3 ms·
One way to do decentralised relatively spam-resistant reviews: * Allow anyone to review. * When viewing other people's reviews, you rank them by how similarly
by londons_explore 4y ago
One way to do decentralised relatively spam-resistant reviews:
* Allow anyone to review.
* When viewing other people's reviews, you rank them by how similarly they review places you and they have both been to. Ie. Do you agree with them? If you agree with them, then weight that users reviews higher. If they rank places highly that you think got 1 star, then rank them negatively.
* This approach can be done on the complete review graph, so you don't need to have reviewed the same places someone else ranked.
* The downside is you can't see any meaningful reviews till you write a few yourself.
- sideshowb 4y agoSo an army of bots can create fake reviews for my business and it will look legit so long as they also give consensus reviews for a bunch of well known places?
- londons_explore 4y agoYep. But as soon as a few real users find your place and rank it highly, then the bot army will only impact scores for users who haven't yet come across such bot scores. Obviously any such scheme is helped by there being not-too-many bot accounts. It's hard to limit bot account creation in a decentralized opensource design. One potentially workable approach is to have users leave a trail of data in some immutable form (eg. a series of reviews left, with scores, timestamps, and text comments), which you later apply some heuristic over (eg. 'all comments containing a web link to a url shortener are probably spam'). All data that is older than the publication of the heuristic is then fairly trustworthy, since bot account owners can't retrospectively edit their spam to make it not match the heuristic.
- londons_explore 4y agoAnother approach: Have each reviewable location publish a key on a sign in their establishment. Make reviewers scan a QR code and sign their comment with that key. That ensures that while the key doesn't leak every review must be written by someone who has physically visited. Obviously you need to trust every physical visitor to not publish the sign, but I suspect that might be sufficient to stop spam nonetheless.