Intelligent application channel dynamic distribution and automatic income management system

Abstract

The invention belongs to the technical field of intelligent enterprise service, and discloses an intelligent application channel dynamic distribution and automatic income management system. Comprising an application obtaining module used for obtaining a user application file; the format auditing module is used for performing format auditing on the user application file and judging whether a format exception instruction is generated or not; the qualification auditing module is used for performing qualification auditing on the user application file if the format exception instruction is not generated, and judging whether a qualification coincidence instruction is generated or not; the plan matching module is used for extracting application feature data from the user application file and matching a corresponding support plan if the generation qualification conformity instruction is generated; the resource allocation module is used for analyzing the application feature data and the support plan and dynamically allocating a support amount; according to the invention, automatic processing and dynamic resource management of the support application are realized, the examination and approval efficiency and the accuracy of resource allocation are improved, and thus the quality and effect of the support service are optimized.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent enterprise services. More specifically, the present invention relates to an intelligent application channel dynamic distribution and automated revenue management system. Background Art

[0002] With the rapid development of technology, especially the wide application of artificial intelligence and automation technologies, all industries are facing higher requirements for efficiency and service quality. In this context, the demand for support services is increasing day by day. Especially during the critical period of economic transformation and social development, how to efficiently and accurately meet the support needs of specific groups has become an urgent problem to be solved. The traditional support application process often relies on manual review, which has problems such as slow processing speed, high error rate, and information asymmetry, resulting in many eligible applicants facing long waits and unnecessary troubles. In addition, due to the limitations of manual operations, the distribution of support resources often shows unevenness. To solve these problems, it is particularly important to develop an intelligent system integrating advanced artificial intelligence technologies and automated processes, aiming to improve the processing speed and accuracy of support applications, and through intelligent matching and dynamic adjustment, improve the effectiveness of resource allocation, and achieve dynamic distribution and management of support resources.

[0003] However, although the existing technologies can achieve intelligent support services, they mainly rely on policy matching and push functions. For example, the patent with the publication number CN114118946A discloses an intelligent policy management method and system, including: collecting and obtaining policy documents issued by governments at all levels; multi-dimensionally refining the keywords of the obtained policy documents; automatically matching the refined keywords to obtain preliminary policy structured data; manually reviewing to form formatted policy structure data; data comparison to obtain the matching degree; pushing enterprises with a matching score exceeding the minimum threshold of different types of policies via text messages and in-site messages. Although this invention can effectively collect, analyze, and push policies to improve the actual effect of policy support, it does not involve the automatic approval of support information and the dynamic management of support resources, and has the following deficiencies:

[0004] (1) Low approval efficiency: The lack of an automatic approval mechanism will slow down the application processing process, increase the burden of manual review, extend the application processing time, and thus reduce the efficiency of support services.

[0005] (2) Unreasonable resource allocation: The lack of an effective support resource management method will lead to uneven distribution of support resources, unable to be rationally allocated according to actual needs, affecting the support effect, and unable to quickly respond to support needs, thus reducing the quality of support services.

[0006] In view of this, the present invention proposes an intelligent application channel dynamic distribution and automated revenue management system to solve the above problems. Summary of the Invention

[0007] To overcome the above-mentioned defects of the prior art and to achieve the above object, the present invention provides the following technical solution: an intelligent application channel dynamic distribution and automated revenue management system, comprising:

[0008] An application acquisition module for acquiring user application files;

[0009] A format review module for performing a format review on the user application file to determine whether a format exception instruction is generated. If a format exception instruction is generated, a format suggestion is generated;

[0010] A qualification review module for, if no format exception instruction is generated, performing a qualification review on the user application file to determine whether a qualification compliance instruction is generated;

[0011] A plan matching module for, if a qualification compliance instruction is generated, performing a feature analysis on the user application file, extracting application feature data, calculating an application fit degree based on the application feature data, and matching a corresponding support plan according to the application fit degree;

[0012] A resource allocation module for performing a multi-dimensional analysis on the application feature data and the support plan, and dynamically allocating support amounts by using an improved solution space exploration algorithm;

[0013] A support distribution module for extracting user financial information from the user application file and distributing the support amount to the corresponding user according to the user financial information.

[0014] Further, the method for determining whether a format exception instruction is generated includes:

[0015] Using optical character recognition technology and named entity recognition technology to obtain user application information from the user application file. The user application information includes all text fields in the user application file; preset required fields, analyzing the text fields in the user application information to determine whether all required fields are included in the user application information; if not, generating a format exception instruction, if so, performing a format review on the user application information by using a preset regular expression set, and the regular expression set includes regular expressions corresponding to each text field in the user application information; if the format review passes, no format exception instruction is generated, if the format review fails, a format exception instruction is generated; the method for determining whether the format review passes is: respectively matching each text field in the user application information with the corresponding regular expression. If all text fields conform to the corresponding regular expressions, the format review passes, if there is a text field that does not conform to the corresponding regular expression, the format review fails;

[0016] The method for generating the format suggestion includes:

[0017] Define a template set, which includes the recommended templates corresponding to each text field in the user application information; mark the required fields not included in the user application information and the text fields in the user application information that do not conform to the corresponding regular expressions as recommended fields; obtain the recommended templates corresponding to each recommended field from the recommendation set and generate format recommendations.

[0018] Further, the method for determining whether to generate an eligibility compliance instruction includes:

[0019] Preset a support set, which includes the support requirements corresponding to each support plan; classify all the support requirements in the support set and all the text fields in the user application information into categorical variables, text variables, and numerical variables;

[0020] Set different numerical labels for each categorical variable and mark them as categorical labels; use a pre-trained word embedding model to perform vector conversion on each text variable and convert it into a corresponding text vector; use the categorical labels, text vectors, and numerical variables corresponding to all the text fields in the user application information, and the categorical labels, text vectors, and numerical variables corresponding to the support requirements of a support plan in the support set as a set of analysis data, and the analysis data corresponds to the support plan one by one; input each set of analysis data into a trained eligibility review model respectively to predict the corresponding review label; where the review label is the numerical label corresponding to the review result, and the numerical labels corresponding to different review results are all different, and the review results include review passed and review failed; the eligibility review model includes a application review models, where a is the number of support plans in the support set; the a application review models are all deep neural network models, and the training processes of the a application review models are all the same;

[0021] Analyze the review results corresponding to all the predicted review labels; if all the review results are review failed, do not generate an eligibility compliance instruction and return the user application file to the user; if there is a review result of review passed, generate an eligibility compliance instruction.

[0022] Further, mark the support plan corresponding to the review result of review passed as the passed plan, count the number of passed plans and mark it as the passed quantity; if the passed quantity is 1, match the corresponding support plan with the user application file; if the passed quantity is greater than 1, calculate the application fitness corresponding to each passed plan; compare each application fitness respectively, and match the passed plan with the largest application fitness with the user application file; the method for calculating the application fitness includes:

[0023] For each text field in the application feature data, group it with the support requirements corresponding to each adoption plan as a set of fields; if the text field corresponding to the set of fields is a categorical variable, calculate the first fit degree of the corresponding set of fields; if the text field corresponding to the set of fields is a text variable, calculate the second fit degree of the corresponding set of fields; if the text field corresponding to the set of fields is a numerical variable, calculate the third fit degree of the corresponding set of fields; preset a weight set, which includes a first set and a second set. The first set includes the weight coefficients corresponding to each text field in the application feature data, and the second set includes the weight coefficients corresponding to categorical variables, text variables, and numerical variables;

[0024] Group the sets of fields corresponding to the same adoption plan as a set of adoption sets; separately count the number of the first fit degree, the number of the second fit degree, and the number of the third fit degree corresponding to each set of adoption sets, and mark them as the first quantity, the second quantity, and the third quantity respectively; multiply each first fit degree corresponding to each set of adoption sets by the corresponding weight coefficient in the first set, add them up in sequence, and then divide by the corresponding first quantity to obtain the first mean value; multiply each second fit degree corresponding to each set of adoption sets by the corresponding weight coefficient in the first set, add them up in sequence, and then divide by the corresponding second quantity to obtain the second mean value; multiply each third fit degree corresponding to each set of adoption sets by the corresponding weight coefficient in the first set, add them up in sequence, and then divide by the corresponding third quantity to obtain the third mean value; multiply the first mean value corresponding to each set of adoption sets by the corresponding weight coefficient in the second set to obtain the first total value; multiply the second mean value corresponding to each set of adoption sets by the corresponding weight coefficient in the second set to obtain the second total value; multiply the third mean value corresponding to each set of adoption sets by the corresponding weight coefficient in the second set to obtain the third total value; subtract the corresponding first total value and the corresponding third total value from the second total value corresponding to each set of adoption sets to obtain the application fit degree corresponding to each adoption plan.

[0025] Further, the calculation method of the first fit degree includes:

[0026] Subtract the classification label corresponding to the support requirement from the classification label corresponding to the text field in each set of fields, and take the absolute value to obtain the first fit degree of each set of fields;

[0027] The calculation method of the second fit degree includes:

[0028] Calculate the dot product of the text vectors corresponding to the text fields in each field set and the text vectors corresponding to the corresponding support requirements, and label it as the vector dot product; calculate the norm corresponding to the text vectors of the text fields in each field set, and label it as the first norm; calculate the norm corresponding to the text vectors of the support requirements in each field set, and label it as the second norm; multiply the first norm of each field set by the corresponding second norm to obtain the norm product; multiply the vector dot product of each field set by the corresponding norm product to obtain the second degree of fit of each field set.

[0029] The calculation method of the vector dot product is as follows: multiply each dimension of the text vector corresponding to the text field by the corresponding dimension of the text vector corresponding to the corresponding support requirement to obtain the dimension product; add the products of each dimension in sequence to obtain the vector dot product.

[0030] The calculation method of the first norm is as follows: square each dimension of the text vector corresponding to the text field to obtain the dimension square; add the dimension squares in sequence and then take the square root to obtain the first norm; the calculation method of the second norm is the same as that of the first norm.

[0031] The calculation method of the third degree of fit includes:

[0032] Subtract the corresponding support requirements from the text fields in each field set respectively, and take the absolute value to obtain the absolute difference of each field set; compare the text fields in each field set with the corresponding support requirements to obtain the maximum value corresponding to each field set; divide the absolute difference of each field set by the corresponding maximum value to obtain the third degree of fit of each field set.

[0033] Further, the steps for dynamically allocating support amounts include:

[0034] Step S101: Obtain the amount range according to the approval plan matched by the user application document.

[0035] Step S102: Construct an exploration solution set. The exploration solution set includes d candidate solutions, and each candidate solution corresponds one-to-one to the support amount in the amount range. The initial iteration count t corresponding to the exploration solution set is 0.

[0036] Step S103: Calculate the exploration probability of each candidate solution.

[0037] Step S104: Determine the exploration mode of each candidate solution according to the exploration probability.

[0038] Step S105: Update each candidate solution according to the exploration mode.

[0039] Step S106: Calculate the traction coefficient corresponding to each candidate solution.

[0040] Step S107: Update each candidate solution again based on the traction coefficient;

[0041] Step S108: Compare the iteration number t with the preset iteration threshold T. If t≥T, go to Step S109; if t<T, return to Step S103;

[0042] Step S109: Calculate the support benefit corresponding to each candidate solution, compare each support benefit respectively, and obtain the support amount corresponding to the candidate solution with the maximum support benefit.

[0043] Further, in the said Step S102, each candidate solution in the exploration solution set is defined in a one-dimensional solution space, and the range of the one-dimensional solution space is the amount interval; the generation method of each candidate solution is: generate d random numbers in the interval [0, 1], where d is an integer greater than 1; subtract the minimum value of the amount interval from the maximum value to obtain the interval difference; multiply each random number by the interval difference respectively, and then add the minimum value of the amount interval to generate each candidate solution; the generation method of the exploration speed corresponding to each candidate solution is: preset the speed interval, generate d random values in the interval [0, 1], subtract the minimum value of the speed interval from the maximum value to obtain the speed difference; multiply each random value by the speed difference respectively, and then add the minimum value of the speed interval to generate the exploration speed corresponding to each candidate solution;

[0044] In the said Step S104, the method for judging the exploration mode of each candidate solution includes:

[0045] Preset the exploration threshold, and compare the exploration probability of each candidate solution with the exploration threshold respectively; if the exploration probability is less than the exploration threshold, the exploration mode of the corresponding candidate solution is local exploration; if the exploration probability is greater than or equal to the exploration threshold, the exploration model of the corresponding candidate solution is global exploration.

[0046] Further, in the said Step S103, the method for calculating the exploration probability of each candidate solution includes:

[0047] Calculate the support benefit corresponding to each candidate solution, compare each support benefit respectively, take the support benefit with the largest value as the maximum benefit, and take the support benefit with the smallest value as the minimum benefit; subtract the minimum benefit from the maximum benefit to obtain the benefit difference; subtract the support benefit corresponding to each candidate solution from the maximum benefit respectively to obtain the benefit range corresponding to each candidate solution; divide the benefit range corresponding to each candidate solution by the benefit difference respectively to obtain the exploration probability of each exploration solution;

[0048] The calculation method of the support benefit includes:

[0049] The subsidy amount corresponding to the candidate solution, the classification labels, text vectors, and numerical variable groups corresponding to all text fields in the application feature data are grouped into a set of calculation data, and the calculation data corresponds to the subsidy amount one by one; the calculation data is respectively input into the trained benefit calculation model to predict the corresponding benefit parameters; among them, the benefit calculation model includes an employment calculation model, a revenue calculation model, and a market calculation model, and each model in the benefit calculation model is a deep neural network model, and the training process is the same as that of the application review model; the benefit parameters include employment growth rate, revenue growth rate, and market competitiveness; a coefficient set is preset, and the coefficient set includes the proportionality coefficients corresponding to each parameter in the benefit parameters; each parameter in the benefit parameters is multiplied by the corresponding proportionality coefficient in the coefficient set and added in sequence to obtain the subsidy benefit.

[0050] Further, in step S105, the method for updating each candidate solution includes:

[0051] If the exploration mode is local exploration, the method for updating the candidate solution includes:

[0052] Preset an exploration factor, and the exploration factor includes a step factor, an adjustment factor, and an influence factor; update the exploration speed of each candidate solution and mark it as the updated speed; generate a corresponding neighborhood solution set for each candidate solution, and the neighborhood solution set includes the candidate solution and g candidate solutions adjacent to the candidate solution; compare the subsidy benefits corresponding to each candidate solution in each neighborhood solution set, and mark the candidate solution with the largest subsidy benefit as the local optimal solution corresponding to each neighborhood solution set; subtract each candidate solution from the corresponding local optimal solution and multiply by the step factor to obtain the first step weight; generate 1 random number in the interval [0, 0.5] and mark it as the first coefficient; multiply the first coefficient by each updated speed and then by the adjustment factor to obtain the first adjustment weight; subtract the corresponding previous solution from each candidate solution and multiply by the influence factor to obtain the first influence weight; the previous solution is the corresponding subsidy amount of the candidate solution in the previous iteration process.

[0053] Add the corresponding first step weight, first adjustment weight, and first influence weight to each candidate solution to obtain the updated solution corresponding to each candidate solution, and the updated solution is the updated candidate solution.

[0054] The method for updating the exploration speed of each candidate solution is: generate 1 random number in the interval [0, 1] and mark it as the second coefficient; compare the subsidy benefits corresponding to each candidate solution in the exploration solution set, and mark the candidate solution with the largest subsidy benefit as the best solution; subtract each candidate solution from the best solution and multiply by the adjustment factor to obtain the second adjustment weight; multiply the second coefficient by each exploration speed and then add the corresponding second adjustment weight to obtain the updated speed corresponding to each candidate solution.

[0055] If the exploration mode is global exploration, the methods for updating candidate solutions include:

[0056] Subtract each candidate solution from the best solution, and multiply by the step factor to obtain the second step weight; generate 1 random number in the interval [0, 1], and mark it as the third coefficient; multiply the third coefficient by each update speed, and multiply by the adjustment factor to obtain the third adjustment weight; subtract the exploration speed corresponding to each candidate solution from the update speed of each candidate solution, and multiply by the influence factor to obtain the second influence weight; add the corresponding second step weight, third adjustment weight, and second influence weight to each candidate solution to obtain the updated solution corresponding to each candidate solution.

[0057] Further, in the step S106, the method for calculating the traction coefficient corresponding to each candidate solution includes:

[0058] Subtract the support benefit of each adjacent solution from the support benefit of the corresponding candidate solution in each neighborhood solution set corresponding to each candidate solution, and take the absolute value to obtain the absolute benefit, where the adjacent solution is g candidate solutions adjacent to the candidate solution in the neighborhood solution set; take the opposite number of each absolute benefit, and perform an exponential operation with the natural constant to obtain the traction coefficient of each adjacent solution in the neighborhood solution set where each candidate solution is located for the candidate solution; among them, the opposite number of the absolute benefit is the exponent, and the natural constant is the base.

[0059] In the step S107, the method for updating each candidate solution again includes:

[0060] Subtract each adjacent solution in each neighborhood solution set from the corresponding candidate solution, multiply by the corresponding traction coefficient, and then multiply by the influence factor to obtain the traction ratio corresponding to each candidate solution; add up the traction ratios corresponding to each candidate solution in turn, and then add the corresponding candidate solution to obtain the re-updated solution; the re-updated solution is the candidate solution after being updated again.

[0061] The technical effects and advantages of the intelligent application channel dynamic distribution and automated revenue management system of the present invention:

[0062] Through an intelligent format review and eligibility review mechanism, automatically complete the form and content review of user application documents, improving the approval efficiency; through the calculation of application fit, automatically match the most suitable support plan, improving the application success rate; adopt a solution space exploration algorithm to dynamically allocate support funds, realizing the precise matching and effective utilization of resources, improving the pertinence and satisfaction of support services; realize the automated processing of support applications and dynamic resource management, improving the approval efficiency and the precision of resource allocation, being able to quickly respond to user needs, reducing the burden of manual review, thereby optimizing the quality and effect of support services and meeting the support needs of specific groups during the economic transformation period. Description of the Drawings

[0063] Figure 1 Flow chart of the intelligent application channel dynamic distribution and automated revenue management system according to Embodiment 1 of the present invention;

[0064] Figure 2 Flow chart of the method for dynamically allocating support funds according to Embodiment 1 of the present invention. Specific implementation manners

[0065] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0066] Embodiment 1

[0067] Please refer to Figure 1 As shown, the intelligent application channel dynamic distribution and automated revenue management system in this embodiment includes an application acquisition module, a format review module, a qualification review module, a plan matching module, a resource allocation module, and a support distribution module; each module is connected by wired and / or wireless means to achieve data transmission between modules.

[0068] The application acquisition module is used to acquire user application files.

[0069] The user application file is a relevant file submitted by the user for applying for plan support, including the user's name, contact information, enterprise name, qualification certificate, project name, project content, project objectives, etc.; the user application file is obtained after the user submits it to the system by himself / herself.

[0070] The format review module is used to review the format of the user application file, determine whether a format exception instruction is generated, and if a format exception instruction is generated, generate a format suggestion.

[0071] The method for determining whether a format exception instruction is generated includes:

[0072] Using optical character recognition technology and named entity recognition technology, obtain user application information from the user application file. The user application information includes all text fields in the user application file; preset required fields, which are preset by those skilled in the art according to the requirements of the support application. Examples of required fields include user name, enterprise name, etc.; analyze the text fields in the user application information to determine whether all required fields are included in the user application information; if not, generate a format exception instruction. If so, use a preset regular expression set to perform a format review on the user application information. The regular expression set includes regular expressions corresponding to each text field in the user application information, and the regular expression set is preset by those skilled in the art according to the actual situation; if the format review passes, do not generate a format exception instruction. If the format review fails, generate a format exception instruction; the method for determining whether the format review passes is: match each text field in the user application information with the corresponding regular expression. If all text fields conform to the corresponding regular expressions, the format review passes. If there are text fields that do not conform to the corresponding regular expressions, the format review fails; it should be noted that optical character recognition technology and named entity recognition technology are existing technologies, and the specific process will not be elaborated here.

[0073] The method for generating format suggestions includes:

[0074] Define a template set, which includes suggestion templates corresponding to each text field in the user application information. The template set is preset by those skilled in the art according to the requirements of the support application; examples of suggestion templates are: for the enterprise name, the suggestion template is: it is recommended to use Chinese characters or letters, with a length not exceeding 20 characters. For the contact information, the suggestion template is: ensure that the input phone number is 11 digits. For the email, the suggestion template is: ensure that it contains the "@" symbol and a valid domain name; mark the required fields not included in the user application information and the text fields in the user application information that do not conform to the corresponding regular expressions as suggested fields; obtain the suggestion templates corresponding to each suggested field from the suggestion set and generate format suggestions.

[0075] The eligibility review module is used to, if no format exception instruction is generated, perform an eligibility review on the user application file to determine whether an eligibility compliance instruction is generated.

[0076] The method for determining whether an eligibility compliance instruction is generated includes:

[0077] Preset support set, where the support requirements corresponding to each support plan are included in the support set. The support requirements are, for example, project scale (such as the number of participants, funding requirements, etc.), project nature (such as innovative nature, environmental protection nature, etc.), etc. The support set is pre-set by technicians in the field by collecting the support information set by enterprises themselves; all support requirements in the support set and all text fields in the user application information are divided into categorical variables, text variables, and numerical variables; categorical variables are variables with a finite number of discrete categories, usually representing specific classifications or types, such as project type, project nature, etc.; text variables are free-form variables, usually without a clear classification, such as enterprise name, user name, project goal, etc.; numerical variables are variables represented by numerical values, usually having a clear mathematical meaning, such as the number of participants, funding requirements, etc.

[0078] Set different digital labels for each categorical variable and mark them as categorical labels; use a pre-trained word embedding model (such as Word2Vec, GloVe, FastText, etc.) to perform vector conversion on each text variable and convert it into the corresponding text vector; the categorical labels, text vectors, and numerical variables corresponding to all text fields in the user application information, and the categorical labels, text vectors, and numerical variables corresponding to the support requirements of a support plan in the support set are used as a set of analysis data, and the analysis data corresponds to the support plan one by one; each set of analysis data is respectively input into the trained eligibility review model to predict the corresponding review label; among them, the review label is the digital label corresponding to the review result, and the digital labels corresponding to different review results are all different. The review results include review passed and review not passed; the eligibility review model includes a application review models, where a is the number of support plans in the support set; the a application review models are all deep neural network models, and the training processes of the a application review models are all the same.

[0079] Analyze the review results corresponding to all the predicted review labels; if all the review results are review not passed, no eligibility compliance instruction is generated, and the user application file is returned to the user; if there is a review result of review passed, an eligibility compliance instruction is generated.

[0080] The training process of the application review model includes:

[0081] Collect b groups of analysis data in advance, set corresponding audit labels for the b groups of analysis data, where b is an integer greater than 1, and convert the analysis data and the corresponding audit labels into a corresponding set of feature vectors; the audit labels corresponding to the analysis data are collected by technical personnel in the process of historically judging whether to generate qualification compliance instructions, collect b groups of analysis data, evaluate the corresponding user application information according to the support requirements in each group of analysis data, review whether the user application information meets the corresponding support requirements, and set corresponding audit labels according to the audit results, and set corresponding audit labels for the b groups of analysis data in turn;

[0082] Each set of feature vectors is used as the input of the application review model. The application review model uses a set of predicted review labels corresponding to each set of analysis data as output, and uses the actual review labels corresponding to each set of analysis data as the prediction target. The actual review labels are the pre-set review labels corresponding to the analysis data. The training goal is to minimize the sum of the prediction errors of all analysis data. The calculation formula for the prediction error is η w =(θ w -ε w ) 2 , where η w is the prediction error, w is the group number of the eigenvector corresponding to the analyzed data, θ w is the predicted review label corresponding to the wth group of analysis data, ε w is the actual review label corresponding to the wth group of analysis data; the application review model is trained until the sum of the prediction errors reaches convergence and the training is stopped.

[0083] The plan matching module is used to perform feature analysis on the user's application documents if a qualification compliance instruction is generated, extract application feature data, calculate the application compatibility based on the application feature data, and match the corresponding support plan according to the application compatibility.

[0084] The application feature data is the key data that describes the application project and its background, and provides support for matching the support plan. The application feature data includes project name, project content, project goal, etc. The support plan corresponding to the audit result of passed is marked as a passed plan, and the number of passed plans is counted and marked as the passed number; if the passed number is 1, the corresponding support plan is matched with the user's application document; if the passed number is greater than 1, the application feature data is matched with the support requirements corresponding to each passed plan in a multi-dimensional manner, and the application fit corresponding to each passed plan is calculated; each application fit is compared separately, and the passed plan with the largest application fit is matched with the user's application document; it should be understood that since the generation of qualifications meets the instructions, it means that there is a support plan that matches the user's application document, so the passed number does not exist. The situation of being less than 1.

[0085] The method for calculating the application compliance includes:

[0086] Take each text field in the application feature data and each support requirement corresponding to each approval plan as a set of fields; Exemplarily, the application feature data includes text field A and text field B, the approval plan A includes support requirement A and support requirement B, the approval plan B includes support requirement C and support requirement D, text field A corresponds to support requirement A and support requirement C respectively, and text field B corresponds to support requirement B and support requirement D respectively. Therefore, text field A and support requirement A are taken as a set of fields, text field A and support requirement C are taken as a set of fields, text field B and support requirement B are taken as a set of fields, and text field B and support requirement D are taken as a set of fields; If the text field corresponding to the field set is a categorical variable, calculate the first compliance of the corresponding field set; If the text field corresponding to the field set is a text variable, calculate the second compliance of the corresponding field set; If the text field corresponding to the field set is a numerical variable, calculate the third compliance of the corresponding field set; Preset a weight set, which includes a first set and a second set. The first set includes the weight coefficients corresponding to each text field in the application feature data, and the second set includes the weight coefficients corresponding to categorical variables, text variables, and numerical variables; The weight set is preset by those skilled in the art according to actual compliance requirements;

[0087] Take the field sets corresponding to the same approval plan as a set of approval sets; Count the number of the first compliance corresponding to each set of approval sets and mark it as the first quantity; Count the number of the second compliance corresponding to each set of approval sets and mark it as the second quantity; Count the number of the third compliance corresponding to each set of approval sets and mark it as the third quantity; Multiply each first compliance corresponding to each set of approval sets by the corresponding weight coefficient in the first set, add them up in sequence, and then divide by the corresponding first quantity to obtain the first mean value corresponding to each set of approval sets; Multiply each second compliance corresponding to each set of approval sets by the corresponding weight coefficient in the first set, add them up in sequence, and then divide by the corresponding second quantity to obtain the second mean value corresponding to each set of approval sets; Multiply each third compliance corresponding to each set of approval sets by the corresponding weight coefficient in the first set, add them up in sequence, and then divide by the corresponding third quantity to obtain the third mean value corresponding to each set of approval sets; Multiply the first mean value corresponding to each set of approval sets by the corresponding weight coefficient in the second set to obtain the first total value; Multiply the second mean value corresponding to each set of approval sets by the corresponding weight coefficient in the second set to obtain the second total value; Multiply the third mean value corresponding to each set of approval sets by the corresponding weight coefficient in the second set to obtain the third total value; Subtract the corresponding first total value and the corresponding third total value from the second total value corresponding to each set of approval sets to obtain the application compliance corresponding to each approval plan.

[0088] The calculation method of the first degree of fit includes:

[0089] Subtract the classification label corresponding to each text field in each field set from the classification label corresponding to the corresponding support requirement, take the absolute value, and obtain the first degree of fit of each field set.

[0090] The calculation method of the second degree of fit includes:

[0091] Calculate the dot product of the text vectors corresponding to the text fields in each field set and the text vectors corresponding to the corresponding support requirements, and mark it as the vector dot product; calculate the norm corresponding to the text vectors of the text fields in each field set, and mark it as the first norm; calculate the norm corresponding to the text vectors of the support requirements in each field set, and mark it as the second norm; multiply the first norm of each field set by the corresponding second norm to obtain the norm product; multiply the vector dot product of each field set by the corresponding norm product to obtain the second degree of fit of each field set.

[0092] The calculation method of the vector dot product is: multiply each dimension of the text vector corresponding to the text field by the corresponding dimension of the text vector corresponding to the corresponding support requirement to obtain the dimension product; add the products of each dimension in sequence to obtain the vector dot product.

[0093] The calculation method of the first norm is: square each dimension of the text vector corresponding to the text field to obtain the dimension square; add the dimension squares in sequence and then take the square root to obtain the first norm; the calculation method of the second norm is the same as that of the first norm.

[0094] The calculation method of the third degree of fit includes:

[0095] Subtract each text field in each field set from the corresponding support requirement, take the absolute value, and obtain the absolute difference of each field set; compare each text field in each field set with the corresponding support requirement to obtain the maximum value corresponding to each field set; divide the absolute difference of each field set by the corresponding maximum value to obtain the third degree of fit of each field set.

[0096] The resource allocation module is used to perform multi-dimensional analysis on the application feature data and the support plan, and dynamically allocate the support amount by using the improved solution space exploration algorithm.

[0097] The steps of dynamically allocating the support amount include:

[0098] Step S101: Obtain the amount range according to the approval plan matched by the user application file;

[0099] Step S102: Construct an exploration solution set. The exploration solution set includes d candidate solutions, and each candidate solution corresponds one-to-one with the support amount in the amount interval. The corresponding initial iteration number t of the exploration solution set is 0;

[0100] Step S103: Calculate the exploration probability of each candidate solution;

[0101] Step S104: Determine the exploration mode of each candidate solution according to the exploration probability;

[0102] Step S105: Update each candidate solution according to the exploration mode;

[0103] Step S106: Calculate the traction coefficient corresponding to each candidate solution;

[0104] Step S107: Update each candidate solution again based on the traction coefficient;

[0105] Step S108: Compare the iteration number t with the preset iteration threshold T. If t≥T, go to Step S109. If t<T, return to Step S103;

[0106] Step S109: Calculate the support benefit corresponding to each candidate solution, compare each support benefit respectively, and obtain the support amount corresponding to the candidate solution with the maximum support benefit.

[0107] In the above Step S101, the amount interval is obtained by those skilled in the art by referring to the relevant information of the adoption plan set by the enterprise itself.

[0108] In the above Step S102, each candidate solution in the exploration solution set is defined in a one-dimensional solution space, and the range of the one-dimensional solution space is the amount interval; the generation method of each candidate solution is as follows: generate d random numbers in the interval [0,1], where d is an integer greater than 1; subtract the minimum value from the maximum value of the amount interval to obtain the interval difference; multiply each random number by the interval difference and then add the minimum value of the amount interval to generate each candidate solution; the generation method of the exploration speed corresponding to each candidate solution is as follows: preset a speed interval, and the speed interval is preset by those skilled in the art according to the actual exploration accuracy requirements; generate d random values in the interval [0,1], subtract the minimum value from the maximum value of the speed interval to obtain the speed difference; multiply each random value by the speed difference and then add the minimum value of the speed interval to generate the exploration speed corresponding to each candidate solution.

[0109] In the above Step S103, the method for calculating the exploration probability of each candidate solution includes:

[0110] Calculate the support benefits corresponding to each candidate solution, compare each support benefit separately, take the support benefit with the largest value as the maximum benefit, and take the support benefit with the smallest value as the minimum benefit; subtract the minimum benefit from the maximum benefit to obtain the benefit difference; subtract the support benefits corresponding to each candidate solution from the maximum benefit respectively to obtain the benefit range corresponding to each candidate solution; divide the benefit range corresponding to each candidate solution by the benefit difference to obtain the exploration probability of each exploration solution.

[0111] The calculation method of the support benefit includes:

[0112] Group the support amount corresponding to the candidate solution, the classification labels, text vectors, and numerical variable groups corresponding to all text fields in the application feature data as a set of calculation data, and the calculation data corresponds to the support amount one by one; input the calculation data into the trained benefit calculation model respectively to predict the corresponding benefit parameters; among them, the benefit calculation model includes an employment calculation model, a revenue calculation model, and a market calculation model, and each model in the benefit calculation model is a deep neural network model, and the training process is the same as that of the application review model; the benefit parameters include employment growth rate, revenue growth rate, and market competitiveness; the employment calculation model is used to predict the employment growth rate, the revenue calculation model is used to predict the revenue growth rate, and the market calculation model is used to predict the market competitiveness; those skilled in the art preset a coefficient set according to the planned target orientation corresponding to the approval plan matched by the user application file, and the coefficient set includes the proportional coefficients corresponding to each parameter in the benefit parameters; for example, if the planned target orientation is to promote employment, the proportional coefficient of the employment growth rate is higher, if the planned target orientation is economic growth, the proportional coefficient of the revenue growth rate is higher, and if the planned target orientation is industrial upgrading and long-term competitiveness, the proportional coefficient of the market competitiveness is higher; multiply each parameter in the benefit parameters by the corresponding proportional coefficient in the coefficient set and add them up in turn to obtain the support benefit.

[0113] In the above step S104, the method for judging the exploration mode of each candidate solution includes:

[0114] Preset an exploration threshold, which is preset by those skilled in the art according to the actual situation; compare the exploration probability of each candidate solution with the exploration threshold respectively; if the exploration probability is less than the exploration threshold, the exploration mode of the corresponding candidate solution is local exploration; if the exploration probability is greater than or equal to the exploration threshold, the exploration model of the corresponding candidate solution is global exploration;

[0115] In the above step S105, the method for updating each candidate solution according to the exploration mode includes:

[0116] If the exploration mode is local exploration, the method for updating the candidate solution includes:

[0117] A preset exploration factor, which includes a step factor, an adjustment factor, and an influence factor. The exploration factor is preset by those skilled in the art according to the actual exploration accuracy requirements;

[0118] Update the exploration speed of each candidate solution and mark it as the updated speed; generate a corresponding neighborhood solution set for each candidate solution, where the neighborhood solution set includes the candidate solution and g candidate solutions adjacent to the candidate solution, 1 < g < d; compare the support benefits corresponding to each candidate solution in each neighborhood solution set, and mark the candidate solution with the maximum support benefit as the local optimal solution corresponding to each neighborhood solution set; subtract each candidate solution from the corresponding local optimal solution and multiply by the step factor to obtain the first step weight; generate 1 random number in the interval [0, 0.5] and mark it as the first coefficient; multiply the first coefficient by each updated speed and multiply by the adjustment factor to obtain the first adjustment weight; subtract the corresponding previous solution from each candidate solution and multiply by the influence factor to obtain the first influence weight; the previous solution is the corresponding support amount of the candidate solution in the previous iteration process;

[0119] Add the corresponding first step weight, first adjustment weight, and first influence weight to each candidate solution to obtain the updated solution corresponding to each candidate solution, and the updated solution is the updated candidate solution.

[0120] The method for updating the exploration speed of each candidate solution is: generate 1 random number in the interval [0, 1] and mark it as the second coefficient; compare the support benefits corresponding to each candidate solution in the exploration solution set, and mark the candidate solution with the maximum support benefit as the best solution; subtract each candidate solution from the best solution and multiply by the adjustment factor to obtain the second adjustment weight; multiply the second coefficient by each exploration speed and then add the corresponding second adjustment weight to obtain the updated speed corresponding to each candidate solution.

[0121] It should be noted that the local exploration is guided by the local optimal solution in the neighborhood solution set to ensure that the candidate solution can perform a fine search near the local optimal solution and improve the local exploration ability.

[0122] If the exploration mode is global exploration, the method for updating the candidate solution includes:

[0123] Subtract each candidate solution from the best solution and multiply by the step factor to obtain the second step weight; generate 1 random number in the interval [0, 1] and mark it as the third coefficient; multiply the third coefficient by each updated speed and multiply by the adjustment factor to obtain the third adjustment weight; subtract the corresponding exploration speed from the updated speed of each candidate solution and multiply by the influence factor to obtain the second influence weight; add the corresponding second step weight, third adjustment weight, and second influence weight to each candidate solution to obtain the updated solution corresponding to each candidate solution;

[0124] It should be noted that the global exploration is guided by the best solution in the exploration solution set, and is used to explore new spaces in the one-dimensional solution space, avoid local optima, and improve the global exploration ability.

[0125] In the above step S106, the method for calculating the traction coefficient corresponding to each candidate solution includes:

[0126] For each candidate solution, subtract the support benefit of each adjacent solution in the corresponding neighborhood solution set from the support benefit of the candidate solution, take the absolute value to obtain the absolute benefit, where the adjacent solutions are g candidate solutions adjacent to the candidate solution in the neighborhood solution set; take the opposite of each absolute benefit and perform an exponential operation with the natural constant to obtain the traction coefficient of each adjacent solution in the neighborhood solution set of the candidate solution for the candidate solution; where the opposite of the absolute benefit is the exponent and the natural constant is the base.

[0127] It should be noted that the exponential decay function is used to measure the mutual traction force between candidate solutions. The smaller the difference in support benefits between candidate solutions, the larger the traction coefficient; conversely, the larger the difference in support benefits between candidate solutions, the smaller the traction coefficient. By introducing the traction coefficient, it is possible to prevent candidate solutions from converging to the same support amount in the one-dimensional solution space, thereby enhancing the global search ability of the solution space exploration algorithm and reducing the risk of premature convergence.

[0128] In the above step S107, the method for updating each candidate solution again includes:

[0129] Subtract each adjacent solution in each neighborhood solution set from the corresponding candidate solution, multiply by the corresponding traction coefficient, and then multiply by the influence factor to obtain the traction ratio corresponding to each candidate solution; add up the traction ratios corresponding to each candidate solution in sequence, and then add the corresponding candidate solution to obtain the re-updated solution; the re-updated solution is the candidate solution after being updated again.

[0130] The support distribution module is used to extract the user's financial information from the user application file and distribute the support amount to the corresponding user according to the user's financial information.

[0131] The user's financial information includes, for example, the bank account name, bank account number, and opening bank.

[0132] In this embodiment, through an intelligent format review and eligibility review mechanism, the form and content review of the user application documents are automatically completed, improving the approval efficiency; through the calculation of the application fit, the most suitable support plan is automatically matched, improving the application success rate; the solution space exploration algorithm is used to dynamically allocate support funds, realizing the precise matching and effective utilization of resources, improving the pertinence and satisfaction of the support service; the automation processing of support applications and dynamic resource management are realized, improving the approval efficiency and the accuracy of resource allocation, being able to quickly respond to user needs, reducing the burden of manual review, thereby optimizing the quality and effect of the support service and meeting the support needs of specific groups during the economic transformation period.

[0133] Embodiment 2

[0134] This application also provides an electronic device. The electronic device may include one or more processors and one or more memories. Among them, computer-readable code is stored in the memory, and when the computer-readable code is run by one or more processors, it can execute the intelligent application channel dynamic distribution and automated revenue management system as described above.

[0135] The method or system according to the embodiment of the present application can also be implemented by means of the architecture of the electronic device shown in the present application. The electronic device may include a bus, one or more CPUs, ROM, RAM, a communication port connected to the network, input / output, a hard disk, etc. The storage device in the electronic device, such as ROM or the hard disk, can store the intelligent application channel dynamic distribution and automated revenue management system provided by the present application. Further, the electronic device may also include a user interface. Of course, the architecture shown in the present application is only exemplary, and when implementing different devices, one or more components shown in the electronic device of the present application can be omitted according to actual needs.

[0136] Embodiment 3

[0137] One embodiment of the present application discloses a computer-readable storage medium. Computer-readable instructions are stored on the computer-readable storage medium. When the computer-readable instructions are run by a processor, the intelligent application channel dynamic distribution and automated revenue management system according to the embodiment of the present application described with reference to the above drawings can be executed. The storage medium includes but is not limited to, for example, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and cache memory, etc. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0138] In addition, according to an embodiment of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the present application provides a non-transitory machine-readable storage medium storing machine-readable instructions that can be run by a processor to execute instructions corresponding to the method steps provided by the present application, such as: an intelligent application channel dynamic distribution and automated revenue management system. When the computer program is executed by a central processing unit (CPU), the above functions defined in the method of the present application are executed.

[0139] As described above, the above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

[0140] Finally: The above are only the preferred embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should all be included in the protection scope of the present invention.

Claims

1. An intelligent application channel dynamic distribution and automated revenue management system, characterized in that, Including: An application acquisition module for acquiring user application files; A format review module for reviewing the format of user application files, determining whether to generate a format exception instruction, and if a format exception instruction is generated, generating a format suggestion; A qualification review module for, if no format exception instruction is generated, reviewing the qualifications of user application files to determine whether to generate a qualification compliance instruction; A plan matching module for, if a qualification compliance instruction is generated, analyzing the characteristics of user application files, extracting application feature data, calculating the application fit degree based on the application feature data, and matching a corresponding support plan according to the application fit degree; A resource allocation module for performing multi-dimensional analysis on the application feature data and the support plan, and dynamically allocating the support amount using an improved solution space exploration algorithm; A support distribution module for extracting user financial information from user application files and distributing the support amount to the corresponding users according to the user financial information.

2. The intelligent application channel dynamic distribution and automated revenue management system according to claim 1, wherein, The method for determining whether to generate a format exception instruction includes: Using optical character recognition technology and named entity recognition technology to obtain user application information from user application files, where the user application information includes all text fields in the user application files; presetting required fields, analyzing the text fields in the user application information to determine whether all required fields are included in the user application information; if not, generating a format exception instruction, if so, performing format review on the user application information using a preset regular expression set, where the regular expression set includes regular expressions corresponding to each text field in the user application information; if the format review passes, not generating a format exception instruction, if the format review fails, generating a format exception instruction; the method for determining whether the format review passes is: matching each text field in the user application information with the corresponding regular expression respectively, if all text fields meet the corresponding regular expressions, the format review passes, if there is a text field that does not meet the corresponding regular expression, the format review fails; The method for generating a format suggestion includes: Defining a template set, where the template set includes suggestion templates corresponding to each text field in the user application information; marking the required fields not included in the user application information and the text fields in the user application information that do not meet the corresponding regular expressions as suggested fields; obtaining the suggestion templates corresponding to each suggested field from the suggestion set and generating a format suggestion.

3. The intelligent application channel dynamic distribution and automated revenue management system according to claim 2, wherein The method for determining whether to generate a qualification compliance instruction includes: Presetting a support set, where the support set includes support requirements corresponding to each support plan; classifying all support requirements in the support set and all text fields in the user application information into categorical variables, text variables, and numerical variables; Different numerical tags are set for each categorical variable and marked as categorical labels; a pre-trained word embedding model is used to perform vector transformation on each text variable and convert it into a corresponding text vector; the categorical labels, text vectors, and numerical variables corresponding to all text fields in the user application information, together with the categorical labels, text vectors, and numerical variables corresponding to the support requirements of a support plan in the support set, are used as a set of analysis data, and the analysis data corresponds to the support plan one by one; each set of analysis data is respectively input into the trained eligibility review model to predict the corresponding review label; among them, the review label is the numerical label corresponding to the review result, and the numerical labels corresponding to different review results are all different, and the review results include review passed and review not passed; the eligibility review model includes a application review models, where a is the number of support plans in the support set; the a application review models are all deep neural network models, and the training processes of the a application review models are all the same; Analyze the review results corresponding to all the predicted review labels; if all the review results are review not passed, no eligibility compliance instruction is generated, and the user application file is returned to the user; if there is a review result of review passed, an eligibility compliance instruction is generated.

4. The intelligent application channel dynamic distribution and automated revenue management system according to claim 3, wherein, Mark the support plan corresponding to the review result of review passed as the passed plan, count the number of passed plans and mark it as the passed quantity; if the passed quantity is 1, match the corresponding support plan with the user application file; if the passed quantity is greater than 1, calculate the application fit degree corresponding to each passed plan; Compare each application fit degree respectively, and match the passed plan with the largest application fit degree with the user application file; The method for calculating the application fit degree includes: Each text field in the application feature data and the support requirements corresponding to each passed plan are used as a set of field sets; if the text field corresponding to the field set is a categorical variable, calculate the first fit degree of the corresponding field set; if the text field corresponding to the field set is a text variable, calculate the second fit degree of the corresponding field set; if the text field corresponding to the field set is a numerical variable, calculate the third fit degree of the corresponding field set; a preset weight set is set, and the weight set includes a first set and a second set. The first set includes the weight coefficients corresponding to each text field in the application feature data, and the second set includes the weight coefficients corresponding to the categorical variable, text variable, and numerical variable; Take the set of fields corresponding to the same adoption plan as a set of adoption sets; separately count the number of the first fitness degrees, the number of the second fitness degrees, and the number of the third fitness degrees corresponding to each set of adoption sets, and mark them as the first quantity, the second quantity, and the third quantity respectively; multiply each first fitness degree corresponding to each set of adoption sets by the corresponding weight coefficient in the first set, add them up in sequence, and then divide by the corresponding first quantity to obtain the first mean value; multiply each second fitness degree corresponding to each set of adoption sets by the corresponding weight coefficient in the first set, add them up in sequence, and then divide by the corresponding second quantity to obtain the second mean value; multiply each third fitness degree corresponding to each set of adoption sets by the corresponding weight coefficient in the first set, add them up in sequence, and then divide by the corresponding third quantity to obtain the third mean value; multiply the first mean value corresponding to each set of adoption sets by the corresponding weight coefficient in the second set to obtain the first total value; multiply the second mean value corresponding to each set of adoption sets by the corresponding weight coefficient in the second set to obtain the second total value; multiply the third mean value corresponding to each set of adoption sets by the corresponding weight coefficient in the second set to obtain the third total value; subtract the corresponding first total value from the second total value corresponding to each set of adoption sets, and then subtract the corresponding third total value to obtain the application fitness degree corresponding to each adoption plan.

5. The intelligent application channel dynamic distribution and automated revenue management system according to claim 4, wherein The calculation method of the first fitness degree includes: Subtract the classification label corresponding to the text field in each field set from the classification label corresponding to the corresponding support requirement, and take the absolute value to obtain the first fitness degree of each field set; The calculation method of the second fitness degree includes: Calculate the dot product of the text vector corresponding to the text field in each field set and the text vector corresponding to the corresponding support requirement, and mark it as the vector dot product; calculate the norm corresponding to the text vector of the text field in each field set, and mark it as the first norm; calculate the norm corresponding to the text vector of the support requirement in each field set, and mark it as the second norm; multiply the first norm of each field set by the corresponding second norm to obtain the norm product; multiply the vector dot product of each field set by the corresponding norm product to obtain the second fitness degree of each field set; The calculation method of the vector dot product is: multiply each dimension of the text vector corresponding to the text field by the corresponding dimension of the text vector corresponding to the corresponding support requirement to obtain the dimension product; add up the products of each dimension in sequence to obtain the vector dot product; The calculation method of the first norm is: square each dimension of the text vector corresponding to the text field to obtain the dimension square; add up the dimension squares in sequence and then take the square root to obtain the first norm; the calculation method of the second norm is the same as that of the first norm; The calculation method of the third fitness degree includes: Subtract the corresponding support requirement from the text field in each field set, and take the absolute value to obtain the absolute difference of each field set; compare the text field in each field set with the corresponding support requirement to obtain the maximum value corresponding to each field set; divide the absolute difference of each field set by the corresponding maximum value to obtain the third fitness degree of each field set.

6. The intelligent application channel dynamic distribution and automated revenue management system according to claim 5, characterized in that The steps for dynamically allocating support amounts include: Step S101: Obtain the amount range according to the approved plan matched with the user application document; Step S102: Construct an exploration solution set. The exploration solution set includes d candidate solutions, and each candidate solution corresponds one-to-one with the support amount in the amount range. The corresponding initial iteration count t of the exploration solution set is 0; Step S103: Calculate the exploration probability of each candidate solution; Step S104: Determine the exploration mode of each candidate solution according to the exploration probability; Step S105: Update each candidate solution according to the exploration mode; Step S106: Calculate the traction coefficient corresponding to each candidate solution; Step S107: Update each candidate solution again based on the traction coefficient; Step S108: Compare the iteration count t with the preset iteration threshold T. If t≥T, proceed to Step S109. If t<T, return to Step S103; Step S109: Calculate the support benefit corresponding to each candidate solution, compare each support benefit respectively, and obtain the support amount corresponding to the candidate solution with the maximum support benefit.

7. The intelligent application channel dynamic distribution and automated revenue management system according to claim 6, characterized in that, In the said Step S102, each candidate solution in the exploration solution set is defined in a one-dimensional solution space, and the range of the one-dimensional solution space is the amount range; the generation method of each candidate solution is: generate d random numbers in the interval [0,1], where d is an integer greater than 1; subtract the minimum value from the maximum value of the amount range to obtain the range difference; multiply each random number by the range difference respectively, and then add the minimum value of the amount range to generate each candidate solution; the generation method of the exploration speed corresponding to each candidate solution is: preset a speed range, generate d random values in the interval [0,1], subtract the minimum value from the maximum value of the speed range to obtain the speed difference; Multiply each random value by the speed difference respectively, and then add the minimum value of the speed range to generate the exploration speed corresponding to each candidate solution; In the said Step S104, the method for determining the exploration mode of each candidate solution includes: Preset an exploration threshold, and compare the exploration probability of each candidate solution with the exploration threshold respectively; If the exploration probability is less than the exploration threshold, the exploration mode of the corresponding candidate solution is local exploration; if the exploration probability is greater than or equal to the exploration threshold, the exploration model of the corresponding candidate solution is global exploration.

8. The intelligent application channel dynamic distribution and automated revenue management system according to claim 7, characterized in that In the said Step S103, the method for calculating the exploration probability of each candidate solution includes: Calculate the support benefit corresponding to each candidate solution, compare each support benefit respectively, take the support benefit with the largest value as the maximum benefit, and take the support benefit with the smallest value as the minimum benefit; subtract the minimum benefit from the maximum benefit to obtain the benefit difference; subtract the support benefit corresponding to each candidate solution from the maximum benefit respectively to obtain the benefit range difference corresponding to each candidate solution; divide the benefit range difference corresponding to each candidate solution by the benefit difference respectively to obtain the exploration probability of each exploration solution; The calculation method of the support benefit includes: The support amount corresponding to the candidate solution and the classification labels, text vectors, and numerical variable groups corresponding to all text fields in the application feature data are grouped into a set of calculation data, and the calculation data corresponds to the support amount one by one; the calculation data is respectively input into the trained benefit calculation model to predict the corresponding benefit parameters; among them, the benefit calculation model includes an employment calculation model, a revenue calculation model, and a market calculation model, and each model in the benefit calculation model is a deep neural network model, and the training process is the same as that of the application review model; the benefit parameters include employment growth rate, revenue growth rate, and market competitiveness; a coefficient set is preset, and the coefficient set includes the proportionality coefficients corresponding to each parameter in the benefit parameters; each parameter in the benefit parameters is multiplied by the corresponding proportionality coefficient in the coefficient set and added in sequence to obtain the support benefit.

9. The intelligent application channel dynamic distribution and automated revenue management system according to claim 8, wherein In the step S105, the method for updating each candidate solution includes: If the exploration mode is local exploration, the method for updating the candidate solution includes: Preset exploration factors, which include a step factor, an adjustment factor, and an influence factor; update the exploration speed of each candidate solution and mark it as the updated speed; generate a corresponding neighborhood solution set for each candidate solution, and the neighborhood solution set includes the candidate solution and g candidate solutions adjacent to the candidate solution; compare the support benefits corresponding to each candidate solution in each neighborhood solution set, and mark the candidate solution with the largest support benefit as the local optimal solution corresponding to each neighborhood solution set; subtract each candidate solution from the corresponding local optimal solution and multiply by the step factor to obtain the first step weight; generate 1 random number in the interval [0, 0.5] and mark it as the first coefficient; multiply the first coefficient by each updated speed and then by the adjustment factor to obtain the first adjustment weight; subtract the corresponding previous solution from each candidate solution and multiply by the influence factor to obtain the first influence weight; the previous solution is the support amount corresponding to the candidate solution in the previous iteration process. Add the corresponding first step weight, first adjustment weight, and first influence weight to each candidate solution to obtain the updated solution corresponding to each candidate solution, and the updated solution is the updated candidate solution. The method for updating the exploration speed of each candidate solution is: generate 1 random number in the interval [0, 1] and mark it as the second coefficient; compare the support benefits corresponding to each candidate solution in the exploration solution set, and mark the candidate solution with the largest support benefit as the best solution; subtract each candidate solution from the best solution and multiply by the adjustment factor to obtain the second adjustment weight; multiply the second coefficient by each exploration speed and then add the corresponding second adjustment weight to obtain the updated speed corresponding to each candidate solution. If the exploration mode is global exploration, the method for updating the candidate solution includes: Subtract each candidate solution from the optimal solution, and multiply by the step factor to obtain the second step weight; generate 1 random number in the interval [0,1] and label it as the third coefficient; multiply the third coefficient by each update speed and by the adjustment factor to obtain the third adjustment weight; subtract the exploration speed corresponding to each candidate solution from the update speed of each candidate solution, and multiply by the influence factor to obtain the second influence weight; add the corresponding second step weight, third adjustment weight, and second influence weight to each candidate solution to obtain the update solution corresponding to each candidate solution.

10. The intelligent application channel dynamic distribution and automated revenue management system according to claim 9, characterized in that In the step S106, the method for calculating the traction coefficient corresponding to each candidate solution includes: Subtract the support benefit of each candidate solution from the support benefit of each adjacent solution in the neighborhood solution set corresponding to each candidate solution, and take the absolute value to obtain the absolute benefit, where the adjacent solution is g candidate solutions adjacent to the candidate solution in the neighborhood solution set; take the opposite of each absolute benefit and perform an exponential operation with the natural constant to obtain the traction coefficient of each adjacent solution in the neighborhood solution set where each candidate solution is located for the candidate solution; among them, the opposite of the absolute benefit is the exponent, and the natural constant is the base; In the step S107, the method for updating each candidate solution again includes: Subtract each candidate solution from each adjacent solution in each neighborhood solution set, multiply by the corresponding traction coefficient, and then multiply by the influence factor to obtain the traction ratio corresponding to each candidate solution; add up the traction ratios corresponding to each candidate solution in sequence, and then add the corresponding candidate solution to obtain the re-update solution; the re-update solution is the candidate solution after being updated again.

Images