A new artificial intelligence model developed by researchers at Skoltech and Sberbank’s Artificial Intelligence Laboratory and the Risks Department helps better predict customers’ shopping carts. This allows using it for more accurate recommendations.

The paper — “Label Attention Network for Temporal Sets Prediction: You Were Looking at a Wrong Self-Attention” — was published in the proceedings of the 27th European Conference on Artificial Intelligence, one of the leading conferences on artificial intelligence in Eurasia.

Current approaches are not very effective, as they misinterpret the history of customer behavior. The authors of the article have developed an alternative method. At first, all available information is collected, focusing on the time of events and the relationship between them. Then, this data is transmitted for further processing. Thanks to this approach, the system learns faster and more accurately recognizes the dependencies between different aspects of events.

The proposed LANET model anticipates the actions of customers and companies, which historical data can be visualized as a sequence of grocery baskets. Accurate forecasting of future behavior allows a business to make the right decisions at the moment, reducing the risk of losing a customer. The research holds great practical significance for companies that work in recommendation systems and scientists conducting research in this promising field.

The effectiveness of the proposed model has been tested on various datasets. The experimental results demonstrated that the LANET mode had a significant advantage — the relative increase in forecasting accuracy compared to the best-known approaches in some cases reached 65%. The scientists also analyzed the influence of individual components of the model on the final result and investigated the cause-and-effect relationships between events. The model is publicly available, which opens up wide opportunities for its use in the recommendation systems market.