Recommendation information pushing method, server, and storage medium

Patent metadata

Field	Value
Publication number	US-10609433-B2
Application number	US-201715715840-A
Country	US
Kind code	B2
Filing date	Sep 26, 2017
Priority date	Sep 8, 2015
Publication date	Mar 31, 2020
Grant date	Mar 31, 2020

Abstract

A method for pushing recommendation information is provided, and the method includes: obtaining a meta path that connects a candidate user and a target user in a heterogeneous information network, the meta path comprising a connection between the candidate user and a candidate recommendation-object and having an attribute value; obtaining a user similarity between the target user and the candidate user relative to the meta path; estimating an attribute value of a connection between the candidate recommendation-object and the target user according to the attribute value of the connection between the candidate user and the candidate recommendation-object, an attribute value constraint condition of the meta path, and the user similarity; and sending recommendation information of the candidate recommendation-object to a terminal corresponding to the target user when the estimated attribute value meets a recommendation condition.

First claim

What is claimed is:

1. A method for pushing recommendation information, comprising:
obtaining a meta path that connects a candidate user and a target user in a heterogeneous information network, the meta path comprising a connection between the candidate user and a candidate recommendation-object and having a candidate attribute value;
obtaining a user similarity between the target user and the candidate user relative to the meta path, including:
identifying a training meta path that connects the candidate user, the target user, and a set of training objects;
splitting the training meta path into a plurality of training atomic meta paths, wherein, in a first of the plurality of atomic training meta paths, rating responses to a first of the set of training objects from the target and candidate users are tabulated into a first table, and in a second of the plurality of atomic meta paths, rating responses to a second of the set of training objects from the target and candidate users are tabulated into a second table;
normalizing the rating responses in the first and second tables into a value of 0, 1, or 2; and
obtaining the user similarity between the candidate and target users based on the first and second tables after normalizing the rating responses;

estimating a target attribute value of a connection between the candidate recommendation-object and the target user according to the candidate attribute value of the connection between the candidate user and the candidate recommendation-object, an attribute value constraint condition of the meta path, and the user similarity, wherein the attribute value constraint condition restricts a mathematical difference between the candidate attribute value and the target attribute value to be within a preset range; and
sending recommendation information of the candidate recommendation-object to a terminal corresponding to the target user when the estimated attribute value meets a recommendation condition.

2. The method according to claim 1 , wherein the estimating the target attribute value of the connection between the candidate recommendation-object and the target user comprises:
obtaining a discrete value range of the target attribute value of the connection between the target user and the candidate recommendation-object;
separately obtaining, for each value in the discrete value range, a connection that is between the candidate user and the candidate recommendation-object and that has an attribute value whose value meets the attribute value constraint condition;
calculating, according to a user similarity that is between the candidate user corresponding to the obtained connection and the target user, an attribute value strength corresponding to the value;
calculating a weighted average of values in the discrete value range separately by using a corresponding attribute value strength as a weight; and
obtaining the estimated attribute value of the connection between the candidate recommendation-object and the target user according to the calculated weighted average.

3. The method according to claim 2 , wherein the obtaining the estimated attribute value of the connection between the candidate recommendation-object and the target user according to the calculated weighted average comprises:
calculating a composite weighted average by separately multiplying a weighted average calculated in each meta path by a path weight of the each meta path, to obtain the estimated attribute value of the connection between the candidate recommendation-object and the target user.

4. The method according to claim 2 , wherein the obtaining the estimated attribute value of the connection between the candidate recommendation-object and the target user according to the calculated weighted average comprises:
calculating a composite weighted average by separately multiplying a weighted average calculated in each meta path by a path weight of the each meta path relative to path weights corresponding to the target user and the meta path, to obtain the estimated attribute value of the connection between the candidate recommendation-object and the target user.

5. The method according to claim 4 , further comprising:
obtaining a real attribute value of the connection between the candidate recommendation-object and the target user;
initializing the path weights corresponding to the target user and the meta path to obtain the initialized path weights; and
adjusting, according to the user similarity, the initialized path weights in a direction towards an average of path weights corresponding to the candidate user and the meta path, so that a difference between the real attribute value and the estimated attribute value meets a minimization condition.

6. The method according to claim 1 , wherein the candidate recommendation-object is a network resource including at least one of a movie, a music recording, a book, and a piece of goods; and the attribute value is a rating value.

7. A server, comprising: a memory storing instructions; and a processor coupled to the memory and, when executing the instructions, configured for:
obtaining a meta path that connects a candidate user and a target user in a heterogeneous information network, the meta path comprising a connection between the candidate user and a candidate recommendation-object and having a candidate attribute value;
obtaining a user similarity between the target user and the candidate user relative to the meta path, including:
identifying a training meta path that connects the candidate user, the target user, and a set of training objects;
splitting the training meta path into a plurality of training atomic meta paths, wherein, in a first of the plurality of atomic training meta paths, rating responses to a first of the set of training objects from the target and candidate users are tabulated into a first table, and in a second of the plurality of atomic meta paths, rating responses to a second of the set of training objects from the target and candidate users are tabulated into a second table;
normalizing the rating responses in the first and second tables into a value of 0, 1, or 2; and
obtaining the user similarity between the candidate and target users based on the first and second tables after normalizing the rating responses;

estimating a target attribute value of a connection between the candidate recommendation-object and the target user according to the candidate attribute value of the connection between the candidate user and the candidate recommendation-object, an attribute value constraint condition of the meta path, and the user similarity, wherein the attribute value constraint condition restricts a mathematical difference between the candidate attribute value and the target attribute value to be within a preset range; and
sending recommendation information of the candidate recommendation-object to a terminal corresponding to the target user when the estimated attribute value meets a recommendation condition.

8. The server according to claim 7 , wherein the estimating the target attribute value of the connection between the candidate recommendation-object and the target user comprises:
obtaining a discrete value range of the target attribute value of the connection between the target user and the candidate recommendation-object;
separately obtaining, for each value in the discrete value range, a connection that is between the candidate user and the candidate recommendation-object and that has an attribute value whose value meets the attribute value constraint condition;
calculating, according to a user similarity that is between the candidate user corresponding to the obtained connection and the target user, an attribute value strength corresponding to th

Assignees

• Tencent Tech Shenzhen Co Ltd

Inventors

• Shi Chuan
• He Peng
• Yi Lingling
• Zhang Zhiqiang

Classifications

• G06Q30/0631

  Recommending goods or services · CPC title

• G06F16/958

  Organisation or management of web site content, e.g. publishing, maintaining pages or automatic linking · CPC title

• G06F16/00

  Information retrieval; Database structures therefor; File system structures therefor · CPC title

• G06Q30/0255

  based on user history · CPC title

• G06F16/9535

  Search customisation based on user profiles and personalisation · CPC title