For those who don’t know, Mafengwo is a Chinese unicorn startup in travel industry, expected to be valued at $2.5 billion. It operates in the same space as TripAdvisor, Lonely Planet and travel tour agencies.
So what is it all about?
It Began with a WeChat Subscription Account Post
It all began with a post on WeChat’s subscription account called “小声比比” (Check out my post explaining WeChat subscription account if you want to know how it works).
It blew up on WeChat, with more than 100k+ pageviews (might be magnitudes higher, but this is the largest counter that WeChat provides), and almost 10k upvotes.
In the post, the authors detailed several allegations against Mafengwo:
1. Official Accounts on Mafengwo copying reviews from other websites
Mafengwo allegedly copied 18 million reviews from other websites, out of 21 millions reviews in total.
The list of websites that it allegedly copied from:
- Ctrip (a travel company in China)
- eLong (also a travel company in China)
- Meituan (on-demand delivery platform in China)
So how does the authors know that these reviews were copied, not genuine?
Basically they extracted a sample of the reviews on Mafengwo and cross-referenced other websites using exact word matching. Turned out that 85% of the reviews were fake.
This is supported by other evidences such as:
- inconsistent identity of user (different genders for different reviews)
- Google-translated reviews from Yelp
- Random, non-relevant information in reviews (algorithm mistakes from scrapping other websites)
- Majority of reviews being posted between 2015 and 2016, and the trend suddenly stopped
- Majority of reviews being posted during week days instead of weekends
- Majority of reviews being posted during working hours instead of non-working hours
2. Fake Accounts Winning lottery and luck draws
The post claimed that the same accounts which posted fake reviews were also actively winning lucky draws.
Chances are these accounts were either operated by internal employees or completely bogus.
Who Is Behind the Post and the Allegation?
Hooray.ai (乎睿数据) provided the technical support for extracting the data in the post.
From the bio in the post, they claim to be a team of graduates from UPenn, Cornell and other top universities, with experience working at Google, Facebook and other top companies.
They also claimed to have published in top AI and NLP journals such as AAI, NAACL and ACL.
How Did Mafengwo Respond?
It responded with a statement and a lawsuit (attached screenshots below).
The main points in the statement are:
- It has conducted checks on reviews and acted on the fake reviews after the accusation was made on 21 Oct.
- Reviews only account for 2.91% of all data in Mafengwo, other data are genuine.
- Before 2016, Mafengwo has encouraged reviews from users, however that has been de-prioritized since.
- Food and restaurant reviews are extracted from travel reviews, so it is not fair to compare with other food review sites.
- We have always been harsh against fake reviews and ads in reviews, they come from 3rd parties.
- We will pursue legal actions.
Also I was surprised to see that WeChat actually has a “dispute” feature in subscription accounts.
This allows for companies to file a “challenge” to posts, in cases of disputes like this one:
When you click on the post, this screen shows up before you can read the post, it basically says “this post is filed for dispute”, and you choose to read the post, or read the statement from the company that filed the dispute.
How Did the Original Accuser Respond Back?
It responded back with a follow-up post on WeChat subscription account.
This time it talked about how Mafengwo is cleaning up the fake accounts to “clear the evidence”. Note that these are official accounts by Mafengwo, not normal user accounts.
The post also gave more evidence of “fake comments” in the tour guides, using some analysis on frequency of certain sentences:
It is also worth noting that in the comment section of the follow-up post, the authors claimed that they have the original data as proof and these data have been “certified by notary public”.
I find this quite interesting as I didn’t know about this previously and I always wondered how you can prove that the content of a website at a particular point in time. Now I know how.
As this story is still developing and the verdict is not out yet, it is too early for a full conclusion.
I have no doubt that the data presented in the post are genuine. However, the legality and ethical issues with these fake reviews are still up to questioning. After all, they could be justified as an automated way of generating contents for growth hacking, and it is up to the websites that they scrap from to file challenges and complains.
We will just have to wait and see how this all unfolds in the the near future. And I will keep this post updated as well.
Meanwhile, feel free to checkout my other posts on tech in China.