A method and device for correcting industry categories

Abstract

The present application relates to the field of data processing technology, and in particular to an industry category correction method and device, which obtain attribute information of an operation object and a pre-labeled industry category; input the attribute information into a trained industry category prediction model, perform feature extraction on the attribute information through a convolution layer, obtain each feature value of the attribute information, determine the feature value that meets the preset feature value condition from each feature value through a pooling layer, and determine the probability value of the operation object corresponding to each preset industry category through a classification output layer; determine the predicted industry category to which the operation object belongs according to the probability value corresponding to each industry category; and correct the labeled industry category according to the predicted industry category of the operation object, so that the predicted industry category obtained by the industry category prediction model can improve the efficiency of industry category correction.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to an industry category correction method and device. Background Art

[0002] At present, the industry category of an enterprise is pre-marked during industrial and commercial registration. When some enterprises change their business projects, their industry category will not be updated in real time, which will cause the pre-marked industry category to become invalid. Therefore, it is necessary to revise the industry category to which the enterprise belongs.

[0003] In the prior art, when the industry category to which an enterprise belongs is corrected, it is usually corrected manually. However, with the increase in the number of enterprises and the increasingly detailed division of industry categories, this manual correction method in the prior art is not efficient. Summary of the invention

[0004] The embodiments of the present application provide an industry category correction method and device to improve the efficiency of industry category correction.

[0005] The specific technical solutions provided by the embodiments of this application are as follows:

[0006] A method for correcting industry categories, including:

[0007] Obtaining attribute information of the operation object and pre-marked industry category;

[0008] Input the attribute information into a trained industry category prediction model, perform feature extraction on the attribute information through a convolution layer to obtain each feature value of the attribute information, determine the feature value that meets the preset feature value condition from the feature values ​​through a pooling layer, and determine the probability value that the operation object corresponds to each preset industry category through a classification output layer, wherein the network structure of the industry category prediction model at least includes an input layer, a convolution layer, a pooling layer, and a classification output layer;

[0009] Determining the predicted industry category to which the operation object belongs according to the probability values ​​corresponding to the various industry categories;

[0010] The marked industry category is corrected according to the predicted industry category of the operation object.

[0011] Optionally, obtain the property information of the operation object, including:

[0012] Obtain various business information of the operation object;

[0013] If it is determined that all the business information is included in the preset business database, the name information and business scope information of the operation object are obtained, and word segmentation processing is performed on the business information, the name information and the business scope information to obtain the attribute information of the operation object;

[0014] If it is determined that at least one of the business information is not included in the business database, word segmentation is performed on the business information to obtain attribute information of the operation object.

[0015] Optionally, performing feature extraction on the attribute information through a convolution layer to obtain each feature value of the attribute information specifically includes:

[0016] Performing vector conversion on each segmented text in the attribute information to generate a word vector corresponding to each segmented text;

[0017] Perform feature extraction on each of the word vectors using a convolution kernel of a first length value of the convolution layer to determine a first feature value corresponding to each of the word vectors;

[0018] Performing feature extraction on each of the word vectors using a convolution kernel of a second length value of the convolution layer to determine a second feature value corresponding to each of the word vectors;

[0019] Performing feature extraction on each of the word vectors using a convolution kernel of a third length value of the convolution layer to determine a third feature value corresponding to each of the word vectors;

[0020] The first eigenvalue, the second eigenvalue and the third eigenvalue are combined to generate eigenvalues ​​corresponding to the word vectors.

[0021] Optionally, determining a feature value satisfying a preset feature value condition from the feature values ​​through a pooling layer specifically includes:

[0022] Selecting a preset number of eigenvalues ​​with the largest values ​​from the eigenvalues ​​through a pooling layer;

[0023] The determined preset several feature values ​​are feature-concatenated to determine the feature value of the attribute information.

[0024] Optionally, after determining the probability value of the operation object corresponding to each preset industry category through the classification output layer, the method further includes:

[0025] Inputting the attribute information into the trained first prediction model, the second prediction model and the third prediction model respectively, and determining the first probability value, the second probability value and the third probability value of the operation object corresponding to each preset industry category;

[0026] Determine the total probability value of the operation object corresponding to each industry category according to each probability value, each first probability value, each second probability value and each third probability value, and the corresponding weights;

[0027] Then, according to the probability values ​​corresponding to the various industry categories, determining the predicted industry category to which the operation object belongs, specifically includes:

[0028] The predicted industry category to which the operation object belongs is determined according to the total probability values ​​corresponding to the various industry categories.

[0029] Optionally, determining the predicted industry category to which the operation object belongs according to the probability values ​​corresponding to the various industry categories specifically includes:

[0030] The industry category corresponding to the maximum value of the probability values ​​corresponding to the preset industry categories is used as the predicted industry category to which the operation object belongs.

[0031] Optionally, according to the determined industry category of the operation object, the pre-marked industry category of the operation object is corrected, specifically including:

[0032] If it is determined that the predicted industry subcategory in the predicted industry category is different from the marked industry subcategory in the marked industry category, the marked industry category is corrected according to the predicted industry category.

[0033] Optionally, the marked industry category is corrected according to the predicted industry category, specifically including:

[0034] Determining whether the predicted industry category in the predicted industry category is a preset specific industry category;

[0035] If it is determined that the industry category predicted by the row is the specific industry category, keywords are extracted from the attribute information, and based on the extracted keywords, an industry subcategory containing the keywords is searched from a preset industry category database, and if it is determined that the found industry subcategory is different from the marked industry subcategory, the marked industry subcategory is corrected to the found industry subcategory;

[0036] If it is determined that the predicted industry category is not the preset specific industry category, the marked industry category will be corrected to the predicted industry category.

[0037] An industry category correction device, comprising:

[0038] An acquisition module is used to obtain the attribute information of the operation object and the pre-marked industry category;

[0039] A prediction module, used to input the attribute information into a trained industry category prediction model, perform feature extraction on the attribute information through a convolution layer to obtain each feature value of the attribute information, determine a feature value that satisfies a preset feature value condition from the feature values ​​through a pooling layer, and determine a probability value that the operation object corresponds to each preset industry category through a classification output layer, wherein the network structure of the industry category prediction model at least includes a convolution layer, a pooling layer, and a classification output layer;

[0040] A first determination module, configured to determine the predicted industry category to which the operation object belongs according to the probability values ​​corresponding to the various industry categories;

[0041] A correction module is used to correct the marked industry category according to the predicted industry category of the operation object.

[0042] Optionally, the acquisition module is specifically used to:

[0043] Obtain various business information of the operation object;

[0044] If it is determined that all the business information is included in the preset business database, the name information and business scope information of the operation object are obtained, and word segmentation processing is performed on the business information, the name information and the business scope information to obtain the attribute information of the operation object;

[0045] If it is determined that at least one of the business information is not included in the business database, word segmentation is performed on the business information to obtain attribute information of the operation object.

[0046] Optionally, when extracting features of the attribute information through a convolution layer to obtain each feature value of the attribute information, the prediction module is specifically used to:

[0047] Performing vector conversion on each segmented text in the attribute information to generate a word vector corresponding to each segmented text;

[0048] Perform feature extraction on each of the word vectors using a convolution kernel of a first length value of the convolution layer to determine a first feature value corresponding to each of the word vectors;

[0049] Performing feature extraction on each of the word vectors using a convolution kernel of a second length value of the convolution layer to determine a second feature value corresponding to each of the word vectors;

[0050] Performing feature extraction on each of the word vectors using a convolution kernel of a third length value of the convolution layer to determine a third feature value corresponding to each of the word vectors;

[0051] The first eigenvalue, the second eigenvalue and the third eigenvalue are combined to generate eigenvalues ​​corresponding to the word vectors.

[0052] Optionally, when determining a feature value satisfying a preset feature value condition from the feature values ​​through the pooling layer, the prediction module is specifically used to:

[0053] Selecting a preset number of eigenvalues ​​with the largest values ​​from the eigenvalues ​​through a pooling layer;

[0054] The determined preset several feature values ​​are feature-concatenated to determine the feature value of the attribute information.

[0055] Optionally, after determining the probability value of the operation object corresponding to each preset industry category through the classification output layer, the method further includes:

[0056] A second determination module is used to input the attribute information into the trained first prediction model, the second prediction model and the third prediction model respectively, and determine the first probability value, the second probability value and the third probability value of the operation object corresponding to each preset industry category;

[0057] A third determination module, configured to determine a total probability value of the operation object corresponding to each industry category according to each probability value, each first probability value, each second probability value and each third probability value, and corresponding weights;

[0058] Then the first determination module is specifically used for:

[0059] The predicted industry category to which the operation object belongs is determined according to the total probability values ​​corresponding to the various industry categories.

[0060] Optionally, the first determining module is specifically used for:

[0061] The industry category corresponding to the maximum value of the probability values ​​corresponding to the preset industry categories is used as the predicted industry category to which the operation object belongs.

[0062] Optionally, the correction module is specifically used to:

[0063] If it is determined that the predicted industry subcategory in the predicted industry category is different from the marked industry subcategory in the marked industry category, the marked industry category is corrected according to the predicted industry category.

[0064] Optionally, when the marked industry category is corrected according to the predicted industry category, the correction module is specifically used to:

[0065] Determining whether the predicted industry category in the predicted industry category is a preset specific industry category;

[0066] If it is determined that the industry category predicted by the row is the specific industry category, keywords are extracted from the attribute information, and based on the extracted keywords, an industry subcategory containing the keywords is searched from a preset industry category database, and if it is determined that the found industry subcategory is different from the marked industry subcategory, the marked industry subcategory is corrected to the found industry subcategory;

[0067] If it is determined that the predicted industry category is not the preset specific industry category, the marked industry category will be corrected to the predicted industry category.

[0068] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned industry category correction method when executing the program.

[0069] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the above-mentioned industry category correction method.

[0070] In an embodiment of the present application, attribute information of an operation object and a pre-labeled industry category are obtained, the attribute information is input into a trained industry category prediction model, feature extraction is performed on the attribute information through a convolutional layer in a convolutional neural network, and each feature value of the attribute information is obtained, a feature value satisfying a preset feature value condition is determined from each feature value through a pooling layer, and a probability value of the operation object corresponding to each preset industry category is determined through a classification output layer, and the predicted industry category to which the operation object belongs is determined according to the probability value corresponding to each industry category, and the labeled industry category is corrected according to the predicted industry category of the operation object. In this way, the industry category to which the operation object belongs is predicted through a pre-trained industry category prediction model, and the labeled industry category is corrected based on the predicted industry category, so that the manual correction method can be converted into an automatic correction method, thereby improving the efficiency of industry category correction, and, for refined industry categories, compared with the manual correction method in the prior art, prediction through the industry category prediction model can improve the accuracy of correction. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 This is a flow chart of an industry category correction method in an embodiment of the present application;

[0072] Figure 2 Another flow chart of an industry category correction method in an embodiment of the present application;

[0073] Figure 3 This is a flow chart of a method for correcting the industry category of an enterprise in an embodiment of the present application;

[0074] Figure 4 This is a schematic diagram of the structure of the industry category correction device in the embodiment of the present application;

[0075] Figure 5 Schematic diagram of the structure of an electronic device in an embodiment of the present application. DETAILED DESCRIPTION

[0076] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0077] At present, in the context of economic globalization, fierce market competition has brought great opportunities and challenges to the survival and development of enterprises. my country's economic management starts from reality, combines the development needs of different industries, provides multi-faceted guidance and assistance to enterprises, and creates better development conditions for enterprises to the greatest extent possible to promote the improvement of the national economy. In order to better manage enterprises in different industries and at different levels, it is necessary to strengthen the classified guidance of industry categories.

[0078] At present, the industry category of an enterprise is pre-marked during industrial and commercial registration. With market changes and intensified competition, some enterprises have changed their business projects but have not updated their industry category in real time, resulting in the invalidation of the industry category. Therefore, it is necessary to revise the industry category to which the enterprise belongs.

[0079] Traditionally, industry categories are manually labeled. When faced with a large number of industry category labeling tasks, manual labeling is tedious and has large errors. Therefore, it is necessary to use an intelligent method to automatically correct industry categories in real time. However, with the increase in the number of enterprises and the increasingly detailed division of industry categories, there are more and more types, the similarity of words between categories is high, and the situation of enterprises may change in real time. Therefore, this correction method in the existing technology is not efficient, and the review and update of industry category labels are extremely difficult.

[0080] In order to solve the above problems, an industry category correction method is provided in an embodiment of the present application, which obtains attribute information of an operation object and a pre-labeled industry category, inputs the attribute information into a trained industry category prediction model, performs feature extraction on the attribute information through a convolution layer, obtains each feature value of the attribute information, determines the feature value that meets the preset feature value condition from each feature value through a pooling layer, and determines the probability value of the operation object corresponding to each preset industry category through a classification output layer, determines the predicted industry category to which the operation object belongs according to the probability value corresponding to each industry category, and corrects the labeled industry category according to the predicted industry category of the operation object. In this way, the industry category of the operation object is predicted by a pre-trained industry category prediction model, and the labeled industry category is corrected based on the predicted industry category, which can convert the manual correction method into an automatic correction method, improve the efficiency of industry category correction, and the accuracy of industry category prediction can be improved through the method in the embodiment of the present application.

[0081] Based on the above embodiments, see Figure 1 As shown, it is a flow chart of an industry category correction method in an embodiment of the present application, which specifically includes:

[0082] Step 100: Obtain attribute information of the operation object and the pre-labeled industry category.

[0083] In the embodiment of the present application, since the historical database contains a plurality of different operation objects, the attribute information of the operation object is obtained from the historical database, and the industry category pre-marked for the operation object is obtained.

[0084] Among them, the operation object can be, for example, an enterprise. The industry category includes at least a major industry category and a minor industry category. The minor industry category is represented by a 4-digit code. The major industry category represents 97 major categories in the industry category, such as the retail industry. The minor industry category represents 1,381 minor categories in the industry category. Information such as the name and business scope of the enterprise also has a certain relevance to the industry category.

[0085] For example, when the operation object is an enterprise, the business information of the enterprise is the main products, the name information is the enterprise name, and the business scope information is the business scope. The main products, enterprise name and business scope are all Chinese texts and are stored in the business database. Duplicate data is deleted and missing and abnormal data are processed.

[0086] The following is a detailed description of the steps for obtaining the attribute information of the operation object in the embodiment of the present application, specifically including:

[0087] S1: Obtain the business information of the operation object.

[0088] In an embodiment of the present application, various business information of the operation object is obtained, and because the business information is the most important factor in determining the industry category of the operation object, after obtaining the various business information of the operation object, each business information is first processed. If it is determined that complex information such as model numbers and English letters appears in the business information, the business information is preliminarily processed through the business classification code to generate a corresponding business classification code.

[0089] Among them, business information can be, for example, the main commodities of the enterprise, which are commodities whose amount accounts for more than 50% of all commodities sold within a preset time, and the main commodities are sorted from large to small according to the amount. In addition, when complex information such as model number and English letters appear in the commodity name of the main commodity, the commodity is first preliminarily processed through the commodity classification code to generate the corresponding commodity classification code name.

[0090] S2: If it is determined that all business information is included in the preset business database, the name information and business scope information of the operation object are obtained, and the business information, name information and business scope information are segmented to obtain the attribute information of the operation object.

[0091] In an embodiment of the present application, it is determined whether all business information of the operation object is included in the preset business database. If it is determined that all business information is included in the preset business database, then when determining the industry category of the operation object, the name information and business scope information of the operation object are required to assist in classification. Therefore, the name information and business scope information of the operation object are obtained, and the business information, name information and business scope information are segmented, and only the first three words of the business scope information are retained after segmentation, and the stop word list of the above-mentioned segmentation operation is maintained, and stop words, punctuation and special symbols are removed to obtain the attribute information of the operation object.

[0092] For example, when the operation object is an enterprise, the business database is an auxiliary commodity library. The commodity classification code name or concise commodity name is judged according to the auxiliary commodity library to determine whether the main commodities of the enterprise are all commodities in this auxiliary commodity library. If the main commodities are all commodities in the auxiliary commodity library, the enterprise's corporate name and business scope information are obtained, and the main commodities, corporate name and business scope information of the enterprise are segmented.

[0093] Among them, the auxiliary product library stores main products such as consulting and services.

[0094] S3: If it is determined that at least one of the business information is not included in the business database, perform word segmentation processing on the business information to obtain attribute information of the operation object.

[0095] In an embodiment of the present application, it is determined whether all the business information of the operation object is included in the preset business database. If it is determined that at least one of the business information is not included in the business database, there is no need to obtain the name information and business scope information of the operation object, and only word segmentation processing is performed on the business information.

[0096] For example, assuming that the operation object is an enterprise and the business database is an auxiliary commodity library, determine whether the main commodities of the enterprise are all commodities in this auxiliary commodity library. If the main commodities exist outside the auxiliary commodity library, the main commodities are segmented to obtain the feature word list corresponding to each enterprise.

[0097] Among them, when performing word segmentation processing, the word segmentation processing can be performed through the Jieba word segmentation tool, which is not limited in the embodiments of the present application.

[0098] Step 110: Input the attribute information into the trained industry category prediction model, perform feature extraction on the attribute information through the convolution layer to obtain the eigenvalues ​​of the attribute information, determine the eigenvalues ​​that meet the preset eigenvalue conditions from the eigenvalues ​​through the pooling layer, and determine the probability value of the operation object corresponding to the preset industry categories through the classification output layer.

[0099] Among them, the network structure of the industry category prediction model at least includes an input layer, a convolution layer, a pooling layer and a classification output layer.

[0100] In an embodiment of the present application, after acquiring the attribute information of the operation object, based on the trained industry category prediction model, the attribute information of the operation object is used as an input parameter, and the attribute information of the operation object is input through the input layer of the industry category prediction model. The attribute information of the operation object is extracted through the convolution layer of the industry category prediction model to determine the characteristic values ​​of the attribute information. The characteristic values ​​that meet the preset characteristic value conditions are selected from the characteristic values ​​through the pooling layer of the industry category prediction model, and the selected characteristic values ​​are input into the classification output layer of the industry category prediction model to determine the probability value of the attribute information corresponding to the preset industry categories.

[0101] Specifically, after the attribute information is input into the trained industry category prediction model, first, since the text data has a local area feature structure, in order to learn this feature and combine these local features to form more complex and abstract features, in the embodiment of the present application, the attribute information is feature extracted through a convolutional layer to obtain each feature value of the attribute information, specifically including:

[0102] S1: Perform vector conversion on each word segmentation text in the attribute information to generate a word vector corresponding to each word segmentation text.

[0103] In the embodiment of the present application, each word segmentation text in the attribute information is randomly vectorized to obtain the corresponding word vectors.

[0104] S2: Perform feature extraction on each word vector through the convolution kernel of the first length value of the convolution layer to determine the first eigenvalue corresponding to each word vector.

[0105] In the embodiment of the present application, feature extraction is performed on each word vector through a convolution kernel with a first length value of the convolution layer to obtain a first feature value corresponding to each word vector.

[0106] The convolution kernel of the first length value may be, for example, 3-gram.

[0107] S3: Perform feature extraction on each word vector through the convolution kernel of the second length value of the convolution layer to determine the second eigenvalue corresponding to each word vector.

[0108] In the embodiment of the present application, feature extraction is performed on each word vector through a convolution kernel of a second length value of the convolution layer to obtain a second feature value corresponding to each word vector.

[0109] The convolution kernel of the first length value may be, for example, 4-gram.

[0110] S4: Perform feature extraction on each word vector through the convolution kernel of the third length value of the convolution layer to determine the third eigenvalue corresponding to each word vector.

[0111] In the embodiment of the present application, feature extraction is performed on each word vector through a convolution kernel of a third length value of the convolution layer to obtain a third feature value corresponding to each word vector.

[0112] The convolution kernel of the first length value may be, for example, 5-gram.

[0113] S5: Merge the first eigenvalue, the second eigenvalue, and the third eigenvalue to generate eigenvalues ​​corresponding to each word vector.

[0114] In the embodiment of the present application, the first eigenvalue, the second eigenvalue and the third eigenvalue are merged to obtain the eigenvalues ​​corresponding to each word vector, that is, each word vector corresponds to three eigenvalues.

[0115] Then, after obtaining each feature value of the attribute information, a pooling layer is used to determine a feature value that meets a preset feature value condition from each feature value. In an embodiment of the present application, a possible implementation method is provided, which specifically includes:

[0116] S1: Select a preset number of eigenvalues ​​with the largest values ​​from each eigenvalue through the pooling layer.

[0117] In an embodiment of the present application, in order to retain more effective features of the attribute information, a K-max pooling pooling method is adopted. K-max pooling can select the K eigenvalues ​​with the largest values ​​from all the eigenvalues ​​of the attribute information, and retain the original order of the selected K eigenvalues. In this way, by retaining more feature information, a better prediction effect can be achieved.

[0118] Furthermore, texts of different lengths can be converted into fixed-length representations after passing through the pooling layer. In the embodiments of the present application, feature values ​​can also be selected through the mean pooling method and the maximum pooling method.

[0119] Among them, the maximum pooling representation selects only the eigenvalue with the largest value from a series of eigenvalues ​​of the convolutional layer when extracting features.

[0120] S2: performing feature concatenation on the determined preset number of feature values ​​to determine the feature value of the attribute information.

[0121] In the embodiment of the present application, after selecting a preset number of characteristic values ​​of the attribute information, the determined preset number of characteristic values ​​are concatenated to determine the characteristic value of the attribute information.

[0122] Finally, after determining the characteristic value of the attribute information, the characteristic value is input into the classification output layer to determine the probability value of the operation object corresponding to each preset industry category.

[0123] The classification output layer may be, for example, softmax.

[0124] Furthermore, after the probability values ​​of the attribute information corresponding to the preset industry categories are determined by the industry category prediction model, in order to better extract the characteristics of the attribute information, a multi-model fusion method can be adopted to make up for the shortcomings of a single model, including:

[0125] S1: Input the attribute information into the trained first prediction model, the second prediction model and the third prediction model respectively, and determine the first probability value, the second probability value and the third probability value of the operation object corresponding to each preset industry category.

[0126] In an embodiment of the present application, the attribute information of the operation object is input into a trained first prediction model to determine the first probability value of the operation object corresponding to each preset industry category, the attribute information of the operation object is input into a trained second prediction model to determine the second probability value of the operation object corresponding to each preset industry category, and the attribute information of the operation object is input into a trained third prediction model to determine the third probability value of the operation object corresponding to each preset industry category.

[0127] S2: Determine the total probability value of the operation object corresponding to each industry category according to each probability value, each first probability value, each second probability value and each third probability value, and the corresponding weights.

[0128] In an embodiment of the present application, the total probability value of the operation object corresponding to each industry category is determined based on each probability value and the corresponding weight, each first probability value and the corresponding weight, each second probability value and the corresponding weight, and each third probability value and the corresponding weight.

[0129] For example, assuming that the preset industry categories are A and B, the probability values ​​are A1 and B1, the weight corresponding to the industry category prediction model is 3 / 6, the first probability values ​​are A2 and B2, the weight corresponding to the first prediction model is 2 / 6, the second probability values ​​are A3 and B3, the weight corresponding to the second prediction model is 1 / 12, the third probability values ​​are A4 and B4, the weight corresponding to the third prediction model is 1 / 12, therefore, the total probability value of determining that the operation object corresponds to industry category A can be expressed as:

[0130]

[0131] The total probability value of the operation object corresponding to industry category B can be expressed as:

[0132]

[0133] It should be noted that in the embodiments of the present application, there is no restriction on the number of preset industry categories, and there is no restriction on the number of prediction models.

[0134] Among them, the prediction model can be, for example, a convolutional neural network (Text Convolutional Neural Networks, TextCNN), a bidirectional recurrent neural network (Bi-directional Long Short-Term Memory, Bi-LSTM), K-max pooling TextCNN, a bidirectional recurrent neural network based on an attention mechanism (Attention-Based Bi-directional Long Short-Term Memory, Attention-Based Bi-LSTM), etc., which is not limited in the embodiments of the present application.

[0135] It should be noted that first, it is necessary to process the attribute information of each operation object in the test set, and input the attribute information after data processing into each prediction model for training, and adjust the parameters of each prediction model according to the prediction effect. Then, the prediction effect of each prediction model in the test set is sorted, and the top N prediction models with the smallest prediction error are determined, and the models are fused using the average weighted method.

[0136] For example, the top three prediction models with the smallest prediction errors can be selected to form a fusion model, and the model fusion is performed using the weighted average method. Assuming that the rankings are 1, 2, and 3, the weights assigned to the three models are 3 / 6, 2 / 6, and 1 / 6, respectively.

[0137] After determining the total probability value of the operation object corresponding to each industry category, the predicted industry category to which the operation object belongs is determined based on the total probability value corresponding to each industry category.

[0138] Step 120: Determine the industry category to which the operation object belongs based on the probability values ​​corresponding to each industry category.

[0139] In the embodiment of the present application, after determining the probability value of the operation object corresponding to each industry category, the industry category to which the operation object belongs can be determined based on the probability value corresponding to each industry category.

[0140] Among them, when determining the industry category to which the operation object belongs according to the probability value corresponding to each industry category, in an embodiment of the present application, a possible implementation method is provided, which specifically includes:

[0141] The industry category corresponding to the maximum value of the probability values ​​corresponding to the preset industry categories is used as the predicted industry category to which the operation object belongs.

[0142] In the embodiment of the present application, the maximum value is selected from the probability values ​​corresponding to each preset industry category as the predicted industry category to which the operation object belongs.

[0143] Step 130: Correct the marked industry category according to the predicted industry category of the operation object.

[0144] In the embodiment of the present application, after determining the predicted industry category of the operation object, the marked industry category is corrected according to the predicted industry category. The steps of correcting the marked industry category in the embodiment of the present application are described in detail below, specifically including:

[0145] If it is determined that the predicted industry subcategory in the predicted industry category is different from the industry subcategory marked in the marked industry category, the marked industry category shall be corrected according to the predicted industry category.

[0146] In the embodiment of the present application, it is determined whether the predicted industry subcategory in the predicted industry category is the same as the marked industry subcategory in the marked industry category, which can be specifically divided into the following two situations:

[0147] The first case: the predicted industry subcategory is the same as the marked industry subcategory.

[0148] Specifically, if it is determined that the predicted industry subcategory in the predicted industry category is the same as the marked industry subcategory in the marked industry category, the marked industry category will not be corrected and the original result will be maintained.

[0149] In an embodiment of the present application, if it is determined that the predicted industry subcategory in the predicted industry category is the same as the marked industry subcategory in the marked industry category, then it is determined that the marked industry subcategory is correct, and therefore, there is no need to correct the marked industry subcategory.

[0150] The second situation: the predicted industry subcategory is different from the marked industry subcategory.

[0151] Specifically, if it is determined that the predicted industry subcategory in the predicted industry category is different from the marked industry subcategory in the marked industry category, the marked industry category will be corrected according to the predicted industry category.

[0152] In the embodiment of the present application, if it is determined that the predicted industry subclass is different from the marked industry subclass, then it is determined that the pre-marked industry subclass may be wrong, and therefore, the marked industry subclass is corrected according to the predicted industry subclass.

[0153] Furthermore, since some industry categories in the industry categories are the industries with the largest proportion and the most subcategories, and the names of different subcategories are similar to each other, it is difficult to distinguish them. Therefore, for these special specific industry categories, when the predicted industry subcategories are different from the marked industry subcategories, the keywords in the attribute information can be used as the key description words of the attribute information, and keyword matching can be performed separately for these specific industry categories, and the industry category prediction model based on K-maxpooling Text CNN can be trained to optimize the prediction accuracy and assist in determining whether the industry category of the operation object needs to be corrected, including:

[0154] A1: Determine whether the predicted industry category in the predicted industry category is a preset specific industry category.

[0155] In an embodiment of the present application, when the predicted industry subcategory is different from the marked industry subcategory, it is determined whether the predicted industry major category in the predicted industry category is a preset specific industry category.

[0156] Among them, the preset specific industry categories may be, for example, retail industry, wholesale industry, etc., and the number and types of specific industry categories are not limited in the embodiments of the present application.

[0157] A2: If it is determined that the predicted industry category is a specific industry category, keywords are extracted from the attribute information, and based on the extracted keywords, the industry subcategory containing the keywords is found from the preset industry category database. If it is determined that the found industry subcategory is different from the marked industry subcategory, the marked industry subcategory is corrected to the found industry subcategory;

[0158] In the embodiment of the present application, it is determined whether the predicted industry category is a preset specific industry category. If it is determined that the predicted industry category is a preset specific industry category, keyword extraction is performed on the attribute information of the operation object to obtain the keywords of the attribute information, and the first 1 to 2 participles of the feature words contained in the attribute information and the keywords are matched in order from front to back, and the preset industry category database is searched to see whether there is an industry subcategory containing the keyword, the first word appearing in the keyword library is found, and the corresponding industry subcategory is located as the prediction result, which can be specifically divided into the following two cases:

[0159] The first case: Find the industry subcategory containing the keyword.

[0160] Specifically, if it is determined that an industry subcategory containing the keyword is found, then a determination is made as to whether the found industry subcategory is the same as the marked industry subcategory; if they are determined to be the same, then the marked industry subcategory is determined to be correct and there is no need to correct the marked industry subcategory; if it is determined that the found industry subcategory is different from the marked industry subcategory, then the marked industry subcategory is corrected to the found industry subcategory.

[0161] Among them, in order to extract the keywords of each industry subcategory in a specific industry category, TF-IDF can be used to set a threshold to select some words, and then further screen them according to the description in the "National Economic Industry Classification" to obtain descriptive words that only appear in this subcategory, summarize them into keywords, and maintain the keyword table after the data is updated.

[0162] The second situation: no industry subcategory containing the keyword is found.

[0163] Specifically, if it is determined that the industry subcategory containing the keyword is not found, the marked industry subcategory will be corrected based on the predicted industry subcategory.

[0164] A3: If it is determined that the predicted industry category is not the preset specific industry category, the marked industry category will be revised to the predicted industry category.

[0165] In an embodiment of the present application, if it is determined that the predicted industry category is not the preset specific industry category, the marked industry category is corrected to the predicted industry category.

[0166] Furthermore, when the marked industry subcategory is different from the predicted industry subcategory, or the found industry subcategory is different from the predicted industry subcategory, it is necessary to correct it according to the predicted industry subcategory or the found industry subcategory.

[0167] However, since some industry categories are very similar and there are other categories, similar records are made for some industry subcategories within the acceptable range for users, including:

[0168] First, similar industry subcategories are counted, and similar sets containing similar industry subcategories are generated, and then each similar set is stored.

[0169] Then, determine whether the marked industry subcategories and the predicted industry subcategories are in the same similarity set. If the marked industry subcategories and the predicted industry subcategories are in the same similarity set, then it is determined that the marked industry subcategories and the predicted industry subcategories are similar industries, and there is no need to correct the marked industry subcategories. If it is determined that the marked industry subcategories and the predicted industry subcategories are not in the same similarity set, then it is determined that the two are not similar industries, and the marked industry subcategories are corrected according to the predicted industry subcategories.

[0170] In the embodiment of the present application, the attribute information of the operation object and the pre-labeled industry category are obtained, the attribute information is input into the trained industry category prediction model, the feature extraction of the attribute information is performed through the convolution layer to obtain each feature value of the attribute information, the feature value that meets the preset feature value condition is determined from each feature value through the pooling layer, and the probability value of the operation object corresponding to each preset industry category is determined through the classification output layer, and the predicted industry category to which the operation object belongs is determined according to the probability value corresponding to each industry category, and the labeled industry category is corrected according to the predicted industry category of the operation object. In this way, in order to solve the problem of automatic review and update of industry subcategories, it is necessary to fully The characteristics of the main commodities are mined separately, and the company name and business scope are combined as auxiliary information. The natural language processing technology is used to propose an industry category correction method. The industry correction method in the embodiment of the present application can solve the correction problem of industry subcategories, and improve the correction accuracy by combining rules, including combining keywords and deep learning to process wholesale and retail industries, etc., mining the characteristics of the main commodities, using deep learning algorithms to predict the industry to which the company belongs, and then comparing it with the original marked subcategories according to the rules to determine whether correction is needed and the category of correction, thereby achieving good application results. The rules and multi-model fusion method are combined to effectively train the model, and the industry category is corrected in real time according to the corrected industry screening process to improve the prediction accuracy.

[0171] Based on the above embodiments, see Figure 2 As shown, it is another flow chart of an industry category correction method in an embodiment of the present application, which specifically includes:

[0172] Step 200: Obtain the attribute information of the enterprise and the pre-labeled industry category.

[0173] The attribute information at least includes a characteristic vocabulary of the enterprise.

[0174] Step 201: Input the attribute information into the trained industry category model to determine the probability value of the enterprise corresponding to each industry category.

[0175] Step 202: Determine the predicted industry category to which the enterprise belongs based on the probability values ​​corresponding to each industry category.

[0176] Among them, the predicted industry categories at least include predicted major industry categories and predicted minor industry categories.

[0177] Step 203: Determine whether the predicted industry subcategory is the same as the marked industry subcategory. If so, execute step 204; if not, execute step 211.

[0178] Step 204: Determine whether the predicted industry category is a preset specific industry category. If so, execute step 205; if not, execute step 208.

[0179] Step 205: extract keywords from the attribute information, and based on the extracted keywords, search for industry subcategories containing the keywords from a preset industry category database.

[0180] Step 206: Determine whether an industry subcategory containing the keyword can be found, if so, execute step 207, if not, execute step 211.

[0181] Step 207: Determine whether the found industry subcategory is the same as the marked industry subcategory. If so, execute step 210; if not, execute step 209.

[0182] Step 208: Determine whether the searched industry subcategory or the predicted industry subcategory is in the same similarity set. If so, execute step 211; if not, execute step 209.

[0183] Step 209: Correct the marked industry subcategories using the found industry subcategories or predicted industry subcategories.

[0184] Step 210: There is no need to correct the marked industry subcategories through the found industry subcategories.

[0185] Step 211: There is no need to revise the marked industry sub-categories through the predicted industry sub-categories.

[0186] In the embodiments of the present application, the pooling layer of TextCNN is improved by utilizing auxiliary commodity libraries, keywords and other rules to retain more effective features, thereby making the feature mining of the main commodities more accurate. In addition, in view of the large proportion of wholesale and retail industries and the fact that they are more difficult to distinguish, the prediction accuracy of the wholesale and retail industries can be improved by combining keyword matching with the K-max pooling TextCNN model. The model can be effectively trained by combining rules and multi-model fusion methods, and the industry category can be corrected in real time according to the corrected industry screening process to improve the prediction accuracy.

[0187] Based on the above embodiments, see Figure 3 As shown in FIG. 1 , it is a flow chart of a method for correcting the industry category of an enterprise in an embodiment of the present application, which specifically includes:

[0188] 1. Data.

[0189] In the embodiment of the present application, the business database includes at least the enterprise's main products, enterprise name, business scope and marked industry subcategories.

[0190] 2. Data preprocessing.

[0191] In the embodiments of the present application, the data is sampled, filtered and standardized.

[0192] 3. Feature engineering.

[0193] In the embodiment of the present application, the main commodities are first processed. If it is determined that the main commodities contain complex information such as model numbers and English letters, the commodities are preliminarily processed through commodity classification codes to generate corresponding commodity classification code names.

[0194] Then, an auxiliary commodity library is established, which stores main commodities of consulting, service and other types. If it is determined that the main commodities of the enterprise are all commodities in the auxiliary commodity library, and it is determined that the enterprise needs the enterprise name and business scope to assist in industry classification, the enterprise name and business scope are obtained from the business database, and the main commodities, enterprise name and business scope are segmented. If it is determined that at least one of the main commodities of the enterprise is not a commodity in the auxiliary commodity library, only the main commodities are segmented.

[0195] Finally, each segmentation obtained through the segmentation process is transformed into a dimensionality reduction method, and the customized segmentation is used to reduce the number of segmentations and remove ultra-low frequency words to obtain each feature word.

[0196] 4. Multi-model fusion.

[0197] In an embodiment of the present application, when predicting the industry category to which an enterprise belongs, the traditional TextCNN model algorithm is improved to retain more effective features, K-max pooling TextCNN is used to predict the industry category, and multiple models are fused.

[0198] Among them, the multiple models may be, for example, Bi-LSTM and Attention-Based Bi-LSTM.

[0199] 5. Keyword matching.

[0200] In the embodiment of the present application, the keywords of the enterprise are extracted through TF-IDF. Due to the particularity of the wholesale and retail industries, keyword matching is performed separately for the wholesale and retail industries. In addition, since the main products with the largest sales amount have the greatest impact on the industry classification of the enterprise, these main product keywords can be used to more accurately find some industry categories. According to the keyword table of the wholesale and retail industry, the first 1 to 2 words in the enterprise feature word list and the keywords are first matched in order from front to back, and the first word that appears in the keyword library is found, and the corresponding industry subcategory is located as the prediction result.

[0201] 6. Wholesale and retail industry forecasting model.

[0202] In the embodiment of the present application, if it is determined that no successful match has been made, the unsuccessful matched enterprises will enter the wholesale and retail industry model based on K-maxpooling Text CNN for industry prediction.

[0203] 7. Industry correction and screening.

[0204] In an embodiment of the present application, if it is determined that the prediction or matching result is consistent with the original annotation, and the prediction or matching result and the meta-annotation belong to the same similarity set in the industry similarity table, there is no need to correct the industry subcategory; if it is determined that the prediction or matching result is inconsistent with the original annotation, and the prediction or matching result and the original annotation do not belong to the same similarity set in the industry similarity table, it is necessary to correct the industry subcategory.

[0205] In the embodiment of the present application, the real-time industry correction method based on multi-model fusion can judge in real time whether the enterprise industry sub-category is consistent with the original annotation. If it is inconsistent, the industry sub-category recommendation result with the highest probability is provided. Compared with the traditional method, the correction scheme of the industry sub-category can be automatically provided in real time. In the correction process, mainly based on the latest main commodity data, combined with industry category keywords and wholesale and retail industry characteristics, multi-model fusion methods such as K-max pooling Text CNN and Attention-Based Bi-LSTM are used to more effectively utilize the advantages of each model and improve the correction accuracy. In this way, the traditional Text CNN model algorithm is improved, more effective features are retained, and K-max pooling Text CNN is used to predict industry categories. In addition, the multi-model fusion method is used to make up for the shortcomings of a single model, better mine the data features of the main commodities, and make the feature mining of the main commodities more accurate. In addition, the main commodity data is extracted, and the rules such as some industry category keywords, the particularity of the wholesale and retail industry, and the similarity of industry sub-categories are used. Combined with the multi-model fusion method, the industry sub-categories with thousands of categories are directly predicted, which can improve the efficiency of the industry category correction of the enterprise.

[0206] Based on the same inventive concept, an industry category correction device is also provided in the embodiment of the present application. The industry category correction device can be a hardware structure, a software module, or a hardware structure plus a software module. Figure 4 The figure is a schematic diagram of the structure of the industry category correction device in the embodiment of the present application, which specifically includes:

[0207] An acquisition module 400 is used to acquire attribute information of an operation object and a pre-marked industry category;

[0208] The prediction module 410 is used to input the attribute information into the trained industry category prediction model, perform feature extraction on the attribute information through a convolution layer to obtain each feature value of the attribute information, determine the feature value that meets the preset feature value condition from the feature values ​​through a pooling layer, and determine the probability value that the operation object corresponds to each preset industry category through a classification output layer, wherein the network structure of the industry category prediction model at least includes a convolution layer, a pooling layer and a classification output layer;

[0209] A first determination module 420, configured to determine the predicted industry category to which the operation object belongs according to the probability values ​​corresponding to the various industry categories;

[0210] The correction module 430 is used to correct the marked industry category according to the predicted industry category of the operation object.

[0211] Optionally, the acquisition module 400 is specifically used for:

[0212] Obtain various business information of the operation object;

[0213] If it is determined that all the business information is included in the preset business database, the name information and business scope information of the operation object are obtained, and word segmentation processing is performed on the business information, the name information and the business scope information to obtain the attribute information of the operation object;

[0214] If it is determined that at least one of the business information is not included in the business database, word segmentation is performed on the business information to obtain attribute information of the operation object.

[0215] Optionally, when extracting features of the attribute information through a convolution layer to obtain each feature value of the attribute information, the prediction module 410 is specifically used to:

[0216] Performing vector conversion on each segmented text in the attribute information to generate a word vector corresponding to each segmented text;

[0217] Perform feature extraction on each of the word vectors using a convolution kernel of a first length value of the convolution layer to determine a first feature value corresponding to each of the word vectors;

[0218] Performing feature extraction on each of the word vectors using a convolution kernel of a second length value of the convolution layer to determine a second feature value corresponding to each of the word vectors;

[0219] Performing feature extraction on each of the word vectors using a convolution kernel of a third length value of the convolution layer to determine a third feature value corresponding to each of the word vectors;

[0220] The first eigenvalue, the second eigenvalue and the third eigenvalue are combined to generate eigenvalues ​​corresponding to the word vectors.

[0221] Optionally, when determining a feature value that satisfies a preset feature value condition from the feature values ​​through the pooling layer, the prediction module 410 is specifically used to:

[0222] Selecting a preset number of eigenvalues ​​with the largest values ​​from the eigenvalues ​​through a pooling layer;

[0223] The determined preset several feature values ​​are feature-concatenated to determine the feature value of the attribute information.

[0224] Optionally, after determining the probability value of the operation object corresponding to each preset industry category through the classification output layer, the method further includes:

[0225] The second determination module 440 is used to input the attribute information into the trained first prediction model, the second prediction model and the third prediction model respectively, and determine the first probability value, the second probability value and the third probability value of the operation object corresponding to each preset industry category;

[0226] A third determination module 450 is used to determine the total probability value of the operation object corresponding to each industry category according to each probability value, each first probability value, each second probability value and each third probability value, and corresponding weights;

[0227] Then the first determination module 420 is specifically used for:

[0228] The predicted industry category to which the operation object belongs is determined according to the total probability values ​​corresponding to the various industry categories.

[0229] Optionally, the first determining module 420 is specifically configured to:

[0230] The industry category corresponding to the maximum value of the probability values ​​corresponding to the preset industry categories is used as the predicted industry category to which the operation object belongs.

[0231] Optionally, the correction module 430 is specifically used for:

[0232] If it is determined that the predicted industry subcategory in the predicted industry category is different from the marked industry subcategory in the marked industry category, the marked industry category is corrected according to the predicted industry category.

[0233] Optionally, when the marked industry category is corrected according to the predicted industry category, the correction module 430 is specifically used to:

[0234] Determining whether the predicted industry category in the predicted industry category is a preset specific industry category;

[0235] If it is determined that the industry category predicted by the row is the specific industry category, keywords are extracted from the attribute information, and based on the extracted keywords, an industry subcategory containing the keywords is searched from a preset industry category database, and if it is determined that the found industry subcategory is different from the marked industry subcategory, the marked industry subcategory is corrected to the found industry subcategory;

[0236] If it is determined that the predicted industry category is not the preset specific industry category, the marked industry category will be corrected to the predicted industry category.

[0237] Based on the above embodiments, see Figure 5 Shown is a schematic diagram of the structure of an electronic device in an embodiment of the present application.

[0238] An embodiment of the present application provides an electronic device, which may include a processor 510 (Center Processing Unit, CPU), a memory 520, an input device 530 and an output device 540, etc. The input device 530 may include a keyboard, a mouse, a touch screen, etc., and the output device 540 may include a display device, such as a liquid crystal display (Liquid Crystal Display, LCD), a cathode ray tube (Cathode Ray Tube, CRT), etc.

[0239] The memory 520 may include a read-only memory (ROM) and a random access memory (RAM), and provides the processor 510 with program instructions and data stored in the memory 520. In the embodiment of the present application, the memory 520 may be used to store the program of any industry category correction method in the embodiment of the present application.

[0240] The processor 510 calls the program instructions stored in the memory 520, and the processor 510 is used to execute any industry category correction method in the embodiments of the present application according to the obtained program instructions.

[0241] Based on the above embodiments, in an embodiment of the present application, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the industry category correction method in any of the above method embodiments is implemented.

[0242] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0243] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0244] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0245] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0246] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A method for correcting industry categories, characterized in that: include: Obtaining attribute information of the operation object and pre-marked industry category; Input the attribute information into a trained industry category prediction model, perform feature extraction on the attribute information through a convolution layer to obtain each feature value of the attribute information, determine the feature value that meets the preset feature value condition from the feature values ​​through a pooling layer, and determine the probability value that the operation object corresponds to each preset industry category through a classification output layer, wherein the network structure of the industry category prediction model at least includes a convolution layer, a pooling layer and a classification output layer; Determining the predicted industry category to which the operation object belongs according to the probability values ​​corresponding to the various industry categories; If it is determined that the predicted industry subcategory in the predicted industry category is different from the marked industry subcategory in the marked industry category, the marked industry category is corrected according to the predicted industry category; According to the predicted industry category, the marked industry category is corrected, including: Determining whether the predicted industry category in the predicted industry category is a preset specific industry category; If it is determined that the predicted industry category is the specific industry category, keywords are extracted from the attribute information, and based on the extracted keywords, an industry subcategory containing the keywords is searched from a preset industry category database, and if it is determined that the found industry subcategory is different from the marked industry subcategory, the marked industry subcategory is corrected to the found industry subcategory; If it is determined that the predicted industry category is not the preset specific industry category, the marked industry category will be corrected to the predicted industry category.

2. The method according to claim 1, characterized in that Get the property information of the operation object, including: Obtain various business information of the operation object; If it is determined that all the business information is included in the preset business database, the name information and business scope information of the operation object are obtained, and word segmentation processing is performed on the business information, the name information and the business scope information to obtain the attribute information of the operation object; If it is determined that at least one of the business information is not included in the business database, word segmentation is performed on the business information to obtain attribute information of the operation object.

3. The method according to claim 2, characterized in that Performing feature extraction on the attribute information through a convolution layer to obtain each feature value of the attribute information specifically includes: Performing vector conversion on each segmented text in the attribute information to generate a word vector corresponding to each segmented text; Perform feature extraction on each of the word vectors using a convolution kernel of a first length value of the convolution layer to determine a first feature value corresponding to each of the word vectors; Performing feature extraction on each of the word vectors using a convolution kernel of a second length value of the convolution layer to determine a second feature value corresponding to each of the word vectors; Performing feature extraction on each of the word vectors using a convolution kernel of a third length value of the convolution layer to determine a third feature value corresponding to each of the word vectors; The first eigenvalue, the second eigenvalue and the third eigenvalue are combined to generate eigenvalues ​​corresponding to the word vectors.

4. The method according to claim 1, characterized in that Determining a feature value that satisfies a preset feature value condition from the feature values ​​through a pooling layer specifically includes: Selecting a preset number of eigenvalues ​​with the largest values ​​from the eigenvalues ​​through a pooling layer; The determined preset several feature values ​​are feature-concatenated to determine the feature value of the attribute information.

5. The method according to claim 1, characterized in that After determining the probability value of the operation object corresponding to each preset industry category through the classification output layer, the method further includes: Inputting the attribute information into the trained first prediction model, the second prediction model and the third prediction model respectively, and determining the first probability value, the second probability value and the third probability value of the operation object corresponding to each preset industry category; Determine the total probability value of the operation object corresponding to each industry category according to each probability value, each first probability value, each second probability value and each third probability value, and the corresponding weights; Then, according to the probability values ​​corresponding to the various industry categories, determining the predicted industry category to which the operation object belongs, specifically includes: The predicted industry category to which the operation object belongs is determined according to the total probability values ​​corresponding to the various industry categories.

6. The method according to claim 1, characterized in that Determining the predicted industry category to which the operation object belongs according to the probability values ​​corresponding to the various industry categories specifically includes: The industry category corresponding to the maximum value of the probability values ​​corresponding to the preset industry categories is used as the predicted industry category to which the operation object belongs.

7. An industry category correction device, characterized in that: include: An acquisition module is used to obtain the attribute information of the operation object and the pre-marked industry category; A prediction module, used to input the attribute information into a trained industry category prediction model, perform feature extraction on the attribute information through a convolution layer to obtain each feature value of the attribute information, determine a feature value that satisfies a preset feature value condition from the feature values ​​through a pooling layer, and determine a probability value that the operation object corresponds to each preset industry category through a classification output layer, wherein the network structure of the industry category prediction model at least includes a convolution layer, a pooling layer, and a classification output layer; A determination module, used to determine the predicted industry category to which the operation object belongs according to the probability values ​​corresponding to the various industry categories; A correction module, for correcting the marked industry category according to the predicted industry category if it is determined that the predicted industry subcategory in the predicted industry category is different from the marked industry subcategory in the marked industry category; Wherein, when the marked industry category is corrected according to the predicted industry category, the correction module is specifically used for: Determining whether the predicted industry category in the predicted industry category is a preset specific industry category; If it is determined that the predicted industry category is the specific industry category, keywords are extracted from the attribute information, and based on the extracted keywords, an industry subcategory containing the keywords is searched from a preset industry category database, and if it is determined that the found industry subcategory is different from the marked industry subcategory, the marked industry subcategory is corrected to the found industry subcategory; If it is determined that the predicted industry category is not the preset specific industry category, the marked industry category will be corrected to the predicted industry category.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images