On the Last Day Before the College Entrance Examination, AI Invigilators Are in Place

Technology plays an increasingly important role in ensuring fairness in exams. With the help of technologies such as computer vision, neural networks, and behavioral analysis, cheating behavior can be effectively identified and order in the examination room can be maintained.

The Liaoning Provincial Admissions and Examination Office issued a document on July 5 stating that the 2020 Liaoning General College Entrance Examination willArtificial intelligence technology will be used to detect suspected violations in examinations.

The article points out that this year's college entrance examination will use AI to conduct preliminary analysis and judgment of all examination room videos to identify suspected violations.

Examination staff at all levels will conduct secondary screening and judgment on suspected violations detected by the AI system and handle them in accordance with relevant regulations.

AI proctoring was launched as early as 2018

In fact, the use of artificial intelligence technology in exam supervision is nothing new.

ATA (American Online), which started with DST (dynamic simulation technique) dynamic simulation technology, uses artificial intelligence and big data technology to help maintain discipline in the examination room.

ATA was founded in 1999 and is headquartered in Beijing. It is the founder and leader of China's intelligent test and assessment services and the largest provider of computer-based intelligent test and assessment services.Committed to providing fair, just, convenient and efficient examination services.

They analyze candidates' movements through tens of thousands of exam room videos throughout history to detect abnormal behavior.

The system uses deep learning algorithms to track candidates in real time and analyze abnormal behaviors in the examination room. It also provides prompts and alarms for violations by analyzing the skeletal behavior of candidates.

Once you find any behavior such as large movements, using mobile phones, leaving your seat, etc.The system will issue a warning and the corresponding video will be uploaded to the cloud and stored as electronic evidence.

In addition, ATA also uses technologies such as intelligent modeling of the examination room and scene semantic recognition to perform semantic segmentation of the examination room seats.Automatically draw a two-dimensional plan of the examination room and generate a behavior heat map of the entire examination room.

The two-dimensional plan corresponds to the actual scene one by one, showing the situation of candidates taking their seats and leaving their seats in real time. The heat map of the whole examination room behavior can help teachers grasp the overall dynamics of the examination room, so that there are no blind spots in the invigilation, which reduces the burden on the invigilators.

This intelligent proctoring system has been used in the 2018 CPA examination.

Neural Networks and Big Data Plagiarism Detection

In addition to monitoring and controlling cheating behavior, another way is to detect plagiarism from answer sheets.

Researchers at the University of Copenhagen in Denmark are conducting a study to try to detect cheating on homework through written analysis using artificial intelligence. The project is called Ghostwriter.

Researchers collected 130,000 Danish written assignments from 10,000 high school students, performed text analysis on them, and trained a detection model that can determine with nearly 90% accuracy whether the assignment is written by the original author or plagiarized. 

Ghostwriter is built around machine learning and neural networks, said Dr. Stephan Lorenzen, a researcher on the project. It uses a "Siamese" neural network,Ability to analyze input text for grammatical errors, common phrases, sentence length, structure, and other characteristics.

The team trained the neural network on a large amount of data, mainly learning from the representation of writing style and making judgments and comparisons.

When a student submits an assignment, the detection model compares it to previous assignments and gives a percentage of similarity between the writing style of the new assignment and the old assignment.

The scores given are weighted averaged as a factor in the calculation. In addition, factors such as submission time are taken into consideration, and a score is finally given, which is expressed as a percentage to indicate the similarity of style and whether there is any suspected cheating.

Ghostwriter can also be used in police work such as analyzing forged documents and forensic document review.

Say goodbye to fancy cheating routines

In one episode of the Korean comedy variety show "SNL Korea", a variety of cheating routines were reproduced, including making notes on nails and bottles, writing answers on arms and legs, and bringing mobile phones into the examination room...

The most intense cheating and anti-cheating showdown in history was staged between the invigilators and the candidates, with plenty of jokes.

The video lists 9 cheating methods

Compared with cheating in real exams, although the plot is exaggerated, it also shows that the teacher is well aware of the students' common cheating methods.

Now with the help of AI in invigilation, the review efficiency is higher and the fairness of the college entrance examination can be better guaranteed.

Finally, I wish every candidate can take the exam with integrity, perform to the best of their ability in a fair examination environment, and achieve ideal results!

-- over--