Deadline: 2019-08-15 Award: $17,500 Open to: Everyone*
The development of user preference modeling has benefited E-commerce a lot, especially for online shopping recommender systems. Leveraging users’ behavior data and other auxiliary information, the platform is able to predict and recommend to users what they are most likely to be interested in.
However, the evolving more accurate and also more complex preference prediction model brings computation problem for systems with large-scale candidate set. Industrial recommender systems usually resort to efficient index structure to solve the problem. For example, the similarity search library FAISS is widely used in many cases along with the vector inner-product based user preference model. Under the efficiency constraint, if contestants in this competition can explore reasonable ways to combine the potential capacity of advanced user preference model and index structure for efficient candidate retrieval, all online recommender service providers can benefit from the mechanism innovation.