Industry category identification method, device, medium and computer program product
By predicting the business scope and name of enterprises layer by layer through a hierarchical classification model, the industry category is automatically identified, which solves the problem of low efficiency in industry classification in existing technologies and achieves faster and more accurate industry category identification.
Patent Information
- Application Number
- CN202110868628.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-07-30
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2041-07-30
AI Technical Summary
In existing technologies, the classification efficiency of national economic industry categories for enterprises is low, especially when the enterprise name and business scope are missing or inaccurate, and manual identification methods are labor-intensive and time-consuming.
A hierarchical classification model is adopted, in which the business scope and name of the enterprise are input into the classification model at each level. Through iterative training and optimization, prediction is made layer by layer to finally determine the target sub-category. Combined with the preset category system mapping relationship, the industry category is automatically identified.
By using hierarchical prediction methods, the search space for industry categories is reduced, the speed and accuracy of classification are improved, the tedious process of manual querying is avoided, and the efficiency of industry classification is enhanced.
Smart Images

Figure CN113591979B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of machine learning technology in financial technology (Fintech), and in particular to an industry category identification method, device, medium and computer program product. Background Art
[0002] With the continuous development of financial technology, especially Internet technology finance, more and more technologies (such as distributed, artificial intelligence, etc.) are applied in the financial field, but the financial industry also puts forward higher requirements on technology, such as higher requirements on the distribution of corresponding to-do items in the financial industry.
[0003] With the development of computer technology, the application of machine learning is becoming more and more extensive. At present, in the national economic industry classification standard, the national economic industry is divided into four categories: category, major category, medium category and minor category. When an enterprise registers at the industrial and commercial level, it is necessary to fill in the corresponding industry category. At present, when the national economic industry category of an enterprise is missing or incorrect, it is often based on the text description of the enterprise name and business scope, and the corresponding category path is manually searched in the national industry classification system. However, when there are many enterprises missing industry categories, the traditional manual judgment method is labor-intensive and takes a long time, which leads to low industry classification efficiency. Summary of the invention
[0004] The main purpose of this application is to provide an industry category identification method, device, medium and computer program product, aiming to solve the technical problem of low industry classification efficiency in the prior art.
[0005] To achieve the above objectives, the present application provides an industry category identification method, the industry category identification method comprising:
[0006] Obtain the business scope and company name of the enterprise to be classified;
[0007] Inputting the business scope and the enterprise name into the first-level category classification model of the corresponding hierarchical classification model respectively to obtain a classification prediction result, wherein the hierarchical classification model includes category classification models of each level;
[0008] The classification prediction results are combined with the business scope and the company name respectively to serve as inputs of the next-layer category classification models corresponding to the business scope and the company name, respectively, to obtain the next classification prediction results, and the cycle is repeated until the last layer of category classification model is reached to obtain the target sub-category recognition result;
[0009] Based on the target subcategory identification result and the mapping relationship between the preset category system, the industry category corresponding to the enterprise to be classified is determined.
[0010] The present application also provides an industry category identification device, which is a virtual device and includes:
[0011] An acquisition module is used to obtain the business scope and enterprise name of the enterprise to be classified;
[0012] A first classification module, used to input the business scope and the enterprise name into the first-level category classification model of the corresponding hierarchical classification model to obtain a classification prediction result, wherein the hierarchical classification model includes the category classification models of each level;
[0013] The second classification module is used to combine the classification prediction result with the business scope and the company name respectively to serve as the input of the next-level category classification model corresponding to the business scope and the company name, obtain the next classification prediction result, and loop until the last-level category classification model is reached to obtain the target sub-category recognition result;
[0014] A determination module is used to determine the industry category corresponding to the enterprise to be classified based on the target subcategory identification result and the mapping relationship between the preset category system.
[0015] The present application also provides an industry category identification device, which is a physical device. The industry category identification device includes: a memory, a processor, and an industry category identification program stored in the memory. The industry category identification program is executed by the processor to implement the steps of the above-mentioned industry category identification method.
[0016] The present application also provides a medium, which is a readable storage medium, on which an industry category identification program is stored, and the industry category identification program is executed by a processor to implement the steps of the above-mentioned industry category identification method.
[0017] The present application also provides a computer program product, including a computer program, which implements the steps of the above-mentioned industry category identification method when executed by a processor.
[0018] The present application provides an industry category identification method, device, medium and computer program product. Compared with the technical means used in the prior art to manually search for corresponding industry categories in the national industry classification system based on text descriptions of company names and business scopes, the present application first obtains the business scope and company name of the company to be classified, and then inputs the business scope and the company name into the first-level category classification model of the respective corresponding hierarchical classification models to obtain classification prediction results, wherein the hierarchical classification model includes category classification models of each level, and further, the classification prediction results are respectively combined with the business scope and the company name as inputs of the next-level category classification model corresponding to the business scope and the company name, to obtain the next classification prediction result, and the cycle is repeated until the last-level category classification model is reached to obtain the target sub-category identification result. It should be noted that The industry categories include four-level category systems, namely, categories, major categories, medium categories and minor categories. Each layer of the category classification model in the hierarchical classification model judges the output of the next layer according to the classification prediction results corresponding to the category classification model of the previous layer. The hierarchical prediction method is adopted, and the prediction target range is relatively concentrated, which reduces the search space of the industry categories and speeds up the speed and accuracy of the classification prediction. Furthermore, based on the mapping relationship between the target minor category recognition results and the preset category system, the industry category corresponding to the enterprise to be classified is determined, and the industry category corresponding to the enterprise to be classified is automatically obtained based on the business scope and the enterprise name. That is, there is no need for manual query and classification, which overcomes the technical defects of the prior art that when there are many enterprises missing industry categories, the traditional manual judgment method is labor-intensive and time-consuming, which leads to low industry classification efficiency, thereby improving the efficiency of industry classification. BRIEF DESCRIPTION OF THE DRAWINGS
[0019] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.
[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without any creative work.
[0021] Figure 1 This is a flow chart of the first embodiment of the industry category identification method of the present application;
[0022] Figure 2 This is a flow chart of the second embodiment of the industry category identification method of the present application;
[0023] Figure 3This is a flowchart of the third embodiment of the industry category identification method of the present application;
[0024] Figure 4 A schematic diagram of the structure of an industry category identification device for the hardware operating environment involved in the industry category identification method in an embodiment of the present application.
[0025] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION
[0026] It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.
[0027] The present application embodiment provides an industry category identification method. In the first embodiment of the industry category identification method of the present application, refer to Figure 1 , the industry category identification method includes:
[0028] Step S10, obtaining the business scope and enterprise name of the enterprise to be classified;
[0029] In this embodiment, it should be noted that the national economic industry classification standard stipulates the classification and coding of economic activities in the whole society. The standard adopts line classification and hierarchical coding method to divide the national economic industry into four categories: categories, major categories, medium categories and minor categories. When an enterprise registers for industrial and commercial grades, it is necessary to fill in the corresponding category. The business scope is the scope of production, operation and service items that the enterprise to be classified can engage in.
[0030] The business scope and enterprise name of the enterprise to be classified are obtained. Specifically, text information corresponding to the business scope of the enterprise to be classified and text information corresponding to the enterprise name of the enterprise to be classified are obtained.
[0031] Step S20, respectively inputting the business scope and the enterprise name into the first-level category classification model of the corresponding hierarchical classification model to obtain a classification prediction result, wherein the hierarchical classification model includes the category classification models of each level;
[0032] In this embodiment, it should be noted that the hierarchical classification model includes a business scope hierarchical classification model and a name hierarchical classification model, wherein the business scope hierarchical classification model and the name hierarchical classification model both include multi-layer category classification models, and the category classification model corresponding to each layer is used to predict the text multi-classification model of the industry category information of the corresponding layer.
[0033] The business scope and the company name are respectively input into the first-layer category classification model of the corresponding hierarchical classification model to obtain a classification prediction result, wherein the hierarchical classification model includes category classification models of each layer. Specifically, the first-layer category classification model in the business scope hierarchical model to be trained is iteratively trained and optimized through the pre-collected business scope to be trained to obtain the first-layer business category classification model, and the training business scope prediction result is output, and the next-layer category classification model in the business scope hierarchical model to be trained is iteratively trained and optimized through the training business scope prediction result output by the upper-layer model and the business scope to be trained, until the last layer of category classification model is trained, the business scope hierarchical model can be obtained. The training process of the name hierarchical classification model can refer to the specific content of the training process of the above-mentioned business scope hierarchical model, which will not be repeated here. Further, the business The scope is input into the first-level category classification model in the business scope hierarchical classification model, and the business scope category probability result is output; the enterprise name is input into the first-level category classification model in the name hierarchical classification model, and the name category probability result is output; then the business scope category probability result and the name category probability result are sorted respectively to obtain the business scope category probability sorting result and the name category probability sorting result; then the sorted business scope category probability sorting result and the name category probability sorting result are added to obtain the category merging result; then category selection is performed in the category merging result or in the business scope category probability result according to the preset selection rule to obtain the classification prediction result, wherein the preset selection rule is a rule of selecting a category with a small ranking order value in the category merging result or selecting a category corresponding to the maximum probability in the business scope category probability result.
[0034] The hierarchical classification model includes a business scope hierarchical classification model and a name hierarchical classification model.
[0035] The step of inputting the business scope and the enterprise name into the first-level category classification model of the respective corresponding hierarchical classification models to obtain the classification prediction result comprises:
[0036] Step S21, inputting the business scope into the first-level category classification model in the business scope hierarchical classification model, and outputting the business scope category probability result;
[0037] In this embodiment, it should be noted that the business scope category probability result is the probability result of the business scope corresponding to the category category.
[0038] The business scope is input into the first-level category classification model in the business scope hierarchical classification model, and the business scope category probability result is output. Specifically, based on the word frequency-inverse document frequency model or the word2vec model, the text information corresponding to the business scope is constructed into a business scope text vector, and then the business scope text vector is classified through the first-level category classification model in the business scope hierarchical classification model to obtain the business scope category probability result corresponding to the business scope text vector.
[0039] Step S22, inputting the enterprise name into the first-level category classification model in the name hierarchical classification model, and outputting the name category probability result;
[0040] In this embodiment, the enterprise name is input into the first-level category classification model in the name hierarchical classification model, and a name category probability result is output. Specifically, based on the word frequency-inverse document frequency model or the word2vec model, the text information corresponding to the enterprise name is constructed into an enterprise name text vector, and then the enterprise name text vector is classified through the first-level category classification model in the name hierarchical classification model to obtain the name category probability result corresponding to the enterprise name text vector. The specific implementation content of step S22 can refer to the specific content in step S21, which will not be repeated here.
[0041] Step S23, merging the business scope category probability result and the name category probability result to obtain a category merging result;
[0042] In this embodiment, the business scope category probability result and the name category probability result are merged to obtain a category merge result. Specifically, the classification probabilities in the business scope category probability result are sorted to obtain a first sorting result, and the classification probabilities in the name category probability result are sorted to obtain a second sorting result. Then the first sorting result and the second sorting result are added to obtain the category merge result. For example, the probabilities corresponding to Class A, Class B, and Class C in the business scope category probability result are (0.15, 0.8, 0.05), respectively, and the sorting results are (2, 1, 3). The probabilities corresponding to Class A, Class B, and Class C in the name category probability result are (0.1, 0.4, 0.5), respectively, and the sorting results are (3, 2, 1). The sum of the sorting results is (5, 3, 4), and the category merge result can be obtained.
[0043] The step of merging the business scope category probability result and the name category probability result to obtain the category merging result includes:
[0044] Step S231, sorting the business scope category probability results to obtain a first sorting result;
[0045] In this embodiment, the business scope category probability results are sorted to obtain a first sorting result. Specifically, the probabilities corresponding to the categories in the business scope category probability results are sorted from large to small to obtain the first sorting result. For example, assuming that the business scope category probability results are (0.15, 0.8, 0.05), the sorting result is (2, 1, 3).
[0046] Step S232, sorting the name category probability results to obtain a second sorting result;
[0047] In this embodiment, the name category probability results are sorted to obtain a second sorting result. Specifically, the probabilities corresponding to the categories in the business scope category probability results are sorted from large to small to obtain the second sorting result.
[0048] Step S233: Add the first sorting result and the second sorting result to obtain the category merging result.
[0049] In this embodiment, the first sorting result and the second sorting result are added to obtain the category merge result. Specifically, the sorting results corresponding to each classification probability in the business scope category probability result and the sorting results corresponding to each classification probability in the name category probability result are added to obtain the category merge result.
[0050] Step S24, selecting a category from the category merging result or the business scope category probability result according to a preset selection rule to obtain a classification prediction result.
[0051] In this embodiment, category selection is performed in the category merging result according to preset selection rules to obtain classification prediction results. Specifically, a category with a smaller ranking order value is selected in the category merging result as the classification prediction result. If there are multiple minimum values in the category merging result, the category corresponding to the maximum probability is selected in the business scope category probability result as the classification prediction result. For example, the business scope category probability result is (0.15, 0.8, 0.05), and the sorting result is (2, 1, 3). The name category probability result is (0.1, 0.4, 0.5), and the sorting result is (3, 2, 1). The sorting addition result is (5, 3, 4), and then the category corresponding to the sorting addition result of 3 is selected.
[0052] Step S30, combining the classification prediction result with the business scope and the company name respectively to serve as inputs of the next-layer category classification model corresponding to the business scope and the company name, obtaining the next classification prediction result, and looping until the last-layer category classification model is reached to obtain the target sub-category recognition result;
[0053] In this embodiment, the classification prediction result is respectively combined with the business scope and the company name to serve as the input of the next-layer category classification model corresponding to the business scope and the company name, respectively, to obtain the next classification prediction result, and the cycle is repeated until the last layer of category classification model is reached to obtain the target subcategory recognition result. Specifically, according to the classification prediction result obtained by the previous layer of category classification model in the hierarchical classification model, the business scope and the classification prediction result are input into the next layer of category classification model in the business scope hierarchical model, and the company name and the classification prediction result are input into the next layer of category classification model in the name hierarchical classification model to obtain the next category prediction result, so as to predict the next layer of category classification model according to the next category prediction result, and the cycle is repeated until the last layer of category classification model is reached to obtain the target subcategory recognition result.
[0054] Step S40, based on the target subcategory identification result and the mapping relationship between the preset category system, determine the industry category corresponding to the enterprise to be classified.
[0055] In this embodiment, it should be noted that, in the national economic industry classification standard, the category is the first-level category in the national economic industry classification standard, and each category is further subdivided into major categories. The major category is the second-level category in the national economic industry classification standard, which is a subdivision of the category. Each major category is further subdivided into medium categories. The medium category is the third-level category in the national economic industry classification standard, which is a subdivision of the major category. The medium category code is represented by three Arabic numerals, the first two digits are the code corresponding to the major category, and the third digit is the medium category sequence code. Each medium category is further subdivided into small categories. The small category is the fourth-level category in the national economic industry classification standard, which is a subdivision of the medium category and the last level directory. The small category code is represented by four Arabic numerals, the first three digits are the code corresponding to the medium category, and the fourth digit is the small category sequence code. The preset category system mapping relationship is the mapping relationship between the classification systems in the industry category, including the mapping relationship between the small category and the medium category, the mapping relationship between the medium category and the major category, the mapping relationship between the major category and the category, and other mapping relationships.
[0056] Based on the mapping relationship between the target subcategory identification result and the preset category system, the industry category corresponding to the enterprise to be classified is determined. Specifically, based on the target subcategory identification result and the mapping relationship between the subcategory and the middle category, the mapping relationship between the middle category and the large category, and the mapping relationship between the large category and the category, the industry category corresponding to the enterprise to be classified is determined.
[0057] The embodiment of the present application provides an industry category identification method. Compared with the technical means used in the prior art to manually search for corresponding industry categories in the national industry classification system based on text descriptions of company names and business scopes, the embodiment of the present application first obtains the business scope and company name of the company to be classified, and then inputs the business scope and the company name into the first-level category classification model of the corresponding hierarchical classification model to obtain a classification prediction result, wherein the hierarchical classification model includes category classification models of each level, and further, the classification prediction result is respectively combined with the business scope and the company name as inputs of the next-level category classification model corresponding to the business scope and the company name to obtain the next classification prediction result, and the cycle is repeated until the last level of category classification model is reached to obtain the target sub-category identification result. It should be noted that the industry category The four-level category system includes categories, major categories, medium categories and minor categories. Each layer of category classification model in the hierarchical classification model judges the output of the next layer according to the classification prediction result corresponding to the category classification model of the previous layer. The hierarchical prediction method is adopted, and the prediction target range is relatively concentrated, which reduces the search space of industry categories and speeds up the speed and accuracy of classification prediction. Furthermore, based on the mapping relationship between the target minor category recognition result and the preset category system, the industry category corresponding to the enterprise to be classified is determined, and the industry category corresponding to the enterprise to be classified is automatically obtained based on the business scope and the enterprise name. That is, there is no need for manual query and classification, which overcomes the technical defects of the prior art that when there are many enterprises missing industry categories, the traditional manual judgment method is labor-intensive and time-consuming, which leads to low industry classification efficiency, thereby improving the efficiency of industry classification.
[0058] Further, refer to Figure 2 Based on the first embodiment of the present application, in another embodiment of the present application, the hierarchical classification model includes a business scope hierarchical classification model and a name hierarchical classification model.
[0059] The classification prediction result is combined with the business scope and the company name respectively to serve as the input of the next-level category classification model corresponding to the business scope and the company name, to obtain the next classification prediction result, and the cycle is repeated until the last-level category classification model is reached. The step of obtaining the target sub-category recognition result includes:
[0060] Step A10, inputting the business scope and the classification prediction result into the second-level category classification model in the business scope classification model, and outputting the business scope major category probability result;
[0061] In this embodiment, the business scope and the classification prediction result are input into the second-layer category classification model in the business scope classification model, and the business scope major category probability result is output. Specifically, the classification prediction result output by the first-layer category classification model in the business scope classification model and the business scope are used as the input of the second-layer category classification model in the business scope classification model to obtain the business scope major category probability result corresponding to the second-layer category classification model.
[0062] Step A20, inputting the enterprise name and the classification prediction result into the second-level category classification model in the name classification model, and outputting the name category probability result;
[0063] In this embodiment, the company name and the classification prediction result are input into the second-layer category classification model in the name classification model, and the name category probability result is output. Specifically, the classification prediction result output by the first-layer category classification model in the name classification model and the company name are used as the input of the second-layer category classification model in the name classification model to obtain the name category probability result corresponding to the second-layer category classification model.
[0064] Step A30, merging the business scope category probability result and the name category probability result to obtain a category merger result;
[0065] In this embodiment, the business scope category probability results and the name category probability results are merged to obtain a category merged result. Specifically, the business scope category probability results are sorted to obtain a third sorted result, and the name category probability results are sorted to obtain a fourth sorted result. Then, the third sorted result and the fourth sorted result are added to obtain the category merged result. The specific implementation content of step A30 can refer to the specific content in step S23, which will not be repeated here.
[0066] Step A40, select a category in the major category merging result or the major category probability result of the business scope according to a preset selection rule, obtain the next category prediction result, and loop until the last layer of category classification model is reached to obtain the target subcategory recognition result.
[0067] In this embodiment, specifically, a classification with a smaller ranking order value is selected from the major category merger results as the classification prediction result of the corresponding level. If there are multiple minimum values in the major category merger results, the classification with the largest probability is selected from the business scope major category probability results as the classification prediction result of the corresponding level, so as to input the category prediction result of the corresponding level, the business scope and the company name into the next level category classification model of the corresponding hierarchical classification model, until the target sub-category recognition result is output.
[0068] The embodiment of the present application provides an industry category identification method, that is, the business scope and the classification prediction result are input into the second-layer category classification model in the business scope classification model, and the business scope major category probability result is output. At the same time, the enterprise name and the classification prediction result are input into the second-layer category classification model in the name classification model, and the name major category probability result is output. Then, the business scope major category probability result and the name major category probability result are merged to obtain a major category merged result. Furthermore, a category is selected in the major category merged result or the business scope major category probability result according to a preset selection rule to obtain the next category prediction result, and the cycle is repeated until the next category prediction result is obtained. Until the last layer of category classification model is reached, the target subcategory identification result is obtained, and the hierarchical classification model is used to predict the enterprises to be classified in the hierarchical order of categories, major categories, medium categories and minor categories. Each layer of category classification model in the hierarchical classification model judges the output of the next layer according to the classification prediction result corresponding to the category classification model of the previous layer, and finally obtains the target subcategory identification result, so that the predicted target range is more concentrated, and the prediction classification efficiency is faster. This lays the foundation for overcoming the technical defects in the prior art that when there are many enterprises missing industry categories, the traditional manual discrimination method has a large workload and takes a long time, which leads to low industry classification efficiency.
[0069] Further, refer to Figure 3 Based on the first embodiment of the present application, in another embodiment of the present application, before the business scope and the enterprise name are respectively input into the first-level category classification model of the corresponding hierarchical classification model to obtain a classification prediction result, wherein the hierarchical classification model includes the steps of each level of category classification model, the industry category identification method further includes:
[0070] Step B10, obtaining sample enterprise names corresponding to each level of categories in the industry classification and sample business scopes corresponding to each level of categories;
[0071] In this embodiment, sample enterprise names corresponding to each level of category in the industry classification and sample business scopes corresponding to each level of category are obtained. Specifically, according to the latest version of the national industry classification standard, the classification information of the latest industry is obtained, that is, sample enterprise names and sample business scopes corresponding to the categories, major categories, medium categories and minor categories, respectively.
[0072] Step B20, performing data cleaning on the sample enterprise names and the sample business scopes to obtain training enterprise names and training business scopes;
[0073] In this embodiment, data cleaning is performed on the sample enterprise names and the sample business scopes to obtain training enterprise names and training business scopes. Specifically, data information such as place names and fixed suffixes in the sample enterprise names are removed to prevent overfitting of the model training, and then the non-business scope text description parts in the sample business scope, such as descriptions of national regulations, are removed.
[0074] Step B30, obtaining a business scope hierarchical model to be trained and a name hierarchical model to be trained;
[0075] Step B40, iteratively training and optimizing the hierarchical model of the business scope to be trained through the training business scope to obtain a hierarchical classification model corresponding to the training business scope;
[0076] In this embodiment, the training business scope is used to iteratively train and optimize the hierarchical model of the business scope to be trained, so as to obtain a hierarchical classification model corresponding to the training business scope. Specifically, the first-layer category classification model in the hierarchical model of the business scope to be trained is iteratively trained through the training business scope to obtain a first-layer category classification model, and the training prediction result of the first layer is output. Then, the training business scope and the training prediction result of the first layer are used as the input of the next-layer category classification model in the hierarchical model of the business scope to be trained, so as to iteratively train the next-layer category classification model in the business scope model to obtain a category classification model of the corresponding layer, until the category classification model of the last layer is trained.
[0077] The step of iteratively training and optimizing the hierarchical model of the business scope to be trained through the training business scope to obtain the hierarchical classification model corresponding to the training business scope includes:
[0078] Step B41, iteratively training the first-layer business category classification model in the to-be-trained business scope hierarchical model through the training business scope, obtaining the first-layer business category classification model, and outputting the training business scope prediction result;
[0079] In this embodiment, the first-layer category classification model in the business scope hierarchical model to be trained is iteratively trained through the training business scope to obtain the first-layer business category classification model, and the training business scope prediction result is output. Specifically, the training business scope is input into the first-layer business category classification model in the business scope hierarchical model to be trained to optimize the first-layer business category classification model, and then determine whether the first-layer business category classification model meets the preset training end conditions, wherein the preset training end conditions include loss function convergence and reaching the maximum iteration number threshold and other conditions. If satisfied, the first-layer business category classification model is obtained, and the training business scope prediction result is output. If not satisfied, return to the execution step: the first-layer business category classification model in the business scope hierarchical model to be trained is iteratively trained through the training business scope to obtain the first-layer business category classification model.
[0080] Step B42, based on the training business scope prediction result and the training business scope, iteratively train and optimize the next layer of business category classification model to obtain the business category classification model of the corresponding level, and output the next training business scope prediction result, and loop until the last layer of business category classification model is trained to obtain the hierarchical classification model corresponding to the training business scope.
[0081] In this embodiment, specifically, the training business scope prediction result output by the first layer of business category classification model in the business scope hierarchical model to be trained is used to iteratively train the next layer of business category classification model with the training business scope, so as to optimize the next layer of business category classification model in the business scope hierarchical model to be trained until the preset training end conditions are met, obtain the category classification model of the corresponding level, and output the next training business scope prediction result, until the business category classification model of the last layer of the business scope hierarchical model to be trained is trained, and then the hierarchical classification model corresponding to the training business scope is obtained.
[0082] Step B50, iteratively train and optimize the name hierarchical model to be trained through the training enterprise name to obtain a hierarchical classification model corresponding to the training enterprise name.
[0083] In this embodiment, the name hierarchical model to be trained is iteratively trained and optimized through the training enterprise name to obtain a hierarchical classification model corresponding to the training enterprise name. Specifically, the first-layer category classification model in the hierarchical name scope model to be trained is iteratively trained through the training enterprise name to obtain a first-layer category classification model, and the training name prediction result is output, and then the training business scope and the training prediction result of the first layer are used as the input of the next-layer category classification model in the business scope hierarchical model to be trained, so as to iteratively train the next-layer category classification model in the business scope model to obtain the category classification model of the corresponding level, until the last layer of category classification model is trained.
[0084] The step of iteratively training and optimizing the to-be-trained name hierarchical model through the training enterprise name to obtain the hierarchical classification model corresponding to the training enterprise name includes:
[0085] Step B51, iteratively training the first-layer name category classification model in the name hierarchy model to be trained by using the training enterprise name, obtaining the first-layer name category classification model, and outputting the training name prediction result;
[0086] In this embodiment, the first-layer name category classification model in the name hierarchy model to be trained is iteratively trained using the training company name to obtain a first-layer name category classification model, and a training name prediction result is output. Specifically, the training company name is input into the first-layer name category classification model in the name hierarchy model to be trained to optimize the first-layer name category classification model, and then determine whether the first-layer name category classification model meets the preset training end condition. If so, the first-layer name category classification model is obtained, and the training name prediction result is output. If not, the execution step is returned to: the first-layer name category classification model in the name hierarchy model to be trained is iteratively trained using the training company name to obtain a first-layer name category classification model, and the training name prediction result is output. The specific implementation content of step B51 can refer to the specific content in step B41, which will not be repeated here.
[0087] Step B52, based on the training name prediction results and the training company names, iteratively train and optimize the next layer of name category classification model to obtain the name category classification model of the corresponding level, and output the next training name prediction result, and loop until the last layer of name category classification model is trained to obtain the hierarchical classification model corresponding to the training company name.
[0088] In this embodiment, specifically, the next layer of name category classification model in the name hierarchical model to be trained is iteratively trained using the training name prediction results output by the first layer of name category classification model in the name hierarchical model to be trained and the training enterprise name, so as to optimize the next layer of name category classification model in the name hierarchical model to be trained until the preset training end conditions are met, and the name category classification model of the corresponding level is obtained, until the last layer of name category classification model is trained, and then the hierarchical classification model corresponding to the training enterprise name is obtained. The specific implementation content of step B52 can refer to the specific content in step B42, which will not be repeated here.
[0089] The embodiment of the present application provides an industry category identification method, that is, obtaining sample enterprise names corresponding to each level of category in the industry classification and sample business scopes corresponding to each level of category, and then performing data cleaning on the sample enterprise names and the sample business scopes to obtain training enterprise names and training business scopes, and further obtaining a hierarchical model of the business scope to be trained and a hierarchical model of the name to be trained, and then iteratively training and optimizing the hierarchical model of the business scope to be trained through the training business scope to obtain a hierarchical classification model corresponding to the training business scope, and further, iteratively training and optimizing the hierarchical model of the name to be trained through the training enterprise names to obtain a hierarchical classification model corresponding to the training enterprise names, and the training process adopts a hierarchical training method, that is, each level only judges the category corresponding to the next level, so that in the process of predicting the enterprise to be classified through the hierarchical classification model, the search space of the category is reduced, thereby improving the efficiency of steel industry classification, and laying a foundation for overcoming the technical defects in the prior art that when there are many enterprises missing industry categories, the use of traditional manual discrimination methods is labor-intensive and time-consuming, thereby resulting in low industry classification efficiency.
[0090] Reference Figure 4 , Figure 4 It is a schematic diagram of the structure of the industry category identification device of the hardware operating environment involved in the embodiment of the present application.
[0091] like Figure 4 As shown, the industry category identification device may include: a processor 1001, such as a CPU, a memory 1005, and a communication bus 1002. The communication bus 1002 is used to realize the connection and communication between the processor 1001 and the memory 1005. The memory 1005 may be a high-speed RBM memory or a stable memory (non-volBtile memory), such as a disk memory. The memory 1005 may optionally be a storage device independent of the aforementioned processor 1001.
[0092] Optionally, the industry category identification device may also include a rectangular user interface, a network interface, a camera, an RF (RBdio Frequency) circuit, a sensor, an audio circuit, a WiFi module, etc. The rectangular user interface may include a display screen (DisplBy), an input submodule such as a keyboard (KeyboBrd), and the optional rectangular user interface may also include a standard wired interface and a wireless interface. The network interface may optionally include a standard wired interface and a wireless interface (such as a WIFI interface).
[0093] Those skilled in the art will understand that Figure 4 The industry category identification device structure shown in the figure does not constitute a limitation on the industry category identification device, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.
[0094] like Figure 4 As shown, the memory 1005 as a computer storage medium may include an operating system, a network communication module, and an industry category identification program. The operating system is a program that manages and controls the hardware and software resources of the industry category identification device, and supports the operation of the industry category identification program and other software and / or programs. The network communication module is used to realize the communication between the components inside the memory 1005, and the communication with other hardware and software in the industry category identification system.
[0095] exist Figure 4 In the industry category identification device shown, the processor 1001 is used to execute the industry category identification program stored in the memory 1005 to implement the steps of any of the above-mentioned industry category identification methods.
[0096] The specific implementation of the industry category identification device of the present application is basically the same as the various embodiments of the above-mentioned industry category identification method, and will not be repeated here.
[0097] The present application also provides an industry category identification device, the industry category identification device comprising:
[0098] An acquisition module is used to obtain the business scope and enterprise name of the enterprise to be classified;
[0099] A first classification module, used to input the business scope and the enterprise name into the first-level category classification model of the corresponding hierarchical classification model to obtain a classification prediction result, wherein the hierarchical classification model includes the category classification models of each level;
[0100] The second classification module is used to combine the classification prediction result with the business scope and the company name respectively to serve as the input of the next-level category classification model corresponding to the business scope and the company name, obtain the next classification prediction result, and loop until the last-level category classification model is reached to obtain the target sub-category recognition result;
[0101] A determination module is used to determine the industry category corresponding to the enterprise to be classified based on the target subcategory identification result and the mapping relationship between the preset category system.
[0102] Optionally, the first classification module is further used for:
[0103] Input the business scope into the first-level category classification model in the business scope hierarchical classification model, and output the business scope category probability result;
[0104] Input the enterprise name into the first-level category classification model in the name hierarchical classification model, and output the name category probability result;
[0105] Merge the business scope category probability result and the name category probability result to obtain a category merger result;
[0106] In the category merging result or the business scope category probability result, category selection is performed according to preset selection rules to obtain a classification prediction result.
[0107] Optionally, the first classification module is further used for:
[0108] Sorting the probability results of the business scope categories to obtain a first sorting result;
[0109] Sorting the name category probability results to obtain a second sorting result;
[0110] The first sorting result and the second sorting result are added together to obtain the category merging result.
[0111] Optionally, the second classification module is further used for:
[0112] Input the business scope and the classification prediction result into the second-level category classification model in the business scope classification model, and output the probability result of the major category of the business scope;
[0113] Input the enterprise name and the classification prediction result into the second-level category classification model in the name classification model, and output the name category probability result;
[0114] Merge the business scope category probability result and the name category probability result to obtain a category merger result;
[0115] In the major category merging result or the major category probability result of the business scope, category selection is performed according to preset selection rules to obtain the next category prediction result, and the cycle is repeated until the last layer of category classification model is reached to obtain the target subcategory recognition result.
[0116] Optionally, the industry category identification device is further used for:
[0117] Obtain the sample enterprise names and sample business scopes corresponding to each level of categories in the industry classification;
[0118] Performing data cleaning on the sample enterprise names and the sample business scopes to obtain training enterprise names and training business scopes;
[0119] Obtain the business scope hierarchical model to be trained and the name hierarchical model to be trained;
[0120] Iteratively train and optimize the hierarchical model of the business scope to be trained through the training business scope to obtain a hierarchical classification model corresponding to the training business scope;
[0121] The hierarchical model of the name to be trained is iteratively trained and optimized through the training enterprise name to obtain a hierarchical classification model corresponding to the training enterprise name.
[0122] Optionally, the industry category identification device is further used for:
[0123] Iteratively training the first-layer category classification model in the to-be-trained business scope hierarchical model through the training business scope to obtain a first-layer business category classification model, and outputting a prediction result of the training business scope;
[0124] Based on the training business scope prediction result and the training business scope, the next layer of business category classification model is iteratively trained and optimized to obtain the business category classification model of the corresponding level, and the next training business scope prediction result is output. The cycle is repeated until the last layer of business category classification model is trained to obtain the hierarchical classification model corresponding to the training business scope.
[0125] Optionally, the industry category identification device is further used for:
[0126] Iteratively training the first-layer category classification model in the to-be-trained name hierarchical model through the training enterprise name to obtain the first-layer name category classification model, and outputting the training name prediction result;
[0127] Based on the training name prediction results and the training company names, the next layer of name category classification model is iteratively trained and optimized to obtain the name category classification model of the corresponding level, and the next training name prediction result is output. The cycle is repeated until the last layer of name category classification model is trained to obtain the hierarchical classification model corresponding to the training company name.
[0128] The specific implementation of the industry category identification device of the present application is basically the same as the various embodiments of the above-mentioned industry category identification method, and will not be repeated here.
[0129] An embodiment of the present application provides a medium, which is a readable storage medium, and the readable storage medium stores one or more programs, and the one or more programs can also be executed by one or more processors to implement the steps of the industry category identification method described in any one of the above items.
[0130] The specific implementation of the readable storage medium of the present application is basically the same as the various embodiments of the above-mentioned industry category identification method, and will not be repeated here.
[0131] An embodiment of the present application provides a computer program product, and the computer program product includes one or more computer programs, and the one or more computer programs can also be executed by one or more processors to implement the steps of any of the above-mentioned industry category identification methods.
[0132] The specific implementation methods of the computer program product of the present application are basically the same as the various embodiments of the above-mentioned industry category identification method, and will not be repeated here.
[0133] The above are only preferred embodiments of the present application, and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent processing scope of the present application.
Claims
1. A method for identifying industry categories, characterized in that: The industry category identification method includes: Obtain the business scope and company name of the enterprise to be classified; Inputting the business scope and the enterprise name into the first-level category classification model of the corresponding hierarchical classification model to obtain a classification prediction result, wherein the hierarchical classification model includes the category classification models of each level; The classification prediction results are combined with the business scope and the company name respectively to serve as inputs of the next-layer category classification models corresponding to the business scope and the company name, respectively, to obtain the next classification prediction results, and the cycle is repeated until the last layer of category classification model is reached to obtain the target sub-category recognition result; Determine the industry category corresponding to the enterprise to be classified based on the target subcategory identification result and the mapping relationship between the preset category system; The hierarchical classification model includes a business scope hierarchical classification model and a name hierarchical classification model. The step of inputting the business scope and the enterprise name into the first-level category classification model of the respective corresponding hierarchical classification models to obtain the classification prediction result includes: Input the business scope into the first-level category classification model in the business scope hierarchical classification model, and output the business scope category probability result; Input the enterprise name into the first-level category classification model in the name hierarchical classification model, and output the name category probability result; Merge the business scope category probability result and the name category probability result to obtain a category merger result; In the category merging result or the business scope category probability result, category selection is performed according to preset selection rules to obtain classification prediction results. The preset selection rule is to select the category with the smallest ranking value in the category merging result or to select the category with the largest probability in the business scope category probability result.
2. The industry category identification method according to claim 1, characterized in that: The step of merging the business scope category probability result and the name category probability result to obtain a category merging result comprises: Sorting the probability results of the business scope categories to obtain a first sorting result; Sorting the name category probability results to obtain a second sorting result; The first sorting result and the second sorting result are added together to obtain the category merging result.
3. The industry category identification method according to claim 1, characterized in that: The hierarchical classification model includes a business scope hierarchical classification model and a name hierarchical classification model. The step of combining the classification prediction result with the business scope and the company name respectively to serve as the input of the next-layer category classification model corresponding to the business scope and the company name, obtaining the next classification prediction result, and looping until the last-layer category classification model is reached to obtain the target sub-category recognition result includes: Input the business scope and the classification prediction result into the second-level category classification model in the business scope classification model, and output the probability result of the major category of the business scope; Input the enterprise name and the classification prediction result into the second-level category classification model in the name classification model, and output the name category probability result; Merge the business scope category probability result and the name category probability result to obtain a category merger result; In the major category merging result or the major category probability result of the business scope, category selection is performed according to preset selection rules to obtain the next classification prediction result, and the cycle is repeated until the last layer of category classification model is reached to obtain the target subcategory recognition result.
4. The industry category identification method according to claim 1, characterized in that: Before the step of inputting the business scope and the enterprise name into the first-level category classification model of the corresponding hierarchical classification model to obtain the classification prediction result, wherein the hierarchical classification model includes the steps of each level of category classification model, the industry category identification method further includes: Obtain the sample enterprise names and sample business scopes corresponding to each level of categories in the industry classification; Performing data cleaning on the sample enterprise names and the sample business scopes to obtain training enterprise names and training business scopes; Obtain the business scope hierarchical model to be trained and the name hierarchical model to be trained; Iteratively train and optimize the hierarchical model of the business scope to be trained through the training business scope to obtain a hierarchical classification model corresponding to the training business scope; The hierarchical model of the name to be trained is iteratively trained and optimized through the training enterprise name to obtain a hierarchical classification model corresponding to the training enterprise name.
5. The industry category identification method according to claim 4, characterized in that: The step of iteratively training and optimizing the to-be-trained business scope hierarchical model through the training business scope to obtain the hierarchical classification model corresponding to the training business scope comprises: Iteratively training the first-layer business category classification model in the to-be-trained business scope hierarchical model through the training business scope to obtain the first-layer business category classification model, and outputting the training business scope prediction result; Based on the training business scope prediction result and the training business scope, the next layer of business category classification model is iteratively trained and optimized to obtain the business category classification model of the corresponding level, and the next training business scope prediction result is output. The cycle is repeated until the last layer of business category classification model is trained to obtain the hierarchical classification model corresponding to the training business scope.
6. The industry category identification method according to claim 4, characterized in that: The step of iteratively training and optimizing the to-be-trained name hierarchical model through the training enterprise name to obtain the hierarchical classification model corresponding to the training enterprise name comprises: Iteratively training the first-layer name category classification model in the name hierarchy model to be trained by using the training enterprise name to obtain the first-layer name category classification model, and outputting the training name prediction result; Based on the training name prediction results and the training company names, the next layer of name category classification model is iteratively trained and optimized to obtain the name category classification model of the corresponding level, and the next training name prediction result is output. The cycle is repeated until the last layer of name category classification model is trained to obtain the hierarchical classification model corresponding to the training company name.
7. An industry category identification device, characterized in that: The industry category identification device includes: a memory, a processor, and an industry category identification program stored in the memory. The industry category identification program is executed by the processor to implement the steps of the industry category identification method as described in any one of claims 1 to 6.
8. A medium, the medium being a readable storage medium, characterized in that: The readable storage medium stores an industry category identification program, and the industry category identification program is executed by a processor to implement the steps of the industry category identification method as described in any one of claims 1 to 6.
9. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the industry category identification method as described in any one of claims 1 to 6 are implemented.
