The Examination and Selection Department used AI to review the question syllabus, and the accuracy of classification of the basic medical examination questions for nursing teachers reached 79.6%.

The biggest fear when preparing for the national examination is that the scope is boundless. Whether the syllabus can keep up with changes in majors and practices directly affects whether candidates can accurately prepare for the examination. The Examination and Selection Department has tried to use AI to organize past test questions and examine the core knowledge covered in the syllabus. Taking the "Basic Medicine" exam for nurses as an example, the AI ​​test question classification hit rate reached 79.6%. In the future, it is expected to provide empirical evidence for updating the proposition syllabus, reducing the preparation pressure and cost of candidates; however, AI is responsible for the preliminary sorting, and the final judgment is still left to experts.

Ouyang Anqi, a special committee member of the Test Development Department of the Examination and Selection Department, and Fan Zhenyou, a project researcher of the Test Research Center, wrote in the 41st issue of the "National Human Resources Forum" that the question outline outlines the preparation scope for candidates and allows the question committee members to follow it to avoid differences in the difficulty of test questions due to different academic backgrounds. The Examination and Selection Department has systematically established a proposition syllabus since 2005, covering thousands of examination subjects.

Please read on...

In the early stages of the establishment of the proposition syllabus, there are more than 700 subjects and 3,000 examination subjects for civil servants and professional personnel examinations. Each syllabus often takes 3 to 6 months from data collection, discussion to review. The current revisions are mostly in conjunction with the consolidation of subjects, or partial adjustments after the exam and question bank committee members find that the content is out of date, and less comprehensive inspections are conducted to see whether it is in line with the current core knowledge.

The examination and selection department therefore tried to introduce AI. The first case uses 610 sample questions and nearly 1,400 knowledge nodes from "Basic Medicine" for nurses, with 122 questions used for calibration and 488 questions used for blind testing. The AI ​​classification hit rate is 79.6%, which is higher than the preset usable threshold of 75%. If only sorted by text search, the hit rate is only 51.3%, which shows that AI makes more accurate judgments after understanding the meaning of the question. 37 of the 488 questions have confidence scores lower than 0.7 and are recommended to be reviewed by experts.

The application topic analyzes the 231 questions in the urban planning technology subject of the college entrance examination, local special examination and railway special examination from 2012 to 2025, and takes stock of the proposition outline nodes of the regular examination and unquestioned questions. Taking legal subjects as an example, 6 of the 32 nodes are high-frequency core, and "Urban Planning Law" appears 31 times; another 6 nodes have not had questions in 12 years, and experts can confirm whether they are still necessary knowledge for current practice.

The Examination and Selection Department stated that AI does not replace proposition majors, but instead handles heavy comparisons first, and then has experts review low-confidence cases and test results. It hopes to gradually extend to other subjects and serve as an empirical basis for improving proposition outlines, reducing the stress and cost of endless preparation for candidates.