Open-domain trending hashtag recommendations

Patent metadata

Field	Value
Publication number	US-12050647-B2
Application number	US-202217877469-A
Country	US
Kind code	B2
Filing date	Jul 29, 2022
Priority date	Jul 29, 2022
Publication date	Jul 30, 2024
Grant date	Jul 30, 2024

Abstract

Techniques for recommending hashtags, including trending hashtags, are disclosed. An example method includes accessing a graph. The graph includes video nodes representing videos, historical hashtag nodes representing historical hashtags, and edges indicating associations among the video nodes and the historical hashtag nodes. A trending hashtag is identified. An edge is added to the graph between a historical hashtag node representing a historical hashtag and a trending hashtag node representing the trending hashtag, based on a semantic similarity between the historical hashtag and the trending hashtag. A new video node representing a new video is added to the video nodes of the graph. A graph neural network (GNN) is applied to the graph, and the GNN predicts a new edge between the trending hashtag node and the new video node. The trending hashtag is recommended for the new video based on prediction of the new edge.

First claim

The invention claimed is:

1. A method comprising:
accessing a graph comprising:
video nodes representing videos,
historical hashtag nodes representing historical hashtags, and
edges indicating associations among the video nodes and the historical hashtag nodes;

identifying a trending hashtag;
adding, to the graph, an edge between a historical hashtag node representing a historical hashtag of the historical hashtags and a trending hashtag node representing the trending hashtag, based on a semantic similarity between the historical hashtag and the trending hashtag;
adding, to the video nodes of the graph, a new video node representing a new video;
utilizing a graph neural network (GNN) applied to a subset of the edges of the graph to predict a new edge between the trending hashtag node and the new video node, the subset of the edges omitting video-to-video edges that are between multiple video nodes in the graph; and
recommending the trending hashtag for the new video based on prediction of the new edge between the trending hashtag node and the new video node.

2. The method of claim 1 , further comprising determining the semantic similarity between the historical hashtag and the trending hashtag, wherein determining the semantic similarity comprises:
determining a first vector representing the trending hashtag;
determining a second vector representing the historical hashtag; and
computing a distance between the first vector and the second vector.

3. The method of claim 2 , wherein determining the first vector representing the trending hashtag comprises:
splitting the trending hashtag into a first word and a second word;
determining a first embedding vector for the first word and a second embedding vector for the second word; and
combining the first embedding vector and the second embedding vector to determine the first vector representing the trending hashtag.

4. The method of claim 3 , wherein computing the distance between the first vector and the second vector comprises computing a cosine distance between the first vector representing the trending hashtag and the second vector representing the historical hashtag.

5. The method of claim 1 , further comprising:
receiving trending hashtags used on a social media platform; and
adding, to the graph, a respective trending hashtag node representing each trending hashtag in the trending hashtags.

6. The method of claim 1 , further comprising:
adding, to the edges of the graph, a video-to-hashtag edge between a particular video node representing a particular video of the videos and an additional historical hashtag node representing an additional historical hashtag of the historical hashtags, based on the additional historical hashtag node being a tag for the particular video.

7. The method of claim 6 , further comprising:
adding, to the edges of the graph, a particular video-to-video edge between the particular video node and an additional video node representing an additional video of the videos, based on the particular video and the additional video having one or more of the historical hashtags in common.

8. The method of claim 1 , further comprising:
adding, to the edges of the graph, a hashtag-to-hashtag edge between a first historical hashtag node representing a first historical hashtag of the historical hashtags and a second historical hashtag node representing a second historical hashtag of the historical hashtags, based on the first historical hashtag and the second historical hashtag being associated with a common video.

9. The method of claim 1 , further comprising training the GNN to predict one or more edges in the graph.

10. The method of claim 1 , wherein adding, to the video nodes of the graph, the new video node representing the new video comprises:
extracting a set of frames from the new video; and
processing the set of frames through a neural network trained to extract frame features.

11. A system comprising:
a memory component; and
a processing device coupled to the memory component, the processing device to perform operations comprising:
accessing a graph comprising:
video nodes representing videos,
historical hashtag nodes representing historical hashtags, and
edges indicating associations among the video nodes and the historical hashtag nodes;

identifying a trending hashtag;
adding, to the graph, an edge between a historical hashtag node representing a historical hashtag of the historical hashtags and a trending hashtag node representing the trending hashtag, based on a semantic similarity between the historical hashtag and the trending hashtag;
adding, to the video nodes of the graph, a new video node representing a new video;
utilizing a graph neural network (GNN) applied to a subset of the edges of the graph to predict a new edge between the trending hashtag node and the new video node, the subset of the edges omitting video-to-video edges that are between multiple video nodes in the graph; and
recommending the trending hashtag for the new video based on prediction of the new edge between the trending hashtag node and the new video node.

12. The system of claim 11 , the operations further comprising determining the semantic similarity between the historical hashtag and the trending hashtag, wherein determining the semantic similarity comprises:
determining a first vector representing the trending hashtag;
determining a second vector representing the historical hashtag; and
computing a distance between the first vector and the second vector.

13. The system of claim 12 , wherein determining the first vector representing the trending hashtag comprises:
splitting the trending hashtag into a first word and a second word;
determining a first embedding vector for the first word and a second embedding vector for the second word; and
combining the first embedding vector and the second embedding vector to determine the first vector representing the trending hashtag.

14. The system of claim 11 , the operations further comprising:
adding, to the edges of the graph, a video-to-hashtag edge between a particular video node representing a particular video of the videos and an additional historical hashtag node representing an additional historical hashtag of the historical hashtags, based on the additional historical hashtag node being a tag for the particular video.

15. The system of claim 14 , the operations further comprising:
adding, to the edges of the graph, a video-to-video edge between the particular video node and an additional video node representing an additional video of the videos, based on the particular video and the additional video having one or more of the historical hashtags in common.

16. The system of claim 11 , the operations further comprising:
adding, to the edges of the graph, a hashtag-to-hashtag edge between a first historical hashtag node representing a first historical hashtag of the historical hashtags and a second historical hashtag node representing a second historical hashtag of the historical hashtags, based on the first historical hashtag and the second historical hashtag being associated with a common video.

17. A non-transitory computer-readable medium storing executable instructions, which when executed by a processing device, cause the processing device to perform operations comprising:
accessing a graph comprising:
video nodes representing videos,
historical hashtag nodes representing historical hashtags, and
edges indicating associations among the video nodes and the historical hashtag nodes;

identifying a

Assignees

• Adobe Inc

Inventors

• Sarkhel Somdeb
• Chen Xiang
• Swaminathan Viswanathan
• Mehta Swapneel
• Mitra Saayan
• Rossi Ryan
• Guo Han
• Aminian Ali
• Garg Kshitiz

Classifications

• G06Q10/40

  Business processes related to social networking or social networking services · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06F16/9024 Primary

  Graphs; Linked lists (G06F16/9027 takes precedence) · CPC title

• G06Q50/01

  Physics · mapped topic

• G06Q10/48

  using social graphs · CPC title