Transfer learning for seniority modeling label shortage
Techniques for using transfer learning to address label data shortage in seniority modeling for an online service are disclosed herein. In some embodiments, a computer-implemented method comprises training an initialized neural network using training examples comprising profile data and labels for the profile data, where each label comprises a standardized position title, and the training of the initialized neural network forms a pre-trained neural network. Next, the computer system may train the pre-trained neural network using training examples comprising profile data and labels for the profile data, where the labels comprise a position seniority, and the training of the pre-trained neural network forms a fine-tuned neural network. The computer system may then compute the position seniority for a user based on profile data of the user using the fine-tuned neural network, and use the position seniority of the user in an application of an online service.
1 . A computer-implemented method performed by a computer system having a memory and at least one hardware processor, the computer-implemented method comprising:
selecting a first plurality of reference users and a second plurality of reference users;
filtering out inactive users from the first plurality of reference users or the second plurality of reference users based on interaction data stored in a database;
training an initialized neural network using a first plurality of training examples, the first plurality of training examples comprising profile data of the first plurality of reference users of an online service and a first label for the profile data of each reference user in the first plurality of reference users, the first label comprising a standardized position title, the training of the initialized neural network forming a pre-trained neural network;
training the pre-trained neural network using a second plurality of training examples, the second plurality of training examples comprising profile data of the second plurality of reference users of the online service and a second label for the profile data of each reference user in the second plurality of reference users, the second label comprising a position seniority, the training of the pre-trained neural network forming a fine-tuned neural network;
computing the position seniority for a first target user of the online service based on profile data of the first target user using the fine-tuned neural network;
using the position seniority of the first target user in an application of the online service to select a content item; and
displaying, on a user interface of a computing device, the content item.
2 . The computer-implemented method of claim 1 , wherein the initialized neural network comprises a deep learning model.
3 . The computer-implemented method of claim 2 , wherein the deep learning model comprises a transformer.
4 . The computer-implemented method of claim 1 , wherein the profile data of each reference user in the first plurality of reference users, the profile data of each reference user in the second plurality of reference users, and the profile data of the first target user each comprise a position title, a company identification, a position description, and a position duration.
5 . The computer-implemented method of claim 1 , wherein the training the pre-trained neural network comprises using a semi-supervised learning process to train the pre-trained neural network using the second plurality of training examples as labeled training data and a third plurality of training examples as unlabeled training data.
6 . The computer-implemented method of claim 5 , wherein the semi-supervised learning process comprises a virtual adversarial training process, and the third plurality of training examples comprises perturbations of the profile data of each reference user in the second plurality of reference users.
7 . The computer-implemented method of claim 1 , wherein the using the position seniority of the first target user in the application of the online service comprises:
selecting a job posting from a plurality of job postings based on a determination that the position seniority of the first target user is stored in a database in association with the job posting; and
displaying, on a computing device of the first target user, the selected job posting.
8 . The computer-implemented method of claim 1 , wherein the using the position seniority of the first target user in the application of the online service comprises:
determining that a search query submitted by a second target user includes the position seniority of the first target user;
selecting a profile of the first target user based on the determining that the search query includes the position seniority of the first target user; and
displaying, on a computing device of the second target user, a user interface element that identifies the profile of the first target user based on the selecting the profile of the first target user.
9 . The computer-implemented method of claim 1 , wherein the using the position seniority of the first target user in the application of the online service comprises:
selecting an online course from a plurality of online courses based on a determination that the position seniority of the first target user is stored in a database in association with the online course; and
displaying, on a computing device of the first target user, the selected online course.
10 . A system comprising:
at least one hardware processor; and
a non-transitory machine-readable medium embodying a set of instructions that, when executed by the at least one hardware processor, cause the at least one hardware processor to perform operations, the operations comprising:
select a first plurality of reference users and a second plurality of reference users;
filter out inactive users from the first plurality of reference users or the second plurality of reference users based on interaction data stored in a database;
train an initialized neural network using a first plurality of training examples, the first plurality of training examples comprising profile data of the first plurality of reference users of an online service and a label for the profile data of each reference user in the first plurality of reference users, the label for the first profile data comprising a standardized position title, the training of the initialized neural network forming a pre-trained neural network;
train the pre-trained neural network using a second plurality of training examples, the second plurality of training examples comprising profile data of the second plurality of reference users of the online service and a label for the second profile data of each reference user in the second plurality of reference users, the label for the second profile data comprising a position seniority, the training of the pre-trained neural network forming a fine-tuned neural network;
compute the position seniority for a first target user of the online service based on profile data of the first target user using the fine-tuned neural network;
use the position seniority of the first target user in an application of the online service to select a content item; and
display, on a user interface of a computing device, the content item.
11 . The system of claim 10 , wherein the initialized neural network comprises a deep learning model.
12 . The system of claim 11 , wherein the deep learning model comprises a transformer.
13 . The system of claim 10 , wherein the profile data of each reference user in the first plurality of reference users, the profile data of each reference user in the second plurality of reference users, and the profile data of the first target user each comprise a position title, a company identification, a position description, and a position duration.
14 . The system of claim 10 , wherein the training the pre-trained neural network comprises using a semi-supervised learning process to train the pre-trained neural network using the second plurality of training examples as labeled training data and a third plurality of training examples as unlabeled training data.
15 . The system of claim 14 , wherein the semi-supervised learning process comprises a virtual adversarial training process, and the third plurality of training examples comprises perturbations of the profile data of each reference user in the second plurality of reference users.
16 . The system of claim 10 , wherein the using the position seniority of the first target user in the application of the online service comprises:
select a job posting from a plurality of job postings based on a determination that the position seniority of the first target user is stored in a database in association with the job posting; and
display, on a computing device of the first target user, the selected job posting.
17 . The system of claim 10 , wherein the using the position seniority of the first target user in the application of the online service comprises:
determine that a search query submitted by a second target user includes the position seniority of the first target user;
select a profile of the first target user based on the determining that the search query includes the position seniority of the first target user; and
display, on a computing device of the second target user, a user interface element that identifies the profile of the first target user based on the selecting the profile of the first target user.
18 . The system of claim 10 , wherein the using the position seniority of the first target user in the application of the online service comprises:
select an online course from a plurality of online courses based on a determination that the position seniority of the first target user is stored in a database in association with the online course; and
display, on a computing device of the first target user, the selected online course.
19 . A non-transitory machine-readable medium embodying a set of instructions that, when executed by at least one hardware processor, cause the at least one hardware processor to perform operations, the operations comprising:
select a first plurality of reference users and a second plurality of reference users;
filter out inactive users from the first plurality of reference users or the second plurality of reference users based on interaction data stored in a database;
train an initialized neural network using a first plurality of training examples, the first plurality of training examples comprising profile data of the first plurality of reference users of an online service and a label for the first profile data of each reference user in the first plurality of reference users, the label for the first profile data comprising a standardized position title, the training of the initialized neural network forming a pre-trained neural network;
train the pre-trained neural network using a second plurality of training examples, the second plurality of training examples comprising profile data of the second plurality of reference users of the online service and a label for the second profile data of each reference user in the second plurality of reference users, the label for the second profile data of each reference user comprising a position seniority, the training of the pre-trained neural network forming a fine-tuned neural network;
compute the position seniority for a first target user of the online service based on profile data of the first target user using the fine-tuned neural network;
use the position seniority of the first target user in an application of the online service to select a content item; and
display, on a user interface of a computing device, the content item.
20 . The non-transitory machine-readable medium of claim 19 , wherein the initialized neural network comprises a deep learning model.