CN117972471A - Brain information driven key post talent selection system - Google Patents

Abstract

The invention discloses a brain information driven key post talent selection system, which comprises: the device comprises an acquisition module, a characteristic extraction module and a classification module; the acquisition module is used for acquiring brain information processing motion information and scalp brain electrical signals of the testee; the feature extraction module is used for processing motion information based on brain information and obtaining brain information processing motion feature vectors; preprocessing scalp electroencephalogram signals, and reducing the dimension of the preprocessed scalp electroencephalogram signals based on a trained variation self-encoder model to obtain dimension-reduced electroencephalogram feature vectors; and the classification module is used for carrying out standardization processing on the brain information processing motion feature vector and the brain electrical feature vector after dimension reduction and connecting the brain information processing motion feature vector and the brain electrical feature vector, and establishing a talent selection decision model through the classifier to obtain talent selection results. The invention provides a new thought for research and development of talent selection models, provides a new method and field for brain science research and application, and has important theoretical research and practical application values.

Description

A brain information driven talent selection system for key positions

Technical Field

The present invention belongs to the technical fields of psychology, management science and neuroscience, and specifically relates to a brain information driven key position talent selection system.

Background technique

With the development of society, the application of psychology, which mainly studies psychological activities and personality behaviors, is becoming more and more widespread, and the demand for psychological measurement tools in all walks of life is also increasing. A lot of research has been done at home and abroad in terms of job application tests, talent selection, and job allocation. The earliest related research seen so far is the validity study of multi-method assessment published by Handyside & Duncan in 1954. The subjects were low-level managers of the company, and the results of the assessment were used to decide whether to hire them. The assessment methods included paper-and-pencil tests, two interviews with different examiners, panel interviews, and letters of recommendation. Silzer reported a study on determining predictive indicators of manager success in 1986. The study involved managers from different levels and departments in 1749, who used the California Personality Inventory and a series of cognitive ability tests. There are relatively few tests that can be used for personnel selection in domestic related research. Currently, the commonly used tests in China include: Cattell 16 Personality Questionnaire (16PF), Minnesota Multiphasic Personality Inventory (MMPI), Test of Basic Occupational Aptitudes (GATB), Standard Raven's Mathematics Test (SPM) projective test, thematic apperception test, Rorschach inkblot test, etc.

However, most psychological tests and methods for career prediction are related to social and cultural backgrounds. Without this cultural background, the effectiveness of this method is difficult to guarantee. Moreover, this method must first define various abilities, and at the same time prove that the psychological tests used can accurately measure these abilities. Moreover, traditional tests use question-and-answer methods or charts, which rarely show the dynamic changes of psychological activities. In addition, when studying psychological activities, traditional psychology combines people's worldview and moral values, so it is very difficult to study the relationship between people's psychological activities and their abilities, personality and behavioral characteristics. Due to the different characteristics of each person's psychological activity process, the abilities and action behavior characteristics of the person revealed by it are also different, so a new research method and idea can be proposed.

Therefore, the psychological activity process can be regarded as a brain information processing process, and then a mathematical model of brain information processing movement can be proposed. Relatively mature engineering experiments and theories can be used to analyze and reveal the law of this movement, and this law can be measured and studied to establish an engineering-based psychological activity research system.

In addition, in the field of psychology, the application of brain-computer interface technology is very extensive, which can help researchers gain a deeper understanding of human cognitive and emotional processes. Brain-computer interface (BCI) is a technology that allows human brain activity to be directly converted into computer instructions or control signals. One of the common brain-computer interface technologies is achieved by measuring electroencephalogram (EEG) signals. In psychology, brain-computer interface has been applied in many aspects such as cognitive research, emotion research, and mental illness research, such as the study of cognitive processes of attention, memory, and decision-making, the identification of brain electrical activity patterns under different emotional states, and the diagnosis of diseases such as depression and attention deficit hyperactivity disorder (ADHD). In general, brain-computer interface technology has great potential in psychological research.

The above background shows that the mathematical model based on brain information processing movement and the talent selection decision method based on brain-computer interface have important research value. The thinking activity of the brain is a special information processing movement. By studying brain information processing through the universal laws of movement, we can reveal the movement characteristics and the behavioral characteristics and ability characteristics of this characteristic. In addition, by collecting and analyzing EEG signals, researchers can gain an in-depth understanding of human cognitive and emotional processes, and obtain rich information from them, thereby providing more accurate talent selection decision results. Therefore, the present invention proposes a brain information-driven key position talent selection system based on the mathematical model of brain information processing movement and brain-computer interface technology.

Summary of the invention

The present invention aims to solve the deficiencies of the prior art and proposes a brain information driven talent selection system for key positions, which realizes talent selection for key positions through the mathematical model of brain information processing movement and brain-computer interface technology.

To achieve the above object, the present invention provides the following solutions:

A brain information driven talent selection system for key positions, comprising: an acquisition module, a feature extraction module and a classification module;

The acquisition module is used to collect the brain information processing movement information and scalp EEG signals of the subject;

The feature extraction module is used to obtain a brain information processing motion feature vector based on the brain information processing motion information; preprocess the scalp EEG signal, and perform dimension reduction on the preprocessed scalp EEG signal based on a trained variational autoencoder model to obtain a reduced-dimensional EEG feature vector;

The classification module is used to normalize and connect the brain information processing motion feature vector and the EEG feature vector after dimensionality reduction, establish a talent selection decision model through a classifier, and obtain a talent selection result.

Preferably, the acquisition module includes a brain information processing motion information acquisition unit and a scalp EEG signal acquisition unit;

The brain information processing motion information acquisition unit is used to obtain the brain information processing motion information based on the workload curve of the subject's continuous addition of manual work;

The scalp EEG signal acquisition unit is used to acquire the scalp EEG signals of the subject when the subject continuously performs addition manual operations based on the EEG acquisition system.

Preferably, the brain information processing motion characteristic vector includes inertia characteristic quantity, elasticity characteristic quantity, energy characteristic quantity, deviation degree and ability characteristic quantity of the upper and lower halves respectively.

Preferably, in the feature extraction module, the process of obtaining the brain information processing motion feature vector is:

Based on the workload curve of the subject, the least square method is used to obtain the coefficients of each row of the workload in the first and second half of the test. and/> The formula is as follows:

In the formula, represents the mean of the total workload of the ith subject, q _ij represents the workload of the ith subject in the jth row, represents the mean total workload of each subject,/> represents the mean workload of each subject in the jth row, where i = 1, 2, ..., 4000, representing the number of subjects; j = 1, 2, ..., 30, representing the number of workload rows;

Based on the coefficients of each row and the mean total workload of the i-th subject, we obtain Estimation of the expected workload of the jth row under the condition/> The formula is as follows:

Based on the workload of the subject in the jth row, the estimate of the expected workload in the jth row, and the mean and standard deviation of the workload differences in each row of all subjects, a workload benchmark difference is obtained, and the workload benchmark difference is standardized;

Based on the subjects The brain information processing motion feature vector is obtained by estimating the expected workload of the jth row under the condition and the workload benchmark difference after normalization.

Preferably, in the feature extraction module, the variational autoencoder model includes an encoder and a decoder;

The encoder is used to encode the scalp EEG signal to obtain latent space parameters; wherein the latent space parameters include a mean vector and a variance vector;

The decoder is used to decode the latent space parameters to obtain reconstructed input data.

Preferably, the variational autoencoder model is trained using a loss function including reconstruction loss and KL divergence;

Based on the reconstruction loss, measuring the difference between the decoder output and the original input;

Based on the KL divergence, the difference between the distribution of the encoder output and the standard normal distribution is measured.

Preferably, the formula for the normalization process is:

x: is the original feature vector;

x′: normalized feature vector;

Rx: a 10-dimensional vector consisting of the maximum values of the original features;

Lx: A 10-dimensional vector consisting of the minimum values of the original features;

Rx′: a 10-dimensional vector consisting of the maximum values of each feature after normalization;

Lx′: A 10-dimensional vector consisting of the minimum values of each feature after normalization.

Preferably, the classifier adopts a regularized linear discriminant analysis classifier, and the mathematical expression is as follows:

y＝w ^T u

In the formula, u represents a sample of the input classifier, that is, an n-dimensional feature vector, y is the classification result, and w is the projection matrix w.

Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention realizes the talent selection for key positions through the mathematical model of brain information processing movement and brain-computer interface technology, and is applicable to subjects with various social and cultural backgrounds, various abilities, personalities, and behavioral characteristics, and the results are reliable. The present invention provides a new idea for the research and development of talent selection models, is an important application of neuroscience, psychology, and artificial intelligence in talent selection, and proposes new methods and fields for brain science research and application, and has important theoretical research and practical application value.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

FIG1 is a schematic diagram of the structure of a brain information driven key position talent selection system according to an embodiment of the present invention;

FIG2 is a schematic diagram of a brain information driven key position talent selection system according to an embodiment of the present invention;

FIG3 is a diagram showing the locations of designated electrodes on the scalp of a subject in an embodiment of the present invention;

FIG4 is a schematic diagram of a continuous addition operation in an embodiment of the present invention;

FIG5 is a schematic diagram of a workload curve in an embodiment of the present invention;

FIG6 is an example of a workload curve for different people in an embodiment of the present invention;

FIG. 7 is a schematic diagram of the structure of a variational autoencoder used in an embodiment of the present invention.

Detailed ways

The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

Embodiment 1

As shown in Figure 1-2, a brain information-driven talent selection system for key positions includes: an acquisition module, a feature extraction module, and a classification module;

An acquisition module is used to collect the subject's brain information processing movement information and scalp EEG signals;

A further embodiment is that the acquisition module includes a brain information processing motion information acquisition unit and a scalp EEG signal acquisition unit;

The brain information processing motion information acquisition unit is used to obtain brain information processing motion information based on the workload curve of the continuous addition manual operation of the test subject; specifically, 82 people from the senior manager class of xx University and 164 R&D personnel of xx Company are used as training samples, a total of 246 people, and each of them is asked to perform continuous addition manual operation. The continuous addition operation table is shown in Figure 4, and the workload is 30 rows, divided into 15 rows in the upper and lower halves. The two workload curves of the upper and lower halves obtained are brain information processing motion information, and the schematic diagrams of the workload curves are shown in Figures 5 and 6;

The scalp EEG signal acquisition unit is used to collect the scalp EEG signals of the subject during the continuous addition manual operation based on the 64-channel EEG acquisition system. The position diagram of the subject's scalp electrodes is shown in Figure 3;

A feature extraction module is used to obtain a brain information processing motion feature vector based on brain information processing motion information; preprocess the scalp EEG signal, and reduce the dimension of the preprocessed scalp EEG signal based on the trained variational autoencoder model to obtain a reduced-dimensional EEG feature vector;

A further implementation method is that the brain information processes the motion characteristic vector, including inertia characteristic quantity, elasticity characteristic quantity, energy characteristic quantity, deviation degree and ability characteristic quantity for the upper and lower halves respectively, for a total of 10 characteristics.

A further implementation method is that, in the feature extraction module, the process of obtaining the brain information processing motion feature vector is:

Based on the workload curve of the subjects, the least square method is used to obtain the coefficients of each row of workload in the first and second half of the test. and The formula is as follows:

In the formula, represents the mean of the total workload of the ith subject, q _ij represents the workload of the ith subject in the jth row, represents the mean total workload of each subject,/> represents the mean workload of each subject in the jth row, where i = 1, 2, ..., 4000, representing the number of subjects; j = 1, 2, ..., 30, representing the number of workload rows;

Based on the coefficients of each row and the mean total workload of the i-th subject, we can obtain Estimation of the expected workload of the jth row under the condition/> The formula is as follows:

In particular, q _ij and The correlation coefficient between them can be calculated by the following formula:

Table 1 lists the coefficients of each row in the upper and lower halves. and/> The calculation results are:

Table 1

Based on the workload of the subject in the jth row, the estimate of the expected workload in the jth row, and the mean and standard deviation of the workload differences in each row of all subjects, a workload benchmark difference is obtained, and the workload benchmark difference is standardized;

In the formula,

Δ _j : workload baseline difference, which is the measured variable in factor analysis.

q _j : let the workload of the subject in row j be

The corresponding evaluation benchmark workload of the jth row (estimate of the expected workload)

It is the average value of the difference in each row of 4000 samples (workload of all subjects);

is the standard deviation of each row difference of 4000 samples, j = 1, 2, ..., 30; N = 4000

Based on the subjects The estimated expected workload of the jth row under the condition and the standardized workload baseline difference are used to obtain the brain information processing motion feature vector.

Specifically, the inertial characteristic quantities _Mu and _Ml of the brain information processing movement in the upper and lower halves can be obtained according to the following formula:

The elastic characteristic quantities _Ku and _Kl of the brain information processing movement in the first and second half can be obtained according to the following formula:

The energy characteristic quantities of brain information processing movement in the first and second half, _Eu and _El, can be obtained according to the following formula:

The ability characteristic quantities of brain information processing movement in the first and second half, _AVu and _AVl, can be obtained according to the following formula:

In the formula,

AV _l is the characteristic quantity of brain information processing movement ability in the second half;

AV _u is the ability characteristic of brain information processing movement in the first half;

q _j is the amount of work in the jth row.

The deviations PF _u and PF _l in the first and second half can be calculated according to the following formula:

q _j is the j-th row of work of the subject;

is the evaluation benchmark task of the jth row for the subject.

A further implementation method is that the process of preprocessing the EEG signal includes bandpass filtering, downsampling, baseline correction, artifact filtering, and common average reference. Among them, the bandpass filter intercepts the EEG signal of 0.5-20Hz, downsamples to 100Hz, and the artifact filtering adopts the FastICA method.

A further implementation method is to use a variational autoencoder (VAE) to reduce the dimension of the EEG signal in the feature extraction module: the variational autoencoder structure is shown in Figure 7, which is a deep learning model. The main idea is to encode the input data into parameters of the latent space through a neural network, then sample these parameters to obtain latent variables, and finally decode the latent variables into reconstructed input data through another neural network.

Specifically, the variational autoencoder model includes an encoder and a decoder;

An encoder is used to encode the scalp EEG signal to obtain latent space parameters; wherein the latent space parameters include a mean vector and a variance vector;

A decoder, used for decoding the latent space parameters to obtain reconstructed input data;

A further implementation method is to train the variational autoencoder model, using a loss function including reconstruction loss and KL divergence;

Based on the reconstruction loss, measure the difference between the decoder output and the original input;

Based on the KL divergence, it measures the difference between the distribution of the encoder output and the standard normal distribution.

The optimizer used in this embodiment is the SGD optimizer. In the training cycle, we first obtain the parameters of the latent space through the encoder, then sample the latent variables from these parameters, then obtain the reconstructed input data through the decoder, and finally calculate the loss function and update the model parameters.

After the training is completed, the new test samples can be reduced in dimension. The input data is converted into parameters of the latent space through the encoder, and then the mean vector is taken as the data after dimension reduction. The dimension of the EEG signal feature vector after dimension reduction is 1*20.

The classification module is used to normalize and connect the brain information processing motion feature vector and the EEG feature vector after dimensionality reduction, establish a talent selection decision model through the classifier, and obtain the talent selection result.

A further implementation method is that the value ranges of the features of the brain information processing motion feature vector (10 dimensions) and the reduced EEG feature vector (20 dimensions) are inconsistent, so the features are normalized so that each feature is between [-1, 1].

The formula for normalization is:

x: is the original feature vector;

x′: normalized feature vector;

Rx: a 10-dimensional vector consisting of the maximum values of the original features;

Lx: A 10-dimensional vector consisting of the minimum values of the original features;

Rx′: a 10-dimensional vector consisting of the maximum values of each feature after normalization;

Lx′: A 10-dimensional vector consisting of the minimum values of each feature after normalization.

A further implementation method is to connect the two normalized feature vectors, input them into a classifier, establish a talent selection decision model, and obtain the final talent selection classification result; the classifier uses regularized linear discriminant analysis (RLDA). The mathematical expression is as follows:

y＝w ^T u

In the formula, u represents a sample of the input classifier, that is, an n-dimensional feature vector, y is the classification result, and w is the projection matrix w.

The projection matrix w can be calculated by the following formula

Where _μe and _μn represent the mean of all target training samples and the mean of all non-target training samples respectively, and _∑′w is the regularized intra-class discreteness matrix, which can be calculated as follows:

∑′ _w =(1-λ)∑ _w +λvI

∑ _w is the intra-class scatter matrix, which can be obtained by summing the covariance matrices of the two classes of samples. λ is an adjustable parameter with a value range of (0, 1], I is the identity matrix, trace() represents the trace of the matrix, and the value of d is the dimension of the intra-class scatter matrix ∑ _w .

The classification of RLDA is achieved by comparing the y value with the threshold Tr. The subjects in this embodiment include university senior management class personnel and company R&D personnel. In the present invention, when y>Tr, RLDA determines that the current sample is a "management personnel", and when y＜Tr, the current sample is determined to be a "R&D personnel".

The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should fall within the protection scope determined by the claims of the present invention.

Claims (8)

1. A brain information driven key post talent selection system, comprising: the device comprises an acquisition module, a characteristic extraction module and a classification module;

The acquisition module is used for acquiring brain information processing motion information and scalp brain electrical signals of the testee;

the feature extraction module is used for processing motion information based on the brain information to obtain brain information processing motion feature vectors; preprocessing the scalp electroencephalogram signal, and reducing the dimension of the preprocessed scalp electroencephalogram signal based on a trained variation self-encoder model to obtain a dimension-reduced electroencephalogram characteristic vector;

The classifying module is used for carrying out standardization processing on the brain information processing motion feature vector and the brain electricity feature vector after dimension reduction and connecting the brain information processing motion feature vector and the brain electricity feature vector, and establishing a talent selection decision model through a classifier to obtain talent selection results.

2. The brain information driven key post talent selection system according to claim 1, wherein the acquisition module comprises a brain information processing movement information acquisition unit and a scalp brain electrical signal acquisition unit;

The brain information processing movement information acquisition unit is used for continuously adding a manual work workload curve based on a testee to obtain the brain information processing movement information;

The scalp electroencephalogram signal acquisition unit is used for acquiring the scalp electroencephalogram signal of a testee during continuous addition manual operation based on an electroencephalogram acquisition system.

3. The brain-information-driven key post talent selection system according to claim 2, wherein the brain-information-processing motion feature vector includes an inertial feature, an elastic feature, an energy feature, a degree of deviation, and a capacity feature, respectively, in the upper and lower halves.

4. The brain-information-driven key post talent selection system according to claim 3, wherein in the feature extraction module, the process of obtaining the brain-information-processing motion feature vector is as follows:

Based on the workload curve of the testee, obtaining each row coefficient of the workload in the upper half and the lower half by adopting a least square method And/>The formula is as follows:

in the method, in the process of the invention, Representing the average value of the total workload of the ith testee, q _ij represents the workload of the jth row of the ith testee,Representing the total workload average of each subject,/>Represents the mean of the workload of each subject at row j, where i=1, 2,..4000, represents the number of subjects; j=1, 2,..30, representing the number of workload lines;

based on the line coefficients and the average value of the total workload of the ith testee, obtaining the total workload of the ith testee Estimation of expected workload of jth line under conditions/>The formula is as follows:

Obtaining a working amount reference difference based on the estimated working amount of a j-th row of a testee, the expected working amount of the j-th row and the average value and standard deviation of each row of difference of the working amounts of all testees, and carrying out standardization processing on the working amount reference difference;

Based on the presence of the subject And estimating the expected work amount of the j-th line under the condition and obtaining the brain information processing motion characteristic vector by the work amount reference difference after the normalization processing.

5. The brain-information-driven critical post talent selection system of claim 1, wherein in said feature extraction module, said variation self-encoder model comprises an encoder and a decoder;

The encoder is used for encoding the scalp electroencephalogram signals to obtain potential space parameters; wherein the potential spatial parameters include a mean vector and a variance vector;

The decoder is configured to decode the potential spatial parameters to obtain reconstructed input data.

6. The brain-information-driven critical post talent selection system of claim 5, wherein training the variational self-encoder model uses a loss function comprising reconstruction loss and KL divergence;

Based on the reconstruction loss, measuring a difference of the decoder output from an original input;

based on the KL divergence, a difference between a distribution of the encoder output and a standard normal distribution is measured.

7. The brain-information-driven key post talent selection system of claim 1, wherein the formula of the normalization process is:

x: is the original feature vector;

x': normalized feature vectors;

rx: a 10-dimensional vector of the maximum values of the original features;

lx: a 10-dimensional vector of the minimum values of the original features;

rx': a 10-dimensional vector composed of maximum values of the features after normalization;

Lx: normalized 10-dimensional vector of minimum values of features.

8. The brain-information-driven key post talent selection system of claim 7, wherein the classifier employs a regularized linear discriminant analysis classifier with the mathematical expression:

y＝w^Tu

where u represents one sample of the input classifier, i.e., an n-dimensional feature vector, y is the classification result, and w is the projection matrix w.