Method for recommending target object, computing device and computer storage medium

Abstract

The invention relates to a method for recommending a target object, computing equipment and a computer storage medium. The method comprises the steps of generating input data, wherein the input data at least comprises click features, browsing features, purchase features and target object features of a user for a target object; averagely segmenting the input data into a plurality of sub-input data;predicting a click probability of the user through the first neural network model; predicting, via a second neural network model, a probability of transformation with respect to the target object, the second neural network model and the first neural network model sharing at least an embedding layer; and predicting a recommendation probability about the target object based on the click probabilityof the user, the transition probability about the target object and the third loss function. According to the method and the apparatus, the defect of overfitting caused by sparse sampled user behavior characteristics can be avoided, and optimization of multiple targets associated with the recommended target object is considered at the same time.

Description

technical field [0001] The present disclosure relates generally to machine learning, and in particular, to methods, computing devices, and computer storage media for recommending target objects. Background technique [0002] In the traditional recommendation target object scheme, for example, in a recommendation system for recommending products or content, the product or content is usually recommended based on the estimated click rate ctr. However, in the actual use of the ctr prediction model, the data to be predicted deviates from the training data, and user behavior features are sparse, making the generalization ability of the ctr prediction model challenging and prone to overfitting; in addition, the estimated conversion rate ctr To a certain extent, it reflects the "behavioral relationship" between the user's click preference and browsing preference. Only based on ctr evaluation, the optimal evaluation cannot essentially solve the multi-objective optimization problems o...

AI Technical Summary

Problems solved by technology

However, in the actual use of the ctr prediction model, the data to be predicted deviates from the training data, and user behavior features are sparse, making the generalization ability of the ctr prediction model challenging and prone to overfitting; in addition, the estimated conversion rate ctr To a certain extent, it reflects the "behavioral relationship" between the user's click preference and browsing preference. Only based on ctr evaluation, the optimal evaluation cannot substantially solve the multi-objective optimization problems of the e-commerce platform, such as the optimal commodity transaction and the optimal user experience.

[0003] In summary, the traditional recommendation target object scheme is difficult to solve the shortcomings of overfitting caused by the sparse sampling of user behavior features, and cannot simultaneously optimize multiple goals associated with the recommended target object

Images