Method and apparatus for retrieving similar video and storage medium

Patent metadata

Field	Value
Publication number	US-10853660-B2
Application number	US-201916509289-A
Country	US
Kind code	B2
Filing date	Jul 11, 2019
Priority date	May 11, 2017
Publication date	Dec 1, 2020
Grant date	Dec 1, 2020

Abstract

Embodiments of this application disclose a method for retrieving similar videos performed at a computing device. The computing device obtains video information of a video for which similar videos are to be retrieved, the video information including a video tag and a video title, and trains the video information by using a preset text depth representation model, to convert the video information into a word vector. After selecting, from a video library according to a preset knowledge graph, videos matching the video information, to obtain a first candidate video set, the computing device screens, in the video library, videos similar to the video information according to the word vector, to obtain a second candidate video set and then determines a similar video for the video information from the first candidate video set and the second candidate video set.

First claim

What is claimed is:

1. A method for retrieving similar videos performed at a computing device having one or more processors and memory storing a plurality of programs to be executed by the one or more processors, the method comprising:
obtaining video information of a video for which similar videos are to be retrieved, the video information comprising a video tag and a video title;
selecting, from a video library according to a preset knowledge graph, videos matching the video information, to obtain a first candidate video set;
training the video information by using a preset text depth representation model, to convert the video information into a word vector;
screening, in the video library, videos similar to the video information according to the word vector, to obtain a second candidate video set; and
determining a similar video for the video information from the first candidate video set and the second candidate video set.

2. The method according to claim 1 , wherein the determining a similar video for the video information from the first candidate video set and the second candidate video set comprises:
scoring each video in the first candidate video set, to obtain a first score value;
scoring each video in the second candidate video set, to obtain a second score value;
respectively calculating weighted values of the first score value and the corresponding second score value, to obtain a comprehensive score value of each video; and
determining a video whose comprehensive score value is greater than a preset score value as the similar video for the video information.

3. The method according to claim 1 , wherein the training the video information by using a preset text depth representation model, to convert the video information into a word vector comprises:
performing word segmentation on the video tag and the video title, to obtain video text after word segmentation; and
training the video text after word segmentation by using the preset text depth representation model, to obtain a word vector of the video text after word segmentation; and
the screening, in the video library, videos similar to the video information according to the word vector, to obtain a second candidate video set further comprises:
screening, in the video library, the videos similar to the video information according to the word vector of the video text after word segmentation, to obtain the second candidate video set.

4. The method according to claim 3 , wherein the screening, in the video library, the videos similar to the video information according to the word vector of the video text after word segmentation, to obtain the second candidate video set comprises:
calculating similarity between the video information and each video in the video library according to the word vector of the video text after word segmentation; and
selecting videos whose similarity is higher than a preset threshold, to obtain the second candidate video set.

5. The method according to claim 3 , further comprising:
before training the video text after word segmentation by using the preset text depth representation model:
obtaining a preset corpus set and performing word segmentation on a sentence in the corpus set; and
learning a preset original model according to a sentence after word segmentation, to obtain the text depth representation model.

6. The method according to claim 1 , wherein the selecting, from a video library according to a preset knowledge graph, videos matching the video information, to obtain a first candidate video set comprises:
extracting an entity word from the video tag and the video title, to obtain a seed; and
selecting, from the video library according to the preset knowledge graph, videos matching the seed, to obtain the first candidate video set.

7. The method according to claim 6 , wherein the selecting, from the video library according to the preset knowledge graph, videos matching the seed, to obtain the first candidate video set comprises:
selecting an entity word having a strong association relationship with the seed according to the preset knowledge graph as a candidate word; and
selecting, from the video library, videos whose video information comprises the candidate word, to obtain the first candidate video set.

8. The method according to claim 7 , wherein the selecting an entity word having a strong association relationship with the seed according to the preset knowledge graph as a candidate word comprises:
mapping the seed onto entity words in a preset entity library, the entity library being established according to the preset knowledge graph;
determining relationship degrees between the seed and the entity words in the entity library; and
selecting an entity word whose relationship degree is less than or equal to a set value as the candidate word.

9. The method according to claim 8 , further comprising:
before mapping the seed onto entity words in a preset entity library,
setting a basic thesaurus;
obtaining Internet information according to the basic thesaurus and performing non-entity word cleaning on the Internet information; and
constructing a triple relationship between the entity words according to the basic thesaurus and cleaned Internet information, to obtain the preset entity library.

10. A computing device comprising one or more processors, memory coupled to the one or more processors and a plurality of programs stored in the memory that, when executed by the one or more processors, cause the computing device to perform a plurality of operations comprising:
obtaining video information of a video for which similar videos are to be retrieved, the video information comprising a video tag and a video title;
selecting, from a video library according to a preset knowledge graph, videos matching the video information, to obtain a first candidate video set;
training the video information by using a preset text depth representation model, to convert the video information into a word vector;
screening, in the video library, videos similar to the video information according to the word vector, to obtain a second candidate video set; and
determining a similar video for the video information from the first candidate video set and the second candidate video set.

11. The computing device according to claim 10 , wherein the determining a similar video for the video information from the first candidate video set and the second candidate video set comprises:
scoring each video in the first candidate video set, to obtain a first score value;
scoring each video in the second candidate video set, to obtain a second score value;
respectively calculating weighted values of the first score value and the corresponding second score value, to obtain a comprehensive score value of each video; and
determining a video whose comprehensive score value is greater than a preset score value as the similar video for the video information.

12. The computing device according to claim 10 , wherein the training the video information by using a preset text depth representation model, to convert the video information into a word vector comprises:
performing word segmentation on the video tag and the video title, to obtain video text after word segmentation; and
training the video text after word segmentation by using the preset text depth representation model, to obtain a word vector of the video text after word segmentation; and
the screening, in the video library, videos similar to the video information according to the word vector, to obtain a second candidate video set further comprises:
screening, in the video library, the videos similar to the video i

Assignees

• Tencent Tech Shenzhen Co Ltd

Inventors

• Zhang Yuanyuan
• Yu Qun
• Zhan Fei
• Hua Xiao
• Cheong Wengsan
• Xiong Lei

Classifications

• G06V30/1983

  Syntactic or structural pattern recognition, e.g. symbolic string recognition · CPC title

• G06F16/3338 Primary

  Query expansion · CPC title

• G06V10/82

  using neural networks · CPC title

• G06V10/764

  using classification, e.g. of video objects · CPC title

• G06V20/46

  Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames · CPC title