Prevalence of Fraud in Yelp’s Review System

Yelp is a global phenomenon in that it has infiltrated businesses throughout the world while having a lasting impact on the future of their financial success. However, there are no requirements to make a Yelp account besides owning an email which can be easily made for free. This means that anyone is able to make a review that may influence a business. This makes Yelp a double-edged sword for businesses as it can either have a positive or negative impact. There is a chance that it could bring more customers if the reviews are great. Unfortunately, competition can get dangerous on Yelp if they hire malicious reviewers to reduce the reputation and popularity of a business. This is very commonplace and has provoked Yelp to create a detection algorithm for fraudulent reviews. In a study, it was shown that up to 25% of reviews may be fraudulent (Rahman, et al.).

Figure 1: Spikes in positive reviews for three different businesses (Rahman, et al.)

With the increase of risk using Yelp, how much benefit would those who abuse the service in their favor? It has been studied that every incremental star leads to a 5-9% increase in revenue and every extra-half star increases the chances of selling out by 19%. As seen in figure 1, one factor that can help spot a company that utilizes fake reviews for their own benefit is a spike in positive reviews. Those who are hired to produce fake content are called opinion spammers (Mukherjee, et al.). As seen in the graphs, we can see that the problem of opinion spamming has greatly increased since around 2010. For small businesses, this is extremely suspicious because the span of customers that they can advertise to would be small. Furthermore, if it was the superb quality that brought about many customers, they should have in the past as well and that is not seen pre-2011 for the three businesses.

There are multiple ways that can increase the chances of spotting a fake reviewer. This is interesting as it can be nearly impossible to even tell by reading them manually. It requires a lot of data analysis as it’s not reliable to directly process subjective and objective text yet alone differentiate them. One method called the quasiclique extraction creates graphs that link nodes that represent users. If they are connected, then that means they have reviewed the same business recently. Doing this, you can spot the army of review mercenaries that are supposedly a “clique” (Jain, et al.). It’s ironic to note that some of these cliques were actually Yelp scouts – reviewers hired by Yelp to basically pave the way for future Yelpers. Another method for detecting fake reviews employs a strategy of correlation between users that write positive reviews for businesses that have more negative reviews than positive and vice versa (Akoglu, et al.).

While fraudulent reviews may be positive, they can also be negative. The logic behind this is to reduce the reputation of competitors in order to wipe them out. It is indeed a nasty way to survive in the business world but it works so it’s not surprising to find this. As computer algorithms to detect these type of false reviews are flawed, it can easily flag a legitimate Yelper. As such, this brings about a controversial topic of the Freedom of Speech. By silencing negative reviews, this creates a potential problem as many may not only be dissatisfied in the future, but consumers also should have the right to voice their opinions (Castro).

Works Cited

Akoglu, Leman, Rishi Chandy, and Christos Faloutsos. “Opinion Fraud Detection in Online Reviews by Network Effects.” ICWSM 13 (2013): 2-11.

Castro, Daniel. “Committee on Commerce, Science, and Transportation US Senate November 4, 2015.” (2015).

Jain, Paras, et al. “Poster: Spotting Suspicious Reviews via (Quasi-) clique Extraction.”

Mukherjee, Arjun, et al. “Spotting opinion spammers using behavioral footprints.” Proceedings of the 19th ACM SIGKDD international conference on Knowledge discovery and data mining. ACM, 2013.

Rahman, Mahmudur, et al. “To catch a fake: Curbing deceptive Yelp ratings and venues.” Statistical Analysis and Data Mining: The ASA Data Science Journal 8.3 (2015): 147-161.