Neural networks select tracks – Kommersant FM
[ad_1]
Russian music streaming companies have become more active in mastering neural networks. Personal recommendation systems that reveal new tracks to listeners are used by the largest market players: Yandex Music, VK Music, MTS Music and HiFi streaming Sound. The latest Kommersant FM said that the development of the algorithm itself is considered not as important as working with data. Now the Sound service, according to its representatives, focuses on content analysis and user preferences.
Meanwhile, Yandex Music recently launched a new algorithm called “Unfamiliar.” It specializes in finding artists that are not in the user’s selections. After the update was released, 20% more new tracks from artists unknown to listeners were added to personal playlists, Alexander Safronov, head of recommendations at Yandex Music, told Kommersant FM:
“First of all, “My Wave” looks at the behavior of the users themselves: what tracks they listen to, which ones they skip, where they like or dislike. We also take a look inside the tracks themselves. We have special neural networks that analyze the sound and build an audio vector of the composition. Tracks with similar sounds have similar vectors, and this way we can find similar melodies.
As for the future, I believe in progress and that recommender systems will continue to develop. Probably the most interesting direction here is related to deep neural networks.
Last year, for example, we introduced such models into “My Wave”. With their help, we have learned to better predict the development of a user’s musical tastes. This is very similar to modern language models, like YandexGPT, but only in the context of music.”
According to analytics company Luminate, 120,000 new tracks were released on music platforms every day in 2023, and the number of streaming services is growing by 23% every year. In such conditions, algorithms with artificial intelligence become a necessary conductor of content on streaming, said Dmitry Konnov, executive director of Zvonko Digital. The rest of the music industry, he says, has to get used to the new rules of the game:
“These algorithms are a very important element in attracting and retaining audiences. Of course, we, as music suppliers, would still like to understand how they mix this matter, why sometimes this or that release, which, for example, a record company or an aggregator has made some kind of priority, is completely ignored by the algorithm.
But on the other hand, we understand that this is really know-how that can be kept secret and protected, so we will probably also need to focus more and more on algorithms than on people called A&R managers. As a result of their activities, you, in fact, listen to 90% of the tracks that are included in the Yandex, VK and even Spotify charts.”
AI recommendation systems are actively used not only in the music industry. In the future, large companies with their help will be able to impose goods and services on consumers without much effort and extra expenses on advertising, believes Alexander Serbul, an expert on neural networks at the Bitrix24 service: “Neural networks will teach you so much that people will become dependent. You see, in essence, the human brain is being hacked. You will be offered a sequence of goods that is not logically connected, but it is perceived in the person’s mind in such a way that he cannot refuse and will continue to buy them. These are chains of control.
For example, a person is interested in horror films, or he will be interested in the lectures of Sergei Kapitsa. Nobody understands why, but it is so. And then, it turns out, neural networks will begin to offer something not related to either Kapitsa or horror films. We have opened Pandora’s box. This has already been proven that people will actually buy more. The user is offered a chain, he has never even thought about this connection, but they showed it to him, and then he cannot fall asleep thinking about it and, for example, buys a smart speaker. All this can lead to super profits, they will make money on people. In the worst case, they will start recommending some harmful products for purchase.”
Spotify was the first streaming service to use a recommendation system using neural networks. To do this, back in 2014, the company bought a specialized startup Echo Nest for $100 million. The platform, for example, has learned to identify individual musical elements in songs and take them into account when creating selections of tracks for users.
Everything is clear with us – Telegram channel “Kommersant FM”.
[ad_2]
Source link
