasking individual staff, or even teams, with monitoring customer reviews to provide timely responses”, the researchers note. “Hospitality businesses that attend closely to online reviews perform far better (e.g., in terms of revenue and visibility) than those that do not”.

 In the era of big data, however, there is scope for a more nuanced look at the relationship between online reviews and restaurant success. The most familiar metric from restaurant review sites might be the average customer rating. The authors realised that further insights could be gained from review variance, the extent to which reviews of a restaurant differ widely in terms of ratings and sentiments. So far, they write, “no study appears to have explicitly examined how review variance affects businesses’ survival”. This is a valuable research opportunity, given that “findings regarding review variance in other contexts (e.g., retail) imply a close relationship between these two variables”.

 Ratings and variance aside, the power of online reviews to affect real-world business survival might depend on another aspect: expertise. Prolific restaurant reviewers, especially those with high badge levels and numerous upvotes, tend to be seen as experts. Signalling theory suggests that such “expert” status signals to readers that a review is reliable. Expert reviews are thus likely to more strongly influence a restaurant’s footfall. “Consumers’ perceptions and decisions influenced by different signals”, the authors summarise, “can affect product demand, which ultimately can affect business survival”.

 Since the end of the COVID-19 pandemic, economies around the world have faced new challenges. For the vulnerable restaurant sector, a thorough analysis of the variables most closely linked to business survival would be a lifeline. Online reviews represent an invaluable, yet not fully utilised, resource in this context. “Within an unstable economy and a competitive business environment, timely research on restaurants’ survival is urgently needed”, the researchers tell us.

 To this end, their comprehensive, data-driven study leveraged mass restaurant review records from the major US city of Boston, “a popular tourist destination boasting a dining culture rich with restaurants and cuisine types”. Over half a million online reviews from Yelp, covering almost 3,000 eateries in the Boston area, were analysed. The authors extracted potential key features of these reviews for predicting restaurant survival, including average rating scores, review length, number of reviews, and words indicating positive and negative sentiment.

 Operational data on the restaurants were also acquired to untangle their relationship with review features. “Several business-related factors can affect restaurants’ survival”, the researchers note, “including chain business, price, the number of competitors, customer engagement, review ratings, review length, number of reviews, and business age”. They then designed prediction models to test how review variance and review source (expert or non-expert) affected restaurants’ survival, and they ran a further analysis to tease out the importance of other features.

 An array of findings emerged from the data, with strategic implications for businesses. First, reviews by “Elite” users (Yelp’s certification for highly engaged reviewers, whose opinions are deemed helpful by the site’s community) held the most sway. “Compared with non-expert reviews”, report the authors, “the variables extracted from expert reviews could more accurately predict restaurants’ one-year-ahead survival status”. That is, the models were better at forecasting restaurant survival when using reviews by Elite members, implying that these have more influence on diners’ decisions and in turn the fates of individual businesses.

 Regarding specific features, status as a chain restaurant, price, and business age were the variables most closely related to survival. Among averaged variables, reviewers’ average level of sentiment regarding food prices was the best predictor of restaurants’ fate. Nonetheless, for all variables except average price sentiment, the “the variance was more important than the average”. Specifically, a wide variance in review features for any given restaurant strongly predicted its failure to survive the next one to two years.

 The implications of these findings are clear. “Business owners should pay attention to review variance”, write the researchers. The sophisticated mathematical methods used in their study provide a ready-made approach for restaurant managers and other stakeholders to forecast performance by using variance data from online reviews. “We have provided an additional tool with which to evaluate businesses’ survival”, the researchers note. When a high degree of review variance is observed, they advise managers that “it is important to minimise possible inconsistencies” in product and service delivery.

 A more fine-grained interpretation of variance data is also possible. While variance in overall review sentiment is crucial, it can be broken down into specific aspects of reviews that strongly indicate issues with business development. “At the micro level”, say the authors, “restaurateurs should carefully evaluate customers’ attitudes towards current price levels to see whether costs meet diners’ expectations”. The outsized influence of Elite reviewers, meanwhile, suggests that managers “should encourage diners with expert status to write reviews after visiting a restaurant”.

 The survival prediction tool that this study provides is relevant not only to practitioners but also to investors, for whom restaurants are invariably a risky sector. “By leveraging our prediction model”, the researchers write, “investors can check the consistency of customers’ attitudes towards certain restaurants to clarify businesses’ status and potential”. Armed with this information on review sentiment variance, “investors can then pursue opportunities with a higher presumed return on investment”.

 “Timely and accurate business survival prediction is vital in today’s dynamic business environment”, the authors remind us. For the food sector, this research provides timely insights that are ground-breaking in both detail and relevance. Their importance even extends to the site operators of review platforms, which could, for example, choose to utilise restaurants’ review variance among their indicators. For the hospitality industry overall, the study reaffirms the importance of a data-informed understanding of businesses’ survival prospects in uncertain times.

 Hengyun Li, Anqi Zhou, Xiang (Kevin) Zheng, Jian Xu, and Jing Zhang (2025). Restaurant Survival Prediction Using Machine Learning: Do the Variance and Sources of Customers’ Online Reviews Matter? Tourism Management, Vol. 107, 105038.