A method and system for estimating information flow based on brain-to-brain interaction

Abstract

The present invention provides an information flow estimation method and system based on brain-to-brain interactivity, including: collecting original EEG signals of a training recommender and original EEG signals of a training recommendee; performing real-time preprocessing and wavelet transformation on the original EEG signals to obtain processed EEG signals; obtaining frequency band EEG signals of a preset frequency band from the processed EEG signals; performing inter-group correlation calculation on the frequency band EEG signals corresponding to every two training recommendees to obtain a neural similarity index, and calculating the frequency band EEG signals of the training recommender and the frequency band EEG signals of the training recommendee to obtain a PDC value; training a neural network model according to the neural similarity index and the PDC value of the degree of brain-to-brain connectivity to create a multi-brain connectivity model; and using the multi-brain connectivity model to estimate the direction of brain-to-brain information transmission between two recommendees or between a recommender and a recommendee. The group neural similarity can be effectively monitored, thereby improving the estimation accuracy.

Description

Technical Field

[0001] The present invention relates to the field of brain signal learning, and in particular to an information flow estimation method and system based on brain-to-brain interactivity. Background Art

[0002] Technological development has promoted the leapfrogging of e-commerce forms such as live streaming and video streaming, and competition is becoming increasingly fierce. The e-commerce industry has put forward higher requirements for accurately understanding the needs, preferences, experience, and feedback of the audience. In live streaming recommendations and video recommendations, consumers' interests and preferences are dynamically changing, which is called information flow recommendation and cannot be simply measured. How to accurately capture the preferences of online consumers and let consumers accept the recommendations of anchors is crucial to the development of e-commerce live streaming platforms, and has become a difficult problem that urgently needs to be broken through in the field of e-commerce live streaming.

[0003] Information flow recommendation refers to the recommendation of relevant product information through computer systems and communication networks, which is also the live streaming and video recommendation in daily life. In information flow recommendation, there is information interaction between the recommender and the recommended person, and this interaction will drive the dynamic changes of brain neural activity. However, the current mainstream research technology, methods and implementation difficulties make it extremely difficult to accurately capture the dynamic changes of neural activity inside the brain. In addition, previous studies have focused more on the analysis and identification of brain activities at the individual level, ignoring the calculation of neural similarity between brains at the group level. On the other hand, the measurement of group brain similarity at this stage is still mainly carried out in traditional ways such as questionnaires, self-reports, and behavioral experiments. There is a lack of objective and process-based group neural similarity calculation methods, and thus a lack of information flow recommendation prediction and optimization methods based on brain interactivity and individual brain EEG correlation. Summary of the invention

[0004] In view of the above problems existing in the prior art, a method and system for estimating information flow based on brain-to-brain interactivity are provided.

[0005] The specific technical solutions are as follows:

[0006] A method for estimating information flow based on brain-to-brain interactivity, comprising the following steps:

[0007] Step S1, dividing the experimental subject group into a training recommender group and a training recommendee group according to the task type, and dividing a training recommender and a training recommendee into a group of experimental groups, collecting a first original EEG signal of the training recommender and a second original EEG signal of the training recommendee, wherein the training recommender group and the training recommendee group have the same stimulus source;

[0008] Step S2, performing real-time preprocessing and wavelet transform on the first original EEG signal to obtain a first processed EEG signal;

[0009] Performing real-time preprocessing and wavelet transform on the second original EEG signal to obtain a second processed EEG signal;

[0010] Step S3, acquiring data of a first preset frequency band from the first processed EEG signal, and recording the acquired data of the preset frequency band as the first frequency band EEG signal;

[0011] Acquire data of a second preset frequency band from the second processed EEG signal, and record the acquired data of the preset frequency band as the EEG signal of the second frequency band;

[0012] Step S4, performing inter-group correlation calculation on every two second frequency band EEG signals to obtain a neural similarity index group of the training recommender group, wherein the neural similarity index group includes a neural similarity index between every two second frequency band EEG signals;

[0013] The first-band EEG signals and the second-band EEG signals in each experimental group are calculated according to partial directional coherence analysis to obtain a PDC value group of the degree of brain-to-brain connectivity between the training recommender group and the training recommendee group, wherein the PDC value group includes the PDC value of the degree of brain-to-brain connectivity between the first-band EEG signals and the second-band EEG signals in each experimental group;

[0014] Step S5, training the neural network model according to the neural similarity index group and the PDC value group of the inter-brain connectivity degree to create a multi-brain connectivity model;

[0015] Step S6, using a multi-brain connectivity model to estimate the direction of inter-brain information transmission between at least two to-be-estimated recommendees in the group of to-be-estimated recommendees, and evaluating the stimulus source of the group of to-be-estimated recommendees based on the estimation result;

[0016] A multi-brain connectivity model is used to estimate the direction of inter-brain information transmission between the recommender group to be estimated and the recommended group to be estimated, and the recommender group to be estimated is evaluated based on the estimation results.

[0017] Preferably, the information flow estimation method based on brain-to-brain interactivity further comprises, after step S2:

[0018] Step A1, acquiring data of a first preset channel from a first processed EEG signal, and recording the acquired data of the preset channel as a first channel EEG signal;

[0019] Acquire data of a second preset channel from the second processed EEG signal, and record the acquired data of the preset channel as a second channel EEG signal;

[0020] Step A2, performing connectivity estimation on all second channel EEG signals to obtain first estimation data;

[0021] Connectivity estimation is performed on the first channel EEG signal and the second channel EEG signal to obtain second estimation data.

[0022] Preferably, the information flow estimation method based on brain-to-brain interactivity further includes the following steps in step S5:

[0023] Using the first estimated data and the second estimated data to determine whether the neural network model has completed training;

[0024] If so, a multi-brain connectivity model is created;

[0025] If not, return to step S1.

[0026] Preferably, in the information flow estimation method based on brain-to-brain interactivity, step S1 specifically includes the following steps: using a multi-channel EEG measurement device to synchronously collect the first original EEG signal of the training recommender and the second original EEG signal of the training recommendee in the experimental group.

[0027] Preferably, in the information flow estimation method based on brain-to-brain interactivity, step S3 specifically includes the following steps:

[0028] Acquire the energy value of each first preset time period in the first preset frequency band from the first processed EEG signal, and use all energy values ​​in the first preset frequency band as the EEG signal of the first frequency band;

[0029] The energy value of each second preset time period in the second preset frequency band is acquired from the second processed EEG signal, and all energy values ​​in the second preset frequency band are used as the EEG signal of the second frequency band.

[0030] Preferably, in the information flow estimation method based on brain-to-brain interactivity, the step of obtaining the neural similarity index group in step S4 specifically includes the following steps:

[0031] The second frequency band EEG signals are paired in pairs to obtain second frequency band EEG signal groups, and the first preset calculation tool is used to perform inter-group correlation calculation on each group of second frequency band EEG signal groups to obtain the neural similarity index of each group of second frequency band EEG signal groups.

[0032] Preferably, in the information flow estimation method based on brain-to-brain interactivity, the step of obtaining the PDC value group in step S4 specifically includes the following steps:

[0033] The first-band EEG signals and the second-band EEG signals in each experimental group are input into the following formula to obtain a PDC value group:

[0034]

[0035] Among them, PDC xy When (f) = 1, α x (f) and α y (f) consistency;

[0036] A xy(f) Fourier transform used to represent coefficients of multivariate autoregressive models;

[0037] α x (f) is used to represent A xy(f) The xth column of

[0038] Used to represent A xy(f) The yth row of

[0039] x is used to represent the EEG signal in the first frequency band;

[0040] y is used to represent the EEG signal in the second frequency band;

[0041] PDC xy (f) is used to indicate the degree to which the recommender’s EEG signal guides the recommendee’s EEG signal.

[0042] Preferably, in the information flow estimation method based on inter-brain interactivity, the neural network model includes a first neural network and a second neural network, and the multi-brain connectivity model includes a first multi-brain connectivity model and a second multi-brain connectivity model;

[0043] Step S5 specifically includes the following steps:

[0044] Step S51, using a preset transformation algorithm to perform data transformation on the neural similarity indexes in the neural similarity index group to obtain first transformation data;

[0045] Using a preset conversion algorithm to perform data conversion on the PDC values ​​in the PDC value group to obtain second conversion data;

[0046] Step S52, dividing the first converted data into a first test set and a first training set;

[0047] Dividing the second transformed data into a second test set and a second training set;

[0048] Step S53, obtaining data feature dimensions of the first conversion data and the second conversion data;

[0049] Step S54, when the data feature dimension is less than the preset feature dimension threshold, the first neural network is trained using the first training set and the second training set to obtain a first trained neural network model, and the first test set and the second test set are used to verify the first trained neural network model to create a first multi-brain connectivity model;

[0050] When the data feature dimension is greater than or equal to a preset feature dimension threshold, the second neural network is trained using the first training set and the second training set to obtain a second trained neural network model, and the second trained neural network model is verified using the first test set and the second test set to create a first multi-brain connectivity model.

[0051] Preferably, in the information flow estimation method based on brain-to-brain interactivity, the second neural network is a SVM-AdaBoost classification model, and the SVM-AdaBoost classification model includes multiple SVM classifiers.

[0052] An information flow estimation device based on brain-to-brain interactivity, comprising:

[0053] A collection unit divides the experimental subject group into a training recommender group and a training recommendee group according to the task type, and divides a training recommender and a training recommendee into a group of experimental groups, and collects a first original EEG signal of the training recommender and a second original EEG signal of the training recommendee, wherein the training recommender group and the training recommendee group have the same stimulus source;

[0054] A processing unit connected to the acquisition unit, and configured to perform real-time preprocessing and wavelet transformation on the first original EEG signal to obtain a first processed EEG signal;

[0055] It is also used to perform real-time preprocessing and wavelet transformation on the second original EEG signal to obtain a second processed EEG signal;

[0056] A frequency band data acquisition unit, connected to the processing unit, for acquiring data of a first preset frequency band from the first processed EEG signal, and recording the acquired data of the preset frequency band as a first frequency band EEG signal;

[0057] It is also used to obtain data of a second preset frequency band from the second processed EEG signal, and record the obtained data of the preset frequency band as the EEG signal of the second frequency band;

[0058] a calculation unit connected to the frequency band data acquisition unit, and configured to perform inter-group correlation calculation on every two second frequency band EEG signals to obtain a neural similarity index group of the training recommendee group, wherein the neural similarity index group includes a neural similarity index between every two second frequency band EEG signals;

[0059] Furthermore, according to the partial directional coherence analysis, the first frequency band EEG signals and the second frequency band EEG signals in each experimental group are calculated to obtain a PDC value group of the degree of brain-to-brain connectivity between the training recommender group and the training recommendee group, wherein the PDC value group includes the PDC value of the degree of brain-to-brain connectivity between the first frequency band EEG signals and the second frequency band EEG signals in each experimental group;

[0060] A creation unit is connected to the calculation unit and is used to train the neural network model according to the neural similarity index group and the PDC value group of the inter-brain connection degree to create a multi-brain connectivity model;

[0061] an estimation unit, using a multi-brain connectivity model to estimate the direction of inter-brain information transmission between at least two to-be-estimated recommendees of the to-be-estimated recommendee group, and evaluating the stimulus source of the to-be-estimated recommendee group according to the estimation result;

[0062] A multi-brain connectivity model is used to estimate the direction of inter-brain information transmission between the recommender group to be estimated and the recommended group to be estimated, and the recommender group to be estimated is evaluated based on the estimation results.

[0063] The above technical solution has the following advantages or beneficial effects: by real-time monitoring and obtaining data on group neural similarity in information flow recommendation, it helps e-commerce live broadcast platforms make scientific decisions. At the same time, it is non-invasive, safe, efficient, and low-cost, and can be promoted in the fields of live broadcast recommendation, film and television drama recommendation, advertising recommendation effect evaluation, etc., and has broad market application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] The embodiments of the present invention will be described more fully with reference to the attached drawings, which are provided for illustration and description only and are not intended to limit the scope of the present invention.

[0065] Figure 1 This is a flow chart of an embodiment of a method for estimating information flow based on brain-to-brain interactivity according to the present invention. DETAILED DESCRIPTION

[0066] 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.

[0067] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0068] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments, but they are not intended to limit the present invention.

[0069] The present invention includes a method for estimating information flow based on brain-to-brain interactivity, such as Figure 1 As shown, the following steps are included:

[0070] Step S1, dividing the experimental subject group into a training recommender group and a training recommendee group according to the task type, and dividing a training recommender and a training recommendee into a group of experimental groups, collecting a first original EEG signal of the training recommender and a second original EEG signal of the training recommendee, and the training recommender group and the training recommendee group have the same stimulus source;

[0071] Step S2, performing real-time preprocessing and wavelet transform on the first original EEG signal to obtain a first processed EEG signal;

[0072] Performing real-time preprocessing and wavelet transform on the second original EEG signal to obtain a second processed EEG signal;

[0073] Step S3, acquiring data of a first preset frequency band from the first processed EEG signal, and recording the acquired data of the preset frequency band as the first frequency band EEG signal;

[0074] Acquire data of a second preset frequency band from the second processed EEG signal, and record the acquired data of the preset frequency band as the EEG signal of the second frequency band;

[0075] Step S4, performing inter-subject correlation (ISC) calculation on every two second frequency band EEG signals to obtain a neural similarity index group of the training recommendee group, wherein the neural similarity index group includes a neural similarity index (E) between every two second frequency band EEG signals;

[0076] The first-band EEG signals and the second-band EEG signals in each experimental group are calculated according to partial directional coherence analysis (PDC) to obtain a PDC value group of the degree of brain-to-brain connectivity between the training recommender group and the training recommendee group, wherein the PDC value group includes the PDC value of the degree of brain-to-brain connectivity between the first-band EEG signals and the second-band EEG signals in each experimental group;

[0077] Step S5, training the neural network model according to the neural similarity index group and the PDC value group of the inter-brain connectivity degree to create a multi-brain connectivity model;

[0078] Step S6, using a multi-brain connectivity model to estimate the direction of inter-brain information transmission between at least two to-be-estimated recommendees in the group of to-be-estimated recommendees, and evaluating the stimulus source of the group of to-be-estimated recommendees based on the estimation result;

[0079] A multi-brain connectivity model is used to estimate the direction of inter-brain information transmission between the recommender group to be estimated and the recommended group to be estimated, and the recommender group to be estimated is evaluated based on the estimation results.

[0080] In the above embodiment, by simultaneously monitoring EEG signals of multiple people and performing real-time dynamic analysis of the EEG signals of the group based on the correlation between neural activities between brains, the neural similarity of the group can be effectively monitored.

[0081] In the above embodiment, the multi-brain connectivity model created can estimate the direction of brain-to-brain information transmission between at least two to-be-estimated recommendees in the group of to-be-estimated recommendees, and evaluate the stimulus source of the group of to-be-estimated recommendees based on the estimation result, so as to determine whether the stimulus source meets the user's needs. For example, when the stimulus source is a sales advertisement, it can be determined by the evaluation result whether the sales advertisement can stimulate the purchase desire of the recommendee;

[0082] The multi-brain connectivity model created can estimate the direction of brain-to-brain information transmission between the estimated recommender group and the estimated recommendee group, and evaluate the estimated recommender group based on the estimation results to determine whether the estimated recommender meets user needs. For example, the evaluation results can be used to determine whether the recommender can stimulate the purchasing desire of the recommendee.

[0083] In the above embodiment, by real-time monitoring and obtaining the data of group neural similarity in information flow recommendation, the e-commerce live broadcast platform can be helped to make scientific decisions. At the same time, it has the characteristics of non-invasiveness, safety, efficiency and low cost, and can be promoted in the fields of live broadcast recommendation, film and television drama recommendation, advertising recommendation effect evaluation, etc., and has broad market application prospects.

[0084] In the above embodiment, the task type may be selling a certain product and purchasing the product. At this time, the task type of the training recommender group is "selling a certain product", and therefore, the task type of the training recommendee group is "purchasing the product".

[0085] In the above embodiment, the stimulus source for training the recommender group and the recommended group is the same, the stimulus source may be an emotion-inducing environment, and the emotion-inducing environment may be a recommendation scenario such as a product, an app, or a film or TV series;

[0086] For example, the emotion-inducing environment for training the recommender group and the recommended group can be a recommendation scenario for a certain product, that is, the stimulus source for the recommender group and the recommended group can be an advertisement for the product, but the task types for the recommender group and the recommended group are different. For example, the task type for the recommender group is "selling the product", while the task type for the recommended group is "purchasing the product". Therefore, the recommender group will watch the advertisement for the product with the idea of ​​how to sell the product, while the recommended group will watch the advertisement for the product with the idea of ​​whether they need to buy the product.

[0087] As a preferred implementation, in step S1, the training recommenders and the training recommendees are divided into a set of experimental groups one by one.

[0088] In the above embodiment, after step S2, the method further includes:

[0089] Step A1, acquiring data of a first preset channel from a first processed EEG signal, and recording the acquired data of the preset channel as a first channel EEG signal;

[0090] Acquire data of a second preset channel from the second processed EEG signal, and record the acquired data of the preset channel as a second channel EEG signal;

[0091] Step A2, performing connectivity estimation on all second channel EEG signals to obtain first estimation data;

[0092] Connectivity estimation is performed on the first channel EEG signal and the second channel EEG signal to obtain second estimation data.

[0093] In the above embodiment, the first preset channel and the second preset channel can be the same channel, that is, the first channel EEG signal and the second channel EEG signal of the same preset channel are respectively obtained from the first processed EEG signal and the second processed EEG signal, and then all the second channel EEG signals are subjected to connectivity estimation to obtain first estimation data, and the connectivity estimation of the first channel EEG signal and the second channel EEG signal can be performed to obtain second estimation data, thereby realizing the subsequent judgment of whether the neural network model has completed training.

[0094] In the above embodiment, step S5 also includes the following steps:

[0095] Using the first estimation data and the second estimation data to determine whether the neural network model has completed training;

[0096] If so, to create the multi-brain connectivity model;

[0097] If not, return to step S1.

[0098] In the above embodiment, the first estimation data and the second estimation data are used to determine whether the neural network model has completed training, so as to improve the confidence of the created multi-brain connectivity model.

[0099] For example, using the first estimation data and the second estimation data to determine whether the neural network model has completed training may include the following steps:

[0100] Step A3, using the first estimated data to determine whether the degree of connectivity between every two EEG signals in the second frequency band in the neural network model exceeds a first preset threshold;

[0101] If so, it means that there is an inter-brain interaction between every two EEG signals in the second frequency band;

[0102] If not, it indicates that the interaction between each two EEG signals in the second frequency band is poor, and all processes need to be checked, and the process returns to step S1;

[0103] Step A4, using the second estimated data to determine whether the degree of connectivity between the first frequency band EEG signals and the second frequency band EEG signals in each experimental group in the neural network model exceeds a second preset threshold;

[0104] If yes, it means that there is an inter-brain interaction between the first-band EEG signal and the second-band EEG signal in each experimental group, and it is determined that the neural network model has completed training to create the multi-brain connectivity model;

[0105] If not, it indicates that the interaction between the first-band EEG signals and the second-band EEG signals in each experimental group is poor, and all processes need to be checked and return to step S1.

[0106] It should be noted that the degree of connectivity includes the directionality and consistency of EEG flow.

[0107] Among them, the order of step A3 and step A4 can be swapped. As long as the judgment results of "there is inter-brain interaction between every two EEG signals in the second frequency band" and "there is inter-brain interaction between the EEG signals in the first frequency band and the EEG signals in the second frequency band in each experimental group" are obtained, a multi-brain connectivity model can be created.

[0108] In the above embodiment, step S1 specifically includes the following steps: using a multi-channel EEG measurement device to synchronously collect the first original EEG signal of the training recommender and the second original EEG signal of the training recommendee in the experimental group.

[0109] In the above embodiment, by synchronously collecting data from a group of training recommenders and training recommendees in an experimental group, the first original EEG signal and the second original EEG signal are synchronously acquired, thereby realizing synchronous real-time collection of original EEG signals by multiple people, and then facilitating subsequent synchronous real-time monitoring and dynamic analysis of EEG signals by multiple people, further realizing a process-based, full-process study of group neural similarities, and effectively avoiding the problems of post-sampling, strong subjectivity, and social expectation bias of traditional measurement methods.

[0110] The above-mentioned multi-channel EEG measurement equipment can adopt a 64-conductance electrode cap, using saline or gel electrodes, and the impedance of each electrode point is lower than 5kΩ, thereby achieving good contact between the electrode cap and the scalp, facilitating the collection of EEG signals.

[0111] In the above embodiment, the step of preprocessing the collected first original EEG signal in step S2 includes but is not limited to: amplifying the collected first original EEG signal, analyzing segment interception, noise reduction, artifact removal, and bandpass filtering. Artifact removal includes but is not limited to: removing electrooculography, electromyography, electrocardiography, and power frequency interference.

[0112] Similarly, the step of preprocessing the collected second original EEG signal in step S2 is consistent with the above-mentioned step of preprocessing the first original EEG signal, and will not be elaborated here.

[0113] and performing a wavelet transform on the preprocessed first original EEG signal to obtain a first processed EEG signal, and

[0114] The preprocessed second original EEG signal is subjected to wavelet transformation to obtain a second processed EEG signal.

[0115] In the above embodiment, step S3 specifically includes the following steps:

[0116] Acquire the energy value of each first preset time period in the first preset frequency band from the first processed EEG signal, and use all energy values ​​in the first preset frequency band as the EEG signal of the first frequency band;

[0117] The energy value of each second preset time period in the second preset frequency band is acquired from the second processed EEG signal, and all energy values ​​in the second preset frequency band are used as the EEG signal of the second frequency band.

[0118] In the above embodiment, the first preset frequency band and the second preset frequency band may be the same frequency band;

[0119] For example, at least one of “theta”, “alpha”, “beta”, “gamma”, etc. may be selected as the first preset frequency band.

[0120] In the above embodiment, the first preset time period and the second preset time period may be the same time period;

[0121] For example, the first preset time period and the second preset time period may both be set to 1 second.

[0122] In the above embodiment, by taking all energy values ​​in the first preset frequency band as the first frequency band EEG signals, and taking all energy values ​​in the second preset frequency band as the second frequency band EEG signals, the subsequent group brain neural similarity calculation is realized based on the EEG frequency band energy of specific brain regions, thereby improving the accuracy of the neural similarity index and the PDC value.

[0123] In the above embodiment, obtaining the neural similarity index group in step S4 specifically includes the following steps:

[0124] The second frequency band EEG signals are paired in pairs to obtain second frequency band EEG signal groups, and the first preset calculation tool is used to perform inter-group correlation calculation on each group of second frequency band EEG signal groups to obtain the neural similarity index of each group of second frequency band EEG signal groups.

[0125] In the above embodiment, the first preset calculation tool may adopt the NeuroRA toolkit based on python.

[0126] In the above embodiment, obtaining the PDC value group in step S4 specifically includes the following steps:

[0127] The first-band EEG signals and the second-band EEG signals in each experimental group are input into the following formula to obtain a PDC value group:

[0128]

[0129] Among them, when PDC xy When (f) = 1, α x (f) Consistency;

[0130] A xy(f) is the Fourier transform of the coefficients of the multivariate autoregression model;

[0131] α x (f) is used to represent A xy(f) The xth column of

[0132] Used to represent A xy(f) The yth row of

[0133] x is used to represent the EEG signal of the recommender;

[0134] y is used to represent the consumer's EEG signal;

[0135] PDC xy (f) is used to indicate the degree to which the recommender’s EEG signal guides the recommendee’s EEG signal;

[0136] In the above embodiment, the neural network model includes a first neural network and a second neural network, and the multi-brain connectivity model includes a first multi-brain connectivity model and a second multi-brain connectivity model;

[0137] Step S5 specifically includes the following steps:

[0138] Step S51, using a preset transformation algorithm to perform data transformation on the neural similarity indexes in the neural similarity index group to obtain first transformation data;

[0139] Using a preset conversion algorithm to perform data conversion on the PDC values ​​in the PDC value group to obtain second conversion data;

[0140] In the above embodiment, the neural similarity index and the PDC value can be converted into data through a preset transformation algorithm, so that the data features of the converted data will be smoother when displayed and meet the input requirements of the neural network model.

[0141] Preferably, the preset transformation algorithm may adopt the Box-Cox algorithm.

[0142] Step S52, dividing the first converted data into a first test set and a first training set;

[0143] Dividing the second transformed data into a second test set and a second training set;

[0144] In the above embodiment, because the separation is performed on the same data set, it can not only ensure that the data feature quantities of the two data sets are the same, but also ensure that the test set used for evaluation and the training set used for model training do not affect each other. Finally, the test set can also be used to analyze and judge the accuracy of the model.

[0145] The ratio of the test set and the training set can be customized according to the needs. For example, if there are N data in the data set, these N data are evenly divided into K subsets, and each subset contains N / K data. In each verification, several subsets of the K subsets are used as test sets, and the remaining subsets are used as training sets, and then the test set is used for model evaluation.

[0146] The second conversion data is divided in the same manner as the first conversion data;

[0147] For example, 30% of the data in the first conversion data can be used as the first test set, and 70% of the data in the first conversion data can be used as the first training set;

[0148] Then, 30% of the data in the second conversion data needs to be used as the second test set, and 70% of the data in the second conversion data needs to be used as the second training set.

[0149] Step S53, obtaining data feature dimensions of the first conversion data and the second conversion data;

[0150] Step S54, when the data feature dimension is less than the preset feature dimension threshold, the first neural network is trained using the first training set and the second training set to obtain a first trained neural network model, and the first test set and the second test set are used to verify the first trained neural network model to create a first multi-brain connectivity model;

[0151] When the data feature dimension is greater than or equal to a preset feature dimension threshold, the second neural network is trained using the first training set and the second training set to obtain a second trained neural network model, and the second trained neural network model is verified using the first test set and the second test set to create a first multi-brain connectivity model.

[0152] In the above embodiment, the first neural network and the second neural network are different. For example, the first neural network may be an SVM classifier, and the second neural network may be an SVM-AdaBoost classification model.

[0153] For example, in SVM classifier training, the grid search method can be used to determine the SVM parameters for the training set, and then the radial basis function is selected according to the SVM parameters. For example, when the SVM parameter is determined to be 1, the radial basis function in the SVM classifier can use the RBF Gaussian radial basis kernel function.

[0154] In the above embodiment, the second neural network is a SVM-AdaBoost classification model, and the SVM-AdaBoost classification model includes multiple SVM classifiers.

[0155] In the above embodiment, in order to improve the practicality of the second multi-brain connectivity model, the loss ratio can be set in the second multi-brain connectivity model. The second neural network is an SVM-AdaBoost classification model. The SVM-AdaBoost classification model is a strong classifier model composed of multiple weighted weak classifiers. The SVM-AdaBoost classification model includes multiple SVM classifiers, and the SVM classifier is a strong classifier. Now, adding the SVM classifier as a weak classifier to the AdaBoost model will have a good prediction effect.

[0156] Specifically, in the process of training the second neural network, the data features of the training set are weighted by the loss ratio, and the weights are initialized according to the importance of each data feature in the model. Each data feature corresponds to an SVM classifier, that is, weighted processing is achieved for each SVM classifier in the second neural network; and the weighted SVM is added to the AdaBoost boosting framework through the boosting framework, so that the initial training set will obtain a diverse training set subset, and then the base classifier is generated by the subset to calculate the error rate of SVM classification. The base classifier is composed of multiple SVM features, so the corresponding base classifier can be generated after a given K rounds of training. Although the recognition rate of each base classifier is not very high, the combined classifier has a high recognition rate, thereby improving the recognition rate of the algorithm. The boosted classifier is weighted to generate the final result classifier.

[0157] In the above embodiment, in step S6, the direction of inter-brain information transmission is estimated between at least two to-be-estimated recommendees in the to-be-estimated recommendee group, and the multi-brain connectivity model may use consistency as an output value;

[0158] For example, when the neural similarity index is greater than or equal to the first preset index threshold, it is determined that the group neural similarity is strong, and the multi-brain connectivity model at this time can output "consistency", that is, the evaluation result is: the group of recommended persons to be estimated resonates with the stimulus source;

[0159] When the neural similarity index is greater than or equal to the second preset index threshold and less than the first preset index threshold, it is determined that the group neural similarity is converging and the group neural similarity is strong. At this time, the multi-brain connectivity model can output "consistent", that is, the evaluation result is: the group of recommended persons to be estimated resonates with the stimulus source;

[0160] When the neural similarity index is less than the second preset index threshold, it is determined that the group neural similarity is insufficient. At this time, the multi-brain connectivity model can be output as "inconsistent", that is, the evaluation result is: the group of recommended persons to be estimated does not resonate with the stimulus source.

[0161] The first preset index threshold is greater than the second preset index threshold. The first preset index threshold and the second preset index threshold can be customized. For example, the first preset index threshold can be set to 0.6 and the second preset index threshold can be set to 0.4.

[0162] In the above embodiment, for estimating the direction of inter-brain information transmission between the to-be-estimated recommender group and the to-be-estimated recommendee group, the multi-brain connectivity model may take acceptance or not as an output value;

[0163] For example, when the PDC value is close to the first preset PDC threshold, it is determined that the recommender to be estimated in one experimental group has a significant guiding effect on the recommended person to be estimated. At this time, the multi-brain connectivity model can output "receiving", that is, the evaluation result is: the recommender to be estimated has a high influence on the recommended person to be estimated;

[0164] When the PDC value is close to the second preset PDC threshold, it is determined that the inter-brain neural similarity between the training recommenders and the training recommendees in a group of experimental groups is insufficient. At this time, the multi-brain connectivity model can be output as "not received", that is, the evaluation result is: the influence of the recommender to be estimated on the recommendee to be estimated is low.

[0165] The first preset PDC threshold is greater than the second preset PDC threshold. The first preset PDC threshold and the second preset PDC threshold can be customized. For example, the first preset PDC threshold can be set to 1 and the second preset PDC threshold can be set to 0. Therefore, a PDC value greater than or equal to 0.5 can be considered to be close to the first preset PDC threshold, and a PDC value less than 0.5 can be considered to be close to the second preset PDC threshold.

[0166] In the above embodiment, the recommended scene stimulation material can be presented by the inducing device, and the recommended scene stimulation material can be used as a visual stimulation source, and the recommended scene stimulation material can be processed and stored.

[0167] In the above embodiment, a comprehensive calculation strategy is formed from the inter-brain consistency of the same group in the information flow recommendation scenario and the neural similarity between the recommender and the recommended person, and it has the characteristics of real-time monitoring and quantification, and can scientifically and comprehensively analyze the neural similarity between the recommender and the consumer in the recommendation scenario. At present, the research on EEG recognition is mostly based on amplitude characteristics, but ignores the phase synchronization characteristics between multiple brains and the directionality of inter-brain signal activities in frequency space. The phase synchronization characteristic detects the relationship between signal pairs through the instantaneous phase relationship between signals, and can effectively identify the synchronization of people's neural activities. The directionality of signal activity can be used to determine the directional influence between signals in a multivariate data set, and provides the directionality of inter-brain activity (frequency space) on the basis of analyzing synchronization, which helps to understand the role relationship between brain regions of different individuals. The neural similarity of every two recommended persons in the recommended group can be effectively identified by the correlation coefficient of the neural similarity value, and the neural similarity between the recommender and the recommended person can be effectively identified by the correlation coefficient of the PDC value.

[0168] In addition, the present invention innovatively proposes a method for calculating the neural similarity index between the recommender and the recommendee based on the EEG frequency band energy of a specific brain region and the SVM-AdaBoost classification model. The SVM-AdaBoost model can accurately predict the neural similarity between the recommender and the consumer (recommended) and between all consumers (recommended) in the information flow recommendation context.

[0169] Also included is an information flow estimation device based on brain-to-brain interactivity, which includes:

[0170] The acquisition unit divides the experimental subject group into a training recommender group and a training recommendee group according to the task type, and divides one training recommender and one training recommendee into one experimental group, acquires a first original EEG signal of the training recommender and a second original EEG signal of the training recommendee, and the training recommender group and the training recommendee group have the same stimulus source;

[0171] A processing unit connected to the acquisition unit, and configured to perform real-time preprocessing and wavelet transformation on the first original EEG signal to obtain a first processed EEG signal;

[0172] It is also used to perform real-time preprocessing and wavelet transformation on the second original EEG signal to obtain a second processed EEG signal;

[0173] A frequency band data acquisition unit, connected to the processing unit, for acquiring data of a first preset frequency band from the first processed EEG signal, and recording the acquired data of the preset frequency band as a first frequency band EEG signal;

[0174] It is also used to obtain data of a second preset frequency band from the second processed EEG signal, and record the obtained data of the preset frequency band as the EEG signal of the second frequency band;

[0175] A calculation unit connected to the frequency band data acquisition unit, and used for performing inter-group correlation calculation on every two second frequency band EEG signals to obtain a neural similarity index group of the training recommendee group, wherein the neural similarity index group includes a neural similarity index between every two second frequency band EEG signals;

[0176] Furthermore, according to partial directional coherence analysis, the first-band EEG signals and the second-band EEG signals in each experimental group are calculated to obtain a PDC value group of the degree of brain-to-brain connectivity between the training recommender group and the training recommendee group, the PDC value group including the PDC value of the degree of brain-to-brain connectivity between the first-band EEG signals and the second-band EEG signals in each experimental group;

[0177] A creation unit is connected to the calculation unit and is used to train the neural network model according to the neural similarity index group and the PDC value group of the inter-brain connection degree to create a multi-brain connectivity model;

[0178] an estimation unit, using a multi-brain connectivity model to estimate the direction of inter-brain information transmission between at least two to-be-estimated recommendees of the to-be-estimated recommendee group, and evaluating the stimulus source of the to-be-estimated recommendee group according to the estimation result;

[0179] A multi-brain connectivity model is used to estimate the direction of inter-brain information transmission between the recommender group to be estimated and the recommended group to be estimated, and the recommender group to be estimated is evaluated based on the estimation results.

[0180] In the above embodiment, the information flow estimation device further includes a visualization module for displaying the evaluation result.

[0181] The information flow estimation device has the functions of dynamic prediction and visualization, which can help users intuitively obtain the data of group neural similarity in information flow recommendation, thereby helping e-commerce live broadcast platforms to make scientific decisions. At the same time, it is non-invasive, safe, efficient, and low-cost, and can promote live broadcast recommendation, film and television drama recommendation, advertising recommendation effect evaluation, and other fields, and has broad market application prospects.

[0182] It should be noted that the various embodiments of the information flow estimation device of the present invention are consistent with the various embodiments of the above-mentioned information flow estimation method, and will not be described in detail here.

[0183] The above are only preferred embodiments of the present invention, and are not intended to limit the implementation methods and protection scope of the present invention. Those skilled in the art should be aware that all solutions obtained by equivalent substitutions and obvious changes made using the description and illustrations of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for estimating information flow based on brain-to-brain interactivity, characterized in that: The steps include: Step S1, dividing the experimental subject group into a training recommender group and a training recommendee group according to the task type, and dividing a training recommender and a training recommendee into a group of experimental groups, collecting a first original EEG signal of the training recommender and a second original EEG signal of the training recommendee, wherein the training recommender group and the training recommendee group have the same stimulus source; Step S2, performing real-time preprocessing and wavelet transform on the first original EEG signal to obtain a first processed EEG signal; Performing real-time preprocessing and wavelet transform on the second original EEG signal to obtain a second processed EEG signal; Step S3, acquiring data of a first preset frequency band from the first processed EEG signal, and recording the acquired data of the preset frequency band as a first frequency band EEG signal; Acquire data of a second preset frequency band from the second processed EEG signal, and record the acquired data of the preset frequency band as the EEG signal of the second frequency band; Step S4, performing inter-group correlation calculation on every two EEG signals in the second frequency band to obtain a neural similarity index group of the training recommendee group, wherein the neural similarity index group includes a neural similarity index between every two EEG signals in the second frequency band; The first frequency band EEG signals and the second frequency band EEG signals in each experimental group are calculated according to partial directional coherence analysis to obtain a PDC value group of the degree of inter-brain connectivity between the training recommender group and the training recommendee group, wherein the PDC value group includes the PDC value of the degree of inter-brain connectivity between the first frequency band EEG signals and the second frequency band EEG signals in each experimental group; Step S5, training a neural network model according to the neural similarity index group and the PDC value group of the inter-brain connectivity degree to create a multi-brain connectivity model; Step S6, using the multi-brain connectivity model to estimate the direction of inter-brain information transmission between at least two to-be-estimated recommendees in the to-be-estimated recommendee group, and evaluating the stimulus source of the to-be-estimated recommendee group according to the estimation result; The multi-brain connectivity model is used to estimate the direction of inter-brain information transmission between the recommender group to be estimated and the recommendee group to be estimated, and the recommender group to be estimated is evaluated according to the estimation result.

2. The method for estimating information flow based on brain-to-brain interactivity according to claim 1, characterized in that: After step S2, the method further includes: Step A1, acquiring data of a first preset channel from the first processed EEG signal, and recording the acquired data of the preset channel as a first channel EEG signal; Acquire data of a second preset channel from the second processed EEG signal, and record the acquired data of the preset channel as a second channel EEG signal; Step A2, performing connectivity estimation on all the second channel EEG signals to obtain first estimation data; Connectivity estimation is performed on the first channel EEG signal and the second channel EEG signal to obtain second estimation data.

3. The method for estimating information flow based on brain-to-brain interactivity according to claim 2, characterized in that: The step S5 also includes the following steps: Using the first estimation data and the second estimation data to determine whether the neural network model has completed training; If so, to create the multi-brain connectivity model; If not, return to step S1.

4. The method for estimating information flow based on brain-to-brain interactivity according to claim 1, characterized in that: The step S1 specifically includes the following steps: using a multi-channel EEG measurement device to synchronously collect the first original EEG signal of the training recommender and the second original EEG signal of the training recommendee in the experimental group.

5. The information flow estimation method based on brain-to-brain interactivity according to claim 1, characterized in that: The step S3 specifically comprises the following steps: Acquire the energy value of each first preset time period in the first preset frequency band from the first processed EEG signal, and use all energy values ​​in the first preset frequency band as the EEG signal in the first frequency band; The energy value of each second preset time period in the second preset frequency band is acquired from the second processed EEG signal, and all energy values ​​in the second preset frequency band are used as the EEG signal in the second frequency band.

6. The method for estimating information flow based on brain-to-brain interactivity according to claim 1, characterized in that: The step S4 of obtaining the neural similarity index group specifically includes the following steps: The second-frequency-band EEG signals are paired in pairs to obtain second-frequency-band EEG signal groups, and the first preset calculation tool is used to perform inter-group correlation calculation on each group of second-frequency-band EEG signal groups to obtain a neural similarity index for each group of second-frequency-band EEG signal groups.

7. The method for estimating information flow based on brain-to-brain interactivity according to claim 1, characterized in that: The step S4 of obtaining the PDC value group specifically includes the following steps: Input the first frequency band EEG signal and the second frequency band EEG signal in each experimental group into the following formula to obtain the PDC value group: Among them, PDC xy When (f) = 1, α x (f) and α y (f) consistency; A xy (f) Fourier transform used to represent the coefficients of the multivariate autoregressive model; α x (f) is used to represent A xy (f) column x; Used to represent A xy (f) row y; x is used to represent the EEG signal in the first frequency band; y is used to represent the EEG signal in the second frequency band; PDC xy (f) is used to indicate the degree to which the recommender’s EEG signal guides the recommendee’s EEG signal.

8. The method for estimating information flow based on brain-to-brain interactivity according to claim 1, characterized in that: The neural network model includes a first neural network and a second neural network, and the multi-brain connectivity model includes a first multi-brain connectivity model and a second multi-brain connectivity model; The step S5 specifically comprises the following steps: Step S51, using a preset transformation algorithm to perform data transformation on the neural similarity index in the neural similarity index group to obtain first transformation data; Performing data conversion on the PDC values ​​in the PDC value group using the preset conversion algorithm to obtain second conversion data; Step S52, dividing the first converted data into a first test set and a first training set; Dividing the second transformed data into a second test set and a second training set; Step S53, obtaining data feature dimensions of the first conversion data and the second conversion data; Step S54, when the data feature dimension is less than a preset feature dimension threshold, the first neural network is trained using the first training set and the second training set to obtain a first trained neural network model, and the first test set and the second test set are used to verify the first trained neural network model to create the first multi-brain connectivity model; When the data feature dimension is greater than or equal to a preset feature dimension threshold, the second neural network is trained using the first training set and the second training set to obtain a second trained neural network model, and the second trained neural network model is verified using the first test set and the second test set to create the first multi-brain connectivity model.

9. The method for estimating information flow based on brain-to-brain interactivity according to claim 1, characterized in that: The second neural network is a SVM-AdaBoost classification model, and the SVM-AdaBoost classification model includes multiple SVM classifiers.

10. An information flow estimation device based on brain-to-brain interactivity, characterized in that: include: A collection unit, which divides the experimental subject group into a training recommender group and a training recommendee group according to the task type, and divides a training recommender and a training recommendee into a group of experimental groups, and collects a first original EEG signal of the training recommender and a second original EEG signal of the training recommendee, wherein the training recommender group and the training recommendee group have the same stimulus source; a processing unit connected to the acquisition unit, and configured to perform real-time preprocessing and wavelet transformation on the first original EEG signal to obtain a first processed EEG signal; Also used for performing real-time preprocessing and wavelet transformation on the second original EEG signal to obtain a second processed EEG signal; a frequency band data acquisition unit, connected to the processing unit, and configured to acquire data of a first preset frequency band from the first processed EEG signal, and record the acquired data of the preset frequency band as a first frequency band EEG signal; Also used for acquiring data of a second preset frequency band from the second processed EEG signal, and recording the acquired data of the preset frequency band as the EEG signal of the second frequency band; a calculation unit connected to the frequency band data acquisition unit, and configured to perform inter-group correlation calculation on every two EEG signals in the second frequency band to obtain a neural similarity index group of the training recommendee group, wherein the neural similarity index group includes a neural similarity index between every two EEG signals in the second frequency band; Furthermore, according to partial directional coherence analysis, the first frequency band EEG signals and the second frequency band EEG signals in each group of the experimental group are calculated to obtain a PDC value group of the degree of brain-to-brain connectivity between the training recommender group and the training recommendee group, wherein the PDC value group includes the PDC value of the degree of brain-to-brain connectivity between the first frequency band EEG signals and the second frequency band EEG signals in each group of the experimental group; A creation unit connected to the calculation unit, used to train a neural network model according to the neural similarity index group and the PDC value group of the inter-brain connectivity degree to create a multi-brain connectivity model; an estimating unit, using the multi-brain connectivity model to estimate the direction of brain-to-brain information transmission between at least two to-be-estimated recommendees of the to-be-estimated recommendee group, and evaluating the stimulus source of the to-be-estimated recommendee group according to the estimation result; The multi-brain connectivity model is used to estimate the direction of inter-brain information transmission between the recommender group to be estimated and the recommendee group to be estimated, and the recommender group to be estimated is evaluated according to the estimation result.

Images