Enterprise risk automatic early warning method and system based on classification model

Abstract

The invention discloses an enterprise risk automatic early warning method and system based on a classification model, and the method comprises the steps: S1, generating model training data Mt, S2, carrying out model training: dividing the data Mt into N sub-data sets with the same number, sequentially taking one sub-data set as a test set, taking the other N-1 sub-data sets as training sets, carrying out the multiple times of model training evaluation, and obtaining the model training data Mt; calculating the average value of the accuracy obtained by multiple times of training as a model score index; s3, model selection: selecting logistic regression, a support vector machine, a random forest and a gradient boosting decision tree algorithm to perform model training according to the requirements of the step S2, generating scores of each algorithm model, and screening out a final prediction model according to the scores; and S4, performing risk prediction by using the screened prediction model. According to the method, risk prediction and evaluation of enterprise labor dispute cases are realized through a trained neural network model by using a data mining means, so that pre-warning is facilitated.

Description

Technical Field

[0001] The present invention belongs to the field of computer software technology, and in particular relates to an automatic early warning method and system for enterprise risks based on a classification model. Background Art

[0002] Existing risk warnings for corporate labor dispute cases are basically issued after the fact. They verify the reporter's information and use rules, comparisons and other means to judge the authenticity and risk level of the case. This method cannot quantify the risk level, nor can it conduct pre-emptive prevention and control of companies with potential labor disputes. Summary of the invention

[0003] In order to solve the above problems, the present invention provides an automatic enterprise risk warning method and system based on a classification model, which uses data mining to achieve risk prediction and evaluation of enterprise labor dispute cases through a trained neural network model, so as to provide early warning.

[0004] The technical solution adopted by the present invention is:

[0005] In a first aspect, the present application discloses an automatic early warning method for enterprise risks based on a classification model, comprising the following steps:

[0006] S1. Generate model training data Mt:

[0007] Mt=[ ]in, It is the basic information feature set of the enterprise. is the characteristic set of information on the insurance participation behavior of enterprises, is the structural characteristics set of the enterprise's insured personnel, It is a feature set of historical labor dispute information of an enterprise;

[0008] S2, model training: Divide the model training data Mt into N equal sub-datasets, take one of the sub-datasets as the test set and the remaining N-1 sub-datasets as the training set, perform multiple model training evaluations, and calculate the average of the precision rates obtained from multiple trainings as the model score indicator; where 8≤N≤12;

[0009] S3, model selection: select logistic regression, support vector machine, random forest, gradient boosting decision tree algorithm to perform model training according to the requirements of step S2, and generate scores for each algorithm model, and select the final prediction model according to the scores;

[0010] S4. Use the selected final prediction model to perform risk prediction: obtain relevant data of the target enterprise, generate a feature set Mp of the target enterprise based on the relevant data, use the feature set Mp as the input of the final prediction model, perform enterprise risk prediction, and output the risk value.

[0011] As an optional technical solution, the step S1 of generating the model training data Mt includes the following steps:

[0012] S11, collecting enterprise-related information, the relevant information at least includes enterprise basic information M1, enterprise insurance behavior information M2, enterprise insurance personnel information M3 and enterprise labor dispute case information M4;

[0013] S12: construct a corresponding data feature set based on the relevant information, wherein the data feature set includes an enterprise basic information feature set constructed based on the enterprise basic information M1. , the enterprise insurance behavior information feature set constructed based on the enterprise insurance behavior information M2 , a structural feature set of enterprise insured personnel constructed based on the enterprise insured personnel information M3 ;

[0014] S13, using the bag-of-words model in combination with cosine similarity to calculate text similarity between the non-standardized enterprise names and the standardized enterprise names, performing standardized mapping on the non-standardized enterprise names, and replacing the non-standardized enterprise names in the enterprise labor dispute case information M4 according to the standardized mapping relationship;

[0015] S14, group and merge the persons subject to execution in the replaced labor dispute case information M4, calculate the number of historical labor disputes of the enterprise f14, the total number of persons involved in the case f15, and the total amount involved in the case f16, and form a characteristic set of historical labor dispute information of the enterprise ;

[0016] S15. Use the standardized enterprise name as an important field for data association, and associate it with the enterprise basic information feature set in turn. , Enterprise insurance behavior information feature set , Structural characteristics of enterprise insured personnel And the enterprise historical labor dispute information feature set , generate the model training data Mt.

[0017] As an optional technical solution, the enterprise basic information feature set in step S12 The construction includes: according to the basic information of the enterprise M1, natural number coding is used to convert the enterprise scale f1, enterprise type f2, economic type f3, affiliation f4, industry f5 and industry risk information f6 to form the basic information feature set of the enterprise .

[0018] As an optional technical solution, the enterprise insurance behavior information feature set in step S12 The construction includes: selecting and constructing the enterprise insurance behavior information feature set based on the enterprise insurance information M2 , the enterprise insurance behavior information feature set Including the number of insured persons in the enterprise f7, the average insurance duration of employees in the enterprise f8, the number of months of social security arrears f9, and the change rate of the number of social security payers f10. Among them, features f8, f9, and f10 are The calculation formula is as follows: in, represents the average insurance participation duration of employees in enterprise j, where i represents the i-th employee, ys represents the number of months that employee i has participated in insurance in enterprise j, and n represents the total number of insured employees in the enterprise;

[0019] in, It indicates the number of months in which enterprise j has social security arrears. If the month is marked as "arrears", it is added by 1. The final sum is the number of months in which enterprise j has social security arrears. Indicates the enterprise The number of months of social security arrears;

[0020] in, Indicates the enterprise The change rate of the number of social security contributors; represents the number of insured persons per month in enterprise j, t represents the month, and n represents The length of (the sequence of the number of insured persons per month of enterprise j), represents the i-th time, represents the average value at a time point, yes The value at time i, is the mean value of X.

[0021] As an optional technical solution, the structure feature set of the enterprise insured personnel in step S12 is The construction includes: according to the information of the insured personnel of the enterprise M3, the age structure proportion f11, gender structure proportion f12, and educational structure proportion f13 of the current insured personnel of the enterprise are counted to form the structural characteristic set of the insured personnel of the enterprise .

[0022] As an optional technical solution, step S13 specifically includes:

[0023] S131, extract the non-standard enterprise names in the enterprise labor dispute case information M4, remove the hyphens, left brackets, right brackets and characters that appear continuously at the beginning of each non-standard enterprise name, and form a non-standard enterprise name array A1 after deduplication; extract the standard enterprise names from the enterprise basic information M1, and form a standard enterprise name array A2 after deduplication; then use Jieba word segmentation to perform text segmentation on arrays A1 and A2;

[0024] S132, combining the segmented arrays A1 and A2, and establishing an index dictionary D of the corpus features. The dictionary D assigns a number to each word and records the frequency information of the word;

[0025] S133. Calculate the frequency of each word in the company name :

[0026]

[0027] in, Expressing words In the company name The number of times it appears in Indicates the company name The total number of words in

[0028] S134. Calculate inverse document frequency :

[0029]

[0030] in, Indicates the total number of company names, Indicates that it contains words Total number of company names;

[0031] S135. Convert the company name into a word frequency weighted sparse vector :

[0032]

[0033] in, For company name The word frequency weight sparse vector of Expressing words The word frequency weight value of .

[0034] S136, calculate the cosine similarity between vectors and construct a similarity matrix, that is, the similarity between standard enterprise names and non-standard enterprise names :

[0035]

[0036] in, represents the standard enterprise name vector, represents a non-standard enterprise name vector, Representation vector With vector The cosine of the angle, represents the word frequency weight value of word i in A, Represents the word frequency weight value of word i in B;

[0037] S137, selecting a standard enterprise name that is most similar to the non-standard enterprise name, and forming a standardized mapping relationship of the enterprise names;

[0038] S138. Replace the non-standardized enterprise name in the enterprise labor dispute case information M4 according to the standardized mapping relationship.

[0039] As an optional technical solution, the step S3 of selecting the final prediction model according to the score includes: if the score differences between the algorithm models are greater than 0.1, the model with the highest model score is used as the final prediction model; if the score differences between the algorithm models are less than 0.1, the average value of the prediction values ​​of the algorithm models is used as the final output result of the model.

[0040] In a second aspect, the present application discloses an automatic early warning system for enterprise risks based on a classification model, comprising:

[0041] Model training data generation module, used to generate model training data Mt:

[0042] Mt=[ ]

[0043] in, It is the basic information feature set of the enterprise. is the characteristic set of information on the insurance participation behavior of enterprises, is the structural characteristics set of the enterprise's insured personnel, It is a feature set of historical labor dispute information of an enterprise;

[0044] The model training module is used to divide the model training data Mt into N sub-datasets of equal number, and use one of the sub-datasets as the test set and the remaining N-1 sub-datasets as the training set in turn, perform multiple model training evaluations, and calculate the average of the precision rates obtained from multiple trainings as the model score indicator; wherein 8≤N≤12;

[0045] The model selection module is used to select logistic regression, support vector machine, random forest, and gradient boosting decision tree algorithms to perform model training according to the requirements of the model training module, generate scores for each algorithm model, and select the final prediction model based on the scores;

[0046] The prediction module is used to perform risk prediction using the screened final prediction model, including: obtaining relevant data of the target enterprise, generating a feature set Mp of the target enterprise based on the relevant data, using the feature set Mp as input of the final prediction model, performing enterprise risk prediction, and outputting a risk value.

[0047] As an optional technical solution, the model training data generation module includes:

[0048] A data collection module is used to collect enterprise-related information, which includes at least enterprise basic information M1, enterprise insurance behavior information M2, enterprise insurance personnel information M3, and enterprise labor dispute case information M4;

[0049] A feature construction module is used to construct a corresponding data feature set based on the relevant information, and the data feature set includes an enterprise basic information feature set constructed based on the enterprise basic information M1. , the enterprise insurance behavior information feature set constructed based on the enterprise insurance behavior information M2 , a structural feature set of enterprise insured personnel constructed based on the enterprise insured personnel information M3 ;

[0050] A standardized enterprise name mapping module is used to calculate the text similarity between non-standardized enterprise names and standardized enterprise names by using a bag-of-words model combined with cosine similarity, perform standardized mapping on the non-standardized enterprise names, and replace the non-standardized enterprise names in the enterprise labor dispute case information M4 according to the standardized mapping relationship;

[0051] The feature building module is also used to: group and merge the persons subject to execution in the replaced labor dispute case information M4, calculate the number of historical labor disputes f14, the total number of persons involved in the case f15, and the total amount involved in the case f16, and form a feature set of the historical labor dispute information of the enterprise. ;

[0052] The feature construction module is also used to: according to the basic information of the enterprise M1, use natural number coding to convert the enterprise scale f1, enterprise type f2, economic type f3, affiliation f4, industry f5 and industry risk information f6 to form a basic information feature set of the enterprise ;

[0053] The feature construction module is also used to select and construct the enterprise insurance behavior information feature set according to the enterprise insurance information M2 , the enterprise insurance behavior information feature set It includes the number of insured persons in the enterprise f7, the number of months of social security payment f8, the number of months of social security arrears f9, and the change rate of the number of social security payers f10. Among them, features f8, f9, and f10 are The calculation formula is as follows: in, represents the average insurance participation duration of employees in enterprise j, where i represents the i-th employee, ys represents the number of months that employee i has participated in insurance in enterprise j, and n represents the total number of insured employees in the enterprise;

[0054] in, It indicates the number of months of social security arrears of enterprise j. If the month is marked as "arrears", add 1, and the final sum is the number of months of social security arrears of enterprise j. Indicates the enterprise The number of months of social security arrears;

[0055] in, Indicates the enterprise The change rate of the number of social security contributors; represents the number of insured persons per month in enterprise j, t represents the month, and n represents The length of (the sequence of the number of insured persons per month of enterprise j), represents the i-th time, represents the average value at a time point, yes The value at time i, is the mean value of X.

[0056] The feature construction module is also used to: count the age structure proportion f11, gender structure proportion f12, and educational structure proportion f13 of the current insured persons in the enterprise according to the insured persons information M3 of the enterprise, and form a structural feature set of the insured persons in the enterprise ;

[0057] The model training data generation module is also used to: use the standardized enterprise name as an important field for data association, and sequentially associate the enterprise basic information feature set , Enterprise insurance behavior information feature set , Structural characteristics of enterprise insured personnel And the enterprise historical labor dispute information feature set , generate the model training data Mt.

[0058] As an optional technical solution, the enterprise name standardization mapping module is also used to:

[0059] Extract the non-standard enterprise names in the enterprise labor dispute case information M4, remove the hyphens, left brackets, right brackets and characters that appear consecutively at the beginning of each non-standard enterprise name, and form a non-standard enterprise name array A1 after deleting duplicates; extract the standard enterprise names from the enterprise basic information M1, and form a standard enterprise name array A2 after deleting duplicates; then use Jieba word segmentation to perform text segmentation on arrays A1 and A2;

[0060] Combine the segmented arrays A1 and A2 to create an index dictionary D of the corpus features. Dictionary D assigns a number to each word and records the frequency of occurrence of the word.

[0061] Calculate the frequency of each word in the company name :

[0062]

[0063] in, Expressing words In the company name The number of times it appears in Indicates the company name The total number of words in

[0064] Calculating Inverse Document Frequency :

[0065]

[0066] in, Indicates the total number of company names, Indicates that it contains words Total number of company names;

[0067] S135. Convert the company name into a word frequency weighted sparse vector :

[0068]

[0069] in, For company name The word frequency weight sparse vector of Expressing words The word frequency weight value of .

[0070] Calculate the cosine similarity between vectors and construct a similarity matrix, i.e. the similarity between standard company names and non-standard company names :

[0071] in, represents the standard enterprise name vector, represents a non-standard enterprise name vector, Representation vector With vector The cosine of the angle, represents the word frequency weight value of word i in A, Represents the word frequency weight value of word i in B;

[0072] Select the standard enterprise name that is most similar to the non-standard enterprise name to form a standardized mapping relationship between the enterprise names;

[0073] The non-standardized enterprise names in the enterprise labor dispute case information M4 are replaced according to the standardized mapping relationship.

[0074] The beneficial effects of the present invention are as follows: the present invention collects data such as basic information of enterprises, information on enterprise insurance participation behavior, information on enterprise insured personnel, and information on enterprise labor dispute cases, and uses data mining methods to construct a neural network model to identify enterprises with potential labor disputes and classify their risk levels, and distributes them through an early warning system to achieve preemptive intervention in enterprise labor disputes, which to a large extent fills the gap in pre-emptive risk prevention and control measures for labor dispute risks and greatly reduces the number of reports of enterprise labor dispute cases. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Figure 1 It is a flow chart of an automatic early warning method for enterprise risks based on a classification model in an embodiment.

[0076] Figure 2 It is an architecture diagram of the enterprise risk automatic early warning system based on the classification model in the embodiment. DETAILED DESCRIPTION

[0077] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0078] like Figure 1As shown, the present application discloses an automatic early warning method for enterprise risks based on a classification model, comprising the following steps:

[0079] S1. Generate model training data Mt:

[0080] Mt=[ ]

[0081] in, It is the basic information feature set of the enterprise. is the characteristic set of information on the insurance participation behavior of enterprises, is the structural characteristics set of the enterprise's insured personnel, It is a feature set of historical labor dispute information of an enterprise.

[0082] Furthermore, the generation of model training data Mt specifically includes the following process:

[0083] S11. Collect enterprise-related information, which at least includes enterprise basic information M1, enterprise insurance behavior information M2, enterprise insured personnel information M3 and enterprise labor dispute case information M4.

[0084] S12: construct a corresponding data feature set based on the relevant information, wherein the data feature set includes an enterprise basic information feature set constructed based on the enterprise basic information M1. , the enterprise insurance behavior information feature set constructed based on the enterprise insurance behavior information M2 , a structural feature set of enterprise insured personnel constructed based on the enterprise insured personnel information M3 .

[0085] Among them, the basic information feature set of enterprises The construction includes: according to the basic information of the enterprise M1, natural number coding is used to convert the enterprise scale f1, enterprise type f2, economic type f3, affiliation f4, industry f5 and industry risk information f6 to form the basic information feature set of the enterprise .

[0086] Enterprise insurance behavior information feature set The construction includes: selecting and constructing the enterprise insurance behavior information feature set based on the enterprise insurance information M2 , the enterprise insurance behavior information feature set It includes the number of insured persons in the enterprise f7, the number of months of social security payment f8, the number of months of social security arrears f9, and the change rate of the number of social security payers f10. Among them, features f8, f9, and f10 are The calculation formula is as follows: in, represents the average insurance participation duration of employees in enterprise j, where i represents the i-th employee, ys represents the number of months that employee i has participated in insurance in enterprise j, and n represents the total number of insured employees in the enterprise;

[0087] in, It indicates the number of months in which enterprise j has social security arrears. If the month is marked as "arrears", it is added by 1. The final sum is the number of months in which enterprise j has social security arrears. Indicates the enterprise The number of months of social security arrears;

[0088] in, Indicates the enterprise The change rate of the number of social security contributors; represents the number of insured persons per month in enterprise j, t represents the month, and n represents The length of (the sequence of the number of insured persons per month of enterprise j), represents the i-th time, represents the average value at a time point, yes The value at time i, is the mean value of X.

[0089] Structural characteristics of enterprise insured personnel The construction includes: according to the information of the insured personnel of the enterprise M3, the age structure proportion f11, gender structure proportion f12, and educational structure proportion f13 of the current insured personnel of the enterprise are counted to form the structural characteristic set of the insured personnel of the enterprise Specifically, enterprises No. Proportion of people by age group for:

[0090]

[0091] in, Indicates the age range, which can be divided into The five age groups can also be divided into other age groups according to actual conditions. Taking the five age groups as an example, when hour, That means the enterprise The number of people aged between 20 (inclusive) and 30 (excluding 30) among them.

[0092] enterprise Ratio of male to female staff for:

[0093]

[0094] in, Indicates male or female, For example, for male, Indicates the enterprise of male employees.

[0095] enterprise Proportion of each educational background for:

[0096]

[0097] in, Indicates college degree or below, bachelor degree, master degree or doctorate degree, Taking undergraduate course as an example, That is to say, the enterprise The total number of people with a bachelor's degree.

[0098] S13, using the Bag-of-Words model combined with cosine similarity to calculate the text similarity between the non-standardized enterprise names and the standardized enterprise names, perform standardized mapping on the non-standardized enterprise names, and replace the non-standardized enterprise names in the enterprise labor dispute case information M4 according to the standardized mapping relationship. This process specifically includes:

[0099] S131, extract the non-standard enterprise names in the enterprise labor dispute case information M4, remove the hyphens, left brackets, right brackets and characters that appear continuously at the beginning of each non-standard enterprise name, and form a non-standard enterprise name array A1 after deduplication; extract the standard enterprise names from the enterprise basic information M1, and form a standard enterprise name array A2 after deduplication; then use Jieba word segmentation to perform text segmentation on arrays A1 and A2;

[0100] S132, combining the segmented arrays A1 and A2, and establishing an index dictionary D of the corpus features. The dictionary D assigns a number to each word and records the frequency information of the word;

[0101] S133. Calculate the frequency of each word in the company name :

[0102]

[0103] in, Expressing words In the company name The number of times it appears in Indicates the company name The total number of words in

[0104] S134. Calculate inverse document frequency :

[0105]

[0106] in, Indicates the total number of company names, Indicates that it contains words Total number of company names;

[0107] S135. Convert the company name into a word frequency weighted sparse vector :

[0108]

[0109] in, For company name The word frequency weight sparse vector of Expressing words The word frequency weight value of .

[0110] S136, calculate the cosine similarity between vectors and construct a similarity matrix, that is, the similarity between standard enterprise names and non-standard enterprise names :

[0111]

[0112] in, represents the standard enterprise name vector, represents a non-standard enterprise name vector, Representation vector With vector The cosine of the angle, represents the word frequency weight value of word i in A, Represents the word frequency weight value of word i in B;

[0113] S137, selecting a standard enterprise name that is most similar to the non-standard enterprise name, and forming a standardized mapping relationship of the enterprise names;

[0114] S138. Replace the non-standardized enterprise name in the enterprise labor dispute case information M4 according to the standardized mapping relationship.

[0115] S14, group and merge the persons subject to execution in the replaced labor dispute case information M4, calculate the number of historical labor disputes of the enterprise f14, the total number of persons involved in the case f15, and the total amount involved in the case f16, and form a characteristic set of historical labor dispute information of the enterprise .

[0116] S15. Use the standardized enterprise name as an important field for data association, and associate it with the enterprise basic information feature set in turn. , Enterprise insurance behavior information feature set , Structural characteristics of enterprise insured personnel And the enterprise historical labor dispute information feature set , generate the model training data Mt.

[0117] S2. Model training: Divide the model training data Mt into N equal sub-datasets, use one of the sub-datasets as a test set and the remaining N-1 sub-datasets as training sets, perform multiple model training evaluations, and calculate the average of the precision rates obtained from multiple trainings as the model score indicator; wherein, 8≤N≤12, and N can be 10 in this embodiment.

[0118] S3, model selection: select logistic regression, support vector machine, random forest, and gradient boosting decision tree algorithms to perform model training according to the requirements of step S2, and generate scores for each algorithm model, and select the final prediction model based on the scores.

[0119] Furthermore, if the score differences between the algorithm models are greater than 0.1, the model with the highest model score will be used as the final prediction model; if the score differences between the algorithm models are less than 0.1, the average of the predicted values ​​of the algorithm models will be used as the final output result of the model.

[0120] S4. Use the selected final prediction model to perform risk prediction: obtain relevant data of the target enterprise, generate a feature set Mp of the target enterprise based on the relevant data, use the feature set Mp as the input of the final prediction model, perform enterprise risk prediction, and output a risk value, which is between 0 and 1.

[0121] It should be noted that the generation process of the feature set Mp of the target enterprise can refer to the generation process of the model training data Mt, which will not be repeated here.

[0122] S5. Risk levels are divided into high, medium and low according to risk values. The early warning and prediction platform will issue a risk list, which will be further verified and processed by the handling personnel.

[0123] like Figure 2 As shown, the present application discloses an automatic early warning system for enterprise risks based on a classification model, comprising:

[0124] Model training data generation module, used to generate model training data Mt:

[0125] Mt=[ ]

[0126] in, It is the basic information feature set of the enterprise. is the characteristic set of information on the insurance participation behavior of enterprises, is the structural characteristics set of the enterprise's insured personnel, It is a feature set of historical labor dispute information of an enterprise;

[0127] The model training module is used to divide the model training data Mt into N sub-datasets of equal number, and use one of the sub-datasets as the test set and the remaining N-1 sub-datasets as the training set in turn, perform multiple model training evaluations, and calculate the average of the precision rates obtained from multiple trainings as the model score indicator; wherein 8≤N≤12;

[0128] The model selection module is used to select logistic regression, support vector machine, random forest, and gradient boosting decision tree algorithms to perform model training according to the requirements of the model training module, generate scores for each algorithm model, and select the final prediction model based on the scores;

[0129] The prediction module is used to perform risk prediction using the final prediction model selected, including: obtaining relevant data of the target enterprise, generating a feature set Mp of the target enterprise according to the relevant data, using the feature set Mp as the input of the final prediction model, performing enterprise risk prediction, and outputting a risk value, wherein the risk value is between 0 and 1;

[0130] And the early warning module is used to classify the risk levels into high, medium and low according to the risk value. For example, if the risk value is greater than or equal to 95%, it is classified as high risk, between 90%-95% is classified as medium risk, and less than or equal to 90% is classified as low risk. Other classifications can also be made according to actual conditions.

[0131] As an optional technical solution, the model training data generation module includes:

[0132] A data collection module is used to collect enterprise-related information, which includes at least enterprise basic information M1, enterprise insurance behavior information M2, enterprise insurance personnel information M3, and enterprise labor dispute case information M4;

[0133] A feature construction module is used to construct a corresponding data feature set based on the relevant information, and the data feature set includes an enterprise basic information feature set constructed based on the enterprise basic information M1. , the enterprise insurance behavior information feature set constructed based on the enterprise insurance behavior information M2 , a structural feature set of enterprise insured personnel constructed based on the enterprise insured personnel information M3 ;

[0134] A standardized enterprise name mapping module is used to calculate the text similarity between non-standardized enterprise names and standardized enterprise names by using a bag-of-words model combined with cosine similarity, perform standardized mapping on the non-standardized enterprise names, and replace the non-standardized enterprise names in the enterprise labor dispute case information M4 according to the standardized mapping relationship;

[0135] The feature building module is also used to: group and merge the persons subject to execution in the replaced labor dispute case information M4, calculate the number of historical labor disputes f14, the total number of persons involved in the case f15, and the total amount involved in the case f16, and form a feature set of the historical labor dispute information of the enterprise. ;

[0136] The feature construction module is also used to: according to the basic information of the enterprise M1, use natural number coding to convert the enterprise scale f1, enterprise type f2, economic type f3, affiliation f4, industry f5 and industry risk information f6 to form a basic information feature set of the enterprise .

[0137] The feature construction module is also used to select and construct the enterprise insurance behavior information feature set according to the enterprise insurance information M2 , the enterprise insurance behavior information feature set It includes the number of insured persons in the enterprise f7, the number of months of social security payment f8, the number of months of social security arrears f9, and the change rate of the number of social security payers f10. Among them, features f8, f9, and f10 are The calculation formula is as follows: in, represents the average insurance participation duration of employees in enterprise j, where i represents the i-th employee, ys represents the number of months that employee i has participated in insurance in enterprise j, and n represents the total number of insured employees in the enterprise;

[0138] in, It indicates the number of months in which enterprise j has social security arrears. If the month is marked as "arrears", it is added by 1. The final sum is the number of months in which enterprise j has social security arrears. Indicates the enterprise The number of months of social security arrears;

[0139] in, Indicates the enterprise The change rate of the number of social security contributors; represents the number of insured persons per month in enterprise j, t represents the month, and n represents The length of (the sequence of the number of insured persons per month of enterprise j), represents the i-th time, represents the average value at a time point, yes The value at time i, is the mean value of X.

[0140] The feature construction module is also used to: count the age structure proportion f11, gender structure proportion f12, and educational structure proportion f13 of the current insured persons in the enterprise according to the insured persons information M3 of the enterprise, and form a structural feature set of the insured persons in the enterprise .

[0141] The model training data generation module is also used to: use the standardized enterprise name as an important field for data association, and sequentially associate the enterprise basic information feature set , Enterprise insurance behavior information feature set , Structural characteristics of enterprise insured personnel And the enterprise historical labor dispute information feature set , generate the model training data Mt.

[0142] As an optional technical solution, the enterprise name standardization mapping module is also used to:

[0143] Extract the non-standard enterprise names in the enterprise labor dispute case information M4, remove the hyphens, left brackets, right brackets and characters that appear consecutively at the beginning of each non-standard enterprise name, and form a non-standard enterprise name array A1 after deleting duplicates; extract the standard enterprise names from the enterprise basic information M1, and form a standard enterprise name array A2 after deleting duplicates; then use Jieba word segmentation to perform text segmentation on arrays A1 and A2;

[0144] Combine the segmented arrays A1 and A2 to create an index dictionary D of the corpus features. Dictionary D assigns a number to each word and records the frequency of occurrence of the word.

[0145] Calculate the frequency of each word in the company name :

[0146]

[0147] in, Expressing words In the company name The number of times it appears in Indicates the company name The total number of words in

[0148] Calculating Inverse Document Frequency :

[0149]

[0150] in, Indicates the total number of company names, Indicates that it contains words Total number of company names;

[0151] S135. Convert the company name into a word frequency weighted sparse vector :

[0152]

[0153] in, For company name The word frequency weight sparse vector of Expressing words The word frequency weight value of .

[0154] Calculate the cosine similarity between vectors and construct a similarity matrix, i.e. the similarity between standard company names and non-standard company names :

[0155]

[0156] in, represents the standard enterprise name vector, represents a non-standard enterprise name vector, Representation vector With vector The cosine of the angle, represents the word frequency weight value of word i in A, Represents the word frequency weight value of word i in B;

[0157] Select the standard enterprise name that is most similar to the non-standard enterprise name to form a standardized mapping relationship between the enterprise names;

[0158] The non-standardized enterprise names in the enterprise labor dispute case information M4 are replaced according to the standardized mapping relationship.

[0159] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0160] Those skilled in the art can understand that the modules in the above system embodiments can be distributed in the system according to the description of the embodiments, or can be changed accordingly to be different from one or more systems in each embodiment. The modules in the above embodiments can be combined into one module, or can be further divided into multiple sub-modules (units).

[0161] The above description is only a specific implementation mode of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. All technical solutions that fall within the scope defined by the claims of the present invention fall within the protection scope of the present invention.

Claims

1. An automatic early warning method for enterprise risks based on a classification model, characterized in that: The following steps are involved: S1, generate model training data Mt: Mt=[ ]in, It is the basic information feature set of the enterprise. is the characteristic set of information on the insurance participation behavior of enterprises, is the structural characteristics set of the enterprise's insured personnel, It is a feature set of historical labor dispute information of an enterprise; S2, model training: Divide the model training data Mt into N equal sub-datasets, take one of the sub-datasets as the test set and the remaining N-1 sub-datasets as the training set, perform multiple model training evaluations, and calculate the average of the precision rates obtained from multiple trainings as the model score indicator; where 8≤N≤12; S3, model selection: select logistic regression, support vector machine, random forest, gradient boosting decision tree algorithm to perform model training according to the requirements of step S2, and generate scores for each algorithm model, and select the final prediction model according to the scores; S4. Use the selected final prediction model to perform risk prediction: obtain relevant data of the target enterprise, generate a feature set Mp of the target enterprise based on the relevant data, use the feature set Mp as the input of the final prediction model, perform enterprise risk prediction, and output the risk value.

2. The enterprise risk automatic early warning method based on the classification model according to claim 1 is characterized in that: Generating the model training data Mt in step S1 includes the following steps: S11, collecting enterprise-related information, the relevant information at least includes enterprise basic information M1, enterprise insurance behavior information M2, enterprise insurance personnel information M3 and enterprise labor dispute case information M4; S12: construct a corresponding data feature set based on the relevant information, wherein the data feature set includes an enterprise basic information feature set constructed based on the enterprise basic information M1. , the enterprise insurance behavior information feature set constructed based on the enterprise insurance behavior information M2 , a structural feature set of enterprise insured personnel constructed based on the enterprise insured personnel information M3 ; S13, using the bag-of-words model in combination with cosine similarity to calculate text similarity between the non-standardized enterprise names and the standardized enterprise names, performing standardized mapping on the non-standardized enterprise names, and replacing the non-standardized enterprise names in the enterprise labor dispute case information M4 according to the standardized mapping relationship; S14, group and merge the persons subject to execution in the replaced labor dispute case information M4, calculate the number of historical labor disputes of the enterprise f14, the total number of persons involved in the case f15, and the total amount involved in the case f16, and form a characteristic set of historical labor dispute information of the enterprise ; S15. Use the standardized enterprise name as an important field for data association, and associate it with the enterprise basic information feature set in turn. , Enterprise insurance behavior information feature set , Structural characteristics of enterprise insured personnel And the enterprise historical labor dispute information feature set , generate the model training data Mt.

3. The enterprise risk automatic early warning method based on the classification model according to claim 2 is characterized in that: The enterprise basic information feature set in step S12 The construction includes: according to the basic information of the enterprise M1, natural number coding is used to convert the enterprise scale f1, enterprise type f2, economic type f3, affiliation f4, industry f5 and industry risk information f6 to form the basic information feature set of the enterprise .

4. The enterprise risk automatic early warning method based on the classification model according to claim 2 is characterized in that: The enterprise insurance behavior information feature set in step S12 The construction includes: selecting and constructing the enterprise insurance behavior information feature set based on the enterprise insurance information M2 , the enterprise insurance behavior information feature set It includes the number of insured persons in the enterprise f7, the number of months of social security payment f8, the number of months of social security arrears f9, and the change rate of the number of social security payers f10. The calculation formula is as follows: in, Indicates the enterprise The change rate of the number of social security contributors, represents the number of insured persons per month in enterprise j, t represents the month, and n represents Length, represents the i-th time, represents the average value at a time point, yes The value at time i, is the mean value of X.

5. The enterprise risk automatic early warning method based on classification model according to claim 2 is characterized in that: The structural feature set of the enterprise insured personnel in step S12 The construction includes: according to the information of the insured personnel of the enterprise M3, the age structure proportion f11, gender structure proportion f12, and educational structure proportion f13 of the current insured personnel of the enterprise are counted to form the structural characteristic set of the insured personnel of the enterprise .

6. The enterprise risk automatic early warning method based on classification model according to claim 2 is characterized in that: The step S13 specifically includes: S131, extract the non-standard enterprise names in the enterprise labor dispute case information M4, remove the hyphens, left brackets, right brackets and characters that appear continuously at the beginning of each non-standard enterprise name, and form a non-standard enterprise name array A1 after deduplication; extract the standard enterprise names from the enterprise basic information M1, and form a standard enterprise name array A2 after deduplication; then use Jieba word segmentation to perform text segmentation on arrays A1 and A2; S132, combining the segmented arrays A1 and A2, and establishing an index dictionary D of the corpus features. The dictionary D assigns a number to each word and records the frequency information of the word; S133. Calculate the frequency of each word in the company name : in, Expressing words In the company name The number of times it appears in Indicates the company name The total number of words in S134. Calculate inverse document frequency : in, Indicates the total number of company names, Indicates that it contains words Total number of company names; S135. Convert the company name into a word frequency weighted sparse vector : in, For company name The word frequency weight sparse vector of Expressing words The word frequency weight value of ; S136, calculate the cosine similarity between vectors and construct a similarity matrix, that is, the similarity between standard enterprise names and non-standard enterprise names : in, represents the standard enterprise name vector, represents a non-standard enterprise name vector, Representation vector With vector The cosine of the angle, represents the word frequency weight value of word i in A, Represents the word frequency weight value of word i in B; S137, selecting a standard enterprise name that is most similar to the non-standard enterprise name, and forming a standardized mapping relationship of the enterprise names; S138. Replace the non-standardized enterprise name in the enterprise labor dispute case information M4 according to the standardized mapping relationship.

7. The enterprise risk automatic early warning method based on classification model according to claim 1 is characterized in that: The step S3 selects the final prediction model according to the scores, including: if the score differences between the algorithm models are greater than 0.1, the model with the highest model score is used as the final prediction model; if the score differences between the algorithm models are less than 0.1, the average of the prediction values ​​of the algorithm models is used as the final output result of the model.

8. The enterprise risk automatic early warning system based on the classification model is characterized by: include: Model training data generation module, used to generate model training data Mt: Mt=[ ]in, It is the basic information feature set of the enterprise. is the characteristic set of information on the insurance participation behavior of enterprises, is the structural characteristics set of the enterprise's insured personnel, It is a feature set of historical labor dispute information of an enterprise; The model training module is used to divide the model training data Mt into N sub-datasets of equal number, and use one of the sub-datasets as the test set and the remaining N-1 sub-datasets as the training set in turn, perform multiple model training evaluations, and calculate the average of the precision rates obtained from multiple trainings as the model score indicator; wherein 8≤N≤12; The model selection module is used to select logistic regression, support vector machine, random forest, and gradient boosting decision tree algorithms to perform model training according to the requirements of the model training module, generate scores for each algorithm model, and select the final prediction model based on the scores; The prediction module is used to perform risk prediction using the screened final prediction model, including: obtaining relevant data of the target enterprise, generating a feature set Mp of the target enterprise based on the relevant data, using the feature set Mp as input of the final prediction model, performing enterprise risk prediction, and outputting a risk value.

9. The enterprise risk automatic early warning system based on classification model according to claim 8 is characterized in that: The model training data generation module includes: A data collection module is used to collect enterprise-related information, which includes at least enterprise basic information M1, enterprise insurance behavior information M2, enterprise insurance personnel information M3, and enterprise labor dispute case information M4; A feature construction module is used to construct a corresponding data feature set based on the relevant information, and the data feature set includes an enterprise basic information feature set constructed based on the enterprise basic information M1. , the enterprise insurance behavior information feature set constructed based on the enterprise insurance behavior information M2 , a structural feature set of enterprise insured personnel constructed based on the enterprise insured personnel information M3 ; A standardized enterprise name mapping module is used to calculate the text similarity between non-standardized enterprise names and standardized enterprise names by using a bag-of-words model combined with cosine similarity, perform standardized mapping on the non-standardized enterprise names, and replace the non-standardized enterprise names in the enterprise labor dispute case information M4 according to the standardized mapping relationship; The feature building module is also used to: group and merge the persons subject to execution in the replaced labor dispute case information M4, calculate the number of historical labor disputes f14, the total number of persons involved in the case f15, and the total amount involved in the case f16, and form a feature set of the historical labor dispute information of the enterprise. ; The feature construction module is also used to: according to the basic information of the enterprise M1, use natural number coding to convert the enterprise scale f1, enterprise type f2, economic type f3, affiliation f4, industry f5 and industry risk information f6 to form a basic information feature set of the enterprise ; The feature construction module is also used to select and construct the enterprise insurance behavior information feature set according to the enterprise insurance information M2 , the enterprise insurance behavior information feature set It includes the number of insured persons in the enterprise f7, the number of months of social security payment f8, the number of months of social security arrears f9, and the change rate of the number of social security payers f10. The calculation formula is as follows: in, Indicates the enterprise The change rate of the number of social security contributors, represents the number of insured persons per month in enterprise j, t represents the month, and n represents Length, represents the i-th time, represents the average value at a time point, yes The value at time i, is the mean value of X; The feature construction module is also used to: count the age structure proportion f11, gender structure proportion f12, and educational structure proportion f13 of the current insured persons in the enterprise according to the insured persons information M3 of the enterprise, and form a structural feature set of the insured persons in the enterprise ; The model training data generation module is also used to: use the standardized enterprise name as an important field for data association, and sequentially associate the enterprise basic information feature set , Enterprise insurance behavior information feature set , Structural characteristics of enterprise insured personnel And the enterprise historical labor dispute information feature set , generate the model training data Mt.

10. The enterprise risk automatic early warning system based on classification model according to claim 9 is characterized in that: The enterprise name standardization mapping module is also used for: Extract the non-standard enterprise names in the enterprise labor dispute case information M4, remove the hyphens, left brackets, right brackets and characters that appear consecutively at the beginning of each non-standard enterprise name, and form a non-standard enterprise name array A1 after deleting duplicates; extract the standard enterprise names from the enterprise basic information M1, and form a standard enterprise name array A2 after deleting duplicates; then use Jieba word segmentation to perform text segmentation on arrays A1 and A2; Combine the segmented arrays A1 and A2 to create an index dictionary D of the corpus features. Dictionary D assigns a number to each word and records the frequency of occurrence of the word. Calculate the frequency of each word in the company name : in, Expressing words In the company name The number of times it appears in Indicates the company name The total number of words in Calculating Inverse Document Frequency : in, Indicates the total number of company names, Indicates that it contains words Total number of company names; Convert company names into word frequency weighted sparse vectors : in, For company name The word frequency weight sparse vector of Expressing words The word frequency weight value of ; Calculate the cosine similarity between vectors and construct a similarity matrix, i.e. the similarity between standard company names and non-standard company names : in, represents the standard enterprise name vector, represents a non-standard enterprise name vector, Representation vector With vector The cosine of the angle, represents the word frequency weight value of word i in A, Represents the word frequency weight value of word i in B; Select the standard enterprise name that is most similar to the non-standard enterprise name to form a standardized mapping relationship between the enterprise names; The non-standardized enterprise names in the enterprise labor dispute case information M4 are replaced according to the standardized mapping relationship.

Images