Feature processing method and apparatus for artificial intelligence recommendation model, electronic device, and storage medium

Patent metadata

Field	Value
Publication number	US-12205147-B2
Application number	US-202117491435-A
Country	US
Kind code	B2
Filing date	Sep 30, 2021
Priority date	Aug 29, 2019
Publication date	Jan 21, 2025
Grant date	Jan 21, 2025

Abstract

This application provides a feature processing method and apparatus for an artificial intelligence (AI) recommendation model, an electronic device, and a storage medium. The method includes obtaining input data, and converting a data structure of the input data into a uniform sample structure; determining a feature group identifier and a feature extraction function that correspond to the AI recommendation model; extracting a feature group identifier, a feature identifier and a feature value of the input data from the converted input data according to the feature extraction function; and constructing a sample of the AI recommendation model according to the feature group identifier, the feature identifier and the feature value of the input data.

First claim

What is claimed is:

1. A feature processing method for an artificial intelligence (AI) recommendation model, performed by a server that communicatively configured with an electronic device and a database, the electronic device being configured to display a service application on a graphical interface, the method comprising:
obtaining, at the server, online data to be processed having an online data structure as input data, and converting a data structure of the input data into a uniform sample structure, the online data to be processed being generated on an electronic device in response to an operation performed via the service application, and sent by the electronic device to the server;
determining, by the server, a feature group identifier and a feature extraction function that correspond to the AI recommendation model;
obtaining, by the server, a feature group identifier, a feature identifier and a feature value of the input data by performing feature extraction on the converted input data according to the feature extraction function;
constructing, by the server, a sample of the AI recommendation model based on a result of the feature extraction according to a triplet structure of: the feature group identifier corresponding to the AI recommendation model, the feature identifier, and the feature value of the input data, wherein:
training data of the AI recommendation model includes offline training data having an offline data structure, the offline data structure being different from the online data structure,
the training data is converted into the uniform sample structure to obtain converted training data for the AI recommendation model;
feature extraction is performed on the converted training data to obtain training samples, and each training sample is constructed with the same triplet structure of: the feature group identifier, the feature identifier, and the feature value;
during training, a parameter in the AI recommendation model is adjusted according to a difference between an output parameter of the AI recommendation model and a feature value in the triplet structure of the training sample, to reduce a difference between the output parameter and the feature value of the training sample for subsequent training of the AI recommendation model;

predicting, by the server, a recommendation result by using the AI recommendation model on the sample, and storing the recommendation result in the database; and
sending, from the database by the server, the recommendation result to the electronic device, to display the recommendation result in the service application on the graphical interface.

2. The feature processing method according to claim 1 , wherein the determining a feature group identifier and a feature extraction function that correspond to the AI recommendation model comprises:
obtaining atomic feature registration information and combined feature registration information, the atomic feature registration information comprising a registered atomic feature group identifier and a corresponding feature extraction function, and the combined feature registration information comprising a registered combined feature group identifier and at least two corresponding atomic feature group identifiers;
obtaining a feature group identifier set corresponding to the AI recommendation model, and traversing the feature group identifier set; and
adding, when a traversed feature group identifier matches the atomic feature registration information, a feature extraction function corresponding to the feature group identifier of the atomic feature registration information to a feature extraction function chain.

3. The feature processing method according to claim 2 , further comprising:
determining, when a traversed feature group identifier does not match the atomic feature registration information but matches the combined feature registration information, a target atomic feature group identifier corresponding to the feature group identifier of the combined feature registration information;
determining a target feature extraction function corresponding to the target atomic feature group identifier of the atomic feature registration information; and
adding the target feature extraction function to the feature extraction function chain when the target feature extraction function does not exist in the feature extraction function chain.

4. The feature processing method according to claim 2 , wherein obtaining a feature group identifier, a feature identifier and a feature value of the input data by performing feature extraction on the converted input data according to the feature extraction function comprises:
traversing the feature extraction function chain; and
obtaining an atomic feature group identifier, an atomic feature identifier and an atomic feature value of the input data by performing feature extraction on the converted input data according to a traversed feature extraction function.

5. The feature processing method according to claim 4 , wherein the constructing a sample of the AI recommendation model according to the feature group identifier corresponding to the AI recommendation model and the feature group identifier, the feature identifier and the feature value of the input data comprises:
traversing the feature group identifier set; and
determining, when a traversed feature group identifier matches the atomic feature registration information, the feature group identifier as a first atomic feature group identifier to be added, and adding the first atomic feature group identifier, a corresponding atomic feature identifier and a corresponding atomic feature value to the sample.

6. The feature processing method according to claim 5 , further comprising:
determining, when a traversed feature group identifier does not match the atomic feature registration information but matches the combined feature registration information, the feature group identifier as a first combined feature group identifier to be added;
determining at least two first atomic feature group identifiers to be added corresponding to the first combined feature group identifier of the combined feature registration information, and determining atomic feature identifiers and atomic feature values that correspond to the at least two first atomic feature group identifiers;
obtaining a combined feature identifier by combining the atomic feature identifiers corresponding to the at least two first atomic feature group identifiers, and obtaining a combined feature value by combining the atomic feature values corresponding to the at least two first atomic feature group identifiers; and
adding the first combined feature group identifier, the combined feature identifier and the combined feature value to the sample.

7. The feature processing method according to claim 6 , wherein
the obtaining the combined feature identifier by combining the atomic feature identifiers corresponding to the at least two first atomic feature group identifiers comprises:
obtaining the combined feature identifier by performing a first combination operation on the atomic feature identifiers corresponding to the at least two first atomic feature group identifiers, the first combination operation comprising at least one of a hash operation, a bitwise OR operation, a bitwise AND operation, and a bitwise XOR operation; and
the obtaining the combined feature identifier by combining the atomic feature values corresponding to the at least two first atomic feature group identifiers comprises:
obtaining the combined feature identifier by performing a second combination operation on the atomic feature values corresponding to the at least two first atomic feature group identifiers, the second combination operation comprising at least one

Assignees

• Tencent Tech Shenzhen Co Ltd

Inventors

• Zhang Yafei
• Zhong Junwei
• Xia Feng

Classifications

• G06F18/2148

  characterised by the process organisation or structure, e.g. boosting cascade · CPC title

• G06N20/00 Primary

  Machine learning · CPC title

• G06F18/214

  Generating training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title

• G06Q30/0269 Primary

  based on user profile or attribute · CPC title