When I talk to people about the upcoming Futourist app, the most common response is “Oh, that’s so cool! But how will you prevent frauds and fake reviews?”. The answer to this one is actually quite logical, but still complex. I’ll do my best to sum up everything in the paragraphs below.
The general challenge is that there are many ways to ‘game’ most systems, and, if unconstrained, this ‘gaming’ may only be limited by the abusers’ imaginations and the computational power they wield (e.g. the use of botnets etc.).
Recognizing this, Futourist accept that this is a ‘cat-and-mouse’ game, and effective solutions involve making sure Futourist maintains the upper hand, to ensure that it is not profitable to abuse the system.
How Futourist Itself Mitigates Abuse
A core feature of Futourist protocol mitigates the effect of abuse. This feature is the indexed Pareto distribution model whereby the amount of funds distributed is allocated according to the index position in a sorted list of most influential reviews, and this index position maps into a ‘bucket’ in a Pareto distribution which determines the reward.
This creates two effects that mitigate abuse:
- There is a hard cap on the maximum proportion of funds distributed. If compared to a more direct distribution model, e.g. funds distributed according to % influence, there is no way a single abusive user account can take the vast majority of funds, which reduces maximum potential profits from abuse.
- The distribution can be divided into two parts, a set of a few ‘big winners’, followed by a long tail of ‘also rans’. For the big winners, because of the indexing model, to raise their winnings, they have to move up a step, and this incurs significant cost at this part of the distribution. For the ‘also rans’, the difference in rewards is comparatively much less on each step up the index, so the cost of abusive practices outweigh the expected minor difference in rewards, therefore it is more profitable for them to simply improve their reviews.
As an example, consider the case where the most influential reviewer A has an impact, of say, 1000 points, with a second reviewer, B, has 150 points. At index position 1 and 2, perhaps A is in line to receive $2000 and B gets $1000. (This ratio disparity might initially seem unfair or inefficient but when taking into account the network effects, whereby more popular reviews propagate exponentially, and the fact that all participants understand and accept the rules, we can say it is a fair allocation). For B to move to position 1 within a given amount of time, takes really significant effort, with no guarantee of success, as:
- network effects will also tend to move A’s influence up in the same amount of time,
- it is difficult for B to gauge A’s actual impact points value if B can only see their position
Differentiating Abuse From Unexpected ‘Winning Strategies’
Nevertheless, B may still engage in ‘gaming’ the system. Note here the use of the word ‘gaming’ rather than ‘abuse’. This is because B may also indulge in practices that may be seen not so much as abuse, but as B developing some kind of ‘winning strategy’. For a familiar example, let’s take a Twitter-based strategy. B posts links to their review on many Twitter feeds, resulting in many Twitter users reading the reviews and rating them, and likely some of them sharing. If effective, B’s influence increases perhaps 10-fold, resulting in B moving to position 1. Now, B may be seen to be an abuser, but also can be seen as someone with a clever (albeit perhaps obvious) way of increasing the visibility and influence of their reviews, and therefore increasing exposure of contest partners. So it may well be that such behaviour is desirable within the system, and B should be rewarded more than A.
Use Of Auxiliary Protection Schemes
In a real abusive case, B might employ a botnet to rate their reviews highly, for example. In this case, Futourist will employ auxiliary protection schemes that look at the reputations and identity of users, i.e. verify users and their activities. E.g. did they actually stay at a resort? Does their reviewing/rating follow a suspiciously predictable pattern? Real abusive users will simply be identified and banned, i.e. receive zero tokens from their activities. To remain fair, there will be appeal procedures.
Difficult Abuse Cases
Beyond that, Futourist faces the real ‘conspiratorial’ abuse cases. In these cases, a standard solution is to employ unsupervised machine learning systems which ultimately report to human users.
What happens generally is this: take as much data as possible, and feed it to an unsupervised machine learning system which looks for patterns and categorizes data according to similarity to other, previously seen patterns. This can include user ratings, review text, images (are images copied from other places on the internet, for example?).
In a fraud case, there should always something in the data that looks unusual, for example, high similarity in word patterns between review texts, or a single reviewer being given the highest rating by 10000 new accounts which all signed up in the last 15 minutes and which never rated anything before or after.
These are obvious cases, but, with enough previous data, even small shifts in patterns can ring alarm bells and be investigated.
To support this, Futourist will implement APIs within the OpenHours system to transfer data to in-house or external backend systems. This provides a highly flexible way to allow any kind of analysis, including the application of whatever standard or cutting-edge algorithms as necessary, including the possibility of outsourcing work to external companies with proprietary algorithms or access to additional data sources.
Validation of Futourist’s Abuse mitigation system
Futourist team will research and formulate an initial set of likely abuse scenarios and use them to evaluate inhouse or external abuse mitigation backends. This testing will be an ongoing process.
Video as the primary review type
Going away from machine learning and other heavy technical stuff to a more down to earth topic. Futourist will focus on video reviews. The rewards for video reviews will be the highest, followed by photo and text. We believe video is the single most convenient way to consume information, but what is important here is that they are very, very hard to fake. Video is trustworthy, text is not, it’s as simple as that.
This means a non-existing fake restaurant could never become one of the top restaurants on Futourist. This exact thing happened not long ago on the currently most popular travel review site (see the amusing Forbes article).
On Futourist, it will be very hard to convince the community the restaurant is real if no videos are present. If there are videos, they might be videos of various restaurants and the videos wouldn’t match. The community will catch the fake restaurant and report it (and be rewarded for reporting it!). Even before that, the community would downvote the hell out of videos that seemed even slightly fake. And even if the curators and the entire community all rebelled against Futourist, we would still catch such a fake place in our abuse mitigation system described in the paragraphs above.
To protect our users against diverse abuse scenarios, Futourist will deploy multiple layers of security and protection based on:
- the Futourist protocol itself,
- auxiliary protection systems including the existing OpenHours architecture and infrastructure,
- unsupervised machine learning to identify new abuse patterns,
- possible use of third-party systems,
- promotion of video reviews,
- ultimately, human involvement and decision-making.
Futourist team will investigate new cases and make decisions on how to handle them in future, and report these improvements to partners or users. This increases the community value by letting users know Futourist is protecting them, as well as letting real abusers know that the Futourist system and team are actively working to defeat them, so they are likely not to even try.
This architecture and business model adds value in two ways; firstly by mitigating fraud, but secondarily, it identifies interesting new business models that users have devised, and which may be advantageous for Futourist to adopt or support.
In case you have any further questions, do not hesitate to contact us through our Telegram group.
Join us on our Social Channels: