Intelligent correlation of video and text content using machine learning

1. A method comprising:

determining, by one or more computer processors coupled to memory, a video file for a movie, the video file comprising first video segment and first text corresponding to the first video segment;

generating a first embedded vector using the first video segment and the first text;

determining, using a first machine learning model, a first image relevance score for the first video content based at least in part on the first embedded vector;

determining, using the first machine learning model, a first text score for the first video content based at least in part on the first embedded vector;

determining a first audio segment corresponding to the first video content; and

determining, using the first machine learning model, a first audio score for the first video content based at least in part on the first embedded vector;

determining, based at least in part on the first image relevance score, the first text score, and the first audio segment, that the first video segment depicts content of a first undesired category;

causing presentation of a notification indicating presence of content of the first undesired category at a display during playback of the first video content; and

causing presentation of an option to skip the first video segment.

2. The method of claim 1 , further comprising:

determining that a number of instances content of the first undesired category appear in the first video content is greater than a smoothing threshold; and

determining an increased confidence threshold for the first video content to reduce a number of notifications associated with the first video content.

3. The method of claim 1 , wherein the first text score represents a likelihood the first video segment depicts content of the first undesired category, and the first image relevance score is indicative of a match between the first text and content depicted in the first frame.

4. The method of claim 1 , further comprising:

receiving feedback data, wherein the feedback data comprises an indication that presence of content of the first undesired category is inaccurate; and

using the feedback data associated with the presentation of the notification to retrain the machine learning model.

5. A method comprising:

determining, by one or more computer processors coupled to memory, first video content comprising a first video segment and first text corresponding to the first video segment;

generating a first embedded vector using the first video segment and the first text;

determining, using a first machine learning model, a first image relevance score for the first video content based at least in part on the first embedded vector;

determining, using the first machine learning model, a first text score for the first video content based at least in part on the first embedded vector;

determining, based at least in part on the first image relevance score and the first text score, that the first video segment depicts content of a first undesired category; and

causing presentation of a notification indicating presence of content of the first undesired category at a display.

6. The method of claim 5 , further comprising:

determining a first audio segment corresponding to the first video content; and

determining, using the first machine learning model, a first audio score for the first video content based at least in part on the first embedded vector;

wherein generating the first embedded vector comprises generating the first embedded vector using the first video segment, the first text, and the first audio segment.

7. The method of claim 5 , further comprising:

determining a confidence score based at least in part on the first image relevance score and the first text score; and

determining that the confidence score satisfies a confidence threshold.

8. The method of claim 7 , wherein the confidence score is customizable, the method further comprising:

determining user preferences associated with a user account;

wherein the user preferences comprise the confidence threshold, and wherein the confidence score is associated with the first undesired category.

9. The method of claim 5 , wherein determining that the first video segment depicts content of a first undesired category is performed at a time the first video content is selected for playback.

10. The method of claim 5 , wherein the notification is presented during playback of the first video content, and within a predetermined time interval prior to presentation of the first video segment.

11. The method of claim 10 , further comprising:

during playback of the first video content, causing presentation of an option to skip the first video segment.

12. The method of claim 5 , wherein the first text score represents a likelihood the first video segment depicts content of the first undesired category; and

wherein the first video segment comprises a first frame, and the first image relevance score is indicative of a match between the first text and content depicted in the first frame.

13. The method of claim 5 , further comprising:

determining that a number of instances content of the first undesired category appear in the first video content is greater than a smoothing threshold; and

determining an increased confidence threshold for the first video content to reduce a number of notifications associated with the first video content.

14. The method of claim 5 , wherein the first text is at least one of: subtitle text or closed caption text.

15. The method of claim 5 , further comprising:

using feedback data associated with the presentation of the notification to retrain the machine learning model.

16. The method of claim 15 , further comprising:

receiving the feedback data, wherein the feedback data comprises an indication that presence of content of the first undesired category is inaccurate.

17. A device comprising:

memory that stores computer-executable instructions; and

at least one processor configured to access the memory and execute the computer-executable instructions to:

determine first video content comprising a first video segment and first text corresponding to the first video segment;

generate a first embedded vector using the first video segment and the first text;

determine, using a first machine learning model, a first image relevance score for the first video content based at least in part on the first embedded vector;

determine, using the first machine learning model, a first text score for the first video content based at least in part on the first embedded vector;

determine, based at least in part on the first image relevance score and the first text score, that the first video segment depicts content of a first undesired category; and

cause presentation of a notification indicating presence of content of the first undesired category at a display.

18. The device of claim 17 , wherein the at least one processor is further configured to access the memory and execute the computer-executable instructions to:

determine a confidence score based at least in part on the first image relevance score and the first text score; and

determine that the confidence score satisfies a confidence threshold.

19. The device of claim 17 , wherein the notification is presented during playback of the first video content within a predetermine time interval prior to presentation of the first video segment, and wherein the at least one processor is further configured to access the memory and execute the computer-executable instructions to:

during playback of the first video content, cause presentation of an option to skip the first video segment.

20. The device of claim 17 , wherein the at least one processor is further configured to access the memory and execute the computer-executable instructions to:

determine that a number of instances content of the first undesired category appear in the first video content is greater than a smoothing threshold; and

determine an increased confidence threshold for the first video content to reduce a number of notifications associated with the first video content.