Customer profile learning based on semi-supervised recurrent neural network using partially labeled sequence data

Patent metadata

Field	Value
Publication number	US-11093818-B2
Application number	US-201615095916-A
Country	US
Kind code	B2
Filing date	Apr 11, 2016
Priority date	Apr 11, 2016
Publication date	Aug 17, 2021
Grant date	Aug 17, 2021

Abstract

A method and system are provided. The method includes receiving by a computer having a processor and a memory, sequence data that includes labeled data and unlabeled data. The method further includes generating, by the computer having the processor and the memory, a recurrent neural network model of the sequence data, the recurrent neural network model having a recurrent layer and an aggregate layer. The recurrent neural network model feeds sequences generated from the recurrent layer into the aggregate layer for aggregation, stores temporal dependencies in the sequence data, and generates labels for at least some of the unlabeled data.

First claim

What is claimed is:

1. A method for label learning, comprising:
generating, by a computer, a deep semi-supervised recurrent neural network having an aggregate layer stacked on a recurrent layer such that outputs of the recurrent layer are fed to inputs of the aggregate layer, the recurrent layer comprising a recurrent neural network configured to model and train sequence data that includes labeled data and unlabeled data such that the labeled and unlabeled data are jointly used for training, the aggregate layer configured to aggregate sequences generated from the recurrent layer to predict label classifications for the unlabeled data and comprising a graph regularizer allowing incorporation of different domain knowledge using a graph-based supervised learning process and a dimension-reducing auto encoder that is stacked on the recurrent neural network and is configured to learn a dimensionally reduced representation of the sequence data including the different domain knowledge by reducing an overall number of dimensions of the sequence data including the labeled and unlabeled data directly provided to an output layer of the deep semi-supervised recurrent neural network from the auto encoder;
generating, by the computer, one or more customer behavior predictions using the deep semi-supervised recurrent neural network; and
generating, by the computer, a targeted advertisement responsive to the one or more customer behavior predictions,
wherein the deep semi-supervised recurrent neural network feeds the sequences generated from the recurrent layer into the aggregate layer for aggregation, stores temporal dependencies in the sequence data, and generates labels for at least some of the unlabeled data.

2. The method of claim 1 , wherein the deep semi-supervised recurrent neural network is generated to form a deep semi-supervised recurrent neural network having the recurrent layer and the aggregate layer by applying the graph-based supervised learning process to the unlabeled data.

3. The method of claim 2 , wherein the graph-based supervised learning process is applied to the unlabeled data in the aggregation layer of the deep semi-supervised recurrent neural network.

4. The method of claim 3 , wherein the aggregation layer in the aggregation layer of the deep semi-supervised recurrent neural network and the graph-based supervised learning process provide aggregation of data having different domains.

5. The method of claim 2 , wherein the one or more customer behavior predictions are generated using the deep semi-supervised recurrent neural network, wherein the one or more customer behavior predictions are generated based on a joint utilization of the labeled data and the unlabeled data.

6. The method of claim 2 , wherein the graph-based supervised learning process is applied to the labeled data and the unlabeled data to extract similarity data there between, and wherein the deep semi-supervised recurrent neural network is formed using the similarity data.

7. The method of claim 1 , wherein the generated sequences from the recurrent layer correspond to different domains that are regularized in the aggregate layer based on graph data.

8. The method of claim 1 , further comprising constructing one or more graphs to describe a similarity between data samples in the sequence data as an input to the deep semi-supervised recurrent neural network.

9. The method of claim 8 , wherein the aggregate layer performs data aggregation based on the similarity between the data samples.

10. The method of claim 1 , wherein at least one of the customer behavior predictions comprises at least one response rate prediction.

11. A non-transitory article of manufacture tangibly embodying a computer readable program which when executed causes a computer to perform the steps of claim 1 .

12. A system, comprising:
a computer, having a processor and a memory, configured to:
generate a deep semi-supervised recurrent neural network having an aggregate layer stacked on a recurrent layer such that outputs of the recurrent layer are fed to inputs of the aggregate layer, the recurrent layer comprising a recurrent neural network configured to model and train sequence data that includes labeled data and unlabeled data such that the labeled and unlabeled data are jointly used for training, the aggregate layer configured to aggregate sequences generated from the recurrent layer to predict label classifications for the unlabeled data and comprising a graph regularizer allowing incorporation of different domain knowledge using a graph-based supervised learning process and a dimension-reducing auto encoder that is stacked on the deep semi-supervised recurrent neural network and is configured to learn a dimensionally reduced representation of the sequence data including the different domain knowledge by reducing an overall number of dimensions of the sequence data including the labeled and unlabeled data directly provided to an output layer of the deep semi-supervised recurrent neural network from the auto encoder;
generating one or more customer behavior predictions using the deep semi-supervised recurrent neural network; and
generating a targeted advertisement responsive to the one or more customer behavior predictions,
wherein the deep semi-supervised recurrent neural network feeds the sequences generated from the recurrent layer into the aggregate layer for aggregation, stores temporal dependencies in the sequence data, and generates labels for at least some of the unlabeled data.

13. The system of claim 12 , wherein the computer is implemented as a server using a cloud computing configuration.

14. The system of claim 12 , wherein the computer generates the deep semi-supervised recurrent neural network to form a deep semi-supervised recurrent neural network having the recurrent layer and the aggregate layer by applying the graph-based supervised learning process to the unlabeled data.

15. The system of claim 14 , wherein the graph-based supervised learning process is applied to the unlabeled data in the aggregation layer of the deep semi-supervised recurrent neural network.

16. The system of claim 14 , wherein the one or more customer behavior predictions are generated using the deep semi-supervised recurrent neural network, wherein the one or more customer behavior predictions are generated based on a joint utilization of the labeled data and the unlabeled data.

17. The system of claim 14 , wherein the graph-based supervised learning process is applied to the labeled data and the unlabeled data to extract similarity data there between, and wherein the deep semi-supervised recurrent neural network is formed using the similarity data.

Assignees

• IBM

Inventors

• Li Hongfei
• Sheopuri Anshul
• Yi Jinfeng
• Yu Qi

Classifications

• G06N3/044 Primary

  Recurrent networks, e.g. Hopfield networks · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06N3/0455

  Auto-encoder networks; Encoder-decoder networks · CPC title

• G06N3/0895 Primary

  Weakly supervised learning, e.g. semi-supervised or self-supervised learning · CPC title

• G06N3/0442

  characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU] · CPC title