Fake Review Checker Tools. Can They Be Trusted?
Any small business owner that has had a competitor create a slanderous review about their business or products knows that their revenue is in jeopardy. Reputation management companies will not be of any help as they can only fight fire with fire. This only damages consumer trust in the online reviews they see on the web. Customers are besieged with fake reviews, and it has become such a problem that CNET has produced a guide to help consumers spot fake product reviews on major websites like Amazon, Best Buy, and Walmart. You also know that something is a major problem when the Federal Trade Commission prosecutes a company in New York for paying for fake reviews.
In response to this widespread problem, there are some services online that claim to be able to identify fake reviews. One of these services claims to provide “a new way of filtering product reviews to find out what real users are saying about the products you want to buy”. So how does this actually work, and how accurate is this process?
The Secret: Natural Language Processing
If you are familiar with machine learning, then you already know about natural language processing (NLP). If you aren’t a programmer or hardcore techie, then allow me to explain: NLP encompasses a set of techniques that allow computers to extract information from large blocks of text. Each block of text is fed into a computer, and an algorithm can extract information like sentiment (positivity or negativity), topics, and even the context of a question. If you’ve had a recent interaction with a customer service chatbot, the chatbot would not be able to provide answers to your questions with NLP algorithms.
So how does NLP and machine learning apply to spotting fake reviews? These fake review checker tools crawl known review websites and retailer websites for customer reviews and copy them into a database. These reviews are then checked against another set of reviews that are categorized as real or fake.
Machine learning algorithms for NLP can be used to take a reference dataset (in this case, the supposed authentic reviews) and used to develop a set of indicators that will tell you how likely a suspect review is real or fake. Certain words or phrases in the review indicate a high probability that the review is either real or fake. This then allows a computer to distinguish whether a certain review has a high probability of being a review that was left by a real customer.
There are other metrics that these sites use in addition to just examining the text itself. A fake review checker will examine the number of times phrases have been repeated, the number of users who have only reviewed the product in question versus other products, the number of reviews posted by each user, and other metrics. The metrics are not necessarily consistent across different fake review checker tools.
The Problem with Fake Review Checker Tools
There are several problems with these tools. First, no machine learning model is perfectly accurate. Machine learning algorithms based on NLP tend to be less accurate than models used for predicting relationships between numerical quantities. This lack of accuracy occurs because there are thousands of ways to construct a sentence that communicates the same message, and it is extremely difficult to capture all of those variations in a single set of data.
These fake review checker websites also lack transparency. From where these sets of reviews were obtained is unknown, thus the accuracy of these reference datasets is entirely questionable. This then calls into question the accuracy of the models used by these websites. These websites may use different datasets and models, leading to different results.
The author of this blog post did a quick experiment and found that two different fake review checker sites (fakespot.com and reviewmeta.com) returned consistent products when they were used to examine these bamboo sheets on Amazon, which received about 30,000 reviews. However, not all products on Amazon will receive nearly this many reviews.
When the author used these two tools to examine the reviews on a similar product that had only received 99 reviews, the results were quite different. Fakespot returned a “D” review grade, indicating the reviews were not trustworthy. However, Reviewmeta returned a “Pass” grade, failing only in the “Suspicious Reviewers” category.
A pass rating and fail rating for the same product.
These fake review checker tools are untrustworthy for the majority of products on the market. Unless your product has thousands of reviews with an unimpeachable reputation, no one can be sure whether the reviews they are seeing on your products and business are real.
Rebusify Has the Solution
When one examines the obvious flaws in fake review checker tools, consumers and businesses need a branded solution that links real customers to real reviews about real purchases. The Rebusify Confidence System gives customers the assurance they need to know that the reviews they see online came from real customers. The reviews seen by your customers are immutable, meaning they cannot be changed by the merchant or by Rebusify. This allows the best merchants to outshine their competition.
Rebusify is here to provide honesty and transparency at a time where anyone can create a fake identity and slander your business. Customers deserve to know the truth about your business and your products, and Rebusify is here to help build that trust.