A psychological consultation and healing system based on human-computer dialogue and core algorithms

Abstract

A psychological counseling and healing system based on human-computer dialogue and core algorithms includes a user end and a system terminal. The user end includes a permission management module, a voice collection module, a voice sending module and an interface interaction module. The system terminal includes a psychological healing speech collection module, a voice recognition module, a psychological healing speech matching module, and a psychological healing speech update module. The beneficial effects of the present invention are that it can realize online and offline human-computer dialogue, and judge the user's psychological state according to the keywords extracted from the user's voice, and then match the most relevant psychological healing speech to provide psychological counseling for the user.

Description

Technical Field

[0001] The invention relates to the field of intelligent medical care, and specifically to a psychological consultation and healing system based on human-computer dialogue and core algorithms. Background Art

[0002] Regarding mental health issues, relevant research results show that basically all groups in society have negative emotions and expression disorders, and the incidence of mental illness is as high as about 30%. According to data analysis, the incidence of illness is still on the rise. Negative emotions such as depression, anxiety, and depression affect people's normal study and life, and greatly reduce their health.

[0003] Therefore, the present invention proposes a psychological consultation and healing system based on human-computer dialogue and core algorithms, hoping to provide people with timely and effective physical and mental health care. Summary of the invention

[0004] In view of the above problems, the present invention aims to provide a psychological counseling and healing system based on human-computer dialogue and core algorithms.

[0005] The purpose of the present invention is achieved through the following technical solutions:

[0006] A psychological consultation and healing system based on human-computer dialogue and core algorithms, including a user terminal and a system terminal, wherein:

[0007] The user end includes a permission management module, a voice collection module, a voice sending module and an interface interaction module. The permission management module is used to obtain the user's permissions, including a permission establishment unit and a permission allocation unit. The permission management executor is the system administrator. The system administrator establishes the required permissions through the permission establishment unit, including obtaining the user's personal information and opening the voice permission. The user's permissions are allocated through the permission allocation unit. After the voice permission is opened, the user's voice is transmitted to the voice collection module. The voice collection module will store the user's voice in the database, and in order to ensure the user's privacy, only the user's last historical record voice is saved. If the user has not been online for more than 30 days, all the user's usage records will be cleared, and the user's voice will be sent to the system terminal through the voice sending module. Through the feedback of the system terminal, a psychological consultation and healing system is obtained. The interface interaction module includes an image display unit, a video display unit and a speaker unit to reply to images, videos and audios;

[0008] The system terminal includes a psychological healing speech collection module, a speech recognition module, a psychological healing speech matching module, and a psychological healing speech update module. The psychological healing speech collection module includes a crawling unit, a storage unit, a psychological healing speech feature extraction unit, a scoring unit, and a screening unit. The crawling unit crawls the existing psychological healing speech in the network and database in real time, and stores the value in the MySQL database through the storage unit. The psychological healing speech feature extraction unit performs lightweight information processing to reduce the storage pressure of the MySQL database. The extracted features need to be labeled. According to the features in the labels, each label is scored in the scoring unit. The screening unit filters out the psychological healing speech with a low score value to further release the storage pressure of the MySQL database. The speech recognition module is used to receive the data from the user end. The voice sent by the voice sending module, the voice recognition module includes a voice-to-text conversion unit and a voice-to-text feature extraction unit, wherein the voice-to-text conversion unit realizes the conversion of the user's voice into text, and the voice-to-text feature extraction unit extracts the feature information such as keywords of the user's voice and text, and matches the psychological healing words with the highest relevance to the MySQL database storing the psychological healing words through the psychological healing words matching module, wherein the psychological healing words matching module includes a feature training unit, a testing unit and a verification unit, to respectively complete the training, testing and verification of the psychological healing words, and the psychological healing words updating module will regularly update the MySQL database according to the user scores of each service, clean up the psychological healing words with low scores, and crawl the latest psychological healing words as replacements.

[0009] Furthermore, the permission management module is used to obtain the user's permissions, including a permission establishment unit and a permission allocation unit. The permission management executor is the system administrator. The system administrator establishes the required permissions through the permission establishment unit, including obtaining the user's personal information and enabling voice permissions. The user's personal information includes age, gender, occupation, marital status and main demands. The user permissions are allocated through the permission allocation unit to refine the permission management, improve the granularity of permission settings, regularly update the administrator key, and regularly maintain the administrator permission list.

[0010] Furthermore, the user voice collected by the voice collection module is transmitted to the voice recognition module of the system terminal through the voice sending module using wifi to recognize features in the user voice.

[0011] Furthermore, the psychological healing words feature extraction unit in the psychological healing words collection module performs lightweight information processing to reduce the storage pressure of the MySQL database. The extracted features need to be labeled, including whether it is a psychological healing words, the adaptability of the psychological healing words, the matching degree of the psychological healing words professional terms, and the source of the psychological healing words.

[0012] Furthermore, the psychological healing speech feature extraction unit and the scoring unit in the psychological healing speech collection module are used to extract features from the tags and score them. If the crawling unit crawls a total of I speeches, the feature value of whether the i-th speech in the tag is a psychological healing speech is recorded as α i , where 1≤i≤I, and i is an integer, α i ∈{0,1}, if α i =0, then the speech is not a psychological healing speech. If α i =1, then the speech is not a psychological healing speech, and the fitness of the i-th psychological healing speech in the label is recorded as β i , where β i ∈[0,1], when β i The closer the value of is to 0, the less effective the adaptation of the psychological healing words in the tag is for the user’s voice reply. i The closer the value is to 1, the more effective the adaptation of the i-th psychological healing words in the label is for the user's voice reply. The matching degree of the i-th psychological healing words in the label is recorded as γ i , where γ i ∈[0,1], when γ i The closer the value of is to 0, the lower the credibility of the matching degree of the professional words of psychological healing techniques in the i-th label. i The closer the value of is to 1, the higher the credibility of the matching degree of the professional words of the psychological healing words in the i-th label is. The source of the quantified psychological healing words in the i-th label is recorded as δ i , and in order to ensure that the adaptability label of the above-mentioned psychological healing words and the matching degree of the professional words of the psychological healing words do not fall into the local solution interval, the source of the psychological healing words after expansion and quantification converges to the interval [0,100], where δ i ∈[0,100], when δ i The closer the value of is to 0, the lower the reliability of the matching degree of the professional words of psychological healing techniques in the i-th label. i The closer the value is to 100, the higher the reliability of the matching degree of the professional words of psychological healing techniques in the i-th label is. For this, the objective function and constraints can be obtained as follows:

[0013] max(β),max(γ),max(δ)

[0014] So that:

[0015]

[0016] Assuming that the adaptability of psychological therapy words, the matching degree of psychological therapy words, and the weights of the source of psychological therapy words after quantification are w1, w2, and w3 respectively, w1+w2+w3=1, the weighted normalized labels are as follows: Eβ =w1*β,E γ =w2*γ,E δ =w3*δ

[0017] Among them, E β is the weighted adaptability of psychological healing speech, E γ is the weighted matching degree of professional words in psychological therapy speech, E δ The weighted quantitative psychological healing words are used to obtain the ideal solution of the objective function using multi-criteria decision making and TOPSIS algorithm. * and the non-ideal solution A of the objective function - as follows:

[0018]

[0019]

[0020] in, It is an ideal solution for the adaptability of psychological healing speech. It is an ideal solution for the matching degree of professional words in psychological therapy speech. To quantify the ideal solution for the source of psychological healing words, It is a non-ideal solution to the adaptability of psychological healing speech. It is a non-ideal solution for the matching degree of professional words in psychological therapy speech. In order to quantify the non-ideal solutions of the sources of psychological healing words, the Euclidean distance between the ideal solution and the non-ideal solution is as follows:

[0021]

[0022]

[0023] Then calculate this psychological healing speech as The optimal solution can be written as:

[0024] F=max{F *}

[0025] So that:

[0026] w1,w2,w3∈[0,1]

[0027] w1+w2+w3=1

[0028] Furthermore, the speech recognition module uses a hierarchical digital dynamic network to recognize the speech input by the user. Assuming that the net input n of the mth layer in the hierarchical digital dynamic network is m (t) is denoted as:

[0029]

[0030] in, is the set of layer subscripts directly forward connected to the mth layer, DL m,l is the set of all delays on the tapped delay line between the lth layer and the mth layer, LW m,l is the layer weight between the l input vectors and the mth layer, IW m,l is the input weight between the lth input vector and the mth layer, I m is the set of input vectors connected to the mth layer, DI m,l is the set of all delays on the tapped delay line between the lth input vector and the mth layer, p l (td) is the lth input vector at time td, b m is the bias of the mth layer, then the output of the mth layer is:

[0031] a m (t) = f m (n m (t))

[0032] To train the network, the steepest descent method is used, using the sum of squared errors: Among them, F is the sum of squared errors, e is the error of each iteration, Q is the total number of iterations, τ is the target output, a is the actual output, and the two components of the gradient are:

[0033]

[0034]

[0035] Among them, lw 1,1 is the layer weight between an input vector and the first layer, iw 1,1 is the input weight between the first input vector and the first layer, then the derivative of the network output with respect to the weight is: The network operation formula is: a(t) = lw 1,1 a(t-1)+iw 1,1 p(t), then the derivative of the network output with respect to the weight can be written as:

[0036] Furthermore, the hierarchical digital dynamic network is improved by using the BP back propagation rule. Assuming that U is the set of all output layer numbers, u∈U, the set X is the set of all input layer numbers, x∈X, and x and the performance index F are one-to-one mappings, the gradient is calculated by the chain rule as follows:

[0037]

[0038] For each iteration, the summation term includes the terms corresponding to each output layer. If the performance metric F(x) is not an explicit function of a particular output a u (t), then the explicit derivative is 0, where Using the chain rule, it can be further generalized as follows: To calculate Assume that the net input of layer x is There is where is the k-th input weight of the x-th layer, and a u (t - d) is the output of the u-th layer at the (t - d)-th iteration. Define the sensitivity of the u-th layer as: The explicit derivative can be obtained as:

[0039]

[0040] where represents the sensitivity of the k-th weight of the u-th layer to the i-th weight of the m-th layer.

[0041] Furthermore, the psychological healing speech matching module uses the I speech texts crawled by the psychological healing speech collection module as sample one, denoted as I` = {i1, i2,..., i I}, and the U user speech texts output by the speech recognition module as sample two, denoted as U` = {u1, u2,..., u U}. Then the overall mean of sample one and sample two is denoted as The overall covariance is cov(I`, U`), and the overall Pearson correlation coefficient can be written as:

[0042]

[0043] Define the correlation coefficient threshold Tu, Tu ∈ [0, 1]. If ρ I`,U` > Tu, then select this psychological healing speech to reply to the user.

[0044] Furthermore, if max(ρ I`,U` ) < Tu, it means that the timeliness of the crawled psychological healing speeches can no longer meet the user's needs. At this time, the psychological healing speech update module needs to clean up the speeches with low long-term usage rate in the MySQL database and restart the crawling unit to obtain the latest psychological healing speeches.

[0045] Furthermore, the interface interaction module includes an image display unit, a video display unit, and a speaker unit to reply with images, videos, and audio. Generate a mobile App through Android Studio and obtain the user's speech permission to complete the human-computer dialogue function between the user and the psychological counseling and healing system.

[0046] The beneficial effects of the present invention include: being able to crawl psychological healing words through a web crawler, and adaptively updating the database in real time according to the time period and relevance information, combining multi-criteria decision making and TOPSIS algorithm to increase the credibility of the scores under the conditions of multiple labels and multiple feature values, constructing ideal solutions and non-ideal solutions and calculating their Euclidean distances to obtain more reliable scores, assigning weights to different features, and calculating the optimal value of the objective function under constraints, which can effectively alleviate the psychological healing words with low utility values ​​in the database, release the cache, and increase the quality of user services. In order to extract the features of speech recognition, a hierarchical digital dynamic network is used. Compared with a static network, a hierarchical digital dynamic network can have multiple layers connected to n layers, some of which may form a reply connection through a tapped delay line. The hierarchical digital dynamic network can also have multiple input vectors, which can be connected to any layer of the network. In contrast, a static multi-layer network assumes that there is only one input vector and it is only connected to the first layer, and the layers are connected in numerical order. In a hierarchical digital dynamic network, any layer can be connected to any layer, and this embodiment uses the chain rule in BP back propagation for improvement. Through iteration and weighting, sensitivity values ​​and explicit derivatives are obtained, and feedback expressions are obtained. The original hierarchical digital dynamic network is expanded into a hierarchical digital dynamic back propagation network. Under this network, the features of speech can be effectively identified. Finally, the Pearson correlation coefficient is compared with the threshold to decide whether the psychological healing words should be replied to the user. The present embodiment proposes a psychological counseling and healing system based on human-computer dialogue and core algorithms, which can realize online and offline human-computer dialogue, and judge the user's psychological state based on the keywords extracted from the user's voice, and then match the most relevant psychological healing words to provide psychological counseling for the user. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The invention is further illustrated by the accompanying drawings, but the embodiments in the accompanying drawings do not constitute any limitation to the invention. A person skilled in the art can obtain other drawings based on the following drawings without creative work.

[0048] Figure 1 It is a schematic diagram of the structure of the present invention. DETAILED DESCRIPTION

[0049] The present invention is further described in conjunction with the following examples.

[0050] See also Figure 1 , a psychological consultation and healing system based on human-computer dialogue and core algorithms in this embodiment includes a user terminal and a system terminal, wherein:

[0051] The user end includes a permission management module, a voice collection module, a voice sending module and an interface interaction module. The permission management module is used to obtain the user's permissions, including a permission establishment unit and a permission allocation unit. The permission management executor is the system administrator. The system administrator establishes the required permissions through the permission establishment unit, including obtaining the user's personal information and opening the voice permission. The user's permissions are allocated through the permission allocation unit. After the voice permission is opened, the user's voice is transmitted to the voice collection module. The voice collection module will store the user's voice in the database, and in order to ensure the user's privacy, only the user's last historical record voice is saved. If the user has not been online for more than 30 days, all the user's usage records will be cleared, and the user's voice will be sent to the system terminal through the voice sending module. Through the feedback of the system terminal, a psychological consultation and healing system is obtained. The interface interaction module includes an image display unit, a video display unit and a speaker unit to reply to images, videos and audios;

[0052] The system terminal includes a psychological healing speech collection module, a speech recognition module, a psychological healing speech matching module, and a psychological healing speech update module. The psychological healing speech collection module includes a crawling unit, a storage unit, a psychological healing speech feature extraction unit, a scoring unit, and a screening unit. The crawling unit crawls the existing psychological healing speech in the network and database in real time, and stores the value in the MySQL database through the storage unit. The psychological healing speech feature extraction unit performs lightweight information processing to reduce the storage pressure of the MySQL database. The extracted features need to be labeled. According to the features in the labels, each label is scored in the scoring unit. The screening unit filters out the psychological healing speech with a low score value to further release the storage pressure of the MySQL database. The speech recognition module is used to receive the data from the user end. The voice sent by the voice sending module, the voice recognition module includes a voice-to-text conversion unit and a voice-to-text feature extraction unit, wherein the voice-to-text conversion unit realizes the conversion of the user's voice into text, and the voice-to-text feature extraction unit extracts the feature information such as keywords of the user's voice and text, and matches the psychological healing words with the highest relevance to the MySQL database storing the psychological healing words through the psychological healing words matching module, wherein the psychological healing words matching module includes a feature training unit, a testing unit and a verification unit, to respectively complete the training, testing and verification of the psychological healing words, and the psychological healing words updating module will regularly update the MySQL database according to the user scores of each service, clean up the psychological healing words with low scores, and crawl the latest psychological healing words as replacements.

[0053] Specifically, the permission management module is used to obtain the user's permissions, including a permission establishment unit and a permission allocation unit. The permission management executor is the system administrator. The system administrator establishes the required permissions through the permission establishment unit, including obtaining the user's personal information and enabling voice permissions. The user's personal information includes age, gender, occupation, marital status and main demands. The user permissions are allocated through the permission allocation unit to refine the permission management, improve the granularity of permission settings, regularly update the administrator key, and regularly maintain the administrator permission list.

[0054] Specifically, the user voice collected by the voice collection module is transmitted to the voice recognition module of the system terminal through the voice sending module using wifi to recognize features in the user voice.

[0055] Specifically, the psychological healing words feature extraction unit in the psychological healing words collection module performs lightweight information processing to reduce the storage pressure of the MySQL database. The extracted features need to be labeled, including whether it is a psychological healing words, the adaptability of the psychological healing words, the matching degree of the psychological healing words professional terms, and the source of the psychological healing words.

[0056] Specifically, psychological healing discourse can be divided into three categories: discourse related to developmental psychological problems, discourse related to adaptive psychological problems, and discourse related to obstructive psychological problems.

[0057] Specifically, the psychological healing speech feature extraction unit and the scoring unit in the psychological healing speech collection module are used to extract features from the label and score it. If the crawling unit crawls a total of I speeches, the feature value of whether the i-th speech in the label is a psychological healing speech is recorded as α i , where 1≤i≤I, and i is an integer, α i ∈{0,1}, if α i =0, then the speech is not a psychological healing speech. If α i =1, then the speech is not a psychological healing speech, and the fitness of the i-th psychological healing speech in the label is recorded as β i , where β i ∈[0,1], when β i The closer the value of is to 0, the less effective the adaptation of the psychological healing words in the tag is for the user’s voice reply. i The closer the value is to 1, the more effective the adaptation of the i-th psychological healing words in the label is for the user's voice reply. The matching degree of the i-th psychological healing words in the label is recorded as γ i , where γ i ∈[0,1], when γ i The closer the value of is to 0, the lower the credibility of the matching degree of the professional words of psychological healing techniques in the i-th label. iThe closer the value of is to 1, the higher the credibility of the matching degree of the professional words of the psychological healing words in the i-th label is. The source of the quantified psychological healing words in the i-th label is recorded as δ i , and in order to ensure that the adaptability label of the above-mentioned psychological healing words and the matching degree of the professional words of the psychological healing words do not fall into the local solution interval, the source of the psychological healing words after expansion and quantification converges to the interval [0,100], where δ i ∈[0,100], when δ i The closer the value of is to 0, the lower the reliability of the matching degree of the professional words of psychological healing techniques in the i-th label. i The closer the value is to 100, the higher the reliability of the matching degree of the professional words of psychological healing techniques in the i-th label is. For this, the objective function and constraints can be obtained as follows:

[0058] max(β),max(γ),max(δ)

[0059] So that:

[0060]

[0061] Assuming that the adaptability of psychological therapy words, the matching degree of psychological therapy words, and the weights of the source of psychological therapy words after quantification are w1, w2, and w3 respectively, w1+w2+w3=1, the weighted normalized labels are as follows: E β =w1*β,E γ =w2*γ,E δ =w3*δ

[0062] Among them, E β is the weighted adaptability of psychological healing speech, E γ is the weighted matching degree of professional words in psychological therapy speech, E δ The weighted quantitative psychological healing words are used to obtain the ideal solution of the objective function using multi-criteria decision making and TOPSIS algorithm. * and the non-ideal solution A of the objective function - as follows:

[0063]

[0064]

[0065] in, It is an ideal solution for the adaptability of psychological healing techniques. It is an ideal solution for the matching degree of professional words in psychological therapy speech. To quantify the ideal solution for the source of psychological healing words, It is a non-ideal solution to the adaptability of psychological healing speech. It is a non-ideal solution for the matching degree of professional words in psychological therapy speech. In order to quantify the non-ideal solutions of the sources of psychological healing words, the Euclidean distance between the ideal solution and the non-ideal solution is as follows:

[0066]

[0067]

[0068] Then calculate this psychological healing speech as The optimal solution can be written as:

[0069] F=max{F *}

[0070] So that:

[0071] w1,w2,w3∈[0,1]

[0072] w1+w2+w3=1

[0073] Specifically, the speech recognition module uses a hierarchical digital dynamic network to recognize the speech input by the user. Assuming that the net input n of the mth layer in the hierarchical digital dynamic network is m (t) is denoted as:

[0074]

[0075] in, is the set of layer subscripts directly forward connected to the mth layer, DL m,l is the set of all delays on the tapped delay line between the lth layer and the mth layer, LW m,l is the layer weight between the l input vectors and the mth layer, IW m,l is the input weight between the lth input vector and the mth layer, I m is the set of input vectors connected to the mth layer, DI m,l is the set of all delays on the tapped delay line between the lth input vector and the mth layer, p l (td) is the lth input vector at time td, b m is the bias of the mth layer, then the output of the mth layer is:

[0076] a m (t) = f m (n m (t))

[0077] To train the network, the steepest descent method is used, using the sum of squared errors: Among them, F is the sum of squared errors, e is the error of each iteration, Q is the total number of iterations, τ is the target output, a is the actual output, and the two components of the gradient are:

[0078]

[0079]

[0080] Among them, lw 1,1 is the layer weight between an input vector and the first layer, iw 1,1 is the input weight between the first input vector and the first layer, then the derivative of the network output with respect to the weight is: The network operation formula is: a(t) = lw 1,1 a(t-1)+iw 1,1 p(t), then the derivative of the network output with respect to the weight can be written as:

[0081]

[0082] Specifically, the layered digital dynamic network is improved using the BP back propagation rule. Assume that U is the set of all output layer numbers, u∈U, set X is the set of all input layer numbers, x∈X, and x and the performance index F are one-to-one mappings. By using the chain rule, the gradient is calculated as follows:

[0083]

[0084] For each iteration, the sum includes the terms corresponding to each output layer. If the performance indicator F(x) is not a specific output a u (t), then the explicit derivative is 0, where The chain rule can be further extended: To calculate Assume that the net input of layer number x is have in, is the k-th input weight of the x-th layer, a u (td) is the output of the u-th layer at td iterations, and the sensitivity of the u-th layer is defined as: The explicit derivative can be found:

[0085]

[0086] in It represents the sensitivity of the k-th weight of the u-th layer to the i-th weight of the m-th layer.

[0087] Specifically, the psychological healing conversation matching module uses the I conversation texts crawled by the psychological healing conversation collection module as sample one, denoted as I` = {i1, i2, …, i I}, and the U user voice texts output by the output layer of the voice recognition module as sample two, denoted as U` = {u1, u2, …, u U}. Then the overall mean of sample one and sample two is denoted as The overall covariance is:

[0088]

[0089] Then the overall Pearson correlation coefficient can be written as:

[0090]

[0091] Define the correlation coefficient threshold Tu, Tu ∈ [0, 1]. If ρ I`,U` > Tu, then select this psychological healing conversation to reply to the user.

[0092] Preferably, if max(ρ I`,U` ) < Tu, it means that the timeliness of the crawled psychological healing conversations can no longer meet the needs of users. At this time, the psychological healing conversation update module needs to clean up the conversations with low long-term usage rate in the MySQL database and restart the crawling unit to obtain the latest psychological healing conversations.

[0093] Preferably, to ensure the effectiveness of the psychological healing conversations replied to users, Tu is set to 0.8, that is, when ρ I`,U` ∈ [Tu, 1], the psychological healing of this human-machine conversation is effective.

[0094] Specifically, the interface interaction module includes an image display unit, a video display unit, and a speaker unit to reply with images, videos, and audio. Generate a mobile App through Android Studio and obtain the user's voice permission to complete the human-machine conversation function between the user and the psychological counseling and healing system.

[0095] The beneficial effects of the present invention include: being able to crawl psychological healing words through a web crawler, and adaptively updating the database in real time according to the time period and relevance information, combining multi-criteria decision making and TOPSIS algorithm to increase the credibility of the scores under the conditions of multiple labels and multiple feature values, constructing ideal solutions and non-ideal solutions and calculating their Euclidean distances to obtain more reliable scores, assigning weights to different features, and calculating the optimal value of the objective function under constraints, which can effectively alleviate the psychological healing words with low utility values ​​in the database, release the cache, and increase the quality of user services. In order to extract the features of speech recognition, a hierarchical digital dynamic network is used. Compared with a static network, a hierarchical digital dynamic network can have multiple layers connected to the m layer, some of which may form a reply connection through a tapped delay line. The hierarchical digital dynamic network can also have multiple input vectors, which can be connected to any layer of the network. In contrast, a static multi-layer network assumes that there is only one input vector and it is only connected to the first layer, and the layers are connected in numerical order. In a hierarchical digital dynamic network, any layer can be connected to any layer, and this embodiment uses the chain rule in BP back propagation for improvement. Through iteration and weighting, sensitivity values ​​and explicit derivatives are obtained, and feedback expressions are obtained. The original hierarchical digital dynamic network is expanded into a hierarchical digital dynamic back propagation network. Under this network, the features of speech can be effectively identified. Finally, the Pearson correlation coefficient is compared with the threshold to decide whether the psychological healing words should be replied to the user. The present embodiment proposes a psychological counseling and healing system based on human-computer dialogue and core algorithms, which can realize online and offline human-computer dialogue, and judge the user's psychological state based on the keywords extracted from the user's voice, and then match the most relevant psychological healing words to provide psychological counseling for the user.

[0096] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention, rather than to limit the scope of protection of the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the essence and scope of the technical solution of the present invention.

Claims

1. A psychological consultation and healing system based on human-computer dialogue and core algorithms, including a user terminal and a system terminal, wherein: The user end includes a permission management module, a voice collection module, a voice sending module and an interface interaction module. The permission management module is used to obtain the user's permissions, including a permission establishment unit and a permission allocation unit. The permission management executor is the system administrator. The system administrator establishes the required permissions through the permission establishment unit, including obtaining the user's personal information and opening the voice permission. The user's permissions are allocated through the permission allocation unit. After the voice permission is opened, the user's voice is transmitted to the voice collection module. The voice collection module will store the user's voice in the database, and in order to ensure the user's privacy, only the user's last historical record voice is saved. If the user has not been online for more than 30 days, all the user's usage records will be cleared, and the user's voice will be sent to the system terminal through the voice sending module. Through the feedback of the system terminal, a psychological consultation and healing system is obtained. The interface interaction module includes an image display unit, a video display unit and a speaker unit to reply to images, videos and audios; The system terminal includes a psychological healing speech collection module, a speech recognition module, a psychological healing speech matching module, and a psychological healing speech update module. The psychological healing speech collection module includes a crawling unit, a storage unit, a psychological healing speech feature extraction unit, a scoring unit, and a screening unit. The crawling unit crawls the existing psychological healing speech in real time based on the network and database, and stores the value in the MySQL database through the storage unit. The psychological healing speech feature extraction unit performs lightweight information processing to reduce the storage pressure of the MySQL database. The extracted features need to be labeled. According to the features in the labels, each label is scored in the scoring unit. The screening unit filters out the psychological healing speech with low scores to further release the storage pressure of the MySQL database. The speech recognition module is used to receive the data from the user. The speech recognition module receives the speech sent by the voice sending module of the end, and includes a speech-to-text conversion unit and a speech-to-text feature extraction unit, wherein the speech-to-text conversion unit realizes converting the user's speech into text, and extracts the keyword feature information of the user's speech and text through the speech-to-text feature extraction unit, and matches the psychological healing speech with the highest relevance through the psychological healing speech matching module in the MySQL database storing the psychological healing speech, wherein the psychological healing speech matching module includes a feature training unit, a testing unit and a verification unit, to respectively complete the training, testing and verification of the psychological healing speech, and the psychological healing speech update module will regularly update the MySQL database according to the user's score for each service, clean up the psychological healing speech with low scores, and crawl the latest psychological healing speech as a replacement; The psychological healing speech feature extraction unit and scoring unit in the psychological healing speech collection module are used to extract features and score tags. If the crawling unit crawls a total of I speeches, the feature value of whether the i-th speech in the tag is a psychological healing speech is recorded as α i , where 1≤i≤I, and i is an integer, α i ∈{0,1}, if α i =0, then the speech is not a psychological healing speech. If α i =1, then the speech is a psychological healing speech, and the fitness of the i-th psychological healing speech in the label is recorded as β i , where β i ∈[0,1], when β i The closer the value of is to 0, the less effective the adaptation of the psychological healing words in the tag is for the user’s voice reply. i The closer the value is to 1, the more effective the adaptation of the i-th psychological healing words in the label is for the user's voice reply. The matching degree of the i-th psychological healing words in the label is recorded as γ i , where γ i ∈[0,1], when γ i The closer the value of is to 0, the lower the credibility of the matching degree of the professional words of psychological healing techniques in the i-th label. i The closer the value of is to 1, the higher the credibility of the matching degree of the professional words of the psychological healing words in the i-th label is. The source of the quantified psychological healing words in the i-th label is recorded as δ i , and in order to ensure that the adaptability label of the above-mentioned psychological healing words and the matching degree of the professional words of the psychological healing words do not fall into the local solution interval, the source of the psychological healing words after expansion and quantification converges to the interval [0,100], where δ i ∈[0,100], when δ i The closer the value of is to 0, the lower the reliability of the matching degree of the professional words of psychological healing techniques in the i-th label. i The closer the value is to 100, the higher the reliability of the matching degree of the professional words of psychological healing techniques in the i-th label is. For this, the objective function and constraints can be obtained as follows: max(β),max(γ),max(δ) So that: Assuming that the adaptability of psychological therapy words, the matching degree of psychological therapy words, and the weights of the source of psychological therapy words after quantification are w1, w2, and w3 respectively, w1+w2+w3=1, the weighted normalized labels are as follows: E β =w1*β,E γ =w2*γ,E δ =w3*δ Among them, E β is the weighted adaptability of psychological healing speech, E γ is the weighted matching degree of professional words in psychological therapy speech, E δ The weighted quantitative psychological healing words are used to obtain the ideal solution of the objective function using multi-criteria decision making and TOPSIS algorithm. * and the non-ideal solution A of the objective function - as follows: in, It is an ideal solution for the adaptability of psychological healing speech. It is an ideal solution for the matching degree of professional words in psychological therapy speech. To quantify the ideal solution for the source of psychological healing words, It is a non-ideal solution to the adaptability of psychological healing speech. It is a non-ideal solution for the matching degree of professional words in psychological therapy speech. In order to quantify the non-ideal solutions of the sources of psychological healing words, the Euclidean distance between the ideal solution and the non-ideal solution is as follows: Then calculate this psychological healing speech as The optimal solution can be written as: F=max{F * } So that: w1,w2,w3∈[0,1] w1+w2+w3=1.

2. A psychological consultation and healing system based on human-computer dialogue and core algorithm according to claim 1, characterized in that: The permission management module is used to obtain the user's permissions, including a permission establishment unit and a permission allocation unit. The permission management executor is the system administrator. The system administrator establishes the required permissions through the permission establishment unit, including obtaining the user's personal information and enabling voice permissions. The user's personal information includes age, gender, occupation, marital status and main demands. The user's permissions are allocated through the permission allocation unit to refine the permission management, improve the granularity of permission settings, regularly update the administrator key, and regularly maintain the administrator permission list.

3. A psychological consultation and healing system based on human-computer dialogue and core algorithm according to claim 1, characterized in that: The user voice collected by the voice collection module is transmitted to the voice recognition module of the system terminal through the voice sending module using wifi to recognize the features in the user voice.

4. According to claim 1, a psychological consultation and healing system based on human-computer dialogue and core algorithm is characterized in that: The psychological healing words feature extraction unit in the psychological healing words collection module performs lightweight information processing to reduce the storage pressure of the MySQL database. The extracted features need to be labeled, including whether it is a psychological healing words, the adaptability of the psychological healing words, the matching degree of the psychological healing words professional terms, and the source of the psychological healing words.

5. According to claim 1, a psychological consultation and healing system based on human-computer dialogue and core algorithm is characterized in that: The speech recognition module uses a hierarchical digital dynamic network to recognize the speech input by the user. Assume that the net input n of the mth layer in the hierarchical digital dynamic network is m (t) is denoted as: in, is the set of layer subscripts directly forward connected to the mth layer, DL m,l is the set of all delays on the tapped delay line between the lth layer and the mth layer, LW m,l is the layer weight between the l input vectors and the mth layer, IW m,l is the input weight between the lth input vector and the mth layer, I m is the set of input vectors connected to the mth layer, DI m,l is the set of all delays on the tapped delay line between the lth input vector and the mth layer, p l (td) is the lth input vector at time td, b m is the bias of the mth layer, then the output of the mth layer is: a m (t)=f m (n m (t)) To train the network, the steepest descent method is used, using the sum of squared errors: Among them, F is the sum of squared errors, e is the error of each iteration, Q is the total number of iterations, τ is the target output, a is the actual output, and the two components of the gradient are: Among them, lw 1,1 is the layer weight between an input vector and the first layer, iw 1,1 is the input weight between the first input vector and the first layer, then the derivative of the network output with respect to the weight is: The network operation formula is: a(t) = lw 1,1 a(t-1)+iw 1,1 p(t), then the derivative of the network output with respect to the weight can be written as:

6. A psychological consultation and healing system based on human-computer dialogue and core algorithm according to claim 5, characterized in that: The layered digital dynamic network is improved by using the BP back propagation rule. Assume that U is the set of all output layer numbers, u∈U, set X is the set of all input layer numbers, x∈X, and x is a one-to-one mapping relationship with the performance index F. By using the chain rule, the gradient is calculated as follows: For each iteration, the sum includes the terms corresponding to each output layer. If the performance indicator F(x) is not a specific output a u (t), then the explicit derivative is 0, where The chain rule can be further extended: To calculate Assume that the net input of layer number x is have in, is the k-th input weight of the x-th layer, a u (td) is the output of the u-th layer at td iterations, and the sensitivity of the u-th layer is defined as: The explicit derivative can be found in It represents the sensitivity of the k-th weight of the u-th layer to the i-th weight of the m-th layer.

7. According to claim 1, a psychological consultation and healing system based on human-computer dialogue and core algorithm is characterized in that: The psychological healing speech matching module crawls I speech through the psychological healing speech collection module as sample 1, denoted as I`={i1,i2,…,i I }, the output layer generated by the speech recognition module outputs U user speech texts as sample 2, denoted as U`={u1,u2,…,u U }, then the population means of sample one and sample two are recorded as The overall covariance is cov(I`,U`), then the overall Pearson correlation coefficient can be written as: Define the correlation coefficient threshold Tu, Tu∈[0,1], if ρ I`,U` >Tu, then select the psychological healing words to reply to the user.

8. A psychological consultation and healing system based on human-computer dialogue and core algorithm according to claim 7, characterized in that: If max(ρ I`,U` ) < Tu, it indicates that the timeliness of the crawled psychological healing words cannot meet the user's needs. At this time, the psychological healing words update module needs to clean up the words with low long-term usage rate in the MySQL database and restart the crawling unit to obtain the latest psychological healing words.

9. According to claim 1, a psychological consultation and healing system based on human-computer dialogue and core algorithm is characterized in that: The interface interaction module includes an image display unit, a video display unit and a speaker unit to respond to images, videos and audios. It generates a mobile app through Android Studio and obtains user voice permissions to complete the human-computer dialogue function between the user and the psychological counseling and healing system.

Images