Activity-based inference of title preferences

A system, comprising:

• one or more processors;
• and
• memory storing instructions that, when executed by the one or more processors, cause the system to:
• determine features and labels for use in training a machine learning model, the features including a set of positively labeled features comprising at least i) job titles of job postings associated with job applications submitted by a first set of candidates, ii) job titles of job postings selected for viewing by a candidate in the first set of candidates, and iii) job titles explicitly specified as preferred job titles by candidates in a first set of candidates, and the features including a set of negatively labeled features comprising at least randomly selected job titles;
• input the set of positively labeled features and the set of negatively labeled features as training data for training the machine learning model to predict for each candidate in a second set of candidates a preferred job title;
• subsequent to training the machine learning model, apply the trained machine learning model to features for the second set of candidates to produce predictions of preferred job titles for the second set of candidates;
• store the predictions of preferred job titles in association with the second set of candidates;
• and
• generate job recommendations for each candidate in the second set of candidates based on the predictions of preferred job titles, by:
• inputting the predictions of preferred job titles with the features for the second set of candidates into a second, pre-trained machine learning model;
• receiving, as output from the second, pre-trained machine learning model, scores representing likelihoods of the second set of candidates applying to jobs associated with job postings;
• and
• generating the job recommendations for the second set of candidates based on rankings of the score;
• detect an interaction, by each candidate, with the job recommendation;
• and
• use data representing the detected interaction as additional training data to update the machine learning model to make improved predictions.

A non-transitory computer-readable storage medium storing instructions that when executed by a computer cause the computer to perform a method, the method comprising:

• determining, by one or more computer systems, features and labels for use in training a machine learning model, the features including a set of positively labeled features comprising at least i) job titles of job postings associated with job applications submitted by candidates in the first set of candidates, ii) job titles of job postings selected for viewing by candidates in the first set of candidates, and iii) job titles explicitly specified as preferred job titles by candidates in a first set of candidates, and the features including a set of negatively labeled features comprising at least randomly selected job titles;
• inputting, by the one or more computer systems, the set of positively labeled features and the set of negatively labeled features as training data for training the machine learning model to predict for each candidate in a second set of candidates a preferred job title;
• subsequent to training the machine learning model, applying, by the one or more computer systems, the trained machine learning model to features for the second set of candidates to produce predictions of preferred job titles for each candidate in the second set of candidates;
• storing the predictions of preferred job titles in association with the second set of candidates;
• generating job recommendations fora each candidate in the second set of candidates, by:
• inputting the predictions of preferred job titles with the features for the second set of candidates into a second, pre-trained machine learning model;
• receiving, as output from the second, pre-trained machine learning model, scores representing likelihoods of the second set of candidates applying to jobs associated with job postings;
• and
• generating the job recommendations for the second set of candidates based on rankings of the scores;
• detecting an interaction, by each candidate, with the job recommendation;
• and
• using data representing the detected interaction as additional training data to update the machine learning model to make improved predictions of preferred job titles.