Flow distribution method based on multi-dimensional information fitting

Abstract

The invention discloses a flow distribution method based on multi-dimensional information fitting, which is characterized by comprising the following steps of: analyzing input data and behavior data in a service period, mining a relationship between input and flow, performing equal-frequency binning and chi-square binning processing on the data, and fitting by utilizing a least square method to obtain a flow distribution result; a mathematical model between the industry flow and the satisfaction degree is established, so that a reasonable industry flow distribution standard is obtained, the model can dynamically distribute the flow based on an expected extension rate, and real-time monitoring of the flow is realized. Through the mode, the e-commerce platform can more effectively manage and optimize flow distribution, the satisfaction degree of suppliers is improved, and the operation efficiency of the whole platform is enhanced.

Description

Technical Field

[0001] The present invention belongs to the field of Internet technology, and in particular relates to a flow distribution method based on multi-dimensional information fitting. Background Art

[0002] The popularity of the Internet has greatly changed people's lifestyles and social structures, and has also become an important driving force for the diversification of industries. The rise of e-commerce is an inevitable result of following the historical trend, which has spawned a large number of e-commerce platforms and related service industries and become the mainstream consumption method. On the other hand, the number of different industries is increasing rapidly, and the continuous development of automation and artificial intelligence has further divided and refined the original industries.

[0003] As a large and comprehensive e-commerce platform, it is necessary to systematically and comprehensively build a traffic ecosystem. Among them, it is particularly important to reasonably allocate traffic to different industries. However, how to reasonably allocate traffic to different industries is a complex technology. The current common industry traffic allocation is mostly based on weight allocation. This method cannot effectively establish the correlation between the supplier's production ratio and supplier satisfaction. In addition, due to the obvious differences between different industries, the traffic distribution of light industry and heavy industry is obviously different. The related problems are endless and need to be solved urgently, including:

[0004] (1) The input and return are not proportional: By exploiting loopholes in the platform mechanism and controlling the traffic entrance, traffic far exceeds the input, resulting in a siphon effect in the distribution of the introduced off-site traffic, so that the traffic is controlled by some users and cannot be distributed to other users. In this case, a small number of users who control a large amount of traffic will form an oligopoly, restrict competition and may have an adverse impact on the market.

[0005] (2) Limitations of product display sorting and uneven exposure: When displaying products, they are sorted only according to their richness and adaptability. The sorting method is single, which easily causes some products to occupy the main exposure position, while other products cannot get effective exposure opportunities;

[0006] (3) Traffic monitoring has limitations: The platform's monitoring of traffic in the entire industry only shows a numerical fluctuation. Although it can reflect the overall trend of traffic changes, it cannot specifically identify which specific industries or market segments have large traffic gaps. In addition, due to the lack of effective traffic evaluation, the platform cannot track and measure whether the traffic obtained by different users is in line with their expectations. If we rely on user feedback to adjust the off-site effect diversion, since feedback is often raised after observing traffic problems, and it takes a long time for the platform to respond and implement solutions, this delay means that the problem cannot be solved immediately, which is very likely to increase customer dissatisfaction and cause customer churn.

[0007] To sum up, e-commerce platforms need an effective traffic distribution technology that can not only detect traffic distribution problems in various industries early, but also timely regulate unreasonable traffic distribution and improve platform user satisfaction. Summary of the invention

[0008] In order to solve the existing technical problems, the present invention provides a traffic allocation method based on multi-dimensional information fitting, which analyzes the input data and behavior data within the service cycle, explores the relationship between input and traffic, performs equal frequency binning and chi-square binning on the data, and uses the least squares method third-order polynomial fitting to establish a mathematical model between industry traffic and satisfaction, thereby deriving a reasonable and reliable industry traffic allocation standard. The model can dynamically allocate traffic based on the expected renewal rate and realize real-time monitoring of traffic. In this way, the e-commerce platform can more effectively manage and optimize traffic allocation, improve supplier satisfaction, and enhance the operating efficiency of the entire platform. The present invention has the characteristics of data-driven, dynamic adjustment and real-time monitoring, and is of great significance for the optimization of traffic allocation in the field of e-commerce.

[0009] The technical solution of the present invention is as follows: A flow distribution method based on multi-dimensional information fitting, specifically comprising

[0010] Step 1: Data collection: Set up global tracking points in the e-commerce platform to obtain and record in real time the input data of supplier users and the behavior data of buyer users within a service period; the service period refers to the contract period for supplier users to purchase platform services; the behavior data of buyer users is the data collected during the user's product search, product browsing and product inquiry process, including product ID, product industry ID, product exposure times, product visit times, product inquiry times; the product exposure times refers to the frequency of the product appearing in front of the buyer user; the product visit times refers to the frequency of the buyer user clicking and visiting the product page; the product inquiry times refers to the frequency of the buyer user initiating product inquiries; the product supplier's input data includes renewal identification and daily industry input;

[0011] Step 2: Data preprocessing, mining and constructing the relationship between the behavior data and the investment data to form an initial data set; the information of the initial data set includes industry ID, supplier ID, traffic, and renewal flag; the traffic refers to the number of product exposures, visits, and product inquiries obtained for every 10,000 yuan of industry investment, and the renewal flag refers to the value assigned according to the renewal status of the supplier user during the service period. If the supplier user renews within the service period, the renewal flag is recorded as 1, otherwise it is recorded as 0; the initial data set is distinguished by industry and saved in an array;

[0012] The step 2 specifically comprises:

[0013] Step 2-1: For each product industry that the supplier invests in, calculate the supplier's daily industry input. The daily industry input Among them C i represents the supplier's daily industry input in industry i, n is the total number of product industries the supplier has invested in, C is the supplier's daily input, and the daily input = the supplier's total input / the number of days in the service cycle; i is the number of valid products in industry i; the valid products refer to products with user behaviors, including search behaviors, access behaviors and inquiry behaviors;

[0014] Step 2-2: Calculate the industry traffic that the supplier receives daily, and associate it with the daily industry input in step 2-1-4: Calculate the industry traffic for each industry based on the industry that the supplier has invested in. The industry traffic refers to the total number of exposures, visits, and inquiries received by products in the industry every day.

[0015] Step 2-3: Calculate the flow F corresponding to every 10,000 yuan of industry investment i =f i ×(10000 / C i ), where f i represents the industry traffic that the supplier obtains in industry i every day; Fi represents the traffic that the supplier obtains per 10,000 yuan per day in industry i;

[0016] Step 2-4: Obtain the supplier's renewal ID and associate the renewal ID with the flow F calculated in step 2-3 i , forming the initial data set;

[0017] Step 3: Process the traffic feature data using binning technology: extract the traffic and the corresponding renewal identifier from the initial data set, arrange the extracted data in ascending order according to the traffic, bin the traffic using the equal frequency binning method, and evenly distribute the extracted data in different bins; then merge adjacent bins using the chi-square binning method, and obtain the renewal rate of each bin after the merging stops;

[0018] In step 3, Python's pandas.qcut function is used to perform equal frequency binning, and the initial number of bins is set to 15;

[0019] In step 3, the adjacent bins are merged by using chi-square binning, and the specific steps include:

[0020] Step 3-1: Create a contingency table, wherein the rows of the contingency table are the box numbers and the columns of the contingency table are the renewal identifiers;

[0021] Step 3-2: Count the frequency of occurrence of different renewal marks in different bins and save them in a contingency table, which is recorded as the observed frequency O jm , represents the observation frequency of the jth row and the mth column;

[0022] Step 3-3: Calculate the expected frequency based on the observed frequency in the contingency table, where E km is the expected frequency of the kth row and mth column, H m is the sum of the observed frequencies in the mth column, R k is the sum of the observed frequencies in the kth row, N is the sum of all observed frequencies in the contingency table, and the calculation formula is:

[0023]

[0024] Step 3-4: Analyze and determine whether to merge adjacent boxes based on the observed frequency and expected frequency, specifically including: calculating the chi-square value of adjacent boxes respectively, screening out the minimum chi-square value of adjacent boxes, and the chi-square value of adjacent boxes Where E k represents the expected frequency sum of the kth row; compare the minimum chi-square value with the threshold value, if it is less than the threshold value, merge the data in the adjacent boxes; if it is greater than the threshold value, stop merging the data in the adjacent boxes; the threshold value is the critical value of the chi-square distribution confirmed according to the degrees of freedom and the significance level, the significance level is 0.05, the degrees of freedom = (r-1) × (c-1), where r is the number of rows of the contingency table, and c is the number of columns of the contingency table;

[0025] After stopping the bin merging in step 3-4, the WOE of each bin is calculated by Python's own function. d Value and IV d value, where d represents the bin number, The IV d =(Renewal rate - Non-renewal rate)*WOE d ; If WOE d As the bin number increases, IV d In the interval [0.02, 0.5], the binning results are robust;

[0026] In the steps 3-4, the number of sub-boxes is preferably combined into 8.

[0027] Step 4: Data fitting, generating a mathematical model between traffic value and renewal rate: Obtain the binned data in step 3, select the bin number as the horizontal axis, select the renewal rate of the corresponding bin as the vertical axis, and plot discrete points; Use the least squares method to fit the discrete points into a third-order polynomial function to generate fitting coefficients;

[0028] In step 4, the use of the least squares method specifically refers to using the polynomial function polyfit(x,y,3) of Matlab to implement the least squares method.

[0029] Step 5: Numerical analysis, determining the standard flow value according to the renewal rate: using the polynomial function fitted in step 4, substituting a specific renewal rate to generate a corresponding value, analyzing the bins into which the values ​​fall, obtaining the data intervals of the bins, and selecting the upper limit of the data interval as the standard flow value;

[0030] In step 5, the specific renewal rate is substituted into the corresponding value, and if the value is a decimal, the value is rounded down;

[0031] In step 5, if no corresponding value exists when a specific renewal rate is substituted, the specific renewal rate is compared with the renewal rates of all sub-boxes. If the renewal rate is greater than the renewal rates of all sub-boxes, the maximum value is selected from the minimum value of the function and the maximum value of the function; if the polynomial function does not have a minimum value, the value corresponding to the maximum value of the function is taken;

[0032] In step 5, if there are at least 2 values ​​when substituting into a specific renewal rate, a value is selected according to the type of monotonic interval into which the value falls; if a value falls within the monotonic increasing interval of the function, the average of these values ​​is taken as the standard flow value; if no value falls within the monotonic increasing interval of the function, the average of the values ​​falling within the monotonic decreasing interval of the function is taken as the standard flow value.

[0033] The beneficial effects achieved by the present invention are:

[0034] (1) The present invention provides a method for analyzing user input data and behavior data during a service cycle, mining the relationship between input and traffic, performing equal frequency binning and chi-square binning on the data to complete the optimal binning, and performing polynomial fitting on the scatter point relationship after the optimal binning to obtain an optimal curve, thereby generating a mathematical model that characterizes the relationship between industry traffic and renewal rate. The curve is overlapped with the predetermined renewal rate, and the traffic required by the industry under the premise of meeting a certain renewal rate can be easily obtained, thereby deriving a reasonable and reliable industry traffic allocation standard. The model can dynamically allocate traffic based on the expected renewal rate and realize real-time monitoring of traffic. In this way, the e-commerce platform can more effectively manage and optimize traffic allocation, improve supplier satisfaction, and enhance the operational efficiency of the entire platform, which is of great significance for the optimization of traffic allocation in the field of e-commerce.

[0035] (2) The present invention performs optimal binning of data based on the chi-square test. This method can significantly enhance the interpretability of the mathematical model, improve the predictive ability, and capture the nonlinear relationship between traffic and renewal rate, thereby improving the model's fit. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a flow chart of a flow distribution method based on multi-dimensional information fitting in an embodiment of the present invention;

[0037] Figure 2 This is a schematic diagram of the process of preprocessing industry flow data in an embodiment of the present invention;

[0038] Figure 3 Schematic diagram of the process of feature data processing based on chi-square binning in an embodiment of the present invention;

[0039] Figure 4 It is a schematic diagram of a fitting function curve of multi-dimensional information fitting in an embodiment of the present invention;

[0040] Figure 5 A diagram showing a function curve of traffic distribution analysis based on renewal rate in an embodiment of the present invention;

[0041] Figure 6 FIG2 is a schematic diagram of a function curve of traffic distribution analysis based on renewal rate in an embodiment of the present invention. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0043] Figure 1 A flow distribution method based on multi-dimensional information fitting in an embodiment of the present invention specifically includes:

[0044] Step 101: Data collection, real-time acquisition and storage of traffic data: global tracking points are set up in the e-commerce platform to obtain and record the input data of supplier users and the behavior data of buyer users within a service period; the service period refers to the contract period of supplier users purchasing platform services; the behavior data of buyer users is the data collected during the user's product search, product browsing and product inquiry process, including product ID, product industry ID, product exposure times, product visit times, and product inquiry times; the product supplier's input data includes whether to renew the contract and daily industry input;

[0045] In the embodiment of the present invention, the product exposure times refer to the frequency of a product appearing in front of a buyer user; the product visit times refer to the frequency of a buyer user clicking and visiting a product page; the product inquiry times refer to the frequency of a buyer user initiating a product inquiry;

[0046] Step 102: Data preprocessing, mining the relationship characteristics between behavior data and input data, forming an initial data set: Calculate the daily industry input of supplier users, and after unified data processing, generate the flow data corresponding to the supplier's investment of 10,000 yuan in different industries. The purpose of analyzing the relationship characteristics between behavior data and input data in step 102 is to more accurately mine the correlation between the supplier's input data and the buyer's behavior data within the service cycle. On the one hand, the supplier's input data reflects the input of the service provision, which measures the supply-side capabilities; on the other hand, the buyer's behavior data captures the behavior activities of service users within the same service cycle, which reflects market demand and user preferences. These two types of data essentially reflect different aspects of traffic services. Mining the correlation relationship provides a decision-making basis for the subsequent traffic allocation model, which helps to achieve optimal allocation.

[0047] like Figure 2 As shown, the specific steps of step 102 include:

[0048] Step 102-1: For each product industry that the supplier invests in, calculate the supplier's daily industry input, including:

[0049] Step 102-1-1: Calculate the supplier's daily investment, where the daily investment = the supplier's total investment / the number of days in the service cycle;

[0050] Step 102-1-2: Calculate daily industry proportion: obtain all product industries invested by suppliers, and count the number of valid products in a single industry and in all industries on a daily basis; the valid products refer to products with user behaviors, including search behaviors, visit behaviors and inquiry behaviors; the daily industry proportion = the number of valid products in a single industry / the number of valid products in all industries;

[0051] Step 102-1-3: Calculate the supplier's daily industry input, where C i represents the supplier's daily industry input in industry i, n is the total number of product industries the supplier has invested in, C is the supplier's daily input, and prod i is the number of effective products in industry i;

[0052]

[0053] Step 102-2: Calculate the industry traffic that the supplier receives daily, and associate it with the daily industry investment in step 102-1: Calculate the industry traffic of each industry based on the industry invested by the supplier. The industry traffic refers to the total number of exposures, visits, and inquiries received by the products in the industry every day.

[0054] Step 102-3: Calculate the flow corresponding to each 10,000 yuan of industry investment, where f i represents the industry traffic that the supplier obtains in industry i every day; Fi represents the traffic that the supplier obtains per 10,000 yuan per day in industry i;

[0055] F i =f i ×(10000 / C i ) (Formula 1)

[0056] Step 102-4: Create an initial data set, including renewal status and traffic obtained per 10,000 yuan per day: obtain the renewal status of the supplier during the service period, record the renewal identifier, if the supplier user renews during the service period, the renewal identifier is recorded as 1; if the supplier user does not renew in the server, the renewal identifier is recorded as 0; associate the supplier's renewal identifier with the supplier's traffic obtained per 10,000 yuan per day in step 2-3 to form an initial data set for recording the traffic obtained by the supplier per 10,000 yuan invested in different industries; the information of the initial data set includes industry ID, supplier ID, industry traffic, and renewal identifier;

[0057] Step 102-5: The initial data set is divided by industry and the data is saved in an array for subsequent analysis or processing.

[0058] Step 103: Process the traffic feature data using binning technology: extract industry traffic and corresponding renewal identifiers from the initial data set, arrange the extracted data in ascending order according to industry traffic, bin the industry traffic using an equal-frequency binning method, evenly distribute the extracted data in different bins, and set the initial number of bins for equal-frequency binning to 15; then use the chi-square binning method to merge adjacent bins, and obtain the renewal rate of each bin after the merging stops;

[0059] Binning technology is a data grouping technology used to discretize continuous variables. The principle of the equal-frequency binning method is to sort the data in ascending order, and then evenly divide the data into several intervals, each interval contains an equal amount of data, and each interval corresponds to a bin, so that continuous data can be divided into multiple bins to better present the distribution of data. In the embodiment of the present invention, industry flow is a continuous variable, so the equal-frequency binning method is used to discretize the data, and the adjacent bins are further merged through the chi-square statistic, which effectively converts the continuous industry flow data into a discrete form, which facilitates the subsequent data analysis and model building.

[0060] In the present invention, the initial number of boxes is 15. The purpose of box division is to use the least squares method for linear fitting later. After testing and verification by technicians, for a linear fitting model, the number of boxes is at least 5 times the number of model parameters. Therefore, the initial number of boxes is set to 15.

[0061] In step 103, the pandas.qcut function of Python is used to perform equal frequency binning;

[0062] like Figure 3 As shown, in step 103, the method of merging adjacent bins by using chi-square binning specifically includes:

[0063] Step 103-1: Create a contingency table, wherein the rows of the contingency table are the box numbers and the columns of the contingency table are the renewal identifiers;

[0064] Step 103-2: Count the occurrence frequencies of different renewal marks in different bins respectively, and save them in a contingency table, which is recorded as the observed frequency O. jm , represents the observation frequency of the jth row and the mth column;

[0065] Step 103-3: Calculate the expected frequency based on the observed frequency in the contingency table, where E km is the expected frequency of the kth row and mth column, H m is the sum of the observed frequencies in the mth column, R k is the sum of the observed frequencies in the kth row, N is the sum of all observed frequencies in the contingency table, and the calculation formula is:

[0066]

[0067] Step 103-4: Calculate the chi-square value of adjacent boxes based on the observed frequency and the expected frequency, and merge adjacent boxes by threshold judgment, specifically:

[0068] Step 103-4-1: Calculate the chi-square values ​​of adjacent boxes respectively, and select the minimum chi-square value of adjacent boxes. Where E k represents the expected frequency sum of the k-th row.

[0069] Step 103-4-2: Compare the minimum chi-square value with the threshold value. If it is less than the threshold value, merge the data in the adjacent boxes; if it is greater than the threshold value, stop merging the data in the adjacent boxes; the threshold value is the critical value of the chi-square distribution confirmed based on the degree of freedom and the significance level, the significance level is 0.05, the degree of freedom = (r-1) × (c-1), where r is the number of rows in the contingency table, c is the number of columns in the contingency table, and r is the number of rows in the contingency table. The chi-square critical value can be obtained through the chi-square distribution table or calculated using statistical software.

[0070] After the bin merging stops in step 103-4, the WOE of each bin is calculated. d Value and IV d value, where d represents the bin number, The IV d =(Renewal rate - Non-renewal rate)*WOE d .

[0071] The WOE of each bin d Value and IV d The values ​​can be obtained through Python's own functions. If WOE d As the bin number increases, IV d In the interval [0.02, 0.5], it indicates that the traffic per 10,000 yuan is increased and the renewal amount is increased, which is in line with the intuition and business logic of reality, and also proves that the binning results are robust. The robustness of the binning results means that the number of bins is appropriate, and there is neither too many bins nor too few bins. Too many bins make the model too complex and capture noise, while too few bins cause the model to be underfitted and poor in predicting unknown data. Therefore, the appropriate number of bins is extremely important for building a prediction model with strong generalization ability. In practical applications, the number of bins needs to be determined through multiple iterations and optimizations. In the embodiment of the present invention, after evaluation and testing, reducing the number of bins to 8 is a balance point. This balance point is obtained through cross-validation and comprehensive judgment of the performance indicators of the model (such as accuracy, precision, recall, etc.). In cross-validation, it is stable, has high accuracy, precision and recall, and the WOE value of the bin is tonal. The IV value is in a reasonable range. It also conforms to the business logic, which can ensure that the model has both good explanatory power and high accuracy when predicting unknown data.

[0072] Step 104: Data fitting, generating a mathematical model of the relationship between traffic value and renewal rate: obtain the data after binning in step 103, select the bin number as the horizontal coordinate, select the renewal rate of the corresponding bin as the vertical coordinate, and draw discrete points; use the least squares method to fit the discrete points into a third-order polynomial to generate fitting coefficients. The reason for choosing the third order is that the technicians found during the experiment that too high an order will lead to overfitting of the data, which is not conducive to the generalization ability of the model; overfitting causes the model to perform well on training data, but perform poorly on unknown test data, so it cannot predict new and unknown data well. At the same time, too high an order will also increase the computational complexity of the model and affect the operating efficiency of the model. Therefore, in order to balance the fit, generalization ability and computational efficiency of the model, the technicians chose a third-order polynomial for fitting.

[0073] In step 104, the least square method is used, specifically, the least square method is implemented using the polynomial function polyfit(x,y,3) of Matlab. Table 1 is the input data, where x is the bin number and y is the renewal rate of the corresponding bin. After fitting, the following is formed: Figure 4 The schematic diagram of the fitting function curve of the multi-dimensional information fitting industry flow distribution in the embodiment of the present invention is shown. According to the fitting function schematic diagram, it is easy to find the standard flow value required under a certain renewal rate;

[0074] Table 1

[0075] x 0 1 2 3 4 5 6 7 y 0.4285 0.5714 0.4285 0.4285 0.5714 0.8571 0.5714 0.8571

[0076] Step 105: Numerical analysis, determine the standard flow value according to the renewal rate: use the polynomial function fitted in step 104, substitute the specific renewal rate to generate the corresponding value, analyze the bin where the value falls, obtain the data interval of the bin, and select the upper limit of the data interval as the standard flow value; in step 105, substitute the specific renewal rate to generate the corresponding value, if the value is a decimal, round it down; the bin number x represents a data interval, not a continuous value, so even if the calculation result is a decimal, in fact only existing and complete bins can be selected, so it needs to be rounded down. Figure 4 , when the renewal rate y = 60%, the corresponding bin value x = 4.3, after rounding down the decimal, x = 4, so the upper limit of the data interval of bin 4 is taken as the standard flow value;

[0077] In step 105, if no corresponding value exists when a specific renewal rate is substituted, the specific renewal rate is compared with the renewal rates of all sub-boxes. If the renewal rate is greater than the renewal rates of all sub-boxes, the maximum value is selected from the minimum value of the function and the maximum value of the function; if the polynomial function does not have a minimum value, the value corresponding to the maximum value of the function is taken;

[0078] like Figure 5 A diagram is provided for illustrating a function curve of traffic distribution analysis based on renewal rate in an embodiment of the present invention. The dots in the diagram are scattered points formed by the renewal rates of each sub-box in the coordinate system. If the renewal rate y=70%, there is no corresponding x value, and 70% is higher than the renewal rates of all sub-boxes. The minimum values ​​of the function are calculated to be 0.59 and 4.52, and the maximum value of the function is 3.01. The maximum value of the three values ​​is x=4.52. After rounding down the decimal, x=4. The upper limit value of the data interval corresponding to sub-box No. 4 is taken as the standard traffic value.

[0079] In step 105, if there are at least two values ​​for the specific renewal rate, a value is selected according to the type of monotonic interval into which the value falls; if a value falls within the monotonic increasing interval of the function, the average of these values ​​is taken as the standard flow value; if no value falls within the monotonic increasing interval of the function, the average of the values ​​falling within the monotonic decreasing interval of the function is taken as the standard flow value; Figure 6 The following are two diagrams showing the function curves of the flow distribution analysis based on the renewal rate in the embodiment of the present invention. The dots in the diagram are the scattered points formed by the renewal rate of each sub-box in the coordinate system. If the renewal rate is 70%, there are two corresponding x values, namely x1=0.2 and x2=5.8, where x1 is in the monotonically decreasing interval of the function and x2 is in the monotonically increasing interval of the function. Therefore, x=5.8 is taken. After rounding down the decimal, x=5. The upper limit value of the data interval corresponding to sub-box No. 5 is taken as the standard flow value. The reason for giving priority to the monotonically increasing interval of the function and then the monotonically decreasing interval in the present invention is to follow the positive relationship between the renewal rate and the flow value. The higher the renewal rate, the more flow is allocated to the product.

[0080] The present invention may also have many other implementation modes. The above embodiments do not limit the present invention in any way. Without departing from the spirit and essence of the present invention, technicians familiar with the field may make various corresponding changes and modifications based on the present invention. All other improvements and applications made to the above embodiments in an equivalent transformation manner should fall within the scope of protection of the claims attached to the present invention.

Claims

1. A flow distribution method based on multi-dimensional information fitting, characterized in that: include: Step 1: Data collection: Set up global tracking points in the e-commerce platform to obtain and record in real time the input data of supplier users and the behavior data of buyer users within a service period; the service period refers to the contract period for supplier users to purchase platform services; the behavior data of buyer users is the data collected during the user's product search, product browsing and product inquiry process, including product ID, product industry ID, product exposure times, product visit times, product inquiry times; the product exposure times refers to the frequency of the product appearing in front of the buyer user; the product visit times refers to the frequency of the buyer user clicking and visiting the product page; the product inquiry times refers to the frequency of the buyer user initiating product inquiries; the product supplier's input data includes renewal identification and daily industry input; Step 2: Data preprocessing, mining and constructing the relationship between the behavior data and the investment data to form an initial data set; the information of the initial data set includes industry ID, supplier ID, traffic, and renewal flag; the traffic refers to the number of product exposures, visits, and product inquiries obtained for every 10,000 yuan of industry investment, and the renewal flag refers to the value assigned according to the renewal status of the supplier user during the service period. If the supplier user renews within the service period, the renewal flag is recorded as 1, otherwise it is recorded as 0; the initial data set is distinguished by industry and saved in an array; Step 3: Process the traffic feature data using binning technology: extract the traffic and the corresponding renewal identifier from the initial data set, arrange the extracted data in ascending order according to the traffic, bin the traffic using the equal frequency binning method, and evenly distribute the extracted data in different bins; then merge adjacent bins using the chi-square binning method, and obtain the renewal rate of each bin after the merging stops; Step 4: Data fitting, generating a mathematical model between traffic value and renewal rate: Obtain the binned data in step 3, select the bin number as the horizontal axis, select the renewal rate of the corresponding bin as the vertical axis, and plot discrete points; Use the least squares method to fit the discrete points into a third-order polynomial function to generate fitting coefficients; Step 5: Numerical analysis, determine the standard flow value according to the renewal rate: use the polynomial function fitted in step 4, substitute the specific renewal rate to generate the corresponding numerical value, analyze the bins into which the numerical value falls, obtain the data interval of the bins, and select the upper limit value of the data interval as the standard flow value.

2. A flow distribution method based on multi-dimensional information fitting as claimed in claim 1, characterized in that: The step 2 specifically comprises: Step 2-1: For each product industry that the supplier invests in, calculate the supplier's daily industry input. The daily industry input Among them C i represents the supplier's daily industry input in industry i, n is the total number of product industries the supplier has invested in, C is the supplier's daily input, and the daily input = the supplier's total input / the number of days in the service cycle; i is the number of valid products in industry i; the valid products refer to products with user behaviors, including search behaviors, access behaviors and inquiry behaviors; Step 2-2: Calculate the industry traffic that the supplier receives daily, and associate it with the daily industry input in step 2-1-4: Calculate the industry traffic for each industry based on the industry that the supplier has invested in. The industry traffic refers to the total number of exposures, visits, and inquiries received by products in the industry every day. Step 2-3: Calculate the flow F corresponding to every 10,000 yuan of industry investment i =f i ×(10000 / C i ), where f i represents the industry traffic that the supplier obtains in industry i every day; Fi represents the traffic obtained by the supplier in industry i per 10,000 yuan per day; Step 2-4: Obtain the supplier's renewal ID and associate the renewal ID with the flow F calculated in step 2-3 i , forming the initial data set.

3. A flow distribution method based on multi-dimensional information fitting as claimed in claim 2, characterized in that: In step 3, Python's pandas.qcut function is used to perform equal frequency binning, and the initial number of bins is set to 15; In step 3, the adjacent bins are merged by using chi-square binning, and the specific steps include: Step 3-1: Create a contingency table, wherein the rows of the contingency table are the box numbers and the columns of the contingency table are the renewal identifiers; Step 3-2: Count the frequency of occurrence of different renewal marks in different bins and save them in a contingency table, which is recorded as the observed frequency O jm , represents the observation frequency of the jth row and the mth column; Step 3-3: Calculate the expected frequency based on the observed frequency in the contingency table, where E km is the expected frequency of the kth row and mth column, H m is the sum of the observed frequencies in the mth column, R k is the sum of the observed frequencies in the kth row, N is the sum of all observed frequencies in the contingency table, and the calculation formula is: Step 3-4: Analyze and determine whether to merge adjacent boxes based on the observed frequency and expected frequency, specifically including: calculating the chi-square value of adjacent boxes respectively, screening out the minimum chi-square value of adjacent boxes, and the chi-square value of adjacent boxes Where E k represents the expected frequency sum of the kth row; the minimum chi-square value is compared with the threshold value, if it is less than the threshold value, the data in the adjacent boxes are merged; if it is greater than the threshold value, the data in the adjacent boxes are stopped from being merged; the threshold value is the critical value of the chi-square distribution confirmed according to the degrees of freedom and the significance level, the significance level is 0.05, the degrees of freedom = (r-1) × (c-1), where r is the number of rows in the contingency table, and c is the number of columns in the contingency table.

4. A flow distribution method based on multi-dimensional information fitting as claimed in claim 3, characterized in that: After stopping the merging of bins in steps 3-4, the WOE of each bin is calculated d Value and IV d value, where d represents the bin number, The IV d =(Renewal rate - Non-renewal rate)*WOE d WOE of each bin d Value and IV d The values ​​can be obtained through Python's own functions. If WOE d As the bin number increases, IV d In the interval [0.02, 0.5], the binning results are robust; In the steps 3-4, the number of sub-boxes is preferably combined into 8.

5. A flow distribution method based on multi-dimensional information fitting as claimed in claim 4, characterized in that: In step 4, the use of the least squares method specifically refers to using the polynomial function polyfit(x,y,3) of Matlab to implement the least squares method.

6. A flow distribution method based on multi-dimensional information fitting as claimed in claim 5, characterized in that: In step 5, the specific renewal rate is substituted into the corresponding value, and if the value is a decimal, the value is rounded down; In step 5, if no corresponding value exists when a specific renewal rate is substituted, the specific renewal rate is compared with the renewal rates of all sub-boxes. If the renewal rate is greater than the renewal rates of all sub-boxes, the maximum value is selected from the minimum value of the function and the maximum value of the function; if the polynomial function does not have a minimum value, the value corresponding to the maximum value of the function is taken; In step 5, if there are at least 2 values ​​when substituting into a specific renewal rate, a value is selected according to the type of monotonic interval into which the value falls; if a value falls within the monotonic increasing interval of the function, the average of these values ​​is taken as the standard flow value; if no value falls within the monotonic increasing interval of the function, the average of the values ​​falling within the monotonic decreasing interval of the function is taken as the standard flow value.

Images