Electronic certificate intelligent classification model based on convolutional neural network

Abstract

The application discloses an electronic certificate intelligent classification model based on a convolutional neural network, and relates to the technical field of deep learning.The application comprises a data module, which is used for obtaining electronic certificate data and storing the data in a certificate library, and is further used for converting the obtained data into data available for a convolutional neural network and randomly dividing the data into a training set and a verification set; a construction and training module, which is used for constructing an electronic certificate intelligent classification model based on the convolutional neural network, and is further used for training and optimizing the electronic certificate intelligent classification model by using the data of the training set and the verification set; and a test module, which is used for inputting the data available for the convolutional neural network in the certificate library into an optimal electronic certificate intelligent classification model, outputting a classification result of a subject of a certificate by the optimal electronic certificate intelligent classification model, and evaluating the advantages and disadvantages of the electronic certificate intelligent classification model by calculating the precision rate, the recall rate and the accuracy rate of the output result.The electronic certificate subject classification result output by the application has high accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of deep learning, in particular to an electronic certificate intelligent classification model based on a convolutional neural network. BACKGROUND

[0002] According to the description of the electronic certificate industry standard, there are many electronic certificate subject types, and the businesses involved in different subject types are quite different.

[0003] For example, according to the "National Integrated Online Government Service Platform Electronic Certificate of Certified Public Accountant (C 0222-2020)", the certificate of certified public accountant belongs to the professional qualification subject type certificate, and needs to be checked every year. According to the inspection results, the certified public accountant annual inspection registration form is generated, and the "annual inspection special seal" is affixed. When the certified public accountant changes the practice office, it needs to handle the transfer procedures, and according to the handling results, the certified public accountant change registration form is generated, and the registered public accountant association of the province and above shall affix the electronic seal of the association.

[0004] For another example, in the "National Integrated Online Government Service Platform Electronic Certificate of Ship Business Transport (C 0209-2019)", the ship business transport certificate belongs to the qualification certification subject type certificate, and the requirements for its application, audit, issuance, change in what scene, cancellation in what situation, and verification and application requirements of the old-age certificate are described in detail.

[0005] As can be seen from the industry standards of different subject type electronic certificates, the electronic certificates of each type involve different subsequent businesses in their complete life cycle.

[0006] At present, the subject classification of electronic certificates is mainly artificial pre-set classification. That is, when the certificate is collected, the subject corresponding to the electronic certificate is selected artificially. The subject type of the certificate is more, time-consuming and laborious, and easy to make mistakes.

[0007] In order to better realize the management and application of electronic certificates, an intelligent subject classification model needs to be provided for the electronic certificate system to support different subject type certificate business handling, provide intelligent recommendation services, and actively recommend related handling businesses under the subject for the people, achieve the goal of "government service actively looking for people", and improve the handling efficiency. SUMMARY

[0008] The present application provides an electronic certificate intelligent classification model based on a convolutional neural network to output the certificate classification results corresponding to the input electronic certificate, aiming at the needs and deficiencies of the current technology development.

[0009] The technical solution adopted by the electronic certificate intelligent classification model based on the convolutional neural network of the present application to solve the above technical problems is as follows:

[0010] An electronic certificate intelligent classification model based on a convolutional neural network, the structure of which comprises:

[0011] A data module for obtaining electronic certificate data and storing in a certificate library, and for converting the obtained electronic certificate data into data available for the convolutional neural network and randomly dividing into a training set and a validation set;

[0012] A construction and training module for constructing an electronic certificate intelligent classification model based on a convolutional neural network, and training the electronic certificate intelligent classification model using the data of the training set, and after the training is completed, optimizing the electronic certificate intelligent classification model using the data of the validation set;

[0013] A test module for inputting the optimal electronic certificate intelligent classification model with the data available for the convolutional neural network in the certificate library, and outputting the classification result of the subject of the certificate by the optimal electronic certificate intelligent classification model, and evaluating the pros and cons of the electronic certificate intelligent classification model by calculating the precision P, recall R and accuracy A of the output result.

[0014] Optionally, the data module obtains electronic certificate data and stores in a certificate library, and the specific operation process comprises:

[0015] Downloading electronic certificate data resources to the locally established certificate library;

[0016] Converting the location information of the electronic certificate data stored in the certificate library into a row and storing in a txt file according to the subject category.

[0017] Further optionally, the data module converts the obtained electronic certificate data into data available for the convolutional neural network, and the specific operation process comprises:

[0018] (1) Based on the certificate library, obtaining all the data of the electronic certificate;

[0019] (2) Purifying the obtained electronic certificate data by constructing a regular expression method to remove useless information and obtaining certificate element data;

[0020] (3) Storing the certificate element data of each certificate obtained in the text document as a row of "subject category-element content" as the original test set;

[0021] (4) Establishing a vocabulary table, the length of which is fixed and sorted according to the word frequency;

[0022] (5) Representing the words in the vocabulary table with corresponding ids, and reversing the ids to words;

[0023] (6) converting the categories contained in the vocabulary into corresponding topic ids, returning the value bits, converting the contents of the test set into corresponding id representations, i.e. word vectors, returning x_file and y_file, respectively storing the surface content id and the topic category id;

[0024] (7) converting the values in x_file and y_file into matrix representation as direct input of the convolutional neural network.

[0025] Preferably, when establishing the vocabulary, the vocabulary size is set to 5000, the number of categories is set to 4, the word vector dimension is set to 64, and the sequence length is set to 600.

[0026] Optionally, the electronic certificate intelligent classification model based on the convolutional neural network constructed by the method comprises an input layer, a convolution calculation layer, a pooling layer, and a full connection layer, wherein,

[0027] The input layer is used to receive the vocabulary in the training set,

[0028] The convolution calculation layer is used to ensure that the width and the word dimension size are the same when the filter slides, and the minimum granularity of the text is slid, and the size of the convolution kernel is set to 5 and the number is set to 256,

[0029] The pooling layer is used to compress the text size, convert the word vector into a scalar, and further extract features, and the maximum pooling strategy is selected,

[0030] The full connection layer is used to integrate the extracted features, and through the connected dropout layer, half of the neurons are activated, and the activated part is transmitted to the softmax layer, and the output of the softmax represents the relative probability between different categories.

[0031] Optionally, when the training module uses the data of the training set to train the electronic certificate intelligent classification model, the following needs to be done in advance:

[0032] The training set is evenly divided into multiple training subsets, and each training subset includes 50 training samples,

[0033] Every 10 rounds of writing tensorboard, the maximum value of total iteration rounds is 10, and the result is printed once every 100 rounds;

[0034] After the training of the electronic certificate intelligent classification model is completed, the construction and training module also uses the data of the validation set to adjust the parameters of the electronic certificate intelligent classification model. After the parameter adjustment, if the output result accuracy of the model based on the validation set is improved, the optimization of the electronic certificate intelligent classification model is realized. After multiple parameter adjustments, the optimal electronic certificate intelligent classification model is obtained.

[0035] Optionally, the specific process of the test module for evaluating the electronic certificate intelligent classification model comprises:

[0036] (1) the true classification result of the electronic certificate data in the certificate library is combined with the predicted classification result of the optimal electronic certificate intelligent classification model to divide into four cases of true negative (TN), true positive (TP), false negative (FN) and false positive (FP);

[0037] (2) the precision P is an evaluation index about the prediction result of the model, which represents how many true positive samples are in the predicted positive samples, and the calculation formula is:

[0038] P=TP / (TP+FP) formula (1);

[0039] (3) the recall R is about all samples in the certificate library, which represents how many correct prediction results are in the true positive samples, and the calculation formula is:

[0040] R=TP / (TP+FN) formula (2);

[0041] (4) the accuracy A is the ratio of the number of correct classification samples to the total number of samples, and the accuracy reflects the judgment ability of the classifier on the whole sample, and the calculation formula is:

[0042] A=(TP+TN) / (TP+FN+FP+TN) formula (3).

[0043] Further, in the process of evaluating the electronic certificate intelligent classification model by the test module, the harmonic mean F1 of the precision and the recall can also be calculated, and the precision and the recall are a pair of mutually exclusive quantities, that is, when the precision is low, the recall is always high, and when the recall is low, the precision is always high;

[0044] The calculation formula of the harmonic mean F1 is:

[0045] F1=(2*P*R) / (P+R) formula (4).

[0046] The electronic certificate intelligent classification model based on the convolutional neural network has the beneficial effects compared with the prior art:

[0047] (1) the electronic certificate intelligent classification model can output the subject type corresponding to the electronic certificate based on the input electronic certificate, has high accuracy, and can improve the work efficiency of government personnel;

[0048] (2) the electronic certificate intelligent classification model can be modified to establish a vocabulary table category, so that the user can conveniently customize the modification of the certificate subject type, and the application and intelligent recommendation system of the certificate subject classification conforming to the construction standard of the national integrated online government service platform can be established, and the standard management of the electronic certificate of each industry business is provided. Attached Figure Description

[0049] Appendix Fig. 1 This is a flowchart of the data module in this invention;

[0050] Appendix Fig. 2 This is a flowchart of the construction and training module in this invention;

[0051] Appendix Fig. 3 This is a flowchart of the testing module in this invention. Detailed Implementation

[0052] To make the technical solution, the technical problem solved, and the technical effect of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with specific embodiments.

[0053] Example 1:

[0054] Combined with appendix Figs. 1-3 This embodiment proposes an intelligent classification model for electronic certificates based on convolutional neural networks, which consists of three parts: a data module, a construction and training module, and a testing module.

[0055] (i) Through the data module, electronic certificate data can be acquired and stored in the certificate database. It is also used to convert the acquired electronic certificate data into data that can be used by the convolutional neural network and randomly divide it into the training set and the validation set.

[0056] The data modules involved acquire electronic certificate data and store it in the certificate database. The specific operation process includes:

[0057] Download electronic certificate data resources to the locally established certificate database;

[0058] The location information of the electronic certificate data stored in the certificate database is converted into a single line and stored in a txt file according to the subject category.

[0059] The data module transforms the acquired electronic certificate data into data that can be used by the convolutional neural network. The specific operation process includes:

[0060] (1) Based on the certificate database, obtain all data of electronic certificates;

[0061] (2) By constructing regular expressions, the acquired electronic certificate data is purified to remove useless information and obtain the photo element data;

[0062] (3) Store the photo element data of each certificate in a text document as a row with “subject category - element content”, as the original test set;

[0063] (4) Establish a vocabulary table, the length of the vocabulary table is fixed, and the vocabulary table is sorted by word frequency. It should be noted that when establishing the vocabulary table, the size of the vocabulary table is set to 5000, the number of categories is set to 4, the dimension of the word vector is set to 64, and the sequence length is set to 600;

[0064] (5) The words in the vocabulary table are represented by corresponding ids, and the ids are converted into words;

[0065] (6) The categories contained in the vocabulary table are converted into corresponding topic ids, and the return value is bit. The content of the test set is converted into corresponding id representation, that is, the word vector. The x_file and y_file are returned, which respectively store the content id and the topic category id;

[0066] (7) The values in the x_file and y_file are converted into matrix representation as the direct input of the convolutional neural network.

[0067] (II) Construction and training module, for constructing an electronic certificate intelligent classification model based on convolutional neural network, and using the data of the training set to train the electronic certificate intelligent classification model. After training, the data of the verification set is also used to optimize the electronic certificate intelligent classification model.

[0068] The constructed electronic certificate intelligent classification model based on convolutional neural network includes an input layer, a convolution calculation layer, a pooling layer and a full connection layer, wherein,

[0069] The input layer is used to receive the vocabulary in the training set,

[0070] The convolution calculation layer is used to ensure that the width and word dimension size are the same when the filter slides, and the minimum granularity of the text is slid. The size of the convolution kernel is set to 5, and the number is set to 256,

[0071] The pooling layer is used to compress the text size, convert the word vector into a scalar, and further extract features. The maximum pooling strategy is selected,

[0072] The full connection layer is used to integrate the extracted features, and through the connected dropout layer, half of the neurons are activated. The activated part is transmitted to the softmax layer. The output of the softmax represents the relative probability between different categories.

[0073] When the construction and training module uses the data of the training set to train the electronic certificate intelligent classification model, the following needs to be done in advance:

[0074] Divide the training set into multiple training subsets, and each training subset includes 50 training samples,

[0075] Set to write to tensorboard every 10 rounds, the maximum value of total iteration rounds is 10, and print the result every 100 rounds;

[0076] After the electronic certificate intelligent classification model is trained, the parameter of the electronic certificate intelligent classification model is adjusted by using the data of the verification set by the constructing and training module. After the adjustment, if the output result accuracy of the model based on the verification set is improved, the optimization of the electronic certificate intelligent classification model is realized. After multiple parameter adjustments, the optimal electronic certificate intelligent classification model is obtained.

[0077] (Three) test module, for using the data available for convolutional neural network in the certificate library to input the optimal electronic certificate intelligent classification model, the optimal electronic certificate intelligent classification model outputs the classification result of the subject to which the certificate belongs, and the precision P, recall R and accuracy A of the output result are calculated to evaluate the advantages and disadvantages of the electronic certificate intelligent classification model.

[0078] The specific process of the test module evaluating the electronic certificate intelligent classification model includes:

[0079] (1) The true classification result of the electronic certificate data in the certificate library and the predicted classification result of the optimal electronic certificate intelligent classification model are combined and divided into four cases: true negative TN, true positive TP, false negative FN and false positive FP, as shown in the following table:

[0080]

[0081] (2) The precision P is an evaluation index of the model prediction result, which represents how many true positive samples are in the predicted positive samples. The calculation formula is:

[0082] P = TP / (TP+FP) Formula (1);

[0083] (3) The recall R is about all samples in the certificate library, which represents how many correct prediction results are in the true positive samples. The calculation formula is:

[0084] R = TP / (TP+FN) Formula (2);

[0085] (4) The accuracy A is the ratio of the number of correctly classified samples to the total number of samples. The accuracy reflects the judgment ability of the classifier on the whole sample, and the calculation formula is:

[0086] A = (TP+TN) / (TP+FN+FP+TN) Formula (3).

[0087] Example two:

[0088] Based on example one, the test module evaluating the electronic certificate intelligent classification model can also calculate the harmonic mean F1 of the precision and recall. The precision and recall are a pair of mutually exclusive quantities, that is, when the precision is low, the recall is always high, and when the recall is low, the precision is always high.

[0089] The calculation formula of the harmonic mean F1 is:

[0090] F1=(2*P*R) / (P+R) Formula (4).

[0091] In conclusion, the electronic certificate intelligent classification model based on the convolutional neural network can output the theme type corresponding to the electronic certificate based on the input electronic certificate, has high accuracy, and can improve the work efficiency of government personnel.

[0092] The above application specific examples have described the principles and implementation modes of the present application in detail, and these examples are only used to help understand the core technical content of the present application. Based on the above specific embodiments of the present application, any improvement and modification of the present application made by the person skilled in the art without departing from the principles of the present application shall fall within the patent protection scope of the present application.

Claims

1. A convolutional neural network-based electronic certificate intelligent classification model, characterized in that The structure comprises: A data module is configured to acquire electronic license data and store the data in a license library, and is further configured to convert the acquired electronic license data into data available for a convolutional neural network and randomly divide the data into a training set and a validation set; wherein the conversion of the acquired electronic license data into data available for the convolutional neural network includes the following specific operations: (1) based on the license library, all data of the electronic license is acquired; (2) the acquired electronic license data is purified by constructing a regular expression to remove useless information and acquire surface element data; (3) the surface element data of each acquired license is stored in a text document as a row of "subject category-element content" as an original test set; (4) a vocabulary table is established, the length of the vocabulary table is fixed, and the vocabulary table is sorted according to the word frequency; (5) the words in the vocabulary table are represented by corresponding ids, and the ids are reversed into words; (6) the categories contained in the vocabulary table are converted into corresponding subject ids, and the return value is bit, the content of the test set is converted into corresponding id representation, that is, a word vector, and x_file and y_file are returned to store the surface content id and the subject category id respectively; and (7) the values in x_file and y_file are converted into matrix representation as direct input of the convolutional neural network; A construction and training module is configured to construct an electronic license intelligent classification model based on the convolutional neural network, train the electronic license intelligent classification model using the data of the training set, and after the training is completed, optimize the electronic license intelligent classification model using the data of the validation set; the constructed electronic license intelligent classification model based on the convolutional neural network comprises an input layer, a convolution calculation layer, a pooling layer and a full connection layer, wherein the input layer is configured to receive the word quantity in the training set, the convolution calculation layer is configured to ensure that the width and the word dimension size are the same when the filter slides, the minimum granularity of the text is slid, and the size of the convolution kernel is set to 5 and the number is set to 256, the pooling layer is configured to compress the text size, convert the word vector into a scalar, and further extract features, and the maximum pooling strategy is selected, and the full connection layer is configured to integrate the extracted features, pass through a connected dropout layer, activate half of the neurons, and pass the activated part to a softmax layer, and the output of the softmax represents the relative probability between different categories; A test module is configured to input data available for the convolutional neural network in the license library into an optimal electronic license intelligent classification model, and the optimal electronic license intelligent classification model outputs a classification result of a subject to which a license belongs, and the advantages and disadvantages of the electronic license intelligent classification model are evaluated by calculating the precision P, recall R and accuracy A of the output result.

2. The electronic certificate intelligent classification model based on a convolutional neural network according to claim 1, characterized in that, The data module acquires electronic license data and stores the data in a license library, and the specific operation process includes: Downloading electronic license data resources to the locally established license library; Converting the position information of the electronic license data stored in the license library into a row and storing the information in a txt file according to the subject category. 3.The electronic certificate intelligent classification model based on a convolutional neural network according to claim 1, characterized in that, When establishing the vocabulary table, the size of the vocabulary table is set to 5000, the number of categories is set to 4, the word vector dimension is set to 64, and the sequence length is set to 600. 4.The electronic certificate intelligent classification model based on a convolutional neural network according to claim 1, characterized in that, When the building and training module trains the electronic certificate intelligent classification model using the data of the training set, the following needs to be done in advance: Divide the training set into multiple training subsets, each including 50 training samples, Set to write tensorboard every 10 rounds, the maximum value of total iteration rounds is 10, and print the result every 100 rounds; After the training of the electronic certificate intelligent classification model is completed, the building and training module also adjusts the parameters of the electronic certificate intelligent classification model using the data of the verification set. After the parameter adjustment, if the output result accuracy of the model based on the verification set is improved, the optimization of the electronic certificate intelligent classification model is realized. After multiple parameter adjustments, the optimal electronic certificate intelligent classification model is obtained. 5.The electronic certificate intelligent classification model based on a convolutional neural network according to claim 1, characterized in that, The specific process of the test module evaluating the electronic certificate intelligent classification model includes: (1) Combine the true classification results of the electronic certificate data in the certificate library with the predicted classification results of the optimal electronic certificate intelligent classification model into four cases: true negative (TN), true positive (TP), false negative (FN), and false positive (FP); (2) The precision P is an evaluation index for the model prediction result, which represents how many true positive samples in the predicted positive samples are true positive. The calculation formula is: P = TP / (TP+FP) Formula (1); (3) The recall R is about all samples in the certificate library, which represents how many correct prediction results are in the positive examples in the true situation samples. The calculation formula is: R = TP / (TP+FN) Formula (2); (4) The accuracy A is the ratio of the number of correctly classified samples to the total number of samples. The accuracy reflects the classification ability of the classifier to the entire sample. The calculation formula is: A = (TP+TN) / (TP+FN+FP+TN) Formula (3). 6.The electronic certificate intelligent classification model based on a convolutional neural network according to claim 5, characterized in that, During the process of the test module evaluating the electronic certificate intelligent classification model, the harmonic mean F1 of the precision and recall can also be calculated. The precision and recall are a pair of mutually exclusive quantities, that is, when the precision is low, the recall is always high, and when the recall is low, the precision is always high; The calculation formula of the harmonic mean F1 is: F1 = (2*P*R) / (P+R) Formula (4).

Images