Hashing query and job posting features for improved machine learning model performance
In an example, a first hash function is performed on job posting features extracted from a job posting to obtain hashed job posting features. The hashed job posting features are stored in a forward-index corresponding to the job posting in the database. When a job search query is received from a first member of a social networking service, job search query features are extracted from the job search query and a second hash function is performed on the job search query features. The hashed job posting features and the hashed job search query features are fed to a job posting result ranking model trained via a machine learning algorithm to compare the hashed job posting features to the hashed job search query features to generate an application likelihood score indicating a likelihood that the first member will apply for a job corresponding to the job posting.
1. A system comprising:
a computer-readable medium having instructions stored thereon, which, when executed by a processor, cause the system to:
retrieve, using a first feature extractor, from a job posting database, a plurality of forward indexes, each forward index representing a different job posting;
extract, using the first feature extractor, job posting features from a first of the plurality of forward indexes;
perform, using a first hashing component, a first hash function on to the job posting features to obtain hashed job posting features;
insert, using the first hashing component, the hashed job posting features in a posting features field of the first of the plurality of forward indexes;
storing, using the first hashing component, the first of the plurality of forward indexes in the job posting database;
receive, by a search engine, a job search query from a first member of a social networking service;
perform, by the search engine, a search of the job posting database to identify one or more forward indexes of job postings that match the job search query;
extract job search query features from the job search query;
perform a second hash function on the job search query features;
feed the hashed job posting features and the hashed job search query features to a job posting result ranking model trained via a machine learning algorithm to compare the hashed job posting features to the hashed job search query features for the one or more identified forward indexes to generate an application likelihood score indicating a likelihood that the first member will apply for a job corresponding to the job posting; and
rank the job posting among other job postings based on the application likelihood score.
2. The system of claim 1 , wherein the first hash function and the second hash function are identical.
3. The system of claim 1 , wherein the instructions further cause the system to embed the hashed job search query features in query-coefficient vectors and store the query-coefficient vectors in a second database.
4. The system of claim 1 , wherein the job posting result ranking model is a combination of a global ranking model and a query-based model.
5. The system of claim 1 , wherein the job posting result ranking model uses logistic regression.
6. The system of claim 1 , wherein the job posting result ranking model is optimized via alternating optimization using parallelized coordinate descent.
7. The system of claim 1 , wherein the job posting result ranking model is optimized by optimizing for global features and per-feature queries for each query while holding all other variables fixed.
8. A computerized method comprising:
retrieving, using a first feature extractor, from a job posting database, a plurality of forward indexes, each forward index representing a different job posting;
extracting, using the first feature extractor, job posting features from a first of the plurality of forward indexes;
performing, using a first hashing component, a first hash function on to the job posting features to obtain hashed job posting features;
inserting, using the first hashing component, the hashed job posting features in a posting features field of the first of the plurality of forward indexes;
storing, using the first hashing component, the first of the plurality of forward indexes in the job posting database;
receiving, by a search engine, a job search query from a first member of a social networking service;
performing, by the search engine, a search of the job posting database to identify one or more forward indexes of job postings that match the job search query;
extracting job search query features from the job search query;
performing a second hash function on the job search query features;
feeding the hashed job posting features and the hashed job search query features to a job posting result ranking model trained via a machine learning algorithm to compare the hashed job posting features to the hashed job search query features for the one or more identified forward indexes to generate an application likelihood score indicating a likelihood that the first member will apply for a job corresponding to the job posting; and
ranking the job posting among other job postings based on the application likelihood score.
9. The method of claim 8 , wherein the first hash function and the second hash function are identical.
10. The method of claim 8 , further comprising: embedding the hashed job search query features in query-coefficient vectors and storing the query-coefficient vectors in a second database.
11. The method of claim 8 , wherein the job posting result ranking model is a combination of a global ranking model and a query-based model.
12. The method of claim 8 , wherein the job posting result ranking model uses logistic regression.
13. The method of claim 8 , wherein the job posting result ranking model is optimized via alternating optimization using parallelized coordinate descent.
14. The method of claim 8 , wherein the job posting result ranking model is optimized by optimizing for global features and per-feature queries for each query while holding all other variables fixed.
15. A non-transitory machine-readable storage medium comprising instructions, which when implemented by one or more machines, cause the one or more machines to perform operations comprising:
retrieving, using a first feature extractor, from a job posting database, a plurality of forward indexes, each forward index representing a different job posting;
extracting, using the first feature extractor, job posting features from a first of the plurality of forward indexes;
performing, using a first hashing component, a first hash function on to the job posting features to obtain hashed job posting features;
inserting, using the first hashing component, the hashed job posting features in a posting features field of the first of the plurality of forward indexes;
storing, using the first hashing component, the first of the plurality of forward indexes in the job posting database;
receiving, by a search engine, a job search query from a first member of a social networking service;
performing, by the search engine, a search of the job posting database to identify one or more forward indexes of job postings that match the job search query;
extracting job search query features from the job search query;
performing a second hash function on the job search query features;
feeding the hashed job posting features and the hashed job search query features to a job posting result ranking model trained via a machine learning algorithm to compare the hashed job posting features to the hashed job search query features for the one or more identified forward indexes to generate an application likelihood score indicating a likelihood that the first member will apply for a job corresponding to the job posting; and
ranking the job posting among other job postings based on the application likelihood score.
16. The non-transitory machine-readable storage medium of claim 15 , wherein the first hash function and the second hash function are identical.
17. The non-transitory machine-readable storage medium of claim 15 , further comprising: embedding the hashed job search query features in query-coefficient vectors and storing the query-coefficient vectors in a second database.
18. The non-transitory machine-readable storage medium of claim 15 , wherein the job posting result ranking model is a combination of a global ranking model and a query-based model.
19. The non-transitory machine-readable storage medium of claim 15 , wherein the job posting result ranking model uses logistic regression.
20. The non-transitory machine-readable storage medium method of claim 15 , wherein the job posting result ranking model is optimized via alternating optimization using parallelized coordinate descent.