In the music streaming industry, search engine is a key component to explore the catalog and discover new content. Users can use the search engine to achieve different goals such as listening to music, adding to favorite or playlist, sharing, fact checking… Although users use search with a specific idea in mind, they can also use search to explore new content. In this case, we speak of an exploring mindset. 

The scientific literature has focused on identifying such mindsets. For instance, if the user types “let it be the beatles”, we can deduce that the user has a specific idea in mind and wants to click/play/collect the famous song from The Beatles. Whereas queries like "sad rock song from the 80s" or "song for a road trip in Italy" are often associated with non-focused queries. Regarding this last mindset, search engines based on a classical word matching often fails. Indeed, when one searches for "80s rock", one probably doesn't want a track called rock 80, but a track released in the 80s linked with the rock genre. 

The objective of this internship will be to study how language models can be used to retrieve the appropriate music content (track, artist, playlist..) to our users, then to apply recent techniques such as semantic search, generative AI to a production-like environment. 

The internship will involve an in-depth literature review of the existing approaches, as well as an analysis of the most relevant strategies to adopt for a music search engine which differs from other search engines. We are open to explore and study a wide range of methods, such as Retrieval augmented generation (RAG) and Query Expansion. The intern will also have the opportunity to implement the selected methods on actual data extracted from Deezer service.

The intern will be supervised by one Machine Learning engineer from the search team, who will provide scientific and technical guidance throughout the internship. The intern is nonetheless welcome to propose solutions and work autonomously. For experiments, Deezer ensures access to internal data, cutting-edge technology, and appropriate calculus power. Several previous research interns from our team have tested their algorithms in production, and/or have published results from their work as scientific articles.

• Master / PhD student  with a strong background in machine learning and natural language processing
• Good programming skills and knowledge of machine learning tools (such as scientific Python, Pytorch, elasticsearch etc)
• Knowledge and experience in LLM models and associated techniques, as well as the tools and frameworks used to implement them (LangChain, Vector store, RLHF, …) would be a plus.
• Knowledge and experience with cloud environment would be a plus
• Good data processing and analysis skills
• Creativity and autonomy