Automatic integrated architecture search system and method for click rate prediction model

• Main Authors: WANG XIN, ZHU WENWU, WEI ZHIKUN
• Format: Patent
• Language: Chinese; English
• Published: 21.01.2022
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; PHYSICS

The invention provides an automatic integrated architecture search system for a click rate prediction model, and the system comprises: a feature embedding module search space, which is used for searching the dimensions of embedding vectors of features, and finding a suitable embedding vector dimension for each feature; a feature interaction module search space, which is used for searching a feature interaction sub-network according to the feature combination and searching a feature interaction combination at the same time to obtain a feature interaction result; and a deep network module search space, which is used for incorporating the feature interaction result into an implicit high-order interaction search space so as to carry out implicit high-order interaction in deep networks with different layer numbers to obtain a click rate prediction value. According to the invention, automatic integration of network design of the click rate prediction model is realized, a model architecture with better performance i