If you shop on Amazon, you ABSOLUTELY need to read this!

First of all, sorry for the clickbait-y title. But as a good corporate citizen I wanted to inform you about a potential widespread scam on Amazon that they should have ideally taken care of. I was a tad bit disappointed that issues like these still exist on #1 eCommerce website, in this day and age when AI & ML are making strides.

If you're in a hurry, there is a TLDR at the end. Lets get started.

Context: Yesterday(12/23) my earbuds stopped working and I decided to order one from Amazon. Its like second nature now ; I don't even waste time going around on other websites. So I just went to amazon.com and searched for 'wired earbuds' To my surprise, I noticed a lot of products FEATURED on first page had 5-star reviews. Here's a screenshot of few

As a Data Scientist, it is my duty not to get bamboozled into buying a product which has fake rating/fake reviews (or else shame on me!). Things like that start standing out when you deal with data everyday. So I decided to check a bit more details of one of the products.

A couple of things to notice here (as of Dec 23):

• All 482 reviews seem to have 5-star rating (say What now??).
• All 482 reviews seem to be posted on one single day – December 21st

Now here's the kicker. At the time of writing this article today, I just went back to check how our 'dear' product was doing today(Dec 24). I didn't even consider this would be a thing. And I was shell shocked AGAIN!!

Notice anything different?

• The product has now become an Amazon Choice overnight!!
• Now there are 335 reviews instead of 482(as of yesterday)
• They deleted all the previous fake reviews & posted new ones today. All of them on December 24. And at this point it's not even a surprise, all these are 5-star ratings.

So not only they create fake bot reviews, but these sellers also delete everything & post new ones to fake activity. I guess that leads Amazon to believe the product is doing real good & should essentially be made 'Amazon Choice'. I hope this hypothesis is wrong ; but with the screenshots as proof I'd say that possibility is bleak.

Back to my yesterday's analysis. Clearly there was something fishy here. Out of curiosity I opened one Amazon profile of one of the reviewers and what I noticed left me stunned. Have a look at the profile screenshot below. 'Britanny' seems to have made 115 reviews - 114 of those were 5-star reviews posted on December 21st. Moreover nearly 99% products that she reviewed had overall rating of 5. If Utopia is real, she's living in it.This is when I concluded that these were fake reviews generated by a bot just so that the product could end up being FEATURED on FIRST PAGE results. Now all I wanted to do is to understand the extent of this bot culture.

If you haven't heard of ReviewMeta, it is a website that tells if the ratings/reviews on an Amazon product are DECEPTIVE or REAL. They essentially use NLP along with reviewer based features. Coding up spam classifiers/deception detectors is not that hard ; but as a Data Scientist you should definitely avoid re-inventing the wheel. If there is a good reliable resource, use it. Saves you a lot of time. Just as a callout, Fakespot is another such website.

I quickly looked up this aforementioned product on ReviewMeta and voila!! There it was, in all its glory, a product that was FAKE (non-organic) FEATURED on first page.

Now all I had to do was find out the extent of bot proliferation on Amazon's first page of featured results. I wrote a quick scraper that extracted product information for the 34 'wireless earbuds' that were first page featured on doing a basic Amazon search. All default settings - like I usually shop.

There after for all the product results returned on first page, I looked up the respective rating from ReviewMeta.com and here's the analysis. I spent maybe 10 minutes on this using Tableau so please pardon the rough charts. Zero polishing/touch-up. Just wanted to get the sentiment across.

And just so we're clear, definition of FAKE here implies that a product was inorganically being featured on the first page.

Nearly 45 % products - a staggering 15 out of 34 products featured turned out to be FAKE. One more analysis I wanted to do is finding the Number of comments based on the featured rank for our FAKE '5-star' products. As you can see below, 3 of such products (ranked 3rd,8th & 9th) ACTUALLY made it to TOP 10 featured results. As a consumer who swears by Amazon's customer service quality, needless to say I was shocked & disappointed!

The Github link is here. Please note that it is full of Spaghetti code, but should be self-explanatory. Yesterday was the first Sunday of winter holidays after all :) As next steps, I think it would be worthwhile to run this script everyday & notice the changes. As mentioned above, I suspect these sellers delete & recreate new fake reviews.

Conclusion

I'd like to report this to Amazon but I feel the sample space for this analysis was quite low. So is the compute power on my laptop. Maybe I'll have to write a multiprocessing scraper to see if this phenomenon prevails for other products as well. Happy to coordinate with other Data Scientists on how we can improve the credibility of such analysis.

TLDR : If this post (or may I say Essay ;)) has been a snoozefest, I dearly apologize. One thing I'd like to urge to all the readers is to GET the REVIEWMETA browser add-on ASAP. It will show you adjusted rating as soon as you open any Amazon product link. And full disclaimer - I'm in no way affiliated to RevieMeta ; they're doing God's work (..and partly Amazon's work tbh). You gotta give credit where it's due!

Any comments appreciated. Just hoping that I don't get berated from Amazonians :)

Thank you!!