Method for training media asset tag generation model, media asset tag generation method and device

Abstract

The embodiment of the application provides a kind of media asset label generation model training method, media asset label generation method and device, it is related to media asset recommendation technical field.The method comprises: receiving the user portrait data corresponding to multiple users, and historical dialogue data containing negative feedback operation;Wherein, the user portrait data includes user interest label, media asset search history and media asset viewing history;Based on the user portrait data and the historical dialogue data, a sample training set is constructed;Each training sample in the sample training set includes: sample input data and sample output data;Through the sample training set, a pre-training language model is trained to generate a target media asset label generation model, and the application can improve the recommendation accuracy and user satisfaction after the user gives negative feedback operation to the recommended result through the target media asset label generation model.

Images