Data intelligent sharing method for intelligent sharing system construction

Abstract

The invention relates to the technical field of data classification, in particular to an intelligent data sharing method for construction of an intelligent sharing system. According to the method, each condition attribute is used for splitting total historical service data to obtain different sub-data sets, and the index characteristics of the index attributes in the total historical service data and the sub-data sets of each condition attribute are analyzed according to the difference of confusion degrees of the index characteristics of the index attributes in the sub-data sets of the condition attributes. Obtaining an initial priority index of each condition attribute according to the change trend of the historical business data of each condition attribute and the change trend of the historical business data of the attention attribute, and adjusting the initial priority index in combination with the influence degree of each condition attribute on the attention attribute and the correlation degree of each condition attribute and other condition attributes so as to construct a decision tree; and classifying and sharing the total historical business data. According to the method, the decision tree is constructed by combining the information gain and the importance degree brought by the condition attributes during data set segmentation and the correlation degree of the different condition attributes, and the classification effect of the decision tree based on the attention attributes is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data classification, and in particular to a data intelligent sharing method for building a smart sharing system. Background Art

[0002] The wave of informatization sweeping the world has brought vitality to the construction of data sharing systems. The construction of various shared databases has improved the scientificity, accuracy and efficiency of data analysis. As an important means of data-driven economic and social development, data sharing is of great significance for promoting innovation, improving efficiency and achieving sustainable development.

[0003] Some companies store directly statistical data in databases or data platforms to share company-related data, but the directly statistical data have not been fully mined and utilized, and thus cannot effectively support the analysis and decision-making of the company's business. In order to fully mine the valuable information in the company's related data, the attributes of the company's related data are usually extracted, and different attributes are used to make decisions on the company's related data to build a decision tree and classify the related data. Existing methods determine the priority or priority of each attribute in building a decision tree by the information gain brought by each attribute when splitting the data set; however, there are often complex associations between different attributes in the company's related data, and the degree of association between different attributes and important information of concern varies. For example, insurance companies pay more attention to profits, and the expense rate and loss rate jointly affect profitability. Even if the premium income is high, if the expense rate and loss rate both rise at the same time, it may also lead to a decline in net profit. At the same time, there are differences in the degree of association between insurance income and loss rate and profit, resulting in poor classification effect of the constructed decision tree based on profit information, and low accuracy in explaining the profitability of insurance companies. Summary of the invention

[0004] In order to solve the technical problem that the priority or priority degree of each attribute in participating in building a decision tree is determined based only on the information gain brought by each attribute when splitting the data set, resulting in poor effect of the decision tree on data classification, the purpose of the present invention is to provide a data intelligent sharing method for the construction of an intelligent sharing system, and the technical scheme adopted is as follows:

[0005] The present invention proposes a data intelligent sharing method for building a smart sharing system, the method comprising:

[0006] Obtaining total historical business data, the total historical business data includes historical business data of several years, extracting business features from the total historical business data, and determining indicator attributes, focus attributes, and several conditional attributes in the business features;

[0007] The total historical business data is split by each condition attribute to obtain different sub-data sets for each condition attribute; the initial priority indicator of each condition attribute is obtained according to the difference in the degree of confusion of the indicator characteristics of the indicator attributes in the total historical business data and the sub-data sets of each condition attribute, and the similarity of the change trend of each condition attribute and the historical business data of the attribute of interest;

[0008] According to the influence of each condition attribute on the focus attribute and the correlation between each condition attribute and other condition attributes, the initial priority index is adjusted, and a decision tree is constructed using the adjusted initial priority index to classify and share the total historical business data.

[0009] Furthermore, the obtaining of the initial priority index of each condition attribute includes:

[0010] According to the difference in the degree of confusion of the indicator characteristics of the indicator attributes in the total historical business data and the sub-dataset of each condition attribute, the information gain of each condition attribute is obtained;

[0011] According to the similarity between the change trend of the historical business data of each condition attribute and the concerned attribute, the synergy index between each condition attribute and the concerned attribute is obtained;

[0012] According to the information gain and the coordination index, an initial priority index of each condition attribute is obtained; the information gain and the coordination index are both positively correlated with the initial priority index.

[0013] Furthermore, obtaining the information gain of each condition attribute includes:

[0014] Obtain the information entropy of the indicator characteristics of the indicator attributes in the total historical business data, recorded as the total information entropy;

[0015] Obtain the information entropy of the indicator feature of the indicator attribute in the historical business data in each sub-dataset of each conditional attribute, and record it as the sub-information entropy of the corresponding sub-dataset; take the proportion of the annual quantity in the historical business data in each sub-dataset of each conditional attribute to the annual quantity in the total historical business data as the quantity indicator of each sub-dataset of each conditional attribute;

[0016] Obtaining an overall uncertainty index for each conditional attribute according to the sub-information entropy and the quantity index of all sub-data sets of each conditional attribute;

[0017] The difference between the total information entropy and the overall uncertainty index is normalized to obtain the information gain of each conditional attribute.

[0018] Furthermore, obtaining the synergy index between each condition attribute and the focus attribute includes:

[0019] Record the condition attribute and the focus attribute as analysis attributes, arrange the historical business data of each analysis attribute in order, and obtain the business data sequence of each analysis attribute;

[0020] Obtaining a second-order difference sequence of the business data sequence of each analysis attribute; calculating the mean of the absolute values ​​of the differences between the elements with the same subscript in the second-order difference sequence of each condition attribute and the focus attribute as a change difference indicator between each condition attribute and the focus attribute;

[0021] Obtain the correlation coefficient of the business data sequence between each condition attribute and the attribute of interest, and obtain the synergy index of each condition attribute and the attribute of interest based on the correlation coefficient and the change difference index, wherein the correlation coefficient and the synergy index are positively correlated, and the change difference index and the synergy index are negatively correlated.

[0022] Furthermore, the initial priority index is adjusted according to the influence of each condition attribute on the concerned attribute and the correlation between each condition attribute and other condition attributes, and a decision tree is constructed using the adjusted initial priority index to classify and share the total historical business data, including:

[0023] According to the influence degree of each condition attribute on the concerned attribute and the correlation degree of each condition attribute with other condition attributes, the initial priority index is adjusted to obtain the final priority index of each condition attribute;

[0024] Arrange the final priority indexes of all condition attributes from large to small to obtain an initial priority sequence; select the first several condition attributes from the initial priority sequence to form a final priority sequence;

[0025] According to the order of the condition attributes in the final priority sequence, a decision tree is constructed using the Hunt algorithm to obtain the classification results of the total historical business data, and the obtained classification results are stored in a database.

[0026] Furthermore, the obtaining of the final priority index of each condition attribute includes:

[0027] Obtain the impact index of each conditional attribute on the attribute of interest;

[0028] Obtaining the correlation coefficients of each condition attribute and the business data sequences of other condition attributes respectively, and taking the average of the correlation coefficients of each condition attribute and the business data sequences of other condition attributes as the overall correlation value of each condition attribute and the other condition attributes;

[0029] Performing negative correlation mapping on the overall correlation value, normalizing the product of the mapping result and the impact index of each condition attribute to obtain a priority construction coefficient of each condition attribute;

[0030] The initial priority index is weighted using the priority construction coefficient to obtain a final priority index for each condition attribute.

[0031] Furthermore, obtaining the influence index of each condition attribute on the concerned attribute includes:

[0032] Perform linear analysis on all conditional attributes and attention attributes to construct an initial attention function, which contains several regression coefficients;

[0033] Substituting the historical business data of all conditional attributes and the historical business data of the attention attribute of each year into the initial attention function to obtain the attention fitting value of each year;

[0034] The error function is obtained based on the difference between the historical business data of the attribute of interest in all years and the fitted value of interest; the value of the regression coefficient corresponding to the minimum value of the error function is used as the optimal value of each regression coefficient in the initial function of interest, and the optimal value of the regression coefficient corresponding to each conditional attribute is used as the influence indicator of each conditional attribute.

[0035] Furthermore, the total historical business data is split by each condition attribute to obtain different sub-data sets for each condition attribute, including:

[0036] The product of the maximum value in the historical business data of each condition attribute and the preset weight threshold is used as the standard segmentation threshold of each condition attribute;

[0037] Based on the standard segmentation threshold, the historical business data of each condition attribute is divided into two categories; the set consisting of the historical business data of the year to which each data in each category of historical business data of each condition attribute belongs is used as the sub-data set of each condition attribute.

[0038] Furthermore, the obtaining of the overall uncertainty index of each conditional attribute according to the sub-information entropy and the quantity index of all sub-data sets of each conditional attribute includes:

[0039] The product of the sub-information entropy and the quantity index of each sub-data set of each conditional attribute is used as the local uncertainty index of the corresponding sub-data set; the cumulative sum of the local uncertainty indexes of all sub-data sets of each conditional attribute is used as the overall uncertainty index of each conditional attribute.

[0040] Furthermore, the correlation coefficient is a Pearson correlation coefficient.

[0041] The present invention has the following beneficial effects:

[0042] In an embodiment of the present invention, the difference in the degree of confusion of the indicator characteristics of the indicator attributes in the total historical business data and the sub-data set of each conditional attribute presents the information gain after the total historical business data is divided by each conditional attribute. The similarity of the change trend of each conditional attribute and the historical business data of the focus attribute is used to analyze the influence of each conditional attribute on the focus information. The initial priority index is obtained by combining the two factors to improve the accuracy of the priority information of each conditional attribute for decision analysis. Since the correlation between different attributes in the business data will affect the priority of the conditional attribute in the process of building a decision tree, in order to build a better decision tree, it is necessary to consider the correlation between each conditional attribute and other conditional attributes, and adjust the initial priority index in combination with the influence of each conditional attribute on the focus attribute. The decision tree is constructed using the adjusted initial priority index. When the focus attribute, i.e., the focus information, is considered, the classification effect of the decision tree on the total historical business data is improved, and the accuracy of the interpretation of the focus information is improved, effectively solving the problem that the decision tree constructed only based on information gain has poor classification effect based on the focus information. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0044] Figure 1 A flowchart of a method for intelligent data sharing for building an intelligent sharing system provided by an embodiment of the present invention;

[0045] Figure 2 A flowchart of a method for obtaining an initial priority index provided by an embodiment of the present invention;

[0046] Figure 3 A system structure diagram of a data intelligent sharing system for building an intelligent sharing system provided by one embodiment of the present invention;

[0047] Figure 4 A schematic diagram of a computer device for a data intelligent sharing device for building a smart sharing system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0048] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the specific implementation method, structure, characteristics and effects of a data intelligent sharing method for the construction of a smart sharing system proposed by the present invention in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0049] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0050] The following is a detailed description of a specific scheme of a data intelligent sharing method for building a smart sharing system provided by the present invention in conjunction with the accompanying drawings.

[0051] Embodiment 1:

[0052] This invention proposes a data intelligent sharing method for building a smart sharing system. Figure 1 , which shows a flowchart of the steps of a data intelligent sharing method for building a smart sharing system provided by an embodiment of the present invention, the method comprising:

[0053] Step S1: Obtain total historical business data, which includes historical business data of several years, extract business features from the total historical business data, and determine indicator attributes, focus attributes and several conditional attributes in the business features.

[0054] This plan takes the financial data of a property insurance company as an example for analysis. The total historical business data is the financial data of different attributes of a property insurance company in several years. The total historical business data can be obtained from the corresponding business database, which is a pre-established database specifically for storing business data.

[0055] Business features are extracted from business data, including insurance income, claims ratio, expense ratio, net profit, auto insurance claims ratio, and profitability. Among them, the claims ratio refers to the ratio of the amount of claims paid by a property insurance company to the premium income, and the expense ratio refers to the ratio of various expenses spent by a property insurance company for operation and sales of insurance to the premium income.

[0056] In order to mine high-value data from financial data, this solution needs to build a suitable decision tree to analyze which attributes mainly affect the profitability of a property insurance company and to find out the profit situation of a property insurance company in general.

[0057] The indicator attribute is a decision attribute, which is determined based on the condition attribute. This solution analyzes which attributes mainly affect the profitability of a property insurance company, and the company's profitability can be measured by net profit. In this embodiment, the indicator attribute is whether it is profitable, the focus attribute is net profit, and the condition attributes include: insurance income, claim ratio, expense ratio and auto insurance claim ratio.

[0058] Different conditional attributes have different impacts on a company's profitability. The conditional attributes need to make decisions on the total historical business data in an appropriate order to maximize the purity of the decision category, thereby more accurately explaining the company's profitability and obtaining high-value financial data.

[0059] As an example, the business data table composed of total historical business data is shown in Table 1:

[0060] Table 1, Business data table

[0061]

[0062]

[0063] It should be noted that all data in Table 1 are collectively referred to as total historical business data; the historical business data of each attribute is all data in the column where the attribute is located in Table 1. As an example, the historical business data of insurance income includes: 3800, 6500, 4500, 1800, 2500, 3200, 1200 and 1600; the historical business data of each year is the data in the same row in Table 1.

[0064] The dimensions and magnitudes of the data of different attributes in the total historical business data are different. Directly using the historical business data to participate in the calculation may highlight some indicators with larger magnitudes and exclude some indicators with smaller magnitudes. Therefore, it is necessary to normalize each business data in the total historical business data to obtain normalized business data. In the embodiment of the present invention, the maximum and minimum normalization is used to normalize each business data. Normalization methods such as Norm function and function transformation can also be selected, which are not limited here. The subsequent steps all use normalized business data for analysis.

[0065] Step S2: Use each condition attribute to split the total historical business data separately to obtain different sub-datasets for each condition attribute; obtain the initial priority indicator of each condition attribute based on the difference in the degree of confusion of the indicator characteristics of the indicator attributes in the total historical business data and the sub-datasets of each condition attribute, as well as the similarity between the change trends of each condition attribute and the historical business data of the attribute of interest.

[0066] In order to simplify the analysis of information gain of different conditional attributes, in this embodiment, binary splitting is used for decision splitting using each conditional attribute, that is, two sub-datasets are obtained after the total historical business data is split based on each conditional attribute.

[0067] Preferably, in some possible implementations of the embodiments of the present invention, the method for obtaining a sub-dataset of each conditional attribute includes: taking the product of the maximum value in the historical business data of each conditional attribute and a preset weight threshold as the standard segmentation threshold of each conditional attribute; dividing the historical business data of each conditional attribute into two categories based on the standard segmentation threshold; and taking a set consisting of the historical business data of the year to which each data in each category of historical business data of each conditional attribute belongs as the sub-dataset of each conditional attribute.

[0068] It should be noted that, for the historical business data of each conditional attribute, the historical business data greater than the standard segmentation threshold is regarded as the first category of historical business data for each conditional attribute, and the historical business data less than or equal to the standard segmentation threshold is regarded as the second category of historical business data. In the embodiment of the present invention, the preset weight threshold takes the empirical value of 0.3, and the implementer can set it according to the specific situation. In Table 1, the data in each row of historical business data is the historical business data of the year to which the historical business data belongs.

[0069] The difference in the degree of confusion of the indicator characteristics of the total historical business data and the sub-dataset of each conditional attribute presents the information gain after the total historical business data is divided by each conditional attribute; the similarity between the changing trend of each conditional attribute and the historical business data of the focus attribute reflects the degree of synergy of the conditional attribute on the focus attribute, i.e., net profit, which facilitates the analysis of the impact of the conditional attribute on the company's profitability; the above two factors are comprehensively analyzed to obtain the initial priority indicators, thereby improving the accuracy of the priority information for decision-making analysis of each conditional attribute.

[0070] See also Figure 2 , which shows a flowchart of a method for obtaining an initial priority index provided by an embodiment of the present invention, the method comprising:

[0071] Step S210: Obtain the information gain of each condition attribute according to the difference in the degree of confusion of the indicator characteristics of the indicator attribute in the total historical business data and the sub-data set of each condition attribute.

[0072] The degree of confusion of the indicator characteristics of the indicator attributes in the total historical business data reflects the uncertainty before the total historical business data is split. The degree of confusion of the indicator characteristics of the indicator attributes in the sub-dataset of each conditional attribute presents the uncertainty after the total historical business data is split using each conditional attribute. The difference between the two represents the degree of uncertainty reduction, and the information gain of each conditional attribute is obtained.

[0073] Preferably, in some possible implementation modes of the embodiments of the present invention, the method for obtaining information gain includes: obtaining the information entropy of the indicator characteristics of the indicator attributes in the total historical business data, recorded as the total information entropy; obtaining the information entropy of the indicator characteristics of the indicator attributes in the historical business data in each sub-data set of each condition attribute, recorded as the sub-information entropy of the corresponding sub-data set; taking the proportion of the annual quantity in the historical business data in each sub-data set of each condition attribute to the annual quantity in the total historical business data as the quantity index of each sub-data set of each condition attribute; obtaining the overall uncertainty index of each condition attribute based on the sub-information entropy and quantity index of all sub-data sets of each condition attribute; normalizing the difference between the total information entropy and the overall uncertainty index to obtain the information gain of each condition attribute.

[0074] In this embodiment, the indicator attribute is whether it is profitable, and whether it is profitable includes two indicator features: yes and no. The probability of the indicator feature "yes" of the indicator attribute in the historical business data of the indicator attribute is The probability of the “No” indicator feature in the historical business data of the indicator attribute is The total information entropy is determined using the above two probabilities. The total information entropy represents the uncertainty of the total historical business data when no decision analysis is performed. The method for calculating the information entropy is a well-known technique for those skilled in the art and will not be described in detail here.

[0075] The quantity index reflects the amount of data in the sub-dataset of each conditional attribute; if the sub-information entropy and quantity index of the sub-dataset of each conditional attribute are both larger, it means that the information contained in the data set is richer and more complex, and the uncertainty of the sub-dataset is greater. In an embodiment of the present invention, the method for obtaining the overall uncertainty index is: the product of the sub-information entropy and the quantity index of each sub-dataset of each conditional attribute is used as the local uncertainty index of the corresponding sub-dataset; the cumulative sum of the local uncertainty indexes of all sub-datasets of each conditional attribute is used as the overall uncertainty index of each conditional attribute. The larger the overall uncertainty index, the greater the overall level of uncertainty after splitting the total historical business data using each conditional attribute.

[0076] In a specific implementation of the embodiment of the present invention, the information gain of each conditional attribute is expressed by the formula:

[0077]

[0078] Where K is the information gain of each conditional attribute; G is the total information entropy; H is the total number of sub-datasets of each conditional attribute. Since binary splitting is adopted in this embodiment, H takes the empirical value of 2; I h is the quantity index of the h-th sub-dataset for each conditional attribute; Jh is the sub-information entropy of the h-th sub-data set for each conditional attribute; I h ×J h is the local uncertainty index of the h-th sub-dataset of each conditional attribute; is the overall uncertainty index of each conditional attribute; Norm is the normalization function.

[0079] It should be noted that the larger the information gain K of each conditional attribute is, the higher the homogeneity of the total historical business data divided by the conditional attribute is. After the total historical business data is split using the conditional attribute, the more obvious the reduction in uncertainty is, and the more suitable the conditional attribute is for decision analysis.

[0080] Step S220: Obtain a synergy index between each condition attribute and the focused attribute according to the similarity between the change trends of the historical business data of each condition attribute and the focused attribute.

[0081] Preferably, in some possible implementation modes of the embodiments of the present invention, the method for obtaining the collaborative index includes: recording the conditional attributes and the focus attributes as analysis attributes, arranging the historical business data of each analysis attribute in sequence, and obtaining the business data sequence of each analysis attribute; obtaining the second-order difference sequence of the business data sequence of each analysis attribute; calculating the mean of the absolute values ​​of the differences between the elements with the same subscript in the second-order difference sequence of each conditional attribute and the focus attribute as the change difference index of each conditional attribute and the focus attribute; obtaining the correlation coefficient of the business data sequence of each conditional attribute and the focus attribute, and obtaining the collaborative index of each conditional attribute and the focus attribute based on the correlation coefficient and the change difference index, the correlation coefficient and the collaborative index are positively correlated, and the change difference index and the collaborative index are negatively correlated.

[0082] The historical business data of each analysis attribute refers to the data in the column where the analysis attribute is located in Table 1, including the business data of each analysis attribute in all years. Therefore, the embodiment of the present invention arranges the historical business data of each analysis attribute in chronological order to obtain a business data sequence of the corresponding analysis attribute.

[0083] The second-order difference sequence of the business data sequence reflects the changes in the historical business data of each conditional attribute, and the change difference index shows the similarity between the change trend of the historical business data of each conditional attribute and the change trend of the historical business data of the focus attribute. If the change difference index is smaller, the change trend of the historical business data of each conditional attribute is more similar to the change trend of the historical business data of the focus attribute, and the degree of coordination between each conditional attribute and the historical business data of the focus attribute, i.e., net profit, is greater. When the correlation coefficient of the business data sequence of each conditional attribute and the focus attribute is larger, it means that the net profit is more affected by the historical business data of each conditional attribute, and the degree of coordination of each conditional attribute on the net profit is greater. Therefore, the change difference index is negatively correlated with the coordination index, and the correlation coefficient is positively correlated with the coordination index.

[0084] In the embodiment of the present invention, each condition attribute is negatively correlated with the change difference index of the attribute of interest, and the product of the mapping result and the correlation coefficient of the business data sequence of each condition attribute and the attribute of interest is normalized to obtain the synergy index of each condition attribute and the attribute of interest. In the embodiment of the present invention, the correlation coefficient of the business data sequence of each condition attribute and the attribute of interest and the correlation relationship between the change difference index and the synergy index can also be constructed through other basic mathematical operations, which will not be limited or elaborated here.

[0085] It should be noted that in the embodiment of the present invention, an exponential function with a natural constant as the base is used for negative correlation mapping, specifically: first take the opposite number of the correlation coefficient of the business data sequence of each conditional attribute and the attribute of interest, and use the opposite number as the exponent of the exponential function with a natural constant as the base to achieve negative correlation mapping of the correlation coefficient; select the Norm function for normalization, and other negative correlation mapping and normalization methods can also be selected, which are not limited here. The correlation coefficient in the embodiment of the present invention uses the Pearson correlation coefficient. In other embodiments of the present invention, the correlation coefficient can also be the Spearman correlation coefficient, the Kendall rank correlation coefficient, and the typical correlation coefficient. Among them, the method for obtaining the Pearson correlation coefficient is a well-known technology for those skilled in the art and will not be repeated here.

[0086] Step S230: According to the information gain and the coordination index, an initial priority index of each condition attribute is obtained; the information gain and the coordination index are both positively correlated with the initial priority index.

[0087] If the information gain of each conditional attribute is larger, it means that after the total historical business data is divided using the conditional attribute, the uncertainty of the total historical business data is reduced more obviously, thereby making the divided subsets purer, and the priority of using the conditional attribute for decision-making in the process of building the decision tree is higher. If the synergy index of each conditional attribute and the attribute of interest is larger, it means that the contribution of the conditional attribute in the total historical business data to the profitability of the insurance company is greater, and the more suitable it is for decision-making, the higher the priority of the conditional attribute when building the decision tree. Therefore, the information gain and the synergy index are both positively correlated with the initial priority index. In an embodiment of the present invention, the product of the synergy index of each conditional attribute and the attribute of interest and the information gain is normalized to obtain the initial priority index of each conditional attribute. If the initial priority index of each conditional attribute is larger, the priority of using the conditional attribute for decision-making in the process of building the decision tree is higher.

[0088] In the embodiment of the present invention, the correlation between the collaborative index and the information gain and the initial priority index may also be constructed through other basic mathematical operations, which will not be limited or elaborated herein.

[0089] It should be noted that in the embodiment of the present invention, the Sigmoid function is used to normalize the product of the collaboration index and the information gain. In the embodiment of the present invention, other normalization methods may also be selected, such as function transformation, maximum and minimum normalization, etc., which are not limited here.

[0090] Step S3: According to the influence of each condition attribute on the focus attribute and the correlation between each condition attribute and other condition attributes, the initial priority index is adjusted, and a decision tree is constructed using the adjusted initial priority index to classify and share the total historical business data.

[0091] Since there are often complex associations between different attributes in financial data, for example, the expense ratio and the loss ratio jointly affect profitability. Even if the premium income is high, if both the expense ratio and the loss ratio increase at the same time, the net profit may also decrease. The degree of association between the attributes will affect their priority in the process of building a decision tree, which will lead to deviations in the mined high-value data. In order to build a better decision tree and mine high-value data, it is necessary to consider the correlation between each condition attribute and other condition attributes, and adjust the initial priority index based on the influence of each condition attribute on the attribute of interest, so as to obtain the final priority index of each condition attribute, so that the priority of the condition attribute in the process of building a decision tree is more reasonable.

[0092] Preferably, in some possible implementation modes of the embodiments of the present invention, the method for obtaining the final priority index includes: obtaining the influence index of each condition attribute on the focus attribute; obtaining the correlation coefficient of each condition attribute with the business data sequence of other condition attributes respectively, and taking the mean of the correlation coefficient of each condition attribute with the business data sequence of other condition attributes as the overall correlation value of each condition attribute with the other condition attributes; performing negative correlation mapping on the overall correlation value, normalizing the product of the mapping result and the influence index of each condition attribute to obtain the priority construction coefficient of each condition attribute; and using the priority construction coefficient to weight the initial priority index to obtain the final priority index of each condition attribute.

[0093] This embodiment uses a multivariate regression function to analyze the influence of each condition attribute on the focus attribute to obtain an influence index. The steps for obtaining the influence index are as follows:

[0094] Perform linear analysis on all conditional attributes and attention attributes to construct an initial attention function, which includes several regression coefficients. In a specific implementation of the embodiment of the present invention, the initial attention function is expressed by a formula:

[0095]

[0096] Where R is the historical business data of the attribute of interest; A is the total number of conditional attributes; N a is the historical business data of the ath condition attribute; is the regression coefficient of the ath conditional attribute, are regression coefficients, all of which are unknown quantities.

[0097] It should be noted that since the conditional attributes in this embodiment include insurance income, claims ratio, expense ratio and auto insurance claims ratio, there are a total of 4 conditional attributes and the number of regression coefficients is 5. In order to determine the specific value of the regression coefficient, it is necessary to ensure that the number of years in the total historical business data is much greater than 5, that is, more than 5 years of historical business data of conditional attributes are required to participate in constructing the error function in order to obtain the specific value of the regression coefficient.

[0098] In some possible implementations of the embodiments of the present invention, the method for obtaining the error function includes: substituting the historical business data of all conditional attributes and the historical business data of the attention attribute of each year into the initial attention function to obtain the attention fitting value of each year; and obtaining the error function according to the difference between the historical business data of the attention attribute of all years and the attention fitting value. In a specific implementation of the embodiments of the present invention, the error function RSS is expressed by the formula:

[0099]

[0100] In the formula, U is the total number of years in the total historical business data, R u is the historical business data of the focus attribute of the u-th year in the total historical business data, that is, the net profit of the u-th year; A is the total number of conditional attributes; N u.a is the historical business data of the ath conditional attribute in the uth year in the total historical business data; is the regression coefficient of the ath conditional attribute, is the regression coefficient; The fitted values ​​for each year of interest.

[0101] The value of the regression coefficient corresponding to the minimum value of the error function is taken as the optimal value of each regression coefficient in the initial attention function, and the optimal value of the regression coefficient corresponding to each conditional attribute is taken as the influence indicator of each conditional attribute.

[0102] It should be noted that the larger the optimal value of the regression coefficient of each conditional attribute, that is, the larger the influence index, the greater the influence of the conditional attribute on the attribute of interest, that is, net profit, the more important the conditional attribute is, and the higher its priority in the process of building a decision tree.

[0103] In order to build a better decision tree, the conditional attribute with a smaller correlation with other conditional attributes may provide unique information when segmenting the data set, and improve the division effect of the decision tree to a certain extent. The overall correlation value reflects the overall correlation level between each conditional attribute and other conditional attributes; if the overall correlation value of each conditional attribute with other conditional attributes is smaller, the overall correlation between each conditional attribute and other conditional attributes is weaker, the more suitable it is for decision making, and the higher the priority of this attribute condition in the process of building a decision tree. If the influence index of each conditional attribute is larger, it means that the influence of this conditional attribute on the attribute of interest, that is, net profit, is greater, the conditional attribute is more important, and the priority of this conditional attribute in the process of building a decision tree is higher. Therefore, the overall correlation value is negatively correlated with the priority construction coefficient, and the influence index is positively correlated with the priority construction coefficient.

[0104] The priority construction coefficient is used to adjust the initial priority index of each conditional attribute. The specific method is: use the priority construction coefficient to weight the initial priority index; ensure that the conditional attribute with the greater information gain for net profit, the greater the degree of synergy with net profit and the smaller the degree of correlation with other conditional attributes has a higher priority in constructing the decision tree.

[0105] In a specific implementation of the embodiment of the present invention, the final priority index is expressed by the formula:

[0106]

[0107] In the formula, U is the final priority index of each conditional attribute; L is the initial priority index of each conditional attribute; M is the impact index of each conditional attribute on the attribute of interest; A is the total number of conditional attributes; T a is the correlation coefficient between each condition attribute and the business data sequence of the ath condition attribute other than the condition attribute; is the overall correlation value of each condition attribute with other condition attributes; The priority construction coefficient for each conditional attribute; exp is an exponential function with a natural constant as the base; Norm is a normalization function.

[0108] Arrange the final priority indicators of all condition attributes from large to small to obtain an initial priority sequence; select the first several condition attributes from the initial priority sequence to form a final priority sequence; use the Hunt algorithm to construct a decision tree according to the order of the condition attributes in the final priority sequence to obtain the classification result of the total historical business data, and store the obtained classification result in a database. Among them, the method of constructing a decision tree by the Hunt algorithm is a well-known technology and will not be repeated here; the database used in this embodiment is a Structured Query Language (SQL) database.

[0109] Since financial data involves attributes in multiple dimensions, too many attributes may make the decision tree model complex. For example, net profit and revenue growth rate have a strong correlation. These attribute characteristics will increase the complexity of the model and lead to overfitting of the model. In order to reduce the structural redundancy of the decision tree and make the model easier to explain, several conditional attributes with larger final priority indicators are selected to construct the decision tree.

[0110] In this embodiment, the method for selecting condition attributes in the final priority sequence is: normalize the final priority index of each condition attribute to obtain the normalized priority index of the corresponding condition attribute; for the condition attributes in the initial priority sequence, arrange the condition attributes with normalized priority indexes greater than the preset priority threshold in order to obtain the final priority sequence. It should be noted that the preset priority threshold in this embodiment is an empirical value of 0.5, and the implementer can set it according to the specific situation.

[0111] It should be noted that the conditional attributes in the final priority sequence are also arranged from large to small according to the final priority index; the conditional attribute with a larger final priority index makes decisions on the total historical business data earlier in the process of building the decision tree, that is, the conditional attribute with a larger final priority index has a shorter distance between the corresponding node and the root node in the decision tree.

[0112] The leaf nodes of the decision tree represent the classification results of the total historical business data. The classification results of the total historical business data can be directly stored in a database or a shared platform. Other companies can also access the database or platform to achieve the purpose of sharing data.

[0113] Since most companies are more concerned about the company's profitability, the company's financial profitability is used as the leaf node of the decision tree. In order to explore which conditional attributes are related to the reasons for the loss of a property insurance company, high-value data is obtained; in other embodiments, the number of loss years for each leaf node of the decision tree is counted, and the conditional attributes on the path from the leaf node with the largest number of loss years to the root node of the decision tree are recorded as negative attributes, and the historical business data of each negative attribute is stored in the database separately. Negative attributes are conditional attributes that have a negative impact on the company's profitability. Sharing historical business data of negative attributes is conducive to other insurance companies focusing on these negative attributes and targeted solutions to the negative impact of negative attributes on the profitability of insurance companies.

[0114] So far, the present invention is completed.

[0115] Embodiment 2:

[0116] This invention proposes a data intelligent sharing system for the construction of a smart sharing system. Figure 3 , which shows a system structure diagram of a data intelligent sharing system for building a smart sharing system provided by an embodiment of the present invention, the system comprising:

[0117] The data collection module 410 is used to obtain total historical business data, which includes historical business data of several years, extract business features from the total historical business data, and determine indicator attributes, attention attributes and several conditional attributes in the business features;

[0118] The priority indicator analysis module 420 is used to split the total historical business data by using each condition attribute to obtain different sub-data sets for each condition attribute; according to the difference in the degree of confusion of the indicator characteristics of the indicator attributes in the total historical business data and the sub-data set of each condition attribute, and the similarity between the change trend of each condition attribute and the historical business data of the attribute of interest, obtain the initial priority indicator of each condition attribute;

[0119] The decision tree construction module 430 is used to adjust the initial priority index according to the influence of each condition attribute on the focus attribute and the correlation between each condition attribute and other condition attributes, and use the adjusted initial priority index to construct a decision tree to classify and share the total historical business data.

[0120] It should be noted that the device provided in the above embodiment is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computer device is divided into different functional modules to complete all or part of the functions described above. In addition, the data intelligent sharing system for the construction of a smart sharing system provided in the above embodiment and the data intelligent sharing method embodiment for the construction of a smart sharing system belong to the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0121] Embodiment 3:

[0122] Figure 4 A computer device schematic diagram of a data intelligent sharing device for building a smart sharing system provided by an embodiment of the present invention. For example, Figure 4 As shown, the computer device includes: a memory 501, a processor 502, and a computer program 503 stored in the memory 501 and running on the processor 502, wherein when the processor 502 executes the computer program 503, the computer device can execute any one of the data intelligent sharing methods for building a smart sharing system introduced above.

[0123] In addition, an embodiment of the present application also protects a device, which may include a memory and a processor, wherein the memory stores executable program code, and the processor is used to call and execute the executable program code to execute a data intelligent sharing method for building a smart sharing system provided by an embodiment of the present application.

[0124] In this embodiment, the functional modules of the device can be divided according to the above method example. For example, each functional module can be corresponded, or two or more functions can be integrated into one processing module. The above integrated module can be implemented in the form of hardware. It should be noted that the division of modules in this embodiment is schematic and is only a logical function division. There may be other division methods in actual implementation.

[0125] It should be understood that the device provided in this embodiment is used to execute the above-mentioned data intelligent sharing method for building a smart sharing system, and therefore can achieve the same effect as the above-mentioned implementation method.

[0126] In the case of an integrated unit, the device may include a processing module and a storage module. When the device is applied to a device, the processing module may be used to control and manage the actions of the device. The storage module may be used to support the device to execute mutual program codes, etc.

[0127] The processing module may be a processor or a controller, which may implement or execute various exemplary logic blocks, modules and circuits included in the disclosure of the present application. The processor may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of digital signal processing (DSP) and a microprocessor, etc. The storage module may be a memory.

[0128] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0129] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0130] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A data intelligent sharing method for building a smart sharing system, characterized in that: The method includes: Obtaining total historical business data, the total historical business data includes historical business data of several years, extracting business features from the total historical business data, and determining indicator attributes, focus attributes, and several conditional attributes in the business features; The total historical business data is split by each condition attribute to obtain different sub-data sets for each condition attribute; the initial priority indicator of each condition attribute is obtained according to the difference in the degree of confusion of the indicator characteristics of the indicator attributes in the total historical business data and the sub-data sets of each condition attribute, and the similarity of the change trend of each condition attribute and the historical business data of the attribute of interest; According to the influence of each condition attribute on the focus attribute and the correlation between each condition attribute and other condition attributes, the initial priority index is adjusted, and a decision tree is constructed using the adjusted initial priority index to classify and share the total historical business data.

2. A data intelligent sharing method for building a smart sharing system according to claim 1, characterized in that: The obtaining of the initial priority index of each condition attribute includes: According to the difference in the degree of confusion of the indicator characteristics of the indicator attributes in the total historical business data and the sub-dataset of each condition attribute, the information gain of each condition attribute is obtained; According to the similarity between the change trend of the historical business data of each condition attribute and the concerned attribute, the synergy index between each condition attribute and the concerned attribute is obtained; According to the information gain and the coordination index, an initial priority index of each condition attribute is obtained; the information gain and the coordination index are both positively correlated with the initial priority index.

3. A data intelligent sharing method for building a smart sharing system according to claim 2, characterized in that: The obtaining of the information gain of each conditional attribute includes: Obtain the information entropy of the indicator characteristics of the indicator attributes in the total historical business data, recorded as the total information entropy; Obtain the information entropy of the indicator feature of the indicator attribute in the historical business data in each sub-dataset of each conditional attribute, and record it as the sub-information entropy of the corresponding sub-dataset; take the proportion of the annual quantity in the historical business data in each sub-dataset of each conditional attribute to the annual quantity in the total historical business data as the quantity indicator of each sub-dataset of each conditional attribute; Obtaining an overall uncertainty index for each conditional attribute according to the sub-information entropy and the quantity index of all sub-data sets of each conditional attribute; The difference between the total information entropy and the overall uncertainty index is normalized to obtain the information gain of each conditional attribute.

4. The data intelligent sharing method for building a smart sharing system according to claim 2, characterized in that: The step of obtaining the synergy index between each condition attribute and the focus attribute includes: Record the condition attribute and the focus attribute as analysis attributes, arrange the historical business data of each analysis attribute in order, and obtain the business data sequence of each analysis attribute; Obtaining a second-order difference sequence of the business data sequence of each analysis attribute; calculating the mean of the absolute values ​​of the differences between the elements with the same subscript in the second-order difference sequence of each condition attribute and the focus attribute as a change difference indicator between each condition attribute and the focus attribute; Obtain the correlation coefficient of the business data sequence between each condition attribute and the attribute of interest, and obtain the synergy index of each condition attribute and the attribute of interest based on the correlation coefficient and the change difference index, wherein the correlation coefficient and the synergy index are positively correlated, and the change difference index and the synergy index are negatively correlated.

5. The data intelligent sharing method for building a smart sharing system according to claim 1, characterized in that: The initial priority index is adjusted according to the influence of each condition attribute on the concerned attribute and the correlation between each condition attribute and other condition attributes, and a decision tree is constructed using the adjusted initial priority index to classify and share the total historical business data, including: According to the influence degree of each condition attribute on the concerned attribute and the correlation degree of each condition attribute with other condition attributes, the initial priority index is adjusted to obtain the final priority index of each condition attribute; Arrange the final priority indexes of all condition attributes from large to small to obtain an initial priority sequence; select the first several condition attributes from the initial priority sequence to form a final priority sequence; According to the order of the condition attributes in the final priority sequence, a decision tree is constructed using the Hunt algorithm to obtain the classification results of the total historical business data, and the obtained classification results are stored in a database.

6. A data intelligent sharing method for building a smart sharing system according to claim 5, characterized in that: The obtaining of the final priority index of each condition attribute includes: Obtain the impact index of each conditional attribute on the attribute of interest; Obtaining the correlation coefficients of each condition attribute and the business data sequences of other condition attributes respectively, and taking the average of the correlation coefficients of each condition attribute and the business data sequences of other condition attributes as the overall correlation value of each condition attribute and the other condition attributes; Performing negative correlation mapping on the overall correlation value, normalizing the product of the mapping result and the impact index of each condition attribute to obtain a priority construction coefficient of each condition attribute; The initial priority index is weighted using the priority construction coefficient to obtain a final priority index for each condition attribute.

7. A data intelligent sharing method for building a smart sharing system according to claim 6, characterized in that: The obtaining of the influence index of each condition attribute on the concerned attribute includes: Perform linear analysis on all conditional attributes and attention attributes to construct an initial attention function, which contains several regression coefficients; Substituting the historical business data of all conditional attributes and the historical business data of the attention attribute of each year into the initial attention function to obtain the attention fitting value of each year; The error function is obtained based on the difference between the historical business data of the attribute of interest in all years and the fitted value of interest; the value of the regression coefficient corresponding to the minimum value of the error function is used as the optimal value of each regression coefficient in the initial function of interest, and the optimal value of the regression coefficient corresponding to each conditional attribute is used as the influence indicator of each conditional attribute.

8. The data intelligent sharing method for building a smart sharing system according to claim 1, characterized in that: The total historical business data is split by each condition attribute to obtain different sub-data sets for each condition attribute, including: The product of the maximum value in the historical business data of each condition attribute and the preset weight threshold is used as the standard segmentation threshold of each condition attribute; Based on the standard segmentation threshold, the historical business data of each condition attribute is divided into two categories; the set consisting of the historical business data of the year to which each data in each category of historical business data of each condition attribute belongs is used as the sub-data set of each condition attribute.

9. The data intelligent sharing method for building a smart sharing system according to claim 3, characterized in that: The obtaining of the overall uncertainty index of each conditional attribute according to the sub-information entropy and the quantity index of all sub-data sets of each conditional attribute includes: The product of the sub-information entropy and the quantity index of each sub-data set of each conditional attribute is used as the local uncertainty index of the corresponding sub-data set; the cumulative sum of the local uncertainty indexes of all sub-data sets of each conditional attribute is used as the overall uncertainty index of each conditional attribute.

10. The data intelligent sharing method for building a smart sharing system according to claim 4, characterized in that: The correlation coefficient is the Pearson correlation coefficient.

Images