Methods and systems for generating alternative content using adversarial networks implemented in an application programming interface layer
Methods and systems for using a generative adversarial network to generate personalized content in real-time as a user accesses original content. The methods and systems perform the generation through the use of an application programming interface (“API”) layer. Using the API layer, the methods and systems may generate alternative content as a user accesses original content (e.g., a website, video, document, etc.). Upon receiving this original content, the API layer access the generative adversarial network to create personalized alternative content.
1. A method for generating alternative content using generative adversarial networks implemented in an application programming interface layer, the method comprising:
receiving content for display, in a user interface of a user device, to a user, wherein the content includes a plurality of sections;
identifying a section of the plurality of sections as having a section characteristic, wherein the section characteristic is indicative of the section being of interest to the user;
parsing the section for a content characteristic and metadata describing the content characteristic;
generating a content map for the section based on the parsing, wherein the content map indicates a webpage-based position of the content characteristic in the section, the position being obtained via webpage-based positioning information associated with the content characteristic;
generating a feature input based on the content map and the metadata;
inputting the feature input into a generative adversarial network, wherein the generative adversarial network is trained to generate an output of an alternative section, wherein the alternative section corresponds to the content map and has an alternative content characteristic at the webpage-based position, wherein the generative adversarial network is further trained to generate an additional output of an additional alternative section, and wherein the additional alternative section corresponds to an alternative webpage-based position outside the content map; and
generating for display, in the user interface of the user device, the alternative section, wherein the alternative section replaces the section in the plurality of sections of the content, and wherein the additional alternative section is simultaneously displayed, with the alternative section, outside the content map.
2. The method of claim 1 , wherein identifying the section of the plurality of sections as having the section characteristic further comprises:
identifying a plurality of section characteristics in the section; and
comparing each of the plurality of section characteristics to user characteristics in a user profile to determine a match.
3. The method of claim 1 , wherein the content characteristic is textual data and the alternative content characteristic is different textual data.
4. The method of claim 1 , wherein the content characteristic is textual data and the alternative content characteristic is image data, and wherein the generative adversarial network is trained to translate the textual data into the image data.
5. The method of claim 1 , wherein the content characteristic is an alphanumeric text string and the metadata describing the content characteristic comprises a category of an object corresponding to the alphanumeric text string.
6. The method of claim 1 , wherein the generative adversarial network further comprises an autoregressive language model that performs natural language processing using pre-trained language representations.
7. The method of claim 1 , further comprising parsing the alternative section for the section characteristic, wherein the alternative section is generated for display in response to identification of the section characteristic in the alternative section.
8. The method of claim 1 , wherein the metadata indicates a context of the content characteristic.
9. The method of claim 1 , wherein parsing the section for the content characteristic further comprises:
retrieving a list of content characteristics;
comparing objects in the section to the list of content characteristics; and
identifying the content characteristic based on matching an object of the objects to a listed content characteristic.
10. The method of claim 1 , wherein the additional alternative section is anchored to the alternative section in the user interface.
11. The method of claim 1 , further comprising:
performing a verification that the alternative section comprises the section characteristic; and
determining to generate the alternative section based on the alternative section comprising the section characteristic.
12. A non-transitory, computer-readable medium for generating alternative content using generative adversarial networks implemented in an application programming interface layer, comprising instructions that, when executed by one or more processors, cause operations comprising:
receiving content for display, in a user interface of a user device, wherein the content includes a plurality of sections;
identifying a section of the plurality of sections as having a section characteristic, wherein the section characteristic is indicative of the section being of interest to the user;
parsing the section for a content characteristic and metadata describing the content characteristic;
generating a content map for the section based on the parsing, wherein the content map indicates a webpage-based position of the content characteristic in the section, the position being obtained via webpage-based positioning information associated with the content characteristic;
generating a feature input based on the content map and the metadata;
inputting the feature input into a generative adversarial network, wherein the generative adversarial network is trained to generate an output of an alternative section, wherein the alternative section corresponds to the content map and has an alternative content characteristic at the webpage-based position, wherein the generative adversarial network is further trained to generate an additional output of an additional alternative section, and wherein the additional alternative section corresponds to an alternative webpage-based position outside the content map; and
generating for display, in the user interface of the user device, the alternative section, wherein the alternative section replaces the section in the plurality of sections of the content, and wherein the additional alternative section is simultaneously displayed, with the alternative section, outside the content map.
13. The non-transitory computer readable medium of claim 12 , wherein identifying the section of the plurality of sections as having the section characteristic further comprises:
identifying a plurality of section characteristics in the section; and
comparing each of the plurality of section characteristics to user characteristics in a user profile to determine a match.
14. The non-transitory computer readable medium of claim 12 , wherein the content characteristic is textual data and the alternative content characteristic is different textual data.
15. The non-transitory computer readable medium of claim 12 , wherein the content characteristic is textual data and the alternative content characteristic is image data, and wherein the generative adversarial network is trained to translate the textual data into the image data.
16. The non-transitory computer readable medium of claim 12 , wherein the content characteristic is an alphanumeric text string and the metadata describing the content characteristic comprises a category of an object corresponding to the alphanumeric text string.
17. The non-transitory computer readable medium of claim 12 , wherein the generative adversarial network further comprises an autoregressive language model that performs natural language processing using pre-trained language representations.
18. The non-transitory computer readable medium of claim 12 , further comprising parsing the alternative section for the section characteristic, wherein the alternative section is generated for display in response to identification of the section characteristic in the alternative section.
19. The non-transitory computer readable medium of claim 12 , wherein parsing the section for the content characteristic further comprises:
retrieving a list of content characteristics;
comparing objects in the section to the list of content characteristics; and
identifying the content characteristic based on matching an object of the objects to a listed content characteristic.
20. A system for generating alternative content using generative adversarial networks implemented in an application programming interface layer, the system comprising:
a database configured to store a machine learning model, wherein the machine learning model comprises:
a generative adversarial network, wherein the generative adversarial network is trained to generate outputs of alternative sections, wherein the alternative sections correspond to content maps and have alternative content characteristics at predetermined positions; and
an autoregressive language model that performs natural language processing using pre-trained language representations;
a computing device configured to:
receive content for display, in a user interface of a web browser on a user device, to a user, wherein the content includes a plurality of sections;
identify a section of the plurality of sections as having a section characteristic, wherein the section characteristic is determined to be indicative of the section being of interest to the user based on a comparison of the section characteristic and user profile data for the user;
parse the section for a content characteristic and metadata describing the content characteristic, wherein the metadata indicates a context of the content characteristic, and wherein the content characteristic comprises human-readable text;
generate a content map for the section based on the parsing, wherein the content map indicates a webpage-based position of the content characteristic in the section, the position being obtained via webpage-based positioning information associated with the content characteristic;
generate a feature input based on the content map and the metadata, wherein the feature input comprises a vector array of values indicative of the content map and the metadata;
input the feature input into the generative adversarial network, wherein the generative adversarial network is trained to generate a first output of an alternative section, wherein the alternative section corresponds to the content map and has an alternative content characteristic at the webpage-based position, wherein the generative adversarial network is further trained to generate an additional output of an additional alternative section, and wherein the additional alternative section corresponds to an alternative webpage-based position outside the content map; and
generate for display, in the user interface of the web browser on the user device, the alternative section, wherein the alternative section replaces the section in the plurality of sections of the content, and wherein the additional alternative section is simultaneously displayed, with the alternative section, outside the content map.