Apparatus for determining role fitness while eliminating unwanted bias

1. An apparatus, the apparatus comprising:

a multicore system processing device;

a plurality of parallel multicore processing devices;

an interface device, the interface device providing connection to the multicore system processing device;

a storage device, the storage device storing training data comprising at least one positive example and at least one negative example, the at least one positive example representing a candidate who would be invited to an interview, the at least one negative example representing a candidate without at least one of a relevant skill and an experience who would not be invited to an interview, the at least one positive example and the at least one negative example being used by the plurality of parallel multicore processing devices to train a deep learning model, the deep learning model being used by the multicore system processing device to predict fitness of a specific candidate for a role; and

an interface bus, the interface bus operably coupling the multicore system processing device, the plurality of parallel multicore processing devices, the interface device, and the storage device,

wherein

the training data is generated by stripping a most recent role from a resume associated with a candidate who has a plurality of instances of work experience listed in a listing of professional history,

the at least one positive example is generated by concatenating the most recent role with a remainder of preceding work experience,

the at least one negative example is generated by concatenating a random job description from a randomly selected and different resume with the remainder of preceding work experience based on an unlikelihood that the candidate will be a good match for the random job description,

at least one indicative pronoun is excluded from a process of extracting relevant information-rich natural language, thereby limiting capacity associated with the deep learning model to develop bias against a demographic group, and

the deep learning model comprises a plurality of parallel deep learning natural language model paths comprising sigmoid output neurons at an output stage of at least one portion of the deep learning model with rectified linear unit neurons used between internal layers within the deep learning model.

2. The apparatus, as defined by claim 1 , wherein fitness is represented as a value between zero and one.

3. The apparatus, as defined by claim 2 , wherein the value between zero and one represents a probability that a human recruiter would invite the specific candidate for an interview for the role.

4. The apparatus, as defined by claim 1 , wherein the apparatus is further configured to perform a statistical test to confirm that measures to limit bias are effective.

5. The apparatus, as defined by claim 4 , wherein parsed resume language and a job description associated with a job for which the specific candidate is applying are tokenized, converted into an integer, and processed by a word embedding model.

6. The apparatus, as defined by claim 1 , wherein a scoring process applied by the deep learning model utilizes parallel processing of a plurality of recent work experiences associated with the specific candidate.

7. The apparatus, as defined by claim 6 , wherein the deep learning model utilizes parallel processing by processing the resume associated with the specific candidate and extracting relevant information-rich natural language.

8. The apparatus, as defined by claim 1 , wherein a capacity of the deep learning model to develop bias against a particular gender is limited.

9. The apparatus, as defined by claim 1 , wherein demographic and personally identifiable information are excluded from extracting relevant information-rich natural language.

10. The apparatus, as defined by claim 1 , wherein a capacity of the deep learning model to develop bias against a particular race is limited.

11. A method, the method comprising:

a plurality of parallel multicore processing devices, within a multicore system processing device, training a deep learning model based on training data that is stored within a storage device and that comprises at least one positive example and at least one negative example, the at least one positive example representing a candidate who would be invited to an interview and the at least one negative example representing a candidate without at least one of a relevant skill and an experience who would not be invited to an interview; and

applying the deep learning model to score a level of fitness associated with a specific candidate for a role,

wherein

an interface bus operably couples the multicore system processing device, the plurality of parallel multicore processing devices, the storage device, and an interface device that provides connection to the multicore system processing device,

the training data is generated by stripping a most recent role from a resume associated with a candidate who has a plurality of instances of work experience listed in a listing of professional history,

the at least one positive example is generated by concatenating the most recent role with a remainder of preceding work experience,

the at least one negative example is generated by concatenating a random job description from a randomly selected and different resume with the remainder of preceding work experience based on an unlikelihood that the candidate will be a good match for the random job description,

at least one indicative pronoun is excluded from a process of extracting relevant information-rich natural language, thereby limiting capacity associated with the deep learning model to develop bias against a demographic group, and

the deep learning model comprises a plurality of deep learning natural language model paths comprising sigmoid output neurons at an output stage of at least one portion of the deep learning model with rectified linear unit neurons used between internal layers within the deep learning model.

12. The method of claim 11 , wherein fitness is represented as a value between zero and one.

13. The method of claim 12 , wherein the value between zero and one represents a probability that a human recruiter would invite the specific candidate for an interview for the role.

14. The method of claim 11 , further comprising performing a statistical test to confirm that measures to limit bias are effective.

15. The method of claim 14 , wherein parsed resume language and a job description associated with a job for which the specific candidate is applying are tokenized, converted into an integer, and processed by a word embedding model.

16. The method of claim 11 , wherein a scoring process applied by the deep learning model utilizes parallel processing of a plurality of recent work experiences associated with the specific candidate.

17. The method of claim 16 , wherein the deep learning model utilizes parallel processing by processing the resume associated with the specific candidate and extracting relevant information-rich natural language.

18. The method of claim 11 , wherein a capacity of the deep learning model to develop bias against a particular gender is limited.

19. The method of claim 11 , wherein demographic and personally identifiable information are excluded from extracting relevant information-rich natural language.

20. A non-transitory computer-readable storage medium storing computer-executable instructions that, when executed, cause performance of a method comprising:

a plurality of parallel multicore processing devices, within a multicore system processing device, training a deep learning model based on training data that is stored within a storage device and that comprises at least one positive example and at least one negative example, the at least one positive example representing a candidate who would be invited to an interview and the at least one negative examples representing a candidate without at least one of a relevant skill and an experience who would not be invited to an interview; and

applying the deep learning model to score a level of fitness associated with a specific candidate for a role,

wherein

an interface bus operably couples the multicore system processing device, the plurality of parallel multicore processing devices, the storage device, and an interface device that provides connection to the multicore system processing device,

the training data is generated by stripping a most recent role from a resume associated with a candidate who has a plurality of instances of work experience listed in a listing of professional history,

the at least one positive example is generated by concatenating the most recent role with a remainder of preceding work experience,

the at least one negative example is generated by concatenating a random job description from a randomly selected and different resume with the remainder of preceding work experience based on an unlikelihood that the candidate will be a good match for the random job description,

at least one indicative pronoun is excluded from a process of extracting relevant information-rich natural language, thereby limiting capacity associated with the deep learning model to develop bias against a demographic group, and

the deep learning model comprises a plurality of parallel deep learning natural language model paths comprising sigmoid output neurons at an output stage of at least one portion of the deep learning model with rectified linear unit neurons used between internal layers within the deep learning model.