OFFLINE COMPUTATION OF PARTIAL JOB RECOMMENDATION SCORES
A recommendation system is configured to generate relevance scores for job postings with respect to a subject member profile offline, preemptively, prior to detecting an indication that the associated subject member is logged in into the online social network system. These offline relevance scores are stored in a key-value store for future use at runtime, in response to detecting that the subject member is logged in into the online social network system. The online component of the recommendation system obtains from the key-value store the offline relevance scores, together with other features generated at runtime, and uses these scores as input when executing an online relevance model to generate respective final relevance scores for job postings that are potentially of interest to the subject member. The resulting scores are used to determine which job postings are to be recommended to the subject member.
1 . A computer-implemented method comprising:
generating and storing for future use offline relevance scores for candidate job postings provided in an online social network system with respect to a subject member profile maintained in the online social network system, a relevance score generated with respect to the subject member profile and a certain job posting from the candidate job postings reflecting probability of a member represented by the subject member profile applying for a job represented by the certain job posting;
in response to detecting that a subject member represented by the subject member profile is logged in into the online social network system, using at least one processor:
generating online features with respect to the subject member profile and respective job postings, the respective job postings comprising one or more postings from the candidate job postings,
executing an online relevance model using the online features together with the offline relevance scores as input into the online relevance model to generate respective runtime relevance scores for the respective job postings, and
from the respective job postings selecting a presentation set based on the respective online relevance values generated for each job posting from the respective job postings; and
causing displaying of a reference to a job posting from the presentation set on a display device associated with the subject member.
2 . The method of claim 1 , wherein the generating of the offline relevance scores comprises utilizing field data extracted from the subject member profile and the candidate job postings.
3 . The method of claim 1 , wherein the generating of the online features comprises generating a relative fitness value, the relative fitness value indicating a likelihood that the subject member is hired for a job represented by a particular job posting from the candidate set of job postings in relationship to a likelihood that another member represented by another profile from the candidate set of member profiles is hired for the particular job.
4 . The method of claim 3 , wherein the generating of the online features comprises utilizing a cap value that limits a number of member profiles for which the particular job can be recommended.
5 . The method of claim 4 , comprising utilizing a number of further member profiles, for which the particular job has been recommended.
6 . The method of claim 1 , wherein the generating of the offline relevance scores comprises executing an offline relevance model.
7 . The method of claim 6 , wherein the online relevance model utilizes an approach that is different from an approach utilized by the offline relevance model.
8 . The method of claim 6 , comprising training the offline relevance model using training data collected with respect to members interacting with job postings in the online social network system;
9 . The method of claim 1 , comprising training the online relevance model using the online features.
10 . The method of claim 1 , comprising training the offline relevance model using content-based features.
11 . A computer-implemented system comprising:
an offline relevance scores generator, implemented using at least one processor, to generate offline relevance scores for candidate job postings provided in an online social network system with respect to a subject member profile maintained in the online social network system, a relevance score generated with respect to the subject member profile and a certain job posting from the candidate job postings reflecting probability of a member represented by the subject member profile applying for a job represented by the certain job posting;
a presentation set generator, implemented using at least one processor, in response to detecting that a subject member represented by the subject member profile is logged in into the online social network system, to:
generate online features with respect to the subject member profile and respective job postings, the respective job postings comprising one or more postings from the candidate job postings,
execute an online relevance model using the online features together with the offline relevance scores as input into the online relevance model to generate respective runtime relevance scores for the respective job postings, and
from the respective job postings select a presentation set based on the respective online relevance values generated for each job posting from the respective job postings; and
a presentation module, implemented using at least one processor, to cause displaying of a reference to a job posting from the presentation set on a display device associated with the subject member.
12 . The system of claim 11 , wherein the presentation set generator generates the offline relevance scores utilizing field data extracted from the subject member profile and the candidate job postings.
13 . The system of claim 11 , wherein the presentation set generator is to generate a relative fitness value, the relative fitness value indicating a likelihood that the subject member is hired for a job represented by a particular job posting from the candidate set of job postings in relationship to a likelihood that another member represented by another profile from the candidate set of member profiles is hired for the particular job, the relative fitness value is a feature from the online features.
14 . The system of claim 13 , wherein the presentation set generator is to generate a feature from the online features utilizing a cap value that limits a number of member profiles for which the particular job can be recommended.
15 . The system of claim 14 , wherein the presentation set generator is to generate a feature from the online features a number of further member profiles, for which the particular job has been recommended.
16 . The system of claim 11 , wherein the offline relevance scores generator executes an offline relevance model to generate the offline relevance scores comprises.
17 . The system of claim 16 , wherein the online relevance model utilizes an approach that is different from an approach utilized by the offline relevance model.
18 . The system of claim 16 , wherein the offline relevance model is trained using training data collected with respect to members interacting with job postings in an online social network system.
19 . The system of claim 11 , wherein the online relevance model is trained using the online features.
20 . A machine-readable non-transitory storage medium having instruction data executable by a machine to cause the machine to perform operations comprising:
generating and storing for future use offline relevance scores for candidate job postings provided in an online social network system with respect to a subject member profile maintained in the online social network system, a relevance score generated with respect to the subject member profile and a certain job posting from the candidate job postings reflecting probability of a member represented by the subject member profile applying for a job represented by the certain job posting;
in response to detecting that a subject member represented by the subject member profile is logged in into the online social network system:
generating online features with respect to the subject member profile and respective job postings, the respective job postings comprising one or more postings from the candidate job postings,
executing an online relevance model using the online features together with the offline relevance scores as input into the online relevance model to generate respective runtime relevance scores for the respective job postings, and
from the respective job postings selecting a presentation set based on the respective online relevance values generated for each job posting from the respective job postings; and
causing displaying of a reference to a job posting from the presentation set on a display device associated with the subject member.