Wants to Work in Suwon but Recommended Jobs in Gangwon-do... The Reality of the '12.8 Billion Won' AI Matching System

[Image source=Yonhap News]

[Asia Economy Reporter Lee Ji-eun] The Ministry of Employment and Labor's AI-based job matching system on the employment information site 'Worknet,' built with a budget of 12.8 billion KRW, has shown very poor performance.

According to the audit report on the "Establishment and Management Status of the Employment Referral Information Network," released by the Board of Audit and Inspection on the 12th, the application rate for the AI matching system was 9.45% during the first phase (July 2020 to July 2021) and 12.99% during the second phase (July 2021 to October 2021), falling far short of the initially predicted 59% and 70.2%, respectively.

This was even lower than the application rates of Worknet's existing service, the "Big Data Job Recommendation Service," which were 15.22% and 18.36%.

Analysis of approximately 1.67 million job postings viewed and 200,000 job applications submitted from job information recommended by the AI matching system revealed that the correlation between matching scores and job applications was 0.0061 in the first phase and 0.0768 in the second phase, indicating almost no relationship.

There were also numerous cases where jobs in regions not preferred by job seekers were recommended. Using Suwon City in Gyeonggi-do, which had the highest number of views, as a sample, an analysis of 6,977 recommended job postings showed that 1,020 cases (14.6%) did not match the job seeker's preferred region with the AI-recommended job location.

In some cases, jobs in Hoengseong-gun and Taebaek-si in Gangwon-do were recommended to job seekers who had set Suwon as their preferred location.

The Board of Audit and Inspection recommended that the Minister of Employment and Labor and the President of the Korea Employment Information Service develop improvement measures to enhance the algorithm, including reviewing the appropriateness of job recommendation criteria and recommended target regions.