Marketing activity prediction method based on GBDT and DL fusion model

Abstract

The invention discloses a marketing activity prediction method based on a GBDT and DL fusion model. The method comprises the steps of data preprocessing, semi-supervised positive and negative sample division based on spy technology, prediction model establishment, marketing activity prediction and the like. The method comprises the following steps: firstly, only distinguishing samples in advertisement putting original data into click users, namely positive samples and unmarked users; selecting M% of the positive sample data set, and putting M% of the positive sample data set into an unmarked user data set; and then, carrying out calculation through an iterative EM algorithm. The training device is a LightGBM and DNN fusion model, dense numerical value features are input into the LightGBM, 0 / 1 features are obtained through splitting of a tree model, the 0 / 1 features and category features are input into a neural network together for learning, and a final dichotomy learning device for predicting user clicks is obtained. The result shows that the method not only effectively utilizes the advantages of the gradient boosting decision tree in the aspect of feature construction, but also has the learning ability of deep learning for high-order features in a high-dimensional sparse matrix, and significantly improves the accuracy of user click behavior prediction.

Description

technical field [0001] The present invention relates to the technical field of artificial intelligence in Internet marketing, and more specifically, relates to a marketing prediction method based on a fusion model of GBDT and DL. Background technique [0002] Click-Through Rate (CTR) estimation is a key link in Internet computing advertising; the accuracy of user estimation directly affects the company's marketing advertising revenue. Since the click-through rate is a typical binary classification (that is, click or not click), the classic algorithm of CTR is logistic regression (Logistic Regression, LR for short). [0003] LR is a generalized linear model that maps input values ​​to the [0,1] interval through Logit transformation. The LR algorithm is suitable for parallel computing, but due to the limitations of the linear model's own algorithm, its learning ability for data is limited. In particular, the input data in CTR is generally a high-dimensional sparse matrix for...

Images