Artificial intelligence knowledge management system

Abstract

An artificial intelligence knowledge management system disclosed by the present invention comprises a main device and an external device, the main device comprises a knowledge data acquisition module, a storage module, a main processing module, a main information transmission module, a classification module and a data mining module, and the external device comprises a login module, an external processing module and an external information transmission module. According to the artificial intelligence knowledge management system, the sample matrix is iteratively updated according to the iteration formula, the optimal feature matrix is obtained, the optimal feature vector is obtained, the classification accuracy of stored knowledge data is improved, the stored knowledge data are classified, the data are more easily searched during searching, and the searching efficiency is improved. The classified knowledge data is clear in structure and convenient to manage and maintain, and the efficiency and accuracy of data management are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of knowledge management, and particularly to an artificial intelligence knowledge management system. Background Art

[0002] After a user's question, the existing knowledge management system first uses natural language processing technology to understand the user's intention and extract key information, then uses a machine learning model to classify the user into the corresponding knowledge base type, and then retrieves or generates the most relevant answer from the knowledge base according to the text similarity matching or retrieval model. And the system ensures the provision of accurate and timely answers according to the knowledge base and model updated in real time.

[0003] However, it has deficiencies. When transmitting knowledge data, it does not protect the transmitted knowledge data. In this way, during the transmission of knowledge data, it is easy to be attacked by a man-in-the-middle and then the data is tampered with, resulting in incorrect knowledge data. Summary of the Invention

[0004] In view of the deficiencies of the prior art, the present invention provides an artificial intelligence knowledge management system, which solves the problems raised in the above background art.

[0005] To achieve the above object, the present invention is realized through the following technical solutions: An artificial intelligence knowledge management system includes a main device and an external device. The main device includes a knowledge data acquisition module, a storage module, a main processing module, a main information transmission module, a classification module, and a data mining module. The external device includes a login module, an external processing module, and an external information transmission module; Step 1: The knowledge data acquisition module acquires the latest knowledge data on the network; Step 2: The main processing module transports the acquired knowledge data to the inside of the storage module for storage; Step 3: The classification module classifies the knowledge data that needs to be stored inside the storage module; Step 4: The user sends a request through the external information transmission module. The main information transmission module receives the user's sent request and sends the user request to the main processing module; Step 5: The main processing module starts the data mining module to mine the knowledge data inside the storage module, and then sends the mined knowledge data to the external information transmission module through the main information transmission module. Protection will be carried out during data transmission.

[0006] Preferably, the classification steps of the classification module are as follows: Step 21: Extract a feature from the knowledge data inside the storage module, and perform normalization processing on the stored knowledge data through this feature to form compressed code training data: Step 22: Use the compressed code training data to train the encoder according to the following formula that minimizes the objective function of the classification error: where, is the weight coefficient, is the loss function, and the expression of the loss function is , is the class label of the th class, is the th compressed code training data, , is the classification parameter corresponding to the th feature in the th class, is the bias parameter, is the number of features, is the projection matrix, is the projection matrix corresponding to the th feature, is the hash function, is the number of the compressed code training data, is the number of classes of the compressed code training data, and are two normalization functions, which are respectively used to adjust the roles of the classification parameter matrix and the projection matrix , and are two real numbers, which are respectively used to adjust the normalization functions and ; After training, obtain the projection matrix , the classification parameter matrix and the bias matrix , ; and use the hash function: as the binary compressed code encoder; Step 33: Use the classifier based on the binary compressed code to classify the binary compressed code of the knowledge data inside the storage module to obtain the class, and classify the binary compressed code of the knowledge data inside the storage module through the following function: .

[0007] Preferably, the mining method of the data mining module is: Step 31: Extract the data features of the request sent by the user; Step 32: Sequentially extract data features of different categories from the classified knowledge data in the storage module. Step 32: Then, use the cosine algorithm to perform a comparison calculation on the data features of the user's sent request and the data features of different categories. The formula is as follows: , where is the data feature of the user's sent request, is the data feature of different categories, The closer the value is to 1, the higher the similarity. Step 33: After finding the category data with the highest similarity, perform subsequent mining within it.

[0008] Preferably, the method for feature extraction is: Construct an iterative formula for the sample matrix based on the knowledge data inside the storage module, the request data sent by the user, and the data of different categories: Iteratively update the sample matrix according to the iterative formula to obtain the feature matrix, where is a constant, and a random iterative initial value is given , , perform iterative update according to the iterative formula. When the number of iterations reaches the set threshold k, the iteration terminates, and is the feature matrix. k represents the product of matrices, Y is an intermediate parameter, P represents the projection gradient algorithm, and P projects all negative numbers in matrix Z to 0.

[0009] Preferably, data transmission between the external device and the main device is protected. The protection method is: Step 51: The external device registers and logs in through the login module, constructs a word2vec model, and deploys the word2vec model inside the external device and the main device 8. The external device uses the word2vec model to encode the registered user information data to obtain encoded vectors of different user information data. Step two: The external device encrypts the encoded vector of the user information data using the homomorphic encryption algorithm, and transmits the encrypted data to the main device through the external information transmission module. The main information transmission module receives the encrypted user information data and transmits it to the storage module for storage. Step three: The external device sends a request to the main device through the external information transmission module. Step 4: The master device receives the request sent by the external device through the master information transmission module, then mines knowledge data according to the request, converts the mined knowledge data into encoded vectors using the word2vec model, and transmits the encoded vectors after homomorphic encryption; Step 5: The external device decrypts the received ciphertext of the encoded vector and performs an inverse operation on the decryption result using the word2vec model to obtain the calculated authentication result data.

[0010] Preferably, the construction of the word2vec model includes an input layer, a hidden layer, and an output layer. The one-hot encoding result of the user information data is used as the input value of the input layer of the word2vec model. Then, the calculation result of the hidden layer of the word2vec model is: where represents the one-hot encoding result of the th type of user information data, represents the corresponding calculation result of the hidden layer; represents the weight matrix in the hidden layer, the size of is , represents the size of the user feature dictionary, represents the size of the hidden layer; The calculation result of the hidden layer is input to the output layer. For the th type of user information data, the output result of the output layer is: where represents the weight matrix in the output layer, the size of ; The obtained encoded vector result is Beneficial effects

[0011] The present invention provides an artificial intelligence knowledge management system. Compared with the prior art, it has the following beneficial effects: 1. This artificial intelligence knowledge management system iteratively updates the sample matrix according to the iteration formula to obtain the optimal feature matrix and the optimal feature vector, improving the classification accuracy of storing knowledge data, classifying the stored knowledge data, making it easier to search for data when searching, and the classified knowledge data has a clear structure, facilitating management and maintenance, and improving the efficiency and accuracy of data management.

[0012] 2. The artificial intelligence knowledge management system constructs a word2vec model and deploys the word2vec model on an external device. The external device uses the word2vec model to encode the received user information data fields to obtain encoding vectors of different user information data fields. The external device uses a homomorphic encryption algorithm to encrypt the encoding vectors and transmits the encryption result to the main device. The main device stores the ciphertext of the encoding vectors. The main device converts the mined knowledge data into encoding vectors through the word2vec model, performs homomorphic encryption on the encoding vectors, and then transmits them to the external device. The external device decrypts the received ciphertext of the encoding vectors and uses the word2vec model to perform an inverse operation on the decryption result to obtain the knowledge data, thus preventing the data from being tampered with by a man-in-the-middle attack during the process of the external device obtaining the knowledge data. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 It is a schematic diagram of the system of the present invention.

[0014] Figure 2 It is a schematic diagram of the workflow structure of the present invention.

[0015] In the figure: 1. Knowledge data acquisition module; 2. Storage module; 3. Main processing module; 4. Main information transmission module; 5. Classification module; 6. Data mining module; 7. External information transmission module; 8. Main device; 9. External device; 10. Login module; 11. External processing module. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0016] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0017] Please refer to Figure 1-2 , the present invention provides a technical solution: an artificial intelligence knowledge management system, including a main device 8 and an external device 9. The main device 8 includes a knowledge data acquisition module 1, a storage module 2, a main processing module 3, a main information transmission module 4, a classification module 5, and a data mining module 6. The external device 9 includes a login module 10, an external processing module 11, and an external information transmission module 7; Step 1. The knowledge data acquisition module 1 acquires the latest knowledge data on the network; Step 2. The main processing module 3 conveys the acquired knowledge data to the inside of the storage module 2 for storage; Step 3. The classification module 5 classifies the knowledge data that needs to be stored inside the storage module 2; Step 4: The user sends a request through the external information transmission module 7. The main information transmission module 4 receives the request sent by the user and sends the user request to the main processing module 3; Step 5: The main processing module 3 starts the data mining module 6 to mine the knowledge data inside the storage module 2, and then sends the mined knowledge data to the external information transmission module 7 through the main information transmission module 4. Data protection will be carried out during data transmission; The steps for the classification module 5 to classify are as follows: Step 21: Extract a feature from the knowledge data inside the storage module 2, and perform normalization processing on the stored knowledge data through this feature to form compressed code training data: Step 22: Use the compressed code training data to train the encoder according to the following formula that minimizes the objective function of the classification error: Among them, is the weight coefficient, is the loss function, and the expression of the loss function is , is the class label of the th category, is the th compressed code training data, , is the th classification parameter corresponding to the th feature in the th category, is the bias parameter, is the number of features, is the th projection matrix corresponding to the th feature, is the number of compressed code training data, is the number of categories of compressed code training data, and are two normalization functions, which are used to adjust the classification parameter matrix and the projection matrix respectively, and are two real numbers, which are used to adjust the normalization functions and respectively; After training, the projection matrix , the classification parameter matrix and the bias matrix are obtained, ; and the hash function: As a binarized compression code encoder; Step 33: Classify the binarized compression codes of the knowledge data inside the storage module 2 using a classifier based on binarized compression codes to obtain categories. Classify the binarized compression codes of the knowledge data inside the storage module 2 through the following function: ; The mining method of the data mining module 6 is as follows: Step 31: Extract the data features of the request sent by the user; Step 32: Sequentially extract the data features of different categories from the classified knowledge data inside the storage module 2; Step 32: Then, perform a comparison calculation on the data features of the request sent by the user and the data features of different categories through the cosine algorithm. The formula is as follows: , where is the data feature of the request sent by the user, is the data feature of different categories, The closer the value is to 1, the higher the similarity; Step 33: After finding the category data with the highest similarity, perform subsequent mining inside it; Construct an iterative formula for the sample matrix based on the knowledge data inside the storage module 2, the request data sent by the user, and the data of different categories: Iteratively update the sample matrix according to the iterative formula to obtain the feature matrix, where is a constant, and a random iterative initial value is given , , and perform iterative updates according to the iterative formula. When the number of iterations reaches the set threshold k, the iteration terminates, and is the feature matrix, k represents the product of the matrices, Y is an intermediate parameter, P represents the projection gradient algorithm, and P means projecting all negative numbers in the matrix Z to 0; By iteratively updating the sample matrix according to the iterative formula, an optimal feature matrix is obtained, and an optimal feature vector is obtained, improving the classification accuracy of the stored knowledge data. Classifying the stored knowledge data makes it easier to search for data. The classified knowledge data has a clear structure, facilitating management and maintenance, and improving the efficiency and accuracy of data management.

[0018] Furthermore, data transmission between the external device 9 and the main device 8 is protected. The protection method is as follows: Step 51: The external device 9 registers and logs in through the login module 10, constructs a word2vec model, and deploys the word2vec model inside the external device 9 and the main device 8. The external device 9 uses the word2vec model to encode the registered and logged-in user information data to obtain the encoded vectors of different user information data; Step 2: The external device 9 uses the homomorphic encryption algorithm to encrypt the encoded vectors of the user information data, and transmits the encrypted data to the main device 8 through the external information transmission module 7. The main information transmission module 4 receives the encrypted user information data and transmits it to the storage module 2 for storage; Step 3: The external device 9 sends a request to the main device 8 through the external information transmission module 7; Step 4: The main device 8 receives the request sent by the external device 9 through the main information transmission module 4, then mines the knowledge data according to the request, converts the mined knowledge data into encoded vectors using the word2vec model, and performs homomorphic encryption on the encoded vectors and then transmits them; Step 5: The external device 9 decrypts the received ciphertext of the encoded vector, and uses the word2vec model to perform an inverse operation on the decryption result to obtain the calculated authentication result data.

[0019] Constructing the word2vec model includes an input layer, a hidden layer, and an output layer. Taking the one-hot encoding result of the user information data as the input value of the input layer of the word2vec model, the calculation result of the hidden layer of the word2vec model is: where represents the one-hot encoding result of the th type of user information data, represents the corresponding calculation result of the hidden layer; represents the weight matrix in the hidden layer, the size of is , represents the size of the user feature dictionary, represents the size of the hidden layer; Inputting the calculation result of the hidden layer into the output layer, for the th type of user information data, the output result of the output layer is: where represents the weight matrix in the output layer, the size of is By encrypting and protecting the knowledge data sent by the main device 8, it can prevent data from being tampered with by a man-in-the-middle attack during the data transmission process, resulting in incorrect knowledge data obtained by the users of the external device 9.

[0020] Meanwhile, the content not described in detail in this specification belongs to the prior art well-known to those skilled in the art.

[0021] It should be noted that, in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, article or device comprising the element.

[0022] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principle and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An artificial intelligence knowledge management system, characterized in that: The invention comprises a main device (8) and an external device (9), wherein the main device (8) comprises a knowledge data acquisition module (1), a storage module (2), a main processing module (3), a main information transmission module (4), a classification module (5) and a data mining module (6), and the external device (9) comprises a login module (10), an external processing module (11) and an external information transmission module (7); Step 1: the knowledge data acquisition module (1) acquires the latest knowledge data on the network; Step 2: the main processing module (3) transmits the acquired knowledge data to the storage module (2) for storage; Step 3, the classification module (5) classifies the knowledge data to be stored in the storage module (2); Step 4: The user sends a request through the external information transmission module (7), and the main information transmission module (4) receives the user's request and sends the user's request to the main processing module (3); Step 5: The main processing module (3) starts the data mining module (6) to mine the knowledge data inside the storage module (2), and then sends the mined knowledge data to the external information transmission module (7) through the main information transmission module (4). Protection will be performed during data transmission.

2. An artificial intelligence knowledge management system according to claim 1, characterized in that: The classification steps of the classification module (5) are as follows: Step 21: extract a feature from the knowledge data in the storage module (2), and normalize the stored knowledge data using the feature to form compressed code training data: Step 22: Using the compressed code training data, perform encoder training according to the following formula that minimizes the objective function of classification error: in, is the weight coefficient, is the loss function, and the expression of the loss function is , for In the the category identifier of the category, For the compressed code training data, , For the In the category, The classification parameters corresponding to the features are is the bias parameter, is the number of features, is the projection matrix, For the The projection matrix corresponding to the features is is a hash function, is the number of the compressed code training data, is the number of categories of the compressed code training data, and There are two normalization functions, which are used to adjust the classification parameter matrix and the projection matrix The role of and are two real numbers, used to adjust the normalization function and ; After training, the projection matrix is ​​obtained , the classification parameter matrix and the bias matrix , ; and the hash function: As a binary compression code encoder; Step 33: Use a classifier based on binary compression code to classify the binary compression code of the knowledge data in the storage module (2) to obtain a category. Use the following function to classify the binary compression code of the knowledge data in the storage module (2): 。 3. An artificial intelligence knowledge management system according to claim 2, characterized in that: The mining method of the data mining module (6) is: Step 31, extracting data features of the request sent by the user; Step 32, extracting data features of different categories from the knowledge data classified in the storage module (2); Step 32: Then, the data features of the request sent by the user and the data features of different categories are compared and calculated by using the cosine algorithm. The formula is as follows: ,in The data characteristics of the request sent by the user, are the data features of different categories. The closer the value is to 1, the higher the similarity; Step 33: After finding the category data with the highest similarity, perform subsequent mining within it.

4. An artificial intelligence knowledge management system according to claim 3, characterized in that: The feature extraction method is: The iterative formulas for constructing the sample matrix are respectively based on the knowledge data in the storage module (2), the request data sent by the user and the data of different categories: The sample matrix is ​​iteratively updated according to the iterative formula to obtain a feature matrix, where is a constant, giving a random initial value for iteration , , iteratively update according to the iterative formula, when the number of iterations reaches the set threshold k, the iteration is terminated, and the result is That is, k represents the product of the matrices, Y is the intermediate parameter, P represents the projection gradient algorithm, and P means projecting all negative numbers in the matrix Z to 0.

5. An artificial intelligence knowledge management system according to claim 4, characterized in that: The data transmitted between the external device (9) and the main device (8) will be protected, and the protection method is as follows; Step 51, the external device (9) registers and logs in through the login module (10), constructs a word2vec model, and deploys the word2vec model inside the external device (9) and the main device 8, and the external device (9) uses the word2vec model to encode the registered user information data to obtain encoding vectors of different user information data; Step 2: The external device (9) uses a homomorphic encryption algorithm to encrypt the user information data encoding vector, and transmits the encrypted data to the main device (8) through the external information transmission module (7); the main information transmission module (4) receives the user information encrypted data and transmits it to the storage module (2) for storage; Step 3: The external device (9) sends a request to the main device (8) through the external information transmission module (7); Step 4: The main device (8) receives the request sent by the external device (9) through the main information transmission module (4), then mines the knowledge data according to the request, converts the mined knowledge data into a coding vector using the word2vec model, and transmits the coding vector after homomorphic encryption; Step 5: The external device (9) decrypts the received coded vector ciphertext and uses the word2vec model to perform an inverse operation on the decryption result to obtain the calculated authentication result data.

6. An artificial intelligence knowledge management system according to claim 5, characterized in that: The word2vec model includes an input layer, a hidden layer, and an output layer. As the input value of the input layer of the word2vec model, the calculation result of the hidden layer of the word2vec model is: in Indicates One-hot encoding result of class user information data, Indicates the corresponding hidden layer calculation results; represents the weight matrix in the hidden layer, The size is , Indicates the size of the user feature dictionary, Indicates the size of the hidden layer; inputs the calculation results of the hidden layer to the output layer. Class user information data, the output result of the output layer is: in represents the weight matrix in the output layer, The size is ; The resulting encoding vector is .

Images