Systems and methods for correcting errors in caption text
Systems and methods are described to address shortcomings in conventional systems by correcting an erroneous term in on-screen caption text for a media asset. In some aspects, the systems and methods identify the erroneous term in a text segment of the on-screen caption text, and identify one or more video frames of the media asset corresponding to the text segment. The systems and methods further identify a contextual term related to the erroneous term from the one or more video frames. By accessing a knowledge graph, the systems and methods identify a candidate correction based on the contextual term and a portion of the text segment. Lastly, the systems and methods replaces the erroneous term with the candidate correction.
1. A method comprising:
identifying an erroneous term in a text portion of a video asset;
identifying a non-textual visual object in a video frame from the video asset that is associated with the erroneous term, wherein a descriptor of the identified non-textual visual object is not a candidate correction term for the erroneous term, and wherein the non-textual visual object comprises an image of a person;
analyzing the identified non-textual visual object in the video frame, using image recognition, to identify an object category based on the descriptor of the identified non-textual visual object, wherein the identified object category comprises an indication of an identity of the person;
identifying the candidate correction term for the erroneous term that belongs to the identified object category; and
replacing the erroneous term in the text portion of the video asset with the identified candidate correction term.
2. The method of claim 1 , wherein the identifying the erroneous term in the text portion of the video asset comprises performing natural language processing on the text portion of the video asset to compare the text portion of the video asset against a plurality of grammar rules.
3. The method of claim 1 , wherein the identifying the non-textual visual object in the video frame from the video asset that is associated with the erroneous term comprises extracting a first video frame at a position of the video asset corresponding to a position of a time-stamped text portion of the video asset.
4. The method of claim 1 , further comprising:
generating the text portion of the video asset by analyzing an audio stream of a media asset.
5. The method of claim 4 , wherein the text portion of the video asset is time-stamped, and wherein the video frame is extracted at a position of the video asset corresponding to a position of the erroneous term in a time-stamped text portion of the video asset.
6. The method of claim 1 , wherein the replacing the erroneous term in the text portion of the video asset with the identified candidate correction term comprises replacing the erroneous term with the identified candidate correction term while a live broadcast of the video asset is being generated for presentation.
7. The method of claim 1 , wherein the replacing the erroneous term in the text portion of the video asset with the identified candidate correction term comprises replacing the erroneous term with the identified candidate correction term when the video asset is being stored as on-demand content.
8. A method comprising:
identifying an erroneous term in a text portion of a video asset;
identifying an object in a video frame from the video asset, wherein a descriptor of the identified object is not a candidate correction term for the erroneous term;
analyzing the identified object in the video frame, using image recognition, to identify an object category based on the descriptor of the identified object;
identifying the candidate correction term for the erroneous term that belongs to the identified object category, wherein the identifying the candidate correction term that belongs to the identified object category comprises:
extracting a keyword from the text portion of the video asset;
searching a knowledge graph for nodes corresponding to the identified object category and the keyword;
analyzing the nodes for properties associated with the identified object category and the keyword; and
determining at least one other node based at least in part on the properties associated with the identified object category and the keyword, wherein the at least one other node corresponds to the candidate correction term that belongs to the identified object category; and
replacing the erroneous term in the text portion of the video asset with the identified candidate correction term.
9. The method of claim 8 , wherein the identifying the candidate correction term that belongs to the identified object category further comprises:
determining a plurality of candidate correction terms for the erroneous term from the knowledge graph;
assigning a weight to each candidate correction term of the plurality of candidate correction terms based on the determining; and
identifying a candidate correction term associated with a highest weight as the identified candidate correction term.
10. The method of claim 9 , wherein a more recent candidate correction term of the plurality of candidate correction terms is assigned a higher weight, and wherein the more recent candidate correction term is a candidate term associated with a more recent time-stamp, a candidate term that has been updated more recently or a candidate term that has gained popularity in recent searches.
11. A system comprising:
an input/output circuitry configured to receive a video asset; and
a control circuitry configured to:
identify an erroneous term in a text portion of the video asset;
identify a non-textual visual object in a video frame from the video asset that is associated with the erroneous term, wherein a descriptor associated with the identified non-textual visual object is not a candidate correction term for the erroneous term, and wherein the non-textual visual object comprises an image of a person;
analyze the identified non-textual visual object in the video frame, using image recognition, to identify an object category based on the descriptor associated with the identified non-textual visual object, wherein the identified object category comprises an indication of an identity of the person;
identify the candidate correction term for the erroneous term that belongs to the identified object category; and
replace the erroneous term in the text portion of the video asset with the identified candidate correction term.
12. The system of claim 11 , wherein the control circuitry is configured to identify the erroneous term in the text portion of the video asset by performing natural language processing on the text portion of the video asset to compare the text portion of the video asset against a plurality of grammar rules.
13. The system of claim 11 , wherein the control circuitry is configured to identify the non-textual visual object in the video frame from the video asset that is associated with the erroneous term by extracting a first video frame at a position of the video asset corresponding to a position of a time-stamped text portion of the video asset.
14. The system of claim 11 , wherein the control circuitry is further configured to:
generate the text portion of the video asset by analyzing an audio stream of a media asset.
15. The system of claim 14 , wherein the text portion of the video asset is time-stamped, and wherein the control circuitry is further configured to extract the video frame at a position of the video asset corresponding to a position of the erroneous term in a time-stamped text portion of the video asset.
16. The system of claim 11 , wherein the control circuitry is configured to identify the candidate correction term that belongs to the identified object category by:
extracting a keyword from the text portion of the video asset;
searching a knowledge graph for nodes corresponding to the identified object category and the keyword;
analyzing the nodes for properties associated with the identified object category and the keyword; and
determining at least one other node based on the properties associated with the identified object category and the keyword, wherein the at least one other node corresponds to the candidate correction term that belongs to the identified object category.
17. The system of claim 16 , wherein the control circuitry is configured to identify the candidate correction term that belongs to the identified object category by:
determining a plurality of candidate correction terms for the erroneous term from the knowledge graph;
assigning a weight to each candidate correction term of the plurality of candidate correction terms based on the determining; and
identifying a candidate correction term associated with a highest weight as the identified candidate correction term.
18. The system of claim 17 , wherein a more recent candidate correction term of the plurality of candidate correction terms is assigned a higher weight, and wherein the more recent candidate correction term is a candidate term associated with a more recent time-stamp, a candidate term that has been updated more recently or a candidate term that has gained popularity in recent searches.
19. The system of claim 11 , wherein the control circuitry is configured to replace the erroneous term in the text portion of the video asset with the identified candidate correction term by replacing the erroneous term with the identified candidate correction term while a live broadcast of the video asset is being generated for presentation.
20. The system of claim 11 , wherein the control circuitry is configured to replace the erroneous term in the text portion of the video asset with the identified candidate correction term by replacing the erroneous term with the identified candidate correction term when the video asset is being stored as on-demand content.