Method for reducing noise label influence in click rate prediction based on interactive learning

• Main Authors: PAN XINFA, XIANG LIANG
• Format: Patent
• Language: Chinese; English
• Published: 01.11.2022
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; DATA PROCESSING SYSTEMS OR METHODS, SPECIALLY ADAPTED FORADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL, SUPERVISORYOR FORECASTING PURPOSES; HANDLING RECORD CARRIERS; PHYSICS; PRESENTATION OF DATA; RECOGNITION OF DATA; RECORD CARRIERS; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE,COMMERCIAL, FINANCIAL, MANAGERIAL, SUPERVISORY OR FORECASTINGPURPOSES, NOT OTHERWISE PROVIDED FOR

The invention discloses a method for reducing noise label influence in click rate prediction based on interactive learning. The method comprises a data preprocessing step, a training set generation step, a prediction model establishment step and a marketing activity prediction step. In the prediction model establishment step, N small batches of batch data (D1, D2... DN) are sequentially selected to train the deep learning network model A and the deep learning network model B. In each batch of data, the deep learning network model A or the deep learning network model B respectively selects a sample with small loss as a useful knowledge sample. A useful knowledge sample is taught to another deep learning network model for training; and weighting and adding the parameters of the deep learning network model A and the deep learning network model B, and transmitting to the deep learning network model C to complete training of residual data to obtain a final classification deep learning network model. According to t