Online user experience can be greatly improved if users are presented with relevant content based on their actual intent, captured real time in their browsing session. GRU4Rec is an open-source deep learning based software that models the sequential information of content consumed in the session and recommends relevant next-best content accordingly.

About the Innovator

Gravity R&D is a personalization engine vendor, offering a product portfolio called Yusp, including multiple modules and filling multiple needs using the same underlying technology. Yusp has all primary product and marketing modules that comprise the full scope of digital personalization. Yusp never lost any A/B test against competitors thanks to its exceptional algorithmic portfolio. Our research team focuses to solve difficult challenges of online personalization - such as cold-start problem, intent-based recommendation - in order to keep and improve Yusp’s competitive advantages.

/futurium/en/file/gravity-rdjpggravity-rd.jpg

What is the innovation

GRU4Rec is a recurrent neural network adapted for session-based recommendations. Gated RNNs - such as the GRU - are generally good for dealing with sequences, but due to the nature of the recommendation domain we introduced ranking loss functions and an efficient negative sampling method building on session parallel mini-batches. The code is highly optimized to reduce the time of computations. With the follow-up work on improved sampling and loss functions, it is now fully production ready. Work for development of this technolgy was undertaken in the EU-funded CrowdRec project.

Out of the lab - Into the Market

We need to improve our product continuously. We select research topics based on the encountered challenges in our production services. If a research result is promising in the lab, we gradually introduce the technology in production, carefully selecting the right use cases, to test whether the offline measurements are also justified in real-world operation. We capitalize both by staying ahead in the fierce competition with newest technology, and offering better KPIs in our service, which bring us more revenue.

Benefits of participation in the Framework Programme

It both helped the company to maintain its thought leadership in recommender system technology, which also contributes to its valuation, and gives opportunities to collaborate with important industry players. The research team of Gravity, together with some other CrowdRec researchers, became key members of the deep learning research line of recommender systems. Our seminal GRU4Rec paper got many citations, we organized two very successful workshops, collocated with the main RecSys conference, and held also a tutorial.