Why would you use NLP to collect reviews that are astroturfed? Garbage in, garbage out

Just go wirecutter style and actually research the products a minimal amount to give an opinion

This is what a good VPN roundup looks like.

https://www.privacyguides.org/providers/vpn/

This "RT-style" looks exactly how I would expect a "machine written sourced from affiliate blogs" review to look like.

Wirecutter is a great resource, and we'll soon include their lists for our analysis. But most people visit 5 different websites and spend many hours doing research before they buy a product. Our goal is to build a service that gives you a holistic view about what the internet thinks of products.

Again, garbage in, garbage out - how is rampant astroturfing and review fraud taken into account?

Edit: the site highly ranks the ones I see pushed on YouTube, so I’m guessing this is just “how often is this vendor referenced” and not “how quality is the vendor”.

Agreed, aggregated reviews are not helpful when you can't trust the source. For example, I just looked at the NordVPN page and all three of the highlighted YouTube reviews are pushing the NordVPN affiliate link in their description. Can I really trust these YouTube reviews when the creators are getting paid for each referral?

The way I see this going is that a bunch of people are writing VPN reviews because these companies pay high commissions. This site is going to aggregate these untrustworthy reviews, and also try to earn a commission in the future. As the end user, I don't see the value here, because I don't trust anyone in the process.

Fake reviews are certainly a big challenge, even for companies like fakespot that try to detect them.

What we plan to do:

- Running statistical analysis to spot discrepancies and contradictions

- Weight reviews based on the trustworthiness of their sources. E.g. Reddit, HN or Wirecutter are known to be very trustworthy sources.

Disagree that specific forums are more trustworthy than others. To me, a trustworthy source is a known entity that tests themselves and keep a track record of accurate reviews. For specific forums, keeping track of users that post honest reviews would be a good way to start.

> Fake reviews are certainly a big challenge, even for companies like fakespot that try to detect them.

Not that big: If you cut away all 5 and 1 star reviews you'll have stripped roughly 99% of astroturfing (for and against a company).

How do you figure either HN or Reddit (anonymous, unverified accounts) are trustworthy when there’s money to be made by astroturfing? The incentives are stacked against you.

I dare say that Wirecutter used to be a great resource before NYT bought them... not so much these days, it's mostly a pay-to-play machine. https://www.nytimes.com/wirecutter/our-response-to-nextdesk/

Don't even get me started with Wine Spectator and wines in general... https://freakonomics.com/podcast/season-11-episode-6/

Like other have pointed out trust is difficult to obtain these days for review sites but I hope you persist and have a successful product.

How do you propose someone actually review VPN services? All of them work, have many countries available as egress points, and are fast enough.

The only actual differentiation is in attributes that are impossible to measure from the outside: Do they log accesses? Make logs available to governments? Sell it to data brokers for advertising purposes?

That's pretty much the only difference between good and bad providers, and there's no way to tell, so in-depth reviews are an impossible task.

One could include honesty in marketing, that seems like a pretty clear thing that this industry would need and that few providers actually do.

> Using the newest NLP (GPT-3, BERT) technologies

Maybe remember your audience here and drop marketing bullshit like that. If you're using specific techniques, great, tell us. But don't use hand wavy marketing buzzwords.

Good input, fixed it.