When will "AI technology + recruitment scenario" become a reality?

Nowadays, various applications combining "AI technology + recruitment scenarios" have begun to become a reality. In the future, the application of AI technology in the recruitment industry is likely to completely replace professional HR. We can wait and see.

Recently, Lagou.com announced the launch of a new corporate and recruiter identity review mechanism, and introduced Baidu AI collaboration section, introducing facial recognition and other intelligent methods to review and verify corporate HR identities and qualifications. In the future, it will also use Artificial intelligence technology verifies business licenses, work licenses, employment certificates, and is used for risk control on resumes, chat records, etc.

Since AlphaGo defeated the world champion of Go last year, the commercial application of AI technology has accelerated significantly. In fields such as transportation, home services, medical care, commerce, and recruitment, it has begun to replace manual work in completing some tasks. While some practitioners enjoy the convenience brought by AI technology, they also feel pressure.

Robots can screen resumes faster than humans, but they are not flexible enough

A few months ago, a competition hosted by Liepin was called a "human-machine war" in the field of recruitment. The answer is given by the score. This time the challengers came from five senior HR and headhunters from Internet companies. Both parties had to quickly select 10 resumes from 37 million resumes that best matched the needs of the recruitment positions.

This is a competition based on matching job requirements and candidates, including technical positions and product positions. After completing the screening and matching of resumes, the jury needs to score the contestants from six dimensions: function, skills, industry, salary, education, and regional matching. The one with the highest total score (full score of 25 points) will be deemed to win the competition. games.

As the blue screen lights up, the score between humans and AI robots is 18.96:18.60. The results show that it only took the AI ??robot 0.0152 seconds to complete the entire competition, which is 63882 times the average speed of humans; in terms of position matching and regional matching, the matching efficiency of the robot is higher than that of humans; in terms of skill matching, the two It was a tie.

Although it lost to humans with a weak score of 0.36, the AI ??robot still exceeded the expectations of Dai Kebin, founder and CEO of Liepin.com, in terms of "matching people to jobs and understanding people" . “In the task of resume search, areas such as region and salary are relatively simple and direct conditions, so the algorithm can be implemented through simple logic without making mistakes; in terms of industry background and skill requirements, through the use of neural networks and natural language processing, etc. Technology and algorithms have been able to perform more accurate understanding and similarity judgments." Shan Yi, the designer of this Bole robot and chief data officer of Liepin, told China Youth Daily·China Youth Online reporter that the current AI robot has been able to achieve better results. By understanding most of the explicit requirements, such as functions, skills, salary, education, region, etc., the matching level of the algorithm can be comparable to that of professional recruiters; however, in terms of implicit conditions that require face-to-face communication, such as culture, values, temperament, etc. Algorithms cannot yet replace human communication and judgment.

In the results, the gap between robots and humans is the largest in terms of academic qualification matching. The main reason for this gap is that robots cannot identify which academic qualification type "junior college to bachelor's degree" belongs to. This also reflects the limited flexibility of robot thinking. In this regard, Shan Yi explained that for the job requirements of "bachelor's degree or above", when designing the robot algorithm, "junior college to bachelor's degree" was determined to be eligible; but in fact, in the eyes of many recruiters (especially high-end headhunters), "junior college to bachelor's degree" is considered to be eligible. It's not as good as an undergraduate degree. Therefore, this screening result triggered different opinions from several judges present. "Robots are still not enough to make personalized selections for soft indicators of talent based on corporate and HR preferences."

Members of the judging panel One, Alibaba Entertainment Recruitment Expert Zhou Xiaolei believes that in terms of large-scale selection of candidates from massive resumes, the gap between robots and humans is almost negligible, and AI robots can improve the overall recruitment efficiency.

It is nothing new that robots can screen resumes much faster than humans. According to reports, in March this year, in an industry competition organized by SourceCon, a well-known headhunting company in North America, Brilent, a robot based on artificial intelligence that screens and rates job applicants, screened 5,500 resumes in just 3.2 seconds. A suitable candidate was selected, and the accuracy ranked third among the contestants. This team is based on the data structuring and field-refined matching experience accumulated by members at Facebook, and uses AI technology to sort job seekers who meet the "person-job matching", freeing HR from mechanical and tedious resume screening. Can focus more on the subsequent interview selection process.

High-precision job matching: Let artificial intelligence learn how HR does recruitment

In June of this year, Dai Kebin announced that recruitment should explore big data and artificial intelligence. Further improve recruitment efficiency and enrich the recruitment ecosystem; on September 12, Sinovation Ventures Chairman and CEO Kai-fu Lee stated at the "2017 China Artificial Intelligence Summit" that for artificial intelligence to truly replace human labor, it also needs sufficient data volume and accuracy. scene as a premise.

From simple listing and classification of job information to job-job matching systems based on big data mining, in recent years, many industry-wide online recruitment companies have formed their own "talent pools" through data accumulation. ". On this basis, matching people and positions based on resume information or job requirements has become the main application of artificial intelligence technology in the recruitment field at this stage.

As a start-up Internet company that focuses on providing mobile recruitment services to enterprises, Wang Weibao, CEO of Qianxun Mobile Recruitment, once pointed out in an article that for the same position, with sufficient data volume and manual labor Under intervention, whether the text matching technology based on JD (job description) and resume can reach the level of effective selection, and whether it can make the system (machine learning process) output results similar to human selection, we can measure the validity of AI. This process includes two dimensions: matching resume information with job requirements, and matching candidates with company positions.

The application of "AI+ recruitment" in Liepin started in 2014. According to Shan Yi, during the process of R&D and exploration, they found that it is necessary to judge "whether high-precision recommendations can be made based on user behavior data, position and resume content", "whether the job requirements and industry fields are understood in detail", "whether the "Can cross-professional and cross-industry job changers achieve personalized job recommendations based on their experience and abilities?" This has become a challenge that artificial intelligence needs to face when applying it to the recruitment industry. It is understood that most current recruitment websites can use data capture and artificial intelligence semantic analysis to achieve matching in the first dimension, while precise matching in the second dimension puts forward higher requirements for personalized recommendations by artificial intelligence.

“Machine screening first requires the accumulation of knowledge and experience, and the ability to interpret semantics; secondly, it is necessary to learn the behavior of HR and headhunters and understand how they match people and positions.” In Shan Yi’s view, the latter is more The key is to use deep profiling, semantic matching and personalized recommendations based on HR preferences.

In order to make artificial intelligence "think" like HR, Shan Yi, who has 17 years of experience in data mining and system development in domestic and foreign companies, led the team and required machines to actively learn from humans. They designed an algorithm system that is open to the HR of the customer company, allowing HR to give feedback on the suitability of the "resume recommended by the machine". When the feedback behavior and data have accumulated to a certain extent, the machine can learn from the differences in selections. Understand the preferences of different HRs and form a matching model that can handle thousands of different industries and functions. Even for the same position, different recommendation results may be given.

“Now my robot Bole is already working for HR.” According to Shan Yi, in terms of the amount of resumes recommended by Liepin, the business volume recommended by the algorithm far exceeds 50% of the total business volume. %, the accuracy is the same as that of general headhunters.

This intelligent recommendation algorithm is called a "smart system" by them. "Let the robot learn how HR does recruitment, instead of directly telling him 'how to do this, how to do that'." Shan Yi emphasized.

Break the information "gap" between companies and job seekers

Artificial intelligence is increasingly popular in the recruitment industry, and behind it is the growing demand for recruitment business.

According to iResearch statistics, it is expected that by 2018, the number of small and medium-sized enterprises in China will exceed 87 million, and the number of job seekers is expected to exceed 160 million. Searching a large number of resumes and screening potential candidates to match positions in various industries (especially non-high-end positions) has become the most repetitive task of headhunting and HR recruitment.

According to media reports, from July 2016 to June 2017, the application of AI has gradually spread to the recruitment process in 68 countries around the world. In the past year or so, Unilever has Attempts to use artificial intelligence to recruit employees in North America include using algorithms to screen resumes, game tests, face recognition, etc., without even requiring the participation of human interviewers. Domestically, as of July this year, there are no fewer than 10 start-up companies claiming to be leaders in artificial intelligence + recruitment, striving to use technology to solve the problems of high recruitment labor costs, low actual conversion rates, and poor information between the recruitment and job-seeking sides.

The method adopted by Niu Zhipin, an Internet intelligent recruitment platform founded in 2016, is to accurately integrate positions and resumes of enterprises (especially small and medium-sized enterprises) through resume decomposition and personalized recommendations. matching; Teamable, a start-up recruitment platform, uses AI algorithms to mine applicants’ social network data and strives to start from social records to create an accurate closed-loop talent recommendation; while mini-schools, which are vertical in the field of campus recruitment, also use data mining and AI algorithms to design intelligent The matching model automatically screens and recommends resumes for different companies and provides suggestions for job seekers with 0 to 3 years of professional experience.

However, Shan Yi has always emphasized that in the recruitment industry, artificial intelligence is just a tool and cannot replace humans. Instead, it “helps humans make more accurate and well-founded judgments, allowing headhunters and HR to engage in More valuable and creative work. "To some extent, the application of artificial intelligence in the field of recruitment is easier to promote in the recruitment of mid- to low-end talents or positions that do not require high industry experience.

A headhunting consultant who focuses on mid-to-high-end talents in the consumer goods industry admitted frankly to a reporter from China Youth Daily·China Youth Online that despite the need to frequently change keywords every day to search for resumes in the talent pool and make more than a dozen calls with candidates, However, this process requires professional communication experience with distinctive social attributes such as adaptability and adaptability, which is difficult for artificial intelligence to achieve.

“The more high-end talents and important executive positions, the more cautious the company’s HR is, and the more professional headhunting is needed.” Wang Guangyuan, general manager of Beijing Yipin Century Management Consulting Company, also proposed, such as for Candidates for functional positions such as product managers must also be considered through soft indicators such as product design ideas. "This is a weakness for machines." Dai Kebin pointed out that "artificial intelligence cannot immediately replace headhunting. The fundamental reason is insufficient data on both supply and demand sides." He mentioned that job-job matching has put forward relatively new requirements for job seeker data held by recruitment platforms and demand descriptions provided by enterprises. High demand, coupled with the dynamics and flexibility of recruitment, while improving intelligent products, "human factors play an irreplaceable role in artificial intelligence recruitment."

In the current autumn recruitment season Faced with the influx of fresh graduates into the job market and the unbalanced supply and demand in the campus recruitment market, Shanyi expects that in the future, AI technology can provide employment guidance suitable for fresh graduates in advance by analyzing the supply and demand data of market positions and mining the individual needs of enterprises. , broaden their horizons and choices, and match the needs of enterprises and fresh graduates more efficiently.