Multi-model fusion user attribute prediction method

Abstract

The invention discloses a user attribute prediction method based on Stacking multi-model fusion. The method comprises the steps of data collection, feature engineering, model training, cross validation and precision evaluation. Acquiring an advertisement click browsing record of a user in data collection, and cleaning and segmenting the data; in the feature engineering, using a feature correlation thermodynamic diagram to display the correlation among the features, using an XGBoost algorithm to obtain the importance ranking of the features, and screening the features in combination with the correlation and the importance ranking of the features; in the model training process, logistic regression, a random forest, a limit tree and an XGBoost algorithm are used as a first layer of a Stacking model, and LightGBM is used as a second layer of the Stacking model to train features; a five-fold cross validation mode is used during cross validation; in the precision evaluation process, the prediction result is evaluated by using the accuracy rate, the recall rate, the F1 value and the precision rate; according to the invention, the gender and age of the advertisement user can be predicted.

Description

technical field [0001] The invention belongs to the field of machine learning, and relates to a multi-model fusion user attribute prediction method. Background technique [0002] With the rapid development of network technology, online advertising has become one of the main ways for businesses to publicize; publishing advertisements on the Internet has a faster dissemination speed and wider dissemination range, and is more efficient than offline advertising. Therefore, the Internet is full of All kinds of advertisements; in advertising targeting, the user's search content, browsing history and basic attributes play an important role, among which the basic attributes gender and age are very important, but not all users are willing to disclose their age and gender information , so the user's basic attribute data will be missing, so it is necessary to use existing data and related algorithms to make predictions. [0003] At present, the existing research mainly adopts traditio...

AI Technical Summary

Problems solved by technology

[0003] At present, the existing research mainly adopts traditional machine learning methods such as SVM and Bayesian. As an important part of machine learning, integrated learning is gradually applied to the field of user attribute prediction; and other Internet behavior data, combined with Bayesian network algorithm, random forest, SVM and other single machine learning algorithms to predict the gender and age of users; some scholars also use Weibo users as research objects, based on user nicknames, tags, Weibo text, etc. Predict the user's gender and age, but the accuracy rate is only 73.6%; some prediction methods integrate LightGBM and FM algorithms, analyze the installation and usage of smartphone apps, and predict the basic attributes of users. The prediction accuracy of gender is 67.65%

[0004] To sum up, at present, the prediction of the gender and age of advertising users is still in its infancy, and most of them use algorithms commonly used in machine learning such as Naive Bayesian or Support Vector Machine, and the algorithm is relatively simple, resulting in poor prediction effect on attributes

Images