US 12,282,940 B2
Advertisement title rewriting method, apparatus and device, and storage medium
Jiangtao Feng, Beijing (CN); Han Li, Beijing (CN); Jiaze Chen, Beijing (CN); Yunbo Chen, Beijing (CN); Hao Zhou, Beijing (CN); and Lei Li, Beijing (CN)
Assigned to BEIJING YOUZHUJU NETWORK TECHNOLOGY CO., LTD., Beijing (CN)
Appl. No. 18/260,180
Filed by BEIJING YOUZHUJU NETWORK TECHNOLOGY CO., LTD., Beijing (CN)
PCT Filed Dec. 29, 2021, PCT No. PCT/CN2021/142340
§ 371(c)(1), (2) Date Jun. 30, 2023,
PCT Pub. No. WO2022/143732, PCT Pub. Date Jul. 7, 2022.
Claims priority of application No. 202011627693.9 (CN), filed on Dec. 31, 2020.
Prior Publication US 2024/0062253 A1, Feb. 22, 2024
Int. Cl. G06Q 30/02 (2023.01); G06N 3/0442 (2023.01); G06Q 30/0241 (2023.01); G06Q 30/0251 (2023.01)
CPC G06Q 30/0276 (2013.01) [G06N 3/0442 (2023.01); G06Q 30/0251 (2013.01)] 18 Claims
OG exemplary drawing
 
1. An advertisement title rewriting method, being executed by an electronic device, comprising:
replacing preset information in an advertisement title with a corresponding placeholder to convert the advertisement title into a title template;
generating, through a preset neural network model, a rewritten title template according to the title template, an advertisement attribute key-value table and a query cluster; and
replacing the placeholder in the rewritten title template with the corresponding preset information to convert the rewritten title template into a rewritten title;
wherein the preset neural network model is composed of a title encoder, a key-value encoder, a query encoder and a decoder;
wherein generating, through the preset neural network model, the rewritten title template according to the title template, the advertisement attribute key-value table and the query cluster comprises:
converting the title template into a matrix representation and a vector representation of the title template through the title encoder, converting the advertisement attribute key-value table into a matrix representation and a vector representation of the advertisement attribute key-value table through the key-value encoder, converting the query cluster into a matrix representation and a vector representation of the query cluster through the query encoder, and performing decoding and generating the rewriting title template through the decoder;
adding the vector representation of the title template, the vector representation of the advertisement attribute key-value table and the vector representation of the query cluster to generate an initial seed of a preset neural network decoder;
connecting the matrix representation of the title template, the matrix representation of the advertisement attribute key-value table and the matrix representation of the query cluster to create a memory; and
generating, through the preset neural network decoder, the rewritten title template according to the initial seed and the memory.