Method for identifying subject decision confidence through decision self-confident data acquisition normal form

Abstract

The invention provides a method for identifying subject decision confidence through a decision self-confidence data acquisition normal form. The method comprises the steps of decision self-confidence data acquisition and alertness data acquisition. The decision-making self-confidence degree data acquisition is characterized in that electroencephalogram and eye movement data of three decision-making self-confidence degrees of a subject are acquired through three sub-experiments; the alertness data acquisition comprises the step of acquiring electroencephalogram and eye movement data of the alertness of a subject in an interval time period of acquiring decision self-confidence degree data; the alertness acquisition part is used for acquiring electroencephalogram and eye movement data of a subject before a mark appears as alertness data of the subject by using the mark appearing at a random interval time; and finally, performing multi-modal decision fusion, namely performing preprocessing, feature extraction and classification processing on the data of the electroencephalogram mode and the eye movement mode, and synthesizing final classification results to obtain a decision confidence recognition result.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and particularly to a method for identifying the decision confidence of a subject by using a decision confidence data acquisition paradigm. Background Art

[0002] At present, institutions of higher learning focus on the actual combat teaching mode, emphasizing that students conduct training in actual scenarios, requiring them to master precise skills and be able to apply the knowledge they have learned in complex environments. For this reason, training evaluation simulator systems are widely used in colleges and universities to simulate real environments. However, it is difficult to accurately identify the knowledge blind spots of students only by repeated practical operations. At the same time, it is impossible to comprehensively evaluate the skill proficiency, decision confidence, and alertness of students. Therefore, these simulator systems need to accurately analyze and evaluate the performance of students. Only through scientific methods can students be better guided to improve their training levels. Similar needs also exist in teaching simulators in other practical operation fields. Summary of the Invention

[0003] The purpose of the present invention is to provide a method for identifying the decision confidence of a subject by using a decision confidence data acquisition paradigm, aiming at the deficiencies of the above-mentioned existing technologies, and solving the problems of difficult accurate acquisition of data on students' decision confidence and alertness and identification of decision confidence when evaluating students' learning achievements.

[0004] To achieve the above purpose, the present invention adopts the following technical solutions:

[0005] The present invention provides a method for identifying the decision confidence of a subject by using a decision confidence data acquisition paradigm, including decision confidence data acquisition and alertness data acquisition;

[0006] For the decision confidence data acquisition, electroencephalogram (EEG) and eye movement data of three decision confidences of the subject are collected through three sub-experiments;

[0007] For the alertness data acquisition, EEG and eye movement data of the subject's alertness are collected during the interval period of collecting decision confidence data.

[0008] Further, the decision confidence data acquisition is specifically as follows:

[0009] In the first sub-experiment, the subject directly answers questions in a completely unfamiliar situation, and EEG and eye movement data under low decision confidence are collected;

[0010] In the second sub-experiment, after the subject is familiar with the test questions, the subject answers the questions for the second time with the same answering process, and EEG and eye movement data under medium decision confidence are collected;

[0011] In the third sub-experiment, after the subject is completely familiar with the set of test questions, the subject answers the questions for the third time with the same answering process, and EEG and eye movement data under high decision confidence are collected.

[0012] Further, the sub-experiment process includes:

[0013] S1. Start answering questions. After each question appears, the subject needs to start choosing an answer 5 seconds later to ensure that sufficient decision confidence data is collected when the subject makes a single decision;

[0014] S2. After confirming the question options, the subject evaluates his / her decision confidence for this decision, and correspondingly selects "completely unconfident", "relatively confident", and "extremely confident". The evaluation result is used as the label of the decision confidence data for this decision;

[0015] S3. Enter the alertness collection stage, and collect the subject's alertness data and labels;

[0016] Repeat the above S1 to S3 until answering a set of questions is completed.

[0017] Further, a set of the questions specifically are:

[0018] S301. A set of questions consists of 10 general knowledge questions and 25 professional knowledge and skill questions;

[0019] S302. The first 5 are fixed general knowledge questions, and the last 30 are 5 general knowledge questions and 25 professional knowledge and skill questions randomly presented;

[0020] S303. The order of the questions and the order of the options for each answer will be randomly scrambled.

[0021] Further, perform signal preprocessing, feature extraction, and task classification on the EEG signals and eye movement signals respectively.

[0022] The beneficial effects of the present invention are as follows: In the decision confidence collection part, by having the subject choose the answers to general knowledge questions and professional skill questions, the EEG and eye movement data during the subject's thinking and decision-making are collected as the decision confidence data of the subject. In the alertness collection part, signs that appear at random intervals are used to collect the EEG and eye movement data of the subject before the appearance of the signs as the alertness data of the subject. Finally, multi-modal decision fusion is used, that is, the data of two modalities, EEG and eye movement, are respectively preprocessed, feature extracted, and classified, and the final classification results are combined to obtain the decision confidence recognition result. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is a flowchart of a method for identifying a subject's decision confidence as a decision confidence data acquisition paradigm;

[0024] Figure 2 It is a detailed flowchart of a sub-experiment;

[0025] Figure 3 It is a flowchart of multi-modal recognition of decision confidence. Detailed implementation manners

[0026] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0027] Please refer to Figure 1 , the experimental paradigm proposed by the present invention is generally divided into three sub-experiments for a single subject, aiming to collect decision confidence data and alertness data of the subject under different conditions;

[0028] The question settings for each wheel experiment are a total of 35 questions, including 10 common sense single-choice questions and 25 professional knowledge and skill single-choice questions. Some simple common sense questions are set in each set of questions to encourage the subject, increase the confidence of the subject in the first answer (strange), and ensure the balanced distribution of confidence data in the three answers, so that the data set can more comprehensively reveal the cognitive process and ability level of the subject.

[0029] The acquisition process of the overall experiment is divided into the following three sub-experiments given in the order of execution:

[0030] The first sub-experiment (low decision confidence data acquisition): The subject directly answers questions in a completely strange situation; after the first sub-experiment, the questions and answers of this set of questions are sent to the subject, and the subject is informed that the questions and answers will be arranged in a random order in the next answer. After the subject briefly reads the questions and answers once, the subject starts to answer the questions again for about 3 minutes;

[0031] The second sub-experiment (medium decision confidence data acquisition): The subject answers the questions for the second time in the same answering process; after the second sub-experiment, the questions and answers of this set of questions are sent to the subject again for reading, and the subject is informed that when answering the questions next time, the questions and answers will be arranged in a random order. After the subject is completely familiar with the questions, the third sub-experiment is carried out;

[0032] The third sub-experiment (high decision confidence data acquisition): The subject answers the questions for the third time in the same answering process when being completely familiar with this set of questions, ensuring a high confidence in the decision;

[0033] The detailed experimental process within the sub-experiment is as Figure 2 shown. The first 5 questions of each set of questions are fixed as common sense questions for data calibration, and the latter 30 questions are randomly shuffled and presented by 5 common sense questions and 25 professional skill questions, and are carried out according to the following description:

[0034] S1. At time point 101, the test starts. After each question appears on the screen, the subject is required to start choosing an answer 5 seconds later, which is used to ensure that sufficient EEG and eye movement data can be collected when the subject makes a single decision as the data of the subject's decision confidence;

[0035] S2. At time point 102, after the subject confirms the question options, the subject evaluates his / her confidence in this decision. On the screen, the subject can choose "completely unconfident" (low decision confidence), "relatively confident" (medium decision confidence), and "extremely confident" (high decision confidence). The evaluation result is used as the label of the decision confidence data for this time;

[0036] S3. At time point 103, the alertness acquisition stage begins. A symbol will randomly appear on the screen within 3 to 5 seconds after reminding the subject. When the symbol appears, it is recorded as time point 104. The subject needs to click on any area of the screen with the mouse as soon as possible after the symbol appears. From the time when the symbol is clicked to time point 105, the EEG and eye movement data of the subject are collected during the time period from time point 103 to time point 105 as the alertness data, and the time length from time point 104 to time point 105 is used as the label of this alertness data;

[0037] S4. Repeat the process from time point 101 to time point 105 until a single set of questions is answered.

[0038] The detailed process of multimodal recognition decision confidence is as attached Figure 3 as shown, and the signal preprocessing, feature extraction, and classification processes are respectively carried out on the EEG signal and the eye movement signal.

[0039] Signal preprocessing part:

[0040] Euclidean alignment EA: Since there is covariate shift between the EEG data of users, data alignment (EA) in Euclidean space is a simple and effective data alignment method, similar to whitening in machine learning, which makes the average covariance matrix of EEG data be the identity matrix and the distribution more similar;

[0041] Notch filter: There are interference signals in EEG caused by the environment and equipment, such as power frequency interference or changes in electrode impedance. The power frequency interference in China is 50Hz, and generally a 50Hz notch filter is used to filter out the power frequency interference in the EEG signal;

[0042] Common average reference: The common average reference is a spatial filtering algorithm that can solve the problem of single-point failure that may occur when using the signal recorded by the mastoid channel as the reference signal;

[0043] Principal Component Analysis (PCA): The pupil diameter is not only related to the emotion recognition task, but also related to the brightness and other interferences in the environment. The principal component analysis method is used to remove the influence of brightness on the pupil diameter;

[0044] Sliding window method: Generally, the amount of data in electroencephalogram (EEG) is small. The sliding window data augmentation is performed on the EEG to increase the number of experimental data.

[0045] Feature extraction part:

[0046] The EEG signal classification uses the deep learning network EEGNet. The network model will learn how to extract features and which features to extract during the training process;

[0047] We have designed and implemented a total of 9 common eye movement data features, including 3 pupil-related features, 4 eye movement position features, and 2 overall fixation eye movement features; The detailed features are shown in Table 1. Table 1 Eye movement features

[0048]

[0049] Task classification part:

[0050] The EEG signal is identified and classified using the deep learning network EEGNet;

[0051] The eye movement signal is classified using the traditional machine learning classification method, the Linear Discriminant Analysis (LDA) classifier;

[0052] Finally, the classification results of the two methods need to be integrated to obtain the final classification result. In the present invention, the decision result with the largest probability or the one with the highest accuracy after adding the probabilities is selected.

[0053] The above-described embodiments only express the implementation manners of the present invention. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the patent of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention should be based on the appended claims.

Claims

1. A method for identifying the subject's decision confidence using a decision confidence data collection paradigm, characterized by: Including decision confidence data collection and alertness data collection; The decision confidence data collection collects the EEG and eye movement data of the subjects at three levels of decision confidence through three sub-experiments; The alertness data collection collects the EEG and eye movement data of the subject's alertness during the interval between the collection of the decision confidence data.

2. The method for identifying the subject's decision confidence according to the decision confidence data collection paradigm of claim 1, characterized in that: The decision confidence data collection is specifically as follows: In the first sub-experiment, the subjects directly completed the questions in a completely unfamiliar situation, and EEG and eye movement data were collected under low decision confidence; In the second sub-experiment, after becoming familiar with the test questions, the subjects took the test questions for the second time using the same answering process, and collected EEG and eye movement data at a medium level of decision confidence; In the third sub-experiment, the subjects answered the questions for the third time using the same answering process after being fully familiar with the set of questions, and collected EEG and eye movement data under high decision confidence.

3. The method for identifying the subject's decision confidence according to the decision confidence data collection paradigm of claim 2, characterized in that: The sub-experimental process includes: S1. Start answering questions. After each question appears, the subject needs to wait 5 seconds before choosing the answer, so as to ensure that the subject collects enough decision confidence data when making a single decision; S2. After confirming the question options, the subjects evaluated their confidence in their decision-making, and chose "completely unconfident", "relatively confident" and "extremely confident" accordingly. The evaluation results were used as the labels for the confidence data of this decision-making. S3, enter the alertness collection stage, collect the subject's alertness data and labels; Repeat S1 to S3 until a set of questions is completed.

4. The method for identifying the subject's decision confidence according to the decision confidence data collection paradigm of claim 3, characterized in that: A set of said topics specifically includes: One set of questions consists of 10 general knowledge questions and 25 professional knowledge and skills questions; The first 5 questions are fixed general knowledge questions, and the last 30 questions are 5 general knowledge questions and 25 professional knowledge and skills questions randomly selected; The order of questions and options will be randomly shuffled each time.

5. The method for identifying the subject's decision confidence according to the decision confidence data collection paradigm of claim 4, characterized in that: Signal preprocessing, feature extraction and task classification are performed on the EEG signal and the eye movement signal respectively.

Images