My Wave: Yandex Music Adds a Setting for Recommending Kazakh-Language Tracks

Yandex Music has introduced a new update to its smart recommendation system, My Wave — now users can choose tracks exclusively in Kazakh or, for example, select songs without vocals or in a foreign language. Thanks to this, Kazakh musicians will be able to find their audience even faster, and listeners can customize their music selection to suit their preferences.

We learned how the recommendation system works and adapts from Vladimir Gorbunov, General Director of Yandex Plus in Central Asia and the Caucasus.

Vladimir Gorbunov, General Director of Yandex Plus in Central Asia and the Caucasus

What is My Wave

My Wave is the next step in the development of our music recommendations. First, dynamic recommendations appeared with “Yandex Radio,” followed by smart playlists that updated once a day. This was convenient, but over time it became clear that users wanted something more dynamic. Thus, we came up with the idea of creating a music stream that adapts to your preferences in real time.

At the core of My Wave are deep neural models that analyze listening history, track sequences, and other factors. This allows the system not only to respond to your current tastes but also to anticipate where they might evolve, offering music that resonates specifically with you.

Introducing personalized recommendations, such as My Wave, is a key element of our strategy aimed at creating a more personalized and user-friendly experience. Our goal is to make Yandex Plus accessible to every user by providing content that matches their interests and language preferences.

The feature that allows you to customize My Wave for Kazakh-language music is an important step in supporting local content. We understand that a significant part of the audience in Kazakhstan prefers content in Kazakh. At the same time, a large amount of local content is created in Russian and bilingual tracks in Kazakhstan, which is also taken into account in recommendations. This feature helps users find a balance between their cultural preferences and interests.

My Wave not only strengthens the connection with the Kazakh audience but also supports local artists by making their music accessible to a wide audience in a convenient digital format.

About the system and differences from other recommendation algorithms

There are two main groups of recommendation algorithms — collaborative filtering and content analysis. My Wave uses both approaches.

To put it simply, collaborative methods are based on finding users with similar musical preferences. For example, if other listeners whose tastes are similar to yours have added certain tracks to their playlists, there is a high probability that they will appeal to you as well.

Content analysis works differently. Special neural models analyze the audio components of tracks: rhythm, volume, presence of vocals, and many other parameters. These data are transformed into unique audio vectors, which are then compared with vectors of other tracks. This allows finding songs that are similar in sound, style, and performance.

Regarding the selection of Kazakh-language tracks, an important element is determining the language of the song lyrics. Algorithms automatically analyze the text to determine the language in which it is written. This helps us accurately recommend songs to listeners that match their preferences.

About processing feedback and the time it takes for recommendations to adjust to new preferences

One of the main advantages of My Wave is its high responsiveness. When a user likes or skips a track, the algorithm instantly processes these actions and updates recommendations in real time. My Wave adapts immediately, allowing users to receive the most suitable tracks on the spot.

About audio vectors

Audio vectors, as I mentioned earlier, are unique sound fingerprints. We create them using neural networks that analyze the track’s spectrogram, capturing its features. For example, if two tracks have a similar timbre or performance style, the audio model will identify them as similar, even if no one has listened to them before.

About the benefits of implementing Kazakh-language recommendations in My Wave

It is important for us that our users in Kazakhstan feel that the service takes into account their cultural and language preferences. Supporting Kazakh-language recommendations allows us to create a closer connection with the audience, emphasizing that we listen to their interests and strive to make the music that is important to them easily accessible. This also supports local musicians, whose work gets a broader platform for promotion. As a result, users see Yandex Music not just as a tool for listening to music, but as a platform adapted to their culture and interests.

We are also working on expanding the customization options for My Wave for users in other regions. In the future, we plan to add support for selecting any language for recommendations, including Russian, regardless of the region, to make the service even more universal and convenient for all users.

About the role of user behavior in improving recommendations

User behavior plays a key role in refining recommendations. We take into account both explicit feedback, such as likes and dislikes, and implicit feedback — actions that may indirectly indicate preferences. For example, how many tracks were played in a row, whether tracks were skipped, or at which exact point they were stopped. Even the frequency of service usage on certain days can be a signal for analysis.

It is important to note that we interpret implicit feedback cautiously: skipping a track does not always mean it was disliked — it could be related to mood or situation. It is precisely through the analysis of such details and many other factors that we can adapt recommendations so that they best match the user’s current preferences.

About the effect of launching My Wave with Kazakh-language recommendations

We see My Wave not just as a recommendation system, but as a whole tool for supporting music in Kazakh. We already have successful cases where artists increased their audience several times over, and we expect the same to happen with Kazakh-language musicians. My Wave responds to listener feedback, and if it is positive, it will recommend the track to a wider audience. This opens up new opportunities for artists.

Thanks to Yandex Music’s recommendation system, emerging Kazakh musicians are already rapidly gaining recognition and audience through the ISKRA Qazaqstan project. The sixth edition of the playlist has already taken place. The playlist helps young artists attract the attention of a wide audience through personalized recommendations available to more than 20 million active users of the service. With ISKRA Qazaqstan, some performers were able to increase their number of plays by 65 times. For example, plays for one of the artists from Iskra, Jazzdauren, grew from 24,000 to 5 million in three months.