Improved advertisement putting method and system based on field matrix factorization machine

Abstract

The invention requests to protect an improved advertisement putting method based on a field matrix factorization machine. The method comprises the following steps: collecting advertisement data, preprocessing the advertisement data, filling or deleting null value or useless data, carrying out bucket separation processing on continuous features to obtain discrete features, carrying out one-hot coding processing, and planning a training set and a test set; the preprocessed data set is input into an embedding layer, the embedding layer is a layer of full-connection neural network, and high-dimensional sparse one-hot features are converted into fixed-length low-dimensional dense feature vectors; and respectively inputting the output part of the embedded layer into an FmFM layer and a DNN layer to carry out low-order feature crossing and high-order feature crossing, adding the output of the FmFM layer and the output of the DNN layer, obtaining an output value through a sigmoid function, carrying out advertisement putting according to the output value, and carrying out evaluation verification on the model by utilizing a test set. According to the method, the FmFM model is used for improvement, a simpler and more convenient click rate estimation model is obtained, and meanwhile, a better result can be obtained only through fewer parameters.

Description

technical field [0001] The invention belongs to the field of recommendation systems, and specifically relates to the field of click-through rate prediction and the field of deep learning for obtaining dense matrices from coefficient matrices. Background technique [0002] Now we are in an era of information explosion. How to find the information you want in the massive amount of information is extremely difficult for ordinary users. For merchants, placing successful advertisements is an extremely important issue. Advertisement Good advertising can bring huge profits to merchants. On the contrary, high advertising costs will cause merchants to suffer huge losses. For similar problems, click-through rate prediction is an important research work in recommender systems. [0003] Click through rate (CTR) estimation is one of the classic problems in recommender systems. Click-through rate estimation is mainly to find out the items that users are most likely to click and sort them...

AI Technical Summary

Problems solved by technology

[0004] However, these models have some problems. One is that the interaction between features does not consider the relationship between feature domains, and the other is that feature combinations do not consider the relationship between feature combinations and feature combinations.

Images