US 12,488,289 B2
Method for training sequence mining model, method for processing sequence data, and device
Ye Tao, Shenzhen (CN); Huan Jin, Shenzhen (CN); and Hongbo Jin, Shenzhen (CN)
Assigned to Tencent Technology (Shenzhen) Company Limited, Shenzhen (CN)
Filed by TENCENT TECHNOLOGY (SHENZHEN) COMPANY LIMITED, Guangdong (CN)
Filed on Mar. 22, 2022, as Appl. No. 17/701,598.
Application 17/701,598 is a continuation of application No. PCT/CN2020/125898, filed on Nov. 2, 2020.
Claims priority of application No. 202010099547.7 (CN), filed on Feb. 18, 2020.
Prior Publication US 2022/0215298 A1, Jul. 7, 2022
Int. Cl. G06N 20/20 (2019.01); G06F 18/214 (2023.01)
CPC G06N 20/20 (2019.01) [G06F 18/2148 (2023.01); G06F 18/2155 (2023.01)] 20 Claims
OG exemplary drawing
 
1. A computer-implemented method for training a sequence mining model, the method comprising:
obtaining, by processing circuitry of a computer device, a first sequence sample in a target service scenario, the first sequence sample comprising historical sequence data in the target service scenario;
determining, by the processing circuitry of the computer device, a tag status of the first sequence sample, the tag status of the first sequence sample indicating a proportion of the first sequence sample that has corresponding tag information;
selecting, by the processing circuitry of the computer device, at least a sub-model from a plurality of sub-models included in a sequence mining frame according to the tag status, to construct the sequence mining model, the plurality of sub-models including:
a first sub-model configured to generate a latent representation of target sequence data of the first sequence sample,
a second sub-model configured to determine target tag information of the target sequence data based on the latent representation when the tag status indicates that the first sequence sample has no tag information, and
a third sub-model configured to determine the target tag information based on the latent representation when the tag status indicates that the first sequence sample at least partially has tag information; and
training, by the processing circuitry of the computer device, the constructed sequence mining model by using the first sequence sample, wherein
the training the sequence mining model comprises:
performing pre-training of the first sub-model by using the first sequence sample, to obtain a pre-trained first sub-model,
processing the first sequence sample by using the pre-trained first sub-model, to obtain a latent representation of the first sequence sample, and
jointly training the pre-trained first sub-model and at least one of the second sub-model or the third sub-model by using the first sequence sample and the latent representation of the first sequence sample, to obtain the sequence mining model; and
the trained sequence mining model is configured to determine target tag information of target sequence data in the target service scenario.