US 12,443,677 B2
Recommendation method and apparatus based on automatic feature grouping
Bin Liu, Shenzhen (CN); Ruiming Tang, Shenzhen (CN); Huifeng Guo, Shenzhen (CN); Niannan Xue, Shenzhen (CN); Guilin Li, Shenzhen (CN); Xiuqiang He, Shenzhen (CN); and Zhenguo Li, Hong Kong (CN)
Assigned to HUAWEI TECHNOLOGIES CO., LTD., Shenzhen (CN)
Filed by HUAWEI TECHNOLOGIES CO., LTD., Guangdong (CN)
Filed on Oct. 12, 2022, as Appl. No. 17/964,117.
Application 17/964,117 is a continuation of application No. PCT/CN2021/070847, filed on Jan. 8, 2021.
Claims priority of application No. 202010294506.3 (CN), filed on Apr. 14, 2020.
Prior Publication US 2023/0031522 A1, Feb. 2, 2023
Int. Cl. G06F 18/00 (2023.01); G06F 18/2113 (2023.01); G06F 18/214 (2023.01); G06N 3/00 (2023.01); G06N 3/04 (2023.01); G06N 3/08 (2023.01)
CPC G06F 18/2113 (2023.01) [G06F 18/2148 (2023.01); G06N 3/04 (2013.01); G06N 3/08 (2013.01)] 15 Claims
OG exemplary drawing
 
1. A recommendation method based on automatic feature grouping, comprising:
obtaining a plurality of candidate recommended objects and a plurality of association features of each of the plurality of candidate recommended objects;
performing multi-order automatic feature grouping on the plurality of association features of each of the plurality of candidate recommended objects, to obtain a multi-order feature interaction set of each candidate recommended object, wherein each-order feature interaction set in the multi-order feature interaction set of each of the plurality of candidate recommended objects comprises one or more feature interaction groups, each of the one or more feature interaction groups comprises at least one of the plurality of association features of the candidate recommended object, in each feature interaction group in a kth-order feature interaction set in the multi-order feature interaction set, a quantity of association features used in a non-linear mathematical operation is k, and k is an integer greater than 0;
obtaining an interaction feature contribution value of each candidate recommended object through calculation based on the plurality of association features in the multi-order feature interaction set of each of the plurality of candidate recommended objects;
obtaining a prediction score of each candidate recommended object through calculation based on the interaction feature contribution value of each of the plurality of candidate recommended objects; and
determining one or more corresponding candidate recommended objects as one or more target recommended objects, each of the one or more corresponding candidate recommended objects having a prediction score higher than at least some of the other candidate recommended objects in the plurality of candidate recommended objects,
wherein the obtaining an interaction feature contribution value of each candidate recommended object through calculation based on the plurality of association features in the multi-order feature interaction set of each of the plurality of candidate recommended objects comprises:
for the multi-order feature interaction set of each candidate recommended object,
obtaining a kth-order interaction result of each feature interaction group through calculation based on a plurality of association features of each feature interaction group in the kth-order feature interaction set, and
obtaining the interaction feature contribution value of the candidate recommended object through calculation based on a corresponding-order interaction result of a feature interaction group in the multi-order feature interaction set of the candidate recommended object,
wherein the obtaining the interaction feature contribution value of the candidate recommended object through calculation based on a corresponding-order interaction result of a feature interaction group in the multi-order feature interaction set of the candidate recommended object comprises:
obtaining an interaction feature vector based on the corresponding-order interaction result of the feature interaction group in the multi-order feature interaction set of the candidate recommended object; and
inputting the interaction feature vector to a neural network model for calculation, to obtain the interaction feature contribution value of the candidate recommended object, wherein the neural network model is obtained based on a fully connected neural network.