Today in Digital Marketing - Deep-Dive Week: Fake Product Reviews

Episode Date: November 15, 2022

A group of researchers have published a study called “Detecting Fake Review Buyers Using Network Structure.” Tod speaks with one of the authors.NOTE: This week, while Tod's away, we're pla...ying interviews with marketing scientists. We'll be back to the regular newscast show on November 21.  ✨ GO PREMIUM! ✨ ✓ Ad-free episodes ✓ Story links in show notes ✓ Deep-dive weekend editions ✓ Better audio quality ✓ Live event replays ✓ Audio chapters ✓ Earlier release time ✓ Exclusive marketing discounts ✓ and more!Check it out: todayindigital.com/premiumfeed📰 Our Newsletter: Get It (daily or weekly)✉️ Contact Us: Email or Send Voicemail⚾ Pitch Us a Story: Fill in this form📈 Reach Marketers: Book Ad🗞️ Classified Ads: Book Now🤝 Join our Slack: todayindigital.com/slack🙂 Share: Tweet About Us • Rate and Review🎤 Follow: LinkedIn • TikTok • FB Page/Group👨🏻‍💼 Follow Tod: LinkedIn • TikTok------------------------------------🎒UPGRADE YOUR SKILLS• Inside Google Ads with Jyll Saskin Gales• Foxwell Slack Group and Courses👍 TOOLS WE RECOMMEND• Social media mgmt: Sprout Social and Agorapulse• Marketing tools: Appsumo• Podcast recording: Riverside.FM💡 MARKETING SPOTLIGHTIf you like Today in Digital Marketing, you’ll LOVE Stacked Marketer: the free daily newsletter that gives marketers an edge on the competition in just 7 minutes a day.Covering breaking news, tips and tricks, and insights for all major marketing channels like Google, Facebook, TikTok, native ads, SEO and more.Join 32k+ marketers who read it daily. Sign up free now! ------------------------------------Today in Digital Marketing is hosted by Tod Maffin and produced by engageQ digital on the traditional territories of the Snuneymuxw First Nation on Vancouver Island, Canada. Associate Producer: Steph Gunn. Ad Coordination: RedCircle. Production Coordinator: Sarah Guild. Theme Composer: Mark Blevis. Music rights: Source AudioSome links in these show notes may provide affiliate revenue to us. Our Sponsors:* Check out Kinsta: https://kinsta.comPrivacy & Opt-Out: https://redcircle.com/privacy

Transcript
Discussion (0)
Starting point is 00:00:00 Hello and happy Monday. I'm Todd Maffin. This week we are shaking things up a little. My wife and I are out of the country on holiday this week. Our associate producer Steph Gunn has been moved over agency side to help with our client work. Sarah, our production coordinator, is also on assignment there. So for this week, things will be a little different. You will still get an episode every day. But this week, we will be doing deep dives into marketing with five different marketing scientists. These, by the way, are what people on our premium feed get as part of their subscription, usually a couple of months, they come out on Saturday. So if you're not already a premium member, check out all the benefits by going to todayindigital.com slash premium. You can check out right on your phone with Apple Pay or Google Pay if you want. Today, we start with a deep look into fake product reviews. There's a lot we marketers have to compete for. Share of voice, product sales, social media engagement,
Starting point is 00:00:55 and few tools to really showcase the difference between our product and a competitor's. One tool that's proven helpful are online reviews. When we have good ones, sales can pick up. But there too, we compete, not only with reviews of our competitors, but also fake reviews. The fake review industry is huge. Many brands quietly pay for buckets of phony testimonials.
Starting point is 00:01:19 And even though sites like Amazon say they're good at detecting those, the fake ones are still there. We can, though, turn to science for some more clarity. One group of researchers have just published a research paper called Detecting Fake Review Buyers Using Network Structure. Heiss Overgur is one of those researchers. He is an assistant professor of marketing at the Saunders College of Business, and he joins me from his office in Rochester, New York. Dr. Overgur, welcome, or should I say welcome back since you've been here before?
Starting point is 00:01:47 Yeah, thanks for having me back. Not at all. There are hundreds of studies on fake review detection. What makes yours different? So what most of those studies do or try to do is that they create kind of their own data set that has labels to indicate whether or not a fake review, a review is fake or not. And what we see in many of those studies is that they use a variety of approaches. One could be that they just create those fake reviews themselves using either a bot or they try to get people to write fake reviews so that they know whether or not that review is fake or whether or not that review is real, right? And the problem with that is that those are often kind of fabricated and they prove kind of easy to detect in a way. And it allows for not the perfect data points, if you will,
Starting point is 00:02:48 for these kinds of classification methods. And the other end, we see that there's a lot of studies that kind of rely on the platforms to do the filtering for them. Let's say Amazon already has their own team that is detecting fake reviews. And so they delete those reviews and then researchers will use those labels as their classification of fake or real. And then they use that to predict several factors of those reviews and see what is real or fake, right? And what we do differently in our study is that we kind of infiltrate Facebook groups that are soliciting fake reviews and use that to directly identify
Starting point is 00:03:37 which products are buying fake reviews. And instead of focusing on the review level, so instead of predicting just straight, this review is fake, this review is real, we focus a little bit higher level and we say, hey, can we detect what reviews or what sellers on Amazon are recruiting these fake reviews? And for that, we have accurate data because we know for sure, because they are advertising in these Facebook groups, that they are recruiting fake reviews. So in that sense, we have a unique data set for which we are 100% sure that they are recruiting fake reviews and for which
Starting point is 00:04:18 none of that is kind of fabricated. This is happening in real life. So are there brands in Facebook groups that are using their own brand name and going into these groups and saying, hey, listen, we'll pay X dollars for a fake review? Not directly, no. So what they do is that they kind of advertise a picture of a product and a description, but they don't advertise who they are or what the product name is or the product identifier on Amazon. So they kind of rely on the people in the Facebook groups that they are trying to pay for fake reviews to go in themselves and find these Amazon products. And then by communication in a private channel,
Starting point is 00:05:06 they then verify that these people have bought a product, that they have written a fake review, that they have written a review that looks authentic with five stars. And then they reimburse these sellers on these groups. Right. Reimburse the reviewers. Sorry. Yeah. They reimburse the reviewers. Correct. Right. Right. Right. Right. How much does a review go for? So they generally pay the purchase price. So it basically gives the reviewer a free product. And sometimes they include a small commission, but that varies from product to product. But generally, it's a free product. Yeah. And where are these people located? Are these just like housewives in Iowa
Starting point is 00:05:53 or are these in a call room in India? Where are these people located that are doing these reviews? Anywhere. And what they require is that these reviewers have a real or a verified account and that they actually, and they verify also the purchase. So they have to show receipt. And that way, that is their way of going around some of the policies that Amazon has put in place of, you need to have a verified purchase and you need to have a verified account with a true address for you to be able to purchase a project and to leave a review. So they go around all of those things. And basically anyone that shows interest in leaving a fake review for a product can do that. And these Facebook groups are also private. So you have to be led into the Facebook group, essentially.
Starting point is 00:06:52 And there's almost no risk to the brand in terms of, I'm speaking financial risk here, you know, because I mean, yes, you've got to reimburse them for the cost of that product, but you get that sale as revenue for the product. Are there any negative repercussions at all for brands? How often are they caught? Yeah, that varies. So we don't have too much information on whether or not they get caught, but we do know that they tend to stay behind a certain level of how many reviewers they recruit and trying to prevent themselves from getting caught. And they've gone over time. We've talked to some of these sellers and they've gone better at it over time. And the only cost really that they
Starting point is 00:07:39 incur in this case is the production cost and the taxes, right? Because they have to reimburse the seller, this fake review writer. And then what they get back is an increase in sales rank, an increase in rating, and a better visibility on the page. And that shows to be quite profitable for the brands. Do you have business insurance? If not, how would you pay to recover from a cyber attack, fire damage, theft, or a lawsuit? No business or profession is risk-free. Without insurance, your assets are at risk from major financial losses, data breaches, and natural disasters. Get customized coverage today, starting at $19 per month at zensurance.com. Be protected. Be Zen.
Starting point is 00:08:27 Your paper was called Detecting Fake Review Buyers Using Network Structure. What is a network structure and what were your main findings overall for the research? Yeah, so that is indeed the title of the paper. And the network structure is really based on network theory or social network theory in computer science. And that is widely used also in many marketing studies and really focuses on mapping the connection between reviewers and products. So we map out, hey, what product leaves, what reviewer leaves reviews at what products? And then what we find is that because these sellers that are soliciting fake reviews kind of grab reviews from a smaller pool of reviewers, if you will, because they recruit them through Facebook, right? And these are connected groups. So what we find is that by mapping out this structure of network, the structure of reviewers
Starting point is 00:09:32 and buyers, reviewers and sellers, sorry, we can find that the products that are buying fake reviews versus the products that are not buying fake reviews, they are much closer connected to each other by the reviewers they share. And that was our main finding. And that proves to be really predictive of whether or not these products, these sellers are buying fake reviews through these platforms. It would strike me that this is technology or an algorithm, I guess, to be more accurate, that Amazon could develop for themselves. Do you think they have this? I can only assume that they are working on this, but we did make some comparison to what we found
Starting point is 00:10:18 and what products that we highlighted as being fake real buyers and the reviews that they have deleted. And we do find overlap, but not to the extent where we can say that they use our method specifically. Amazon, of course, says the number of fake reviews is a very tiny fraction in the total pool. After having done this research, do you have a sense of how widespread the fake review problem is? Is Amazon right? Is it tiny or is it bigger than we all think? No, it is quite large. So we found about 1,400 products in these Facebook groups for the short amount of time that we gathered them. And we did some kind of like back of the envelope calculation. And we estimate that between 2019 and 2020, over the course of a year,
Starting point is 00:11:15 there's about 4.5 million products purchasing fake reviews through these kind of marketplaces, if you will. And what product categories do you find this to be particularly egregious in? Are there a lot of people selling shoes or a lot of people selling, I don't know, electronics? Are there categories where there just tends to be a lot more volume of fake reviews? No, that is what is really interesting. There's not a specific category, or there's not a specific type of seller or a specific type of product that is more likely to buy these fake reviews. And the quality of products also is kind of widespread. So there's not really any specific type of company that is more likely to do so.
Starting point is 00:12:06 I know you studied Amazon and Amazon alone, but what does your gut tell you about the other bigger players in the e-commerce world? I'm thinking places like Walmart, which is trying to compete with Amazon, or even maybe more of a platform provider like Shopify. I can only imagine that this happens in the same way, right? For the reasons that you've discussed is the competition is so fierce and there's only a number of ways that as a brand or a product, you can stand out. And we know that reviews are a very important part of the consumer decisions. And for that reason, we can only expect that these groups or these kind of
Starting point is 00:12:47 recruitments happen on any major platform. What surprised you the most about your findings? Well, what was surprising to us, we expected that the network structure was going to reveal some interesting patterns and that would be an effective predictor. But we didn't expect it to outperform very sophisticated machine learning metrics such as detecting certain types of text or detecting certain types of words or using the use of pictures. All of those were performing worse as compared to the network structure. And even if we looked at only the top two features from the network that were important, we were still able to really accurately predict this. So that was what we were hoping for,
Starting point is 00:13:44 but that was also surprising in a sense. So what can an average marketer do? I mean, anything, what happens if they, let's say someone who's listening to this competes on Amazon or Walmart or Shopify, and they discover that one of their competitors is likely running fake reviews, like it's just obvious the way it's worded or whatever. Is there anything marketers can do? Is there a way of reporting reviews? I think you can report fake reviews or fake review buyers probably on Amazon.
Starting point is 00:14:16 So that is your best bet, yeah. I do wonder if anything will really happen though. I mean, you know, reporting these days on social, there's so many staff cuts at all these platforms. I don't know. I think, you know, reporting these days on social, there's so many staff cuts at all these platforms. I don't know. I think, I assume it just goes to a bot that decides whether or not it's good. It's frustrating for sure. Yeah. And I assume that they already get a lot of these reports, right? So I can imagine that they use it as some sort of flagging system where if they get a lot of those messages, they might take a closer look.
Starting point is 00:14:47 But in general, Amazon is pretty aggressive in deleting fake reviews, individual fake reviews, and also in deleting individual accounts. Well, it's fascinating research. Thank you very much for your time. I appreciate it. Yeah, thank you so much for having me. Dr. Heiss Uvergur is an assistant professor of marketing at the Saunders College of Business. He joined me from his office in Rochester, New York.

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.