US 12,242,959 B2
Method, apparatus, device and storage medium for embedding user app interest
Huiqiang Zhong, Beijing (CN); Siqi Xu, Beijing (CN); Chenhui Liu, Beijing (CN); Lianghui Chen, Beijing (CN); and Jun Fang, Beijing (CN)
Assigned to Beijing Baidu Netcom Science and Technology Co., Ltd., Beijing (CN)
Filed by Beijing Baidu Netcom Science and Technology Co., Ltd., Beijing (CN)
Filed on Mar. 16, 2021, as Appl. No. 17/202,721.
Claims priority of application No. 202010995356.9 (CN), filed on Sep. 21, 2020.
Prior Publication US 2021/0201149 A1, Jul. 1, 2021
Int. Cl. G06N 3/08 (2023.01); G06F 8/61 (2018.01); G06F 16/22 (2019.01); G06F 16/2458 (2019.01); G06N 3/04 (2023.01)
CPC G06N 3/08 (2013.01) [G06F 8/61 (2013.01); G06F 16/2282 (2019.01); G06F 16/2477 (2019.01); G06N 3/04 (2013.01)] 13 Claims
OG exemplary drawing
 
1. A method for embedding user app interest, the method comprising:
acquiring a user existing app installation list and a user app installation list within a predetermined time window, wherein an app comprises app ID information and app category information; and
inputting the user existing app installation list and the user app installation list within the predetermined time window into a pre-trained user app interest embedding model to obtain a user app interest embedding vector, wherein the pre-trained user app interest embedding model is a neural network, and is trained by:
collecting a set of sample user app installation lists comprising a sample current user app installation list, and a sample previous user app installation list within a sample predetermined time window before a sample current time point, wherein the sample current user app installation list comprises first app ID information and first app category information, and the sample previous user app installation list comprises second app ID information and second app category information;
generating a first sample app embedding vector of the sample current user app installation list and generating a second sample app embedding vector of the sample previous user app installation list, by randomly setting an app ID query vector table and an app category query vector table and splicing the app ID query vector table and the app category query vector table; representing the sample current user app installation list as a first one-hot feature, representing the sample previous user app installation list as a second one-hot feature, determining a first input vector of the sample current app installation list by multiplying the first sample app embedding vector with the first one-hot feature, and determining a second input vector of the sample previous app installation list by multiplying the second sample app embedding vector with the second one-hot feature; and
training the pre-trained user app interest embedding model using the first input vector and the second input vector as an input.