Training method for application MOS model, device, and system

A training method for an application mean opinion score (MOS) model, wherein the method comprises:

• sending, by a central network data analytics function (C-NWDAF) entity, a first subscription request to an edge network data analytics function (E-NWDAF) entity, wherein the first subscription request is used to subscribe to a quality of service MOS level of a target service and a corresponding first network performance indicator, and the first network performance indicator is a network performance indicator of a transmission network that carries the target service;
• receiving, by the C-NWDAF entity, the quality of service MOS level and the first network performance indicator from the E-NWDAF entity;
• establishing, by the C-NWDAF entity, a MOS model of the target service based on the quality of service MOS level and the first network performance indicator;
• and
• obtaining, by the C-NWDAF entity, a second network performance indicator corresponding to the MOS level and/or a first load of a first network function entity corresponding to the MOS level, wherein the second network performance indicator is a performance indicator of a radio network that carries the target service, and the first network function entity is a network data provider function entity;
• wherein the establishing, by the C-NWDAF entity, a MOS model of the target service based on the quality of service MOS level and the first network performance indicator comprises:
• excluding, by the C-NWDAF entity, a sample whose quality of service MOS is poor due to an exception of the terminal device and a sample whose quality of service MOS is poor due to an exception of the application function entity based on the quality of service MOS level, the first network performance indicator, and the second network performance indicator and/or the first load of the first network function entity;
• and
• establishing the MOS model of the target service.