Using a generative adversarial network for query-keyword matching

Patent metadata

Field	Value
Publication number	US-10642846-B2
Application number	US-201715784057-A
Country	US
Kind code	B2
Filing date	Oct 13, 2017
Priority date	Oct 13, 2017
Publication date	May 5, 2020
Grant date	May 5, 2020

Abstract

A computer-implemented technique is described herein for providing a digital content item using a generator component. The generator component corresponds to a sequence-to-sequence neural network that is trained using an adversarial generative network (GAN) system. In one approach, the technique involves: receiving a query from a user computing device over a computer network; generating random information; generating a key term using the generator component based on the query and the random information; selecting at least one content item based on the key term; and sending the content item(s) over the computer network to the user computing device.

First claim

What is claimed is:

1. One or more computing devices for providing a content item to a user, comprising:
hardware logic circuitry implemented by: (a) one or more hardware processors that execute machine-readable instructions stored in a memory, and/or by (b) one or more other hardware logic components that perform operations using a task-specific collection of logic gates, the hardware logic circuitry including:
an interface component configured to receive a query from a user computing device over a computer network, the user computing device operated by the user;
a random noise generator configured to generate random information;
a generator component configured to generate a synthetic key term based on the query and the random information, the generator component corresponding to a sequence-to-sequence neural network that is trained using an adversarial generative network (GAN) system;
a selection component configured to select at least one content item based, at least in part, on the key term; and
a delivery component configured to send said at least one content item over the computer network to the user computing device.

2. The one or more computing devices of claim 1 ,
wherein the random noise generator is configured to generate new random information, and
wherein the generator component is configured to generate a new key term based on the query and the new random information, the new key term differing from the first-mentioned key term.

3. The one or more computing devices of claim 1 , wherein the sequence-to-sequence neural network includes a plurality of processing units arranged in series, each processing unit corresponding to a long short-term memory (LSTM) unit.

4. The one or more computing devices of claim 1 ,
wherein the sequence-to-sequence neural network includes an encoder and a decoder,
wherein the encoder includes a first series of processing units that map the query into an output vector, the output vector representing an encoding result, and
wherein the decoder includes a second series of processing units that map the output vector into a key term.

5. The one or more computing devices of claim 4 , wherein the generator component uses the random information to modify an input vector that is fed to the encoder.

6. The one or more computing devices of claim 4 , wherein the generator component uses the random information to modify the output vector provided by the encoder.

7. The one or more computing devices of claim 1 , wherein the hardware logic circuitry further includes a verification component that is configured to verify that the key term is a valid match for the query.

8. The one or more computing devices of claim 1 , wherein the GAN system includes a discriminator component that receives a candidate query and a candidate key term as input, and which provides, as an output, an indication of whether the candidate key term is a valid match for the candidate query.

9. The one or more computing devices of claim 8 , wherein the discriminator component includes a convolutional neural network.

10. The one or more computing devices of claim 8 , wherein the discriminator component includes:
a first neural network that maps the candidate query into a first vector;
a second neural network that maps the candidate key term into a second vector; and
a classification component that is configured to provide the indication of whether the candidate key term is a valid match for the candidate query, based on the first vector and the second vector.

11. A method, implemented by one or more computing devices, for delivering a content item to a user, comprising:
receiving a query from a user computing device over a computer network, the user computing device being operated by the user;
generating random information;
generating a synthetic key term using a generator component based on the query and the random information, the generator component corresponding to a sequence-to-sequence neural network that is trained using an adversarial generative network (GAN) system;
selecting at least one content item based, at least in part, on the key term; and
sending said at least one content item over the computer network to the computing device,
said receiving, said generating random information, said generating a key term, said selecting, and said sending being performed by said one or more computing devices.

12. The method of claim 11 , wherein each content item corresponds to a digital ad.

13. The method of claim 11 , wherein the method further involves:
repeating said generating of random information to provide new random information; and
repeating said generating of a key term, to provide a new key term based on the query and the new random information, the new key term differing from the first-mentioned key term.

14. The method of claim 11 ,
wherein the sequence-to-sequence neural network includes an encoder and a decoder,
wherein the encoder includes a first series of processing units that map the query into an output vector, the output vector representing an encoding result, and
wherein the decoder includes a second series of processing units that map the output vector into a key term.

15. The method of claim 14 , wherein the generator component uses the random information to modify an input vector that is fed to the encoder.

16. The method of claim 14 , wherein the generator component uses the random information to modify the output vector provided by the encoder.

17. The method of claim 11 ,
wherein the GAN system includes a discriminator component that receives a candidate query and a candidate key term as input, and which provides, as an output, an indication of whether the candidate key term is a valid match for the candidate query, and
wherein the discriminator component includes a convolutional neural network.

18. A computing environment for providing a content item to a user, comprising:
a search framework that includes one or more computing devices; and
a training framework that includes one or more computing devices,
each computing device of the search framework and the training framework including hardware logic circuitry implemented by: (a) one or more hardware processors that execute machine-readable instructions stored in a memory, and/or by (b) one or more other hardware logic components that perform operations using a task-specific collection of logic gates,
the hardware logic circuitry of the search framework including:
an interface component configured to receive a query from a user computing device over a computer network, the user computing device being operated by the user;
a random noise generator configured to generate random information;
a generator component configured to generate a synthetic key term based on the query and the random information, the generator component corresponding to a sequence-to-sequence neural network that is trained using an adversarial generative network (GAN) system;
a selection component configured to select at least one content item based, at least in part, on the key term; and
a delivery component configured to send said at least one content item over the computer network to the user computing device; and

the hardware logic circuitry of the training framework including:
the GAN system, wherein the GAN system includes the generator component in combination with a discriminator component,
the discriminator component being configured to receive a candidate query and a candidate key term as input, and provide, as an output, an indication of whether the

Assignees

• Microsoft Technology Licensing Llc

Inventors

• Gao Bin
• Zhang Ruofei
• Lee Mu-Chu

Classifications

• G06Q30/0256

  User search · CPC title

• G06F16/2343

  Locking methods, e.g. distributed locking or locking implementation details · CPC title

• G06F16/334

  Query execution (filtering based on additional data G06F16/335) · CPC title

• G06N3/082

  modifying the architecture, e.g. adding, deleting or silencing nodes or connections · CPC title

• G06F16/24575 Primary

  using context · CPC title