Data processing methods and systems for determining a service target item

Patent metadata

Field	Value
Publication number	US-12204522-B2
Application number	US-202217674682-A
Country	US
Kind code	B2
Filing date	Feb 17, 2022
Priority date	Jan 7, 2020
Publication date	Jan 21, 2025
Grant date	Jan 21, 2025

Abstract

The present disclosure provides a data processing method performed by an electronic device. The method includes: obtaining first and second service behavior features, and a service scenario feature of an object from service history data associated with the object; generating first and second service embedding vectors and a scenario representation vector according to the first and second service behavior features and the service scenario feature; obtaining first and second weights of the first and second service embedding vectors according to the scenario representation vector and the first and second service embedding vectors; generating first and second service feature vectors according to the first and second service embedding vectors and the first and second weights of the first and second service embedding vectors, respectively; and obtaining an object embedding vector according to the first service feature vector and the second service feature vector.

First claim

What is claimed is:

1. A data processing method performed by an electronic device, the method comprising:
obtaining a first service behavior feature, a second service behavior feature, and a service scenario feature of an object from service history data associated with the object, wherein the service scenario feature comprises a service type, a net type, and a channel ID;
generating a first service embedding vector according to the first service behavior feature, generating a second service embedding vector according to the second service behavior feature, and generating a scenario representation vector according to the service scenario feature;
obtaining a first weight of the first service embedding vector and a second weight of the second service embedding vector according to the scenario representation vector, the first service embedding vector, and the second service embedding vector;
generating a first service feature vector according to the first service embedding vector and the first weight of the first service embedding vector, and generating a second service feature vector according to the second service embedding vector and the second weight of the second service embedding vector;
automatically learning, using a self-attention mechanism, a relevance between the first service feature vector and the second service feature vector by determining a correlation between features of different services to generate an object embedding vector according to the first service feature vector and the second service feature vector, wherein the object embedding vector is obtained by:
generating a first matrix and a second matrix according to the first service feature vector and the second service feature vector;
obtaining n first sub-matrices according to the first matrix and n second sub-matrices according to the second matrix, wherein n is a positive integer greater than 1;
generating n first head matrices according to the n first sub-matrices and the n second sub-matrices;
stitching the n first head matrices; and
generating the service feature group vector according to the stitched n first head matrices and a mapping matrix;

obtaining a plurality of first service candidate items;
determining a first service target item from the plurality of first service candidate items according to the object embedding vector by determining a similarity between the object embedding vector and each of the plurality of first service candidate items; and
presenting the first service target item to a user to provide a personalized recommendation result based on the object embedding vector.

2. The data processing method according to claim 1 , wherein the generating a first service embedding vector according to the first service behavior feature comprises:
vectorizing a feature value in the first service behavior feature, to obtain a feature value vector in the first service behavior feature;
generating a first service representation vector according to the feature value vector in the first service behavior feature; and
performing spatial mapping on the first service representation vector, to generate the first service embedding vector.

3. The data processing method according to claim 1 , wherein the generating a scenario representation vector according to the service scenario feature comprises:
respectively vectorizing feature values in the service type, the net type, and the channel ID, to obtain a feature value vector of the service type, a feature value vector of the net type, and a feature value vector of the channel ID; and
generating the scenario representation vector according to the feature value vector of the service type, the feature value vector of the net type, and the feature value vector of the channel ID.

4. The data processing method according to claim 1 , wherein the generate an object embedding vector according to the first service feature vector and the second service feature vector further comprises:
generating a fourth matrix according to the second service feature vector;
obtaining n fourth sub-matrices according to the fourth matrix;
generating n second head matrices according to the n first sub-matrices, the n second sub-matrices, and the n fourth sub-matrices; and
generating a second service feature group vector according to the n second head matrices.

5. The data processing method according to claim 4 , further comprising:
obtaining a basic attribute feature and a social feature of the object;
generating a basic attribute vector according to the basic attribute feature; and
generating a social embedding vector according to the social feature.

6. The data processing method according to claim 5 , wherein the obtaining an object embedding vector according to the first service feature vector and the second service feature vector further comprises:
obtaining a stitched feature vector according to the first service feature group vector, the second service feature group vector, the basic attribute vector, and the social embedding vector; and
processing the stitched feature vector by using a feed forward neural network (FNN), to obtain the object embedding vector.

7. The data processing method according to claim 1 , wherein the determining a first service target item from the plurality of first service candidate items according to the object embedding vector comprises:
calculating a similarity between the object embedding vector and each of the plurality of first service candidate items; and
selecting the first service target item from the plurality of first service candidate items according to the similarities.

8. The data processing method according to claim 1 , further comprising:
calculating an estimated click through rate (CTR) of the first service target item;
ranking the first service target item according to the estimated CTR; and
presenting the ranked first service target item at the object.

9. The data processing method according to claim 1 , further comprising:
determining an estimated CTR of the first service target item;
determining a service strategy matching the first service target item according to a use environment of the first service target item; and
ranking the first service target item according to the estimated CTR of the first service target item and the service strategy.

10. An electronic device, comprising:
one or more processors; and
a storage apparatus, configured to store one or more programs, the one or more programs, when executed by the one or more processors, causing the electronic device to implement a data processing method including:
obtaining a first service behavior feature, a second service behavior feature, and a service scenario feature of an object from service history data associated with the object, wherein the service scenario feature comprises a service type, a net type, and a channel ID;
generating a first service embedding vector according to the first service behavior feature, generating a second service embedding vector according to the second service behavior feature, and generating a scenario representation vector according to the service scenario feature;
obtaining a first weight of the first service embedding vector and a second weight of the second service embedding vector according to the scenario representation vector, the first service embedding vector, and the second service embedding vector;
generating a first service feature vector according to the first service embedding vector and the first weight of the first service embedding vector, and generating a second service feature vector according to the second service embedding vector and the second weight of the second service embedding vector;
automatically

Assignees

• Tencent Tech Shenzhen Co Ltd

Inventors

• Qiu Zhijie
• Xie Ruobing
• Liu Yi
• Rao Jun
• Zhang Bo

Classifications

• Y02D10/00

  Energy efficient computing, e.g. low power processors, power management or thermal management · CPC title

• G06Q30/0251

  Targeted advertisements · CPC title

• G06Q30/0282

  Rating or review of business operators or products · CPC title

• H04N21/44226

  on social networks · CPC title

• H04N21/252

  Processing of multiple end-users' preferences to derive collaborative data · CPC title