Method for advertisement click rate prediction based on GRU neural network

Abstract

The invention discloses a method for advertisement click rate prediction based on a GRU neural network. The method constructs a time-series data based on advertisements when preprocessing data, , andenhances the integrity and the stability between the series data compared with the user series data; and then text features are digitally encoded by one-hot coding, and only the subscript with a bit value of 1 in the mapped value is taken as the mapped value of the character. In this way, the dimension of the feature is greatly reduced, and the training speed of the model is improved. According tothe method for advertisement click rate prediction based on the GRU neural network, the invention combines a regression algorithm with a deep learning algorithm to form a final prediction method. Firstly, the feature selection is performed with a ridge regression algorithm to reduce the interference of invalid features on model training; then training and predicting are carried out with the improved GRU neural networks based on LSTM neural network. The GRU prediction model can improve the prediction accuracy compared with a RNN prediction model, , and can quickly improve the prediction accuracy and can perform a faster model training compared with the LSTM prediction model.

Description

technical field [0001] The present invention relates to the technical field of search engine network advertisements, in particular to a method for predicting the click-through rate of advertisements based on the GRU neural network in deep learning Background technique [0002] With the rapid popularization and development of the Internet, the role of search engines is increasing day by day. According to the 41st "Statistical Report on China's Internet Development", it can be seen that the number of search engine users in China has reached 640 million. Advertising promotion based on search engines has become a very popular way of advertising promotion. [0003] For the prediction of the click-through rate of online advertisements, the historical click-through-rate value was used as the forecast value at the earliest stage to the predictive click-through rate pre-method based on traditional machine learning algorithms (such as logistic regression, decision tree and Bayesian a...

AI Technical Summary

Problems solved by technology

However, due to the instability of each user's behavior, it is difficult to construct complete and stable user sequence data, which greatly reduces the accuracy of pre-click rate prediction; in addition, the RNN model will have gradient bursts or vanishing problem, which also makes the model's predictions unstable

Images