The largest online recruiter – HeadHunter – has implemented machine learning algorithms to search for personnel. Boris Volfson, CBDO at the company, has told how the artificial intelligence system operates and why it is better than previously used algorithms.
Why machine learning?
Nowadays, investors are putting up large sums in the automation of personnel search processes and applications for company operations and corporate training. Previous year, one has financed a lot of projects related with machine learning and artificial intelligence technologies, as they are not based on hardcoded parameters but rather can take into account context and learn from obtained data. Such systems can thoroughly scan resumes of candidates and offer them the most appropriate job vacancies depending on their work experience and desired labor conditions.
According to Boris Volfson, machine learning systems don’t have disadvantages of already used algorithms. Namely:
- wrong tracking of search filters;
- providing results with irrelevant parameters;
- providing search results without seamless ranking.
Current operation concept of HeadHunter search systems
Machine learning systems analyze not only user resumes and filters but also the user action history: vacancy responses, adding to favorites, switching to the page, time of studying company’s offers, etc. Parameters used for selection take about 100 GB. Besides, the results are thoroughly classified from the most appropriate to those just formally complying with desired conditions.
HeadHunter search system includes three filter blocks:
1. Heuristic – makes selection based on the region and professional sphere.
2. Fast model – uses additional features in order to sort out inappropriate vacancies.
3. Complex ranking algorithm – uses approximately 200 features to show the most appropriate vacancies in the correct order.
Thus, the search system is designed so that each following algorithm compensates disadvantages of the previous one.
Boris Volfson has told that the company implements only efficient solutions that can be fast and cheap realized. So, the implementation of the latest innovations costs more than 2 million rubles. HeadHunter representative believes that this sum will be paid off in six months.
This year, the company is also going to integrate the technology of behavioral characteristics analysis and improve accuracy of search hits by means of tracking the history of previous requests. Perhaps, in future, the system will work with photos and portfolios. But it's a long way off.