Method and apparatus for generating push notifications

Patent metadata

Field	Value
Publication number	US-10757218-B2
Application number	US-201715473442-A
Country	US
Kind code	B2
Filing date	Mar 29, 2017
Priority date	Mar 29, 2017
Publication date	Aug 25, 2020
Grant date	Aug 25, 2020

Abstract

A method for generating one or more push notifications to a user device is described. The method comprises: obtaining history data representing a history of online activities of a user and candidate data representing a set of candidate information; generating, based on the history data and the candidate data, user profile vectors representing a user profile associated with the user and content vectors representing a set of content profiles associated with the set of candidate information; generating, based on a machine learning model trained with a history of online activities, embedding user feature vectors and embedding content feature vectors based on the history data and the candidate data; and providing for transmission information for one or more push notifications including first candidate information of to a user device associated with the user, the first candidate information being determined from the set of candidate information based on the aforementioned vectors.

First claim

What is claimed is:

1. A method performed by one or more servers having one or more processors for generating one or more push notifications to a user device, the method comprising:
obtaining history data representing a history of online activities of a user;
obtaining candidate data representing a set of candidate information to be selected in one or more push notifications to the user;
generating, based on the history data and the candidate data, user profile vectors representing a user profile associated with the user and content vectors representing a set of content profiles associated with the set of candidate information;
generating, by using a neural network, embedding user feature vectors of the user and embedding content feature vectors of the set of candidate information based on the history data and the candidate data, wherein the embedding user feature vectors are generated based on one or more items the user selects within a predetermined time period, and the embedding content feature vectors are generated based on one or more content identifiers associated with the set of candidate information;
determining a first distance between the user profile vectors and the content vectors;
determining a second distance between the embedding user feature vectors and the embedding content feature vectors;
selecting first candidate information from the set of candidate information based on a weighted combination of the first and second distances; and
providing for transmission information for one or more push notifications including the first candidate information to a user device associated with the user.

2. The method of claim 1 , wherein the user profile includes a user keyword profile and a user category profile;
wherein the user keyword profile is generated based on keywords extracted from the history data including at least one of: prior search queries of the user, title and metadata of a first set of webpages previously visited by the user, and title of a first set of information items previously selected by the user;
wherein the user category profile is generated by associating a first set of predetermined categories to at least some of the history data.

3. The method of claim 1 , wherein the set of content profiles include a set of content keyword profiles and set of content category profiles;
wherein the set of content keyword profiles are generated based on keywords extracted from the second data including at least one of: titles and metadata of a second set of webpages linked to the candidate information database, and title of a second set of information items linked to the second set of webpages;
wherein the set of content category profiles are generated by associating a second set of predetermined categories with at least some of the candidate data.

4. The method of claim 1 , wherein the neural network is trained with a history of online activities of a plurality of users tracked by the one or more servers;
wherein the neural network includes a deep neural network model including a hidden layer configured as a skip-gram model; and
wherein the embedding user feature vectors of the user and embedding content feature vectors are generated based on scaling parameters associated with the hidden layer.

5. The method of claim 4 , further comprising: training the skip-gram model using pairings of target and context elements generated from a data sequence of content access and information item selection;
wherein the skip-gram model is trained to provide a probability value of occurrence of a target element when a context element is present; and
wherein the scaling parameters are updated during the training to generate the probability value according to a distribution of the pairings of target and context elements.

6. The method of claim 4 , wherein generating embedding user feature vectors of the user and embedding content feature vectors of the set of candidate information comprises:
creating, based on the scaling parameters, a lookup table that maps the history data and the candidate data to a set of embedding feature vectors; and
obtaining the embedding user feature vectors and the embedding content feature vectors from the lookup table.

7. The method of claim 1 , wherein determining a first distance between the user profile vectors and the content vectors and a second distance between the embedding user feature vectors and the embedding content feature vectors comprises:
determining a first set of cosine distances between the user profile vectors and the content vectors;
determining a second set of cosine distances between the embedding user feature vectors and the embedding content feature vectors;
determining a set of scores for the set of candidate information based on weighted sums of the first and second set of cosine distances; and
ranking the set of candidate information according to the set of scores;
wherein the first candidate information is selected from the set of candidate information based on a result of the ranking.

8. An apparatus for generating one or more push notifications to a user device, the apparatus comprising:
a memory storing a set of instructions; and
one or more processors configured to execute the set of instructions to cause the apparatus to:
obtain history data representing a history of online activities of a user;
obtain candidate data representing a set of candidate information to be selected in one or more push notifications to the user;
generate, based on the history data and the candidate data, user profile vectors representing a user profile associated with the user and content vectors representing a set of content profiles associated with the set of candidate information;
generate, by using a neural network, embedding user feature vectors of the user and embedding content feature vectors of the set of candidate information based on the history data and the candidate data, wherein the embedding user feature vectors are generated based on one or more items the user selects within a predetermined time period, and the embedding content feature vectors are generated based on one or more content identifiers associated with the set of candidate information;
determine a first distance between the user profile vectors and the content vectors and a second distance between the embedding user feature vectors and the embedding content feature vectors;
select first candidate information from the set of candidate information based on a combination of the first and second distances; and
provide for transmission information for one or more push notifications including the first candidate information to a user device associated with the user.

9. The apparatus of claim 8 , wherein the user profile includes a user keyword profile and a user category profile;
wherein the user keyword profile is generated based on keywords extracted from the history data including at least one of: prior search queries of the user, title and metadata of a first set of webpages previously visited by the user, and title of a first set of information items previously selected by the user;
wherein the user category profile is generated by associating a first set of predetermined categories to at least some of the history data.

10. The apparatus of claim 8 , wherein the set of content profiles include a set of content keyword profiles and set of content category profiles;
wherein the set of content keyword profiles are generated based on keywords extracted from the second data including at least one of: titles and metadata of a second set of webpages linked to the candidate information database, and title of a second set of information items linked to the second set of w

Assignees

• Alibaba Group Holding Ltd

Inventors

• Zhao Huasha
• Li Xiaogang
• Zhang Qiong
• Si Luo
• Gu Zhenyu
• Zhang Qiyu

Classifications

• G06N3/048

  Activation functions · CPC title

• G06N7/01

  Probabilistic graphical models, e.g. probabilistic networks · CPC title

• G06N3/08

  Learning methods · CPC title

• G06N3/0499

  Feedforward networks · CPC title

• H04L67/55

  Push-based network services · CPC title