Deep neural networks for targeted content distribution

Patent metadata

Field	Value
Publication number	US-10846587-B2
Application number	US-201715664214-A
Country	US
Kind code	B2
Filing date	Jul 31, 2017
Priority date	Jul 31, 2017
Publication date	Nov 24, 2020
Grant date	Nov 24, 2020

Abstract

Herein are techniques to use an artificial neural network to score the relevance of content items for a target and techniques to rank the content items based on their scores. In embodiments, a computer uses a plurality of expansion techniques to identify expanded targets for a content item. For each of the expanded targets, the computer provides inputs to an artificial neural network to generate a relevance score that indicates a relative suitability of the content item for that target. The computer ranks the expanded targets based on the relevance score generated for each of the expanded targets. Based on the ranking, the computer selects a subset of targets from the available expanded targets as the expanded targets for whom the content item is potentially most relevant. The computer stores an association between the content item and each target in the subset of expanded targets.

First claim

What is claimed is:

1. A method comprising:
using a plurality of expansion techniques to identify an expanded plurality of targets that do not satisfy targeting criteria that are specified by a content delivery campaign for a first content item;
for each target in the expanded plurality of targets that do not satisfy said targeting criteria, providing a plurality of inputs to an artificial neural network to generate a relevance score that indicates a suitability of the first content item for said each target, wherein the plurality of inputs are based on the first content item and said each target;
generating a ranking of the expanded plurality of targets that do not satisfy said targeting criteria based on the relevance score generated for each target of the expanded plurality of targets that do not satisfy said targeting criteria;
selecting, based on the ranking, a subset of targets that do not satisfy said targeting criteria from the expanded plurality of targets that do not satisfy said targeting criteria;
storing an association between each target in the subset of targets that do not satisfy said targeting criteria with the first content item;
wherein the method is performed by one or more computers.

2. The method of claim 1 wherein based on the relevance score generated for each target of the expanded plurality of targets comprises based on the relevance score generated for each target of the expanded plurality of targets for the first content item exceeding the relevance scores generated for the target for one or more other content items that were previously provided to the target.

3. The method of claim 1 wherein:
the artificial neural network comprises a plurality of input neurons arranged in an input layer;
the method further comprises associating each feature of a plurality of features with at least one neuron of the plurality of input neurons;
the plurality of features comprises at least one of each of:
a target feature of a particular target, and
a content item feature of the first content item.

4. The method of claim 3 wherein associating each feature of the plurality of features with the at least one neuron of the plurality of input neurons comprises associating a numeric result of a respective mapping function for each feature of the plurality of features with at least one neuron of the plurality of input neurons.

5. The method of claim 3 wherein the target feature comprises at least one of: a user profile detail, a social graph connection, a topic of interest for a user, or a historic action of the user.

6. The method of claim 3 wherein the content item feature comprises at least one of: content text, content pixels, or content metadata.

7. The method of claim 1 further comprising:
receiving a request that is initiated by a given target of the subset of targets;
sending the first content item to the given target.

8. The method of claim 7 wherein:
receiving the request from the given target comprises receiving at least one context feature that is incidental to a session of the given target;
the method further comprises selecting the first content item based on the at least one context feature.

9. The method of claim 8 wherein the context feature comprises at least one of:
a device attribute, a time of day, a day of week, or a geographic location.

10. A method comprising:
storing a plurality of known targets, a plurality of inventoried content items, and a plurality of historical occurrences, each of which indicates:
a content item of the plurality of inventoried content items that was sent,
a target of the plurality of known targets to whom the content item was sent, and
whether or not the target responded to the content item;

based on the plurality of historical occurrences, training an artificial neural network to generate relevance scores for pairings of a target and a content item;
using a plurality of expansion techniques to identify an expanded plurality of targets for a first content item;
for each target in the expanded plurality of targets, providing a plurality of inputs to the artificial neural network to generate a relevance score that indicates a suitability of the first content item for said each target, wherein the plurality of inputs are based on the first content item and said each target;
generating a ranking of the expanded plurality of targets based on the relevance score generated for each target of the expanded plurality of targets;
selecting, based on the ranking, a subset of targets from the expanded plurality of targets;
storing an association between each target in the subset of targets with the first content item;
wherein the method is performed by one or more computers.

11. The method of claim 10 wherein the first content item is not indicated by any historical occurrence in the plurality of historical occurrences.

12. The method of claim 10 , further comprising:
training a shared artificial neural network based on the plurality of historical occurrences that indicate different targets;
training a plurality of target-specific artificial neural networks, each: (a) corresponding to a different target from a second subset of targets of the plurality of known targets and (b) trained based on historical occurrences that indicate the different target;
wherein providing the plurality of inputs to the artificial neural network for each target in the expanded plurality of targets comprises:
for each target of a first subset of targets of the plurality of known targets, providing the plurality of inputs to the shared artificial neural network; and
for each target of the second subset of targets of the plurality of known targets, providing the plurality of inputs to a corresponding target-specific artificial neural network that is based the shared artificial neural network.

13. The method of claim 10 wherein the target responded to the content item comprises the target clicked on an element of a web page.

14. The method of claim 10 wherein training based on the plurality of historical occurrences comprises training based on historical occurrences whose age does not exceed a threshold.

15. One or more non-transient computer-readable media storing instructions that, when executed by one or more processors, cause:
using a plurality of expansion techniques to identify an expanded plurality of targets that do not satisfy targeting criteria that are specified by a content delivery campaign for a first content item;
for each target in the expanded plurality of targets that do not satisfy said targeting criteria, providing a plurality of inputs to an artificial neural network to generate a relevance score that indicates a suitability of the first content item for said each target, wherein the plurality of inputs are based on the first content item and said each target;
generating a ranking of the expanded plurality of targets that do not satisfy said targeting criteria based on the relevance score generated for each target of the expanded plurality of targets that do not satisfy said targeting criteria;
selecting, based on the ranking, a subset of targets that do not satisfy said targeting criteria from the expanded plurality of targets that do not satisfy said targeting criteria;
storing an association between each target in the subset of targets that do not satisfy said targeting criteria with the first content item.

16. The one or more non-transient computer-readable media of claim 15 wherein the instructions further cause:
storing a plurality of known targets, a plurality of inventoried content items, an

Assignees

• Microsoft Technology Licensing Llc

Inventors

• Liu Haishan
• Gao Huiji
• Zhong Jianling

Classifications

• G06N3/084 Primary

  Backpropagation, e.g. using gradient descent · CPC title

• G06N3/043 Primary

  based on fuzzy logic, fuzzy membership or fuzzy inference, e.g. adaptive neuro-fuzzy inference systems [ANFIS] · CPC title

• G06N3/042

  Knowledge-based neural networks; Logical representations of neural networks · CPC title

• G06N3/08

  Learning methods · CPC title

• G06N3/09

  Supervised learning · CPC title