Multimedia content publishing method and apparatus, electronic device, and storage medium

Patent metadata

Field	Value
Publication number	US-12393628-B2
Application number	US-202118024016-A
Country	US
Kind code	B2
Filing date	Aug 18, 2021
Priority date	Aug 28, 2020
Publication date	Aug 19, 2025
Grant date	Aug 19, 2025

Abstract

A multimedia content publishing method and apparatus, an electronic device and a storage medium. The method includes: determining a multimedia content to be published (S 101 ); obtaining a plurality of candidate topics matching the multimedia content, the candidate topics being topics whose relevancy to the multimedia content meets a predetermined condition among topics included in a topic set (S 102 ); determining a selected target topic among the plurality of candidate topics (S 103 ); and sending multimedia content publishing information containing the target topic to a server in response to a multimedia content publishing request. According to the method, a topic having relatively high relevance to multimedia content can be automatically provided for a user, thus saving time for users to think about and edit topics and improving the accuracy of topic selection.

First claim

What is claimed is:

1. A multimedia content publishing method, comprising:
determining a multimedia content to be published;
obtaining a plurality of candidate topics matching the multimedia content, the candidate topics being topics whose relevancy to the multimedia content meets a predetermined condition among topics included in a topic set;
displaying the plurality of candidate topics matching the multimedia content;
determining a selected target topic among the plurality of candidate topics; and
sending multimedia content publishing information containing the selected target topic to a server in response to a multimedia content publishing request,
wherein the relevancy is determined by a relevancy model, and wherein the relevancy model is obtained by:
obtaining a plurality of historical multimedia content items and at least one user-edited topic corresponding to each historical multimedia content item among the plurality of historical multimedia content items,
determining, for each historical multimedia content item, the at least one user-edited topic corresponding to the each historical multimedia content item as a positive class topic corresponding to the each historical multimedia content item, and determining user-edited topics corresponding to other historical multimedia content items as negative class topics corresponding to the each historical multimedia content item,
generating a plurality of training data groups, wherein each of the plurality of training data groups comprises one of the plurality of historical multimedia content items, the positive class topic corresponding to the historical multimedia content item, and the negative class topics corresponding to the historical multimedia content item, and
training the relevancy model based on the plurality of training data groups to obtain model parameters of the relevancy model.

2. The method according to claim 1 , wherein after obtaining the plurality of candidate topics matching the multimedia content and before determining the selected target topic among the plurality of candidate topics, the method further comprises:
displaying, after a target identifier input into a topic input box is detected, the plurality of candidate topics matching the multimedia content in a topic selection box corresponding to the topic input box.

3. The method according to claim 2 , wherein the determining a selected target topic among the plurality of candidate topics comprises:
displaying the selected target topic selected from the plurality of candidate topics in the topic input box corresponding to the topic selection box in response to a selection operation in the topic selection box.

4. The method according to claim 2 , further comprising:
displaying the plurality of candidate topics matching the multimedia content in a horizontal scrolling manner in response to a first trigger instruction for a horizontal-row display mode; or
displaying the plurality of candidate topics matching the multimedia content in a vertical scrolling manner in response to a second trigger instruction for a vertical-row display mode.

5. A multimedia content publishing method, comprising:
obtaining a multimedia content to be published;
selecting a plurality of candidate topics matching the multimedia content from a prestored topic set by utilizing a relevancy model, and returning the plurality of selected candidate topics to a client;
receiving multimedia content publishing information containing a target topic, the target topic being a topic included in the plurality of candidate topics; and
publishing the multimedia content based on the multimedia content publishing information, wherein the relevancy model is obtained by:
obtaining a plurality of historical multimedia content items and at least one user-edited topic corresponding to each historical multimedia content item among the plurality of historical multimedia content items,
determining, for each historical multimedia content item, the at least one user-edited topic corresponding to the each historical multimedia content item as a positive class topic corresponding to the each historical multimedia content item, and determining user-edited topics corresponding to other historical multimedia content items as negative class topics corresponding to the each historical multimedia content item,
generating a plurality of training data groups, wherein each of the plurality of training data groups comprises one of the plurality of historical multimedia content items, the positive class topic corresponding to the historical multimedia content item, and the negative class topics corresponding to the historical multimedia content item, and
training the relevancy model based on the plurality of training data groups to obtain model parameters of the relevancy model.

6. The method according to claim 5 , wherein the selecting a plurality of candidate topics matching the multimedia content from a prestored topic set by utilizing a relevancy model comprises:
determining relevancy between the multimedia content and each topic in the prestored topic set by utilizing the relevancy model; and
selecting the plurality of candidate topics from the prestored topic set based on the relevancy.

7. The method according to claim 5 , wherein the training the relevancy model based on the plurality of training data groups to obtain model parameters of the relevancy model comprises:
inputting, for each training data group among the plurality of training data groups, the training data group into the relevancy model to be trained, so as to determine first relevancy between the historical multimedia content item corresponding to the training data group and the positive class topic corresponding to the historical multimedia content item, and second relevancy between the historical multimedia content item corresponding to the training data group and each negative class topic corresponding to the historical multimedia content item; and
adjusting, based on a difference value between the first relevancy and any second relevancy in all the second relevancy being less than a predetermined threshold, the model parameters of the relevancy model, and training the relevancy model to be trained again until the difference values between the first relevancy and each second relevancy are each greater than or equal to the predetermined threshold, and stopping training to obtain model parameters of the trained relevancy model.

8. The method according to claim 5 , wherein the determining user-edited topics corresponding to the other historical multimedia content items as negative class topics corresponding to the historical multimedia content item comprises:
determining the positive class topics corresponding to the other historical multimedia content items among the plurality of historical multimedia content items having a topic similarity with the positive class topic of the historical multimedia content item that is less than a predetermined threshold, as the negative class topics corresponding to the historical multimedia content item.

9. The method according to claim 8 , wherein the topic similarity comprises a word overlapping degree, and wherein the method further comprises determining the word overlapping degree based on:
performing, for two topics for which the word overlapping degree is to be calculated, word segmentation processing on each of the two topics to obtain a plurality of topic words corresponding to each topic;
performing intersection processing on the plurality of topic words respectively corresponding to the two topics to obtain a processed first topic word group, and performing union processing on the plurality of topic words respectively corresponding to

Assignees

• Beijing Bytedance Network Tech Co Ltd

Inventors

• Su Kai
• Wang Changhu

Classifications

• G06N3/08

  Learning methods · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06F16/45

  Clustering; Classification · CPC title

• G06F16/43

  Querying · CPC title

• G06F16/583

  using metadata automatically derived from the content · CPC title