Hinge is an interesting example of a recommendation engine. I believe (although I donโt have hard evidence to back this up) that prior to 2018, Hinge was using a user-based collaborative filtering engine to provide matches. However, in 2018 they shifted their algorithm to the Gale-Shapley algorithm, which is a solution to the famous stable marriage problem. I find this to be an interesting and novel approach to a difficult recommendation problem, as it seeks not to maximize the happiness of any one person but rather to maximize the happiness of the entire group of of users. This approach to recommendation is really interesting and I think worthy of further analysis and experimentation.
Attacks on recommenders and false reviews are an unfortunate but rather common problem in this day and age. The most common example is Amazon reviews: many companies employ fake reviews to artificially boost the ratings of their products. In fact, this practice has become so widespread that there is even a website dedicated to analyzing how fake the reviews for any given Amazon product are (www.fakespot.com).
There are multiple potential ways to combat this recommender attacks or fake reviews. One way would be to look for user patterns and try to identify fake review bots used to leave positive reviews. One could even train a machine learning algorithm to identify fake users if there were a labeled dataset of fake and real users. Another approach would be to weight user ratings by the number of reviews they have left, which would potentially solve the issues of attacks like the Promise. If a user joins a website to leave a single review, this would not be highly weighted and would only slightly affect the overall rating.
Neither of these solutions are perfect or will fully solve the problems faced by recommendation engines, but they are a good first step towards creating a robust solution.