The recommendation algorithms themselves allow

[tanjimajuha20]

Since October 2023, major online platforms have been required to disclose how their recommendation algorithms work. Now you can ask the question "Why is this in my feed?" not rhetorically, but concretely, and study point by point what influences recommended products on marketplaces or songs on VK or Yandex.Music.

Fontanka found out what data of Internet users interests Russian companies and why the government demanded that this information be disclosed.

In October, VK, Yandex, australia whatsapp resource Ozon, Wildberries, Sbermarket, M.Video-Eldorado, ivi and Start published rules for using their recommendation algorithms on their resources — as required by the new law that came into force in early October. It applies to social networks, marketplaces and online cinemas. Now large online platforms must explain the basis on which their users' recommendation feeds are compiled.
the online service to change the list and order of data given to a specific user in response to his request. Now almost all sites use such algorithms, from search engines to the application of some small flower shop. As market players emphasized, this is not the plot of the next episode of "Black Mirror" - your data is already being collected, and here it is somehow for your benefit, so that the service is more convenient to use.

Fontanka has reviewed the list of data that companies use to compile their recommendation feed. Most of the items are formal and obvious, like likes and "interesting" and "not interesting" marks, but there are also some curious parameters that companies pay attention to for some reason.

Yandex now has a section called "Recommendation Rules," which lists information for each of the company's services, from Alice to Yandex.Music and Kinopoisk. Moreover, the list clearly states that the company "uses special tests and methods to combat bias" to ensure that "the algorithm does not give unfair advantage or discrimination to certain groups or content."

The search engine itself takes into account the user's search history, clicks on links in the results and the time of return to the search page, information about the user's device, his geolocation, cookies, the language of queries and pages in the search results. In Kinopoisk, recommendations are influenced by the search history, the ratings the user gives, and the addition of films to the collection.

With "Yandex.Go" it's more interesting. Both taxi and delivery take into account the geography of orders and trips, payment methods, the user's location and their tariff preferences. That is, if you often order "comfort", then "comfort" will be offered automatically. What's more interesting is that scooter riders are offered insurance. Yes, if you often ride scooters from "Yandex", then the service will offer "to insure your health and the health of others, as well as your smartphone in case it falls during the trip and gets damaged."