There are many online services which attempt to recommend movies, books, webpages etc that someone will like. In some implementations, recommendations are based upon finding new items that are liked by other users: If you like A, and someone else likes A and B, then B is perhaps a good recommendation for you. The difficulty is that establishing your profile can be a tedious task, as you have to initially indicate several items that you like. On the order of twenty items, perhaps.

A new algorithm, developed by Evgeny Frolov and Ivan Oseledets of the Skolkovo Institute, provides a much less time consuming, and likely more accurate, way of establishing your preferences. It uses information about items that you do not like, as well as about items that you like. Roughly stated, if you do not like item B, and those who like B also like C, then item C is perhaps not a good suggestion for you.

The details of their algorithm are subtle, and designed for efficient operation. It is not just graph searching. See Arxiv 1607.04228 for their paper — linked below — and a press release also linked below.

Best wishes,
Ken Roberts
01-Aug-2016

http://arxiv.org/abs/1607.04228
Evgeny Frolov and Ivan Oseledets — Fifty Shades of Ratings: How to Benefit from a Negative Feedback in Top-N Recommendations Tasks

Press Release — Skoltech scientists have created an algorithm that improves the quality of recommender systems