CN118313868A - Method, device and medium for grading potential value of government enterprise clients - Google Patents

Abstract

The invention provides a method, a device and a medium for grading potential value of government enterprise customers, wherein the method comprises the following steps: acquiring enterprise portrait data, client history subscription data and client industry history subscription data corresponding to government enterprise clients; processing the enterprise portrait data, the client history subscription data and the client industry history subscription data based on Airflow workflow management platform to obtain client portrait feature information; training a Support Vector Machine (SVM) model by adopting the customer portrait characteristic information to obtain a customer potential value prediction model; and grading the potential value of the government enterprise client based on the client potential value prediction model. The method, the device and the medium can solve the problem of how to identify high potential value clients from a huge government and enterprise client group in the prior art, and the related scheme is not available at present.

Description

Method, device and medium for grading potential value of government enterprise clients

Technical Field

The invention relates to the technical field of networks, in particular to a method, a device and a medium for grading potential values of government enterprise customers.

Background

In telecom operators, the main business market is divided into two major categories according To customer categories, one category is the mobile market/personal market, and is facing the public To provide services such as telephone or traffic, broadband, etc., namely, to customer (To customer) business. The other type is the government and enterprise market, and provides special line, cloud computing, seat and other services for government or enterprise units. In the branch companies of operators, whether province companies and city companies, county companies, employees of government and enterprise lines need to face the customer groups of enterprises or government units in the jurisdiction of the branch companies (such as Guangzhou city company just faces the government and enterprise customers of Guangzhou city), the customer groups are generally divided according to industries, such as the internet industry, the automobile industry, the medical industry, the cartoon industry and the like. How to screen out high potential value customers from a huge customer group is important, so that the competitive activities such as bidding or manpower resource allocation of customer managers are occupied and planned reasonably.

However, there is currently no relevant solution for how to identify high potential value customers from a vast group of government enterprise customers.

Disclosure of Invention

The technical problem to be solved by the invention is to provide a grading method, a grading device and a grading medium for potential value of government and enterprise customers aiming at the defects of the prior art, so as to solve the problem of how to identify high potential value customers from a huge government and enterprise customer group in the prior art.

In a first aspect, the present invention provides a method of ranking potential value of government and enterprise customers, the method comprising:

Acquiring enterprise portrait data, client history subscription data and client industry history subscription data corresponding to government enterprise clients;

Processing the enterprise portrait data, the client history subscription data and the client industry history subscription data based on Airflow workflow management platform to obtain client portrait feature information;

training a Support Vector Machine (SVM) model by adopting the customer portrait characteristic information to obtain a customer potential value prediction model;

And grading the potential value of the government enterprise client based on the client potential value prediction model.

Further, the enterprise portrait data includes enterprise public information, visit records and associated business machines, and the acquiring enterprise portrait data corresponding to the government enterprise clients specifically includes:

Acquiring enterprise public information of all known government and/or enterprise clients within a preset range through a third party service interface;

access records and associated business opportunities entered when the government and/or enterprise clients are visited in the field are obtained.

Further, the enterprise public information includes basic information, business activities that the company can engage in, registered capital, stakeholder information, industry scope, financial conditions, intellectual property rights, administrative permissions and penalties, legal litigation and arbitration, market share, and corporate governance.

Further, the workflow management platform based on Airflow processes the enterprise portrait data, the client history subscription data and the client industry history subscription data to obtain client portrait feature information, which specifically includes:

Defining a directed acyclic graph DAG based on Airflow workflow management platforms;

defining a plurality of tasks and execution sequences and dependency relationships among the plurality of tasks in the DAG, wherein the plurality of tasks comprise a data extraction task, a client portrait generation task and a text analysis processing task;

and executing the tasks according to the DAG to obtain the customer portrait characteristic information.

Further, the data extraction task is used for extracting an enterprise information table, a visit record table, an associated business table, a client history subscription table and a history subscription table of the client industry according to the enterprise portrait data, the client history subscription data and the client industry history subscription data;

The client portrait generation task is used for aggregating the enterprise information table, the visit record table, the associated business table, the client history subscription table and the history subscription table of the client industry, and selecting characteristics related to the client potential value prediction from the aggregated data to generate a client portrait;

The text analysis processing task is used for calling a Chinese word segmentation library jieba to segment the features in the customer portrait to extract text contents, and converting the text contents into word vectors to obtain the feature information of the customer portrait.

Further, the training of the support vector machine SVM model by using the customer portrait characteristic information to obtain a customer potential value prediction model specifically includes:

taking the customer portrait characteristic information as a data set, and dividing the data set into a training set, a verification set and a test set;

Training the SVM model by using a cross verification method based on the training set, wherein different kernel function verification results are respectively applied to find out optimal configuration parameters of the SVM model in the training process;

Evaluating and optimizing the SVM model by using a preset evaluation index based on the verification set;

and testing the SVM model by using the test set to finally obtain the customer potential value prediction model.

Further, the method for realizing the grading of the potential value of the government enterprise client based on the client potential value prediction model specifically comprises the following steps:

And inputting customer portrait characteristic information corresponding to the government enterprise customers to be predicted into the customer potential value prediction model to obtain potential value grading labels of the government enterprise customers to be predicted.

In a second aspect, the present invention provides an apparatus for ranking potential value of government and enterprise customers, said apparatus comprising:

The client data acquisition module is used for acquiring enterprise portrait data, client historical contract signing data and client industry historical contract signing data corresponding to government enterprise clients;

The characteristic information acquisition module is connected with the client data acquisition module and is used for processing the enterprise portrait data, the client historical contract signing data and the client industry historical contract signing data based on the Airflow workflow management platform to obtain client portrait characteristic information;

The prediction model training module is connected with the characteristic information acquisition module and is used for training a Support Vector Machine (SVM) model by adopting the customer portrait characteristic information to obtain a customer potential value prediction model;

And the potential value grading module is connected with the prediction model training module and is used for grading the potential value of the government enterprise client based on the client potential value prediction model.

In a third aspect, the present invention provides an apparatus for grading potential value of a political customer, comprising a memory and a processor, the memory having stored therein a computer program, the processor being arranged to run the computer program to implement the method of grading potential value of a political customer of the first aspect described above.

In a fourth aspect, the present invention provides a computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the method of grading potential value of a political client as described in the first aspect above.

The application provides a method, a device and a medium for grading potential value of government enterprise customers, which are characterized in that enterprise portrait data, customer history subscription data and customer industry history subscription data corresponding to the government enterprise customers are firstly obtained; then processing the enterprise portrait data, the client historical contract signing data and the client industry historical contract signing data based on Airflow workflow management platform to obtain client portrait characteristic information; training a Support Vector Machine (SVM) model by adopting the customer portrait characteristic information to obtain a customer potential value prediction model; finally, grading the potential value of the government enterprise customers based on the customer potential value prediction model, and the method can identify the customers with high potential value from a huge government enterprise customer group through grading prediction of the potential value of the government enterprise customers, thereby solving the problem of how to identify the customers with high potential value from the huge government enterprise customer group in the prior art, and the problem of no related scheme at present.

Drawings

FIG. 1 is a flow chart of a method of ranking potential values of government and enterprise customers according to embodiment 1 of the invention;

FIG. 2 is a diagram of a potential value ranking system for government enterprise customers in accordance with an embodiment of the present invention;

FIG. 3 is a schematic diagram of a potential value grading device for a government enterprise client according to embodiment 2 of the present invention;

Fig. 4 is a schematic structural diagram of a grading device for potential value of a government enterprise client according to embodiment 3 of the present invention.

Detailed Description

In order to make the technical scheme of the present invention better understood by those skilled in the art, the following detailed description of the embodiments of the present invention will be given with reference to the accompanying drawings.

It is to be understood that the specific embodiments and figures described herein are merely illustrative of the invention, and are not limiting of the invention.

It is to be understood that the various embodiments of the invention and the features of the embodiments may be combined with each other without conflict.

It is to be understood that only the portions relevant to the present invention are shown in the drawings for convenience of description, and the portions irrelevant to the present invention are not shown in the drawings.

It should be understood that each unit and module in the embodiments of the present invention may correspond to only one physical structure, may be formed by a plurality of physical structures, or may be integrated into one physical structure.

It will be appreciated that the terms "first," "second," and the like in embodiments of the present invention are used to distinguish between different objects or to distinguish between different processes on the same object, and are not used to describe a particular order of objects.

It will be appreciated that, without conflict, the functions and steps noted in the flowcharts and block diagrams of the present invention may occur out of the order noted in the figures.

It is to be understood that the flowcharts and block diagrams of the present invention illustrate the architecture, functionality, and operation of possible implementations of systems, apparatuses, devices, methods according to various embodiments of the present invention. Where each block in the flowchart or block diagrams may represent a unit, module, segment, code, or the like, which comprises executable instructions for implementing the specified functions. Moreover, each block or combination of blocks in the block diagrams and flowchart illustrations can be implemented by hardware-based systems that perform the specified functions, or by combinations of hardware and computer instructions.

It should be understood that the units and modules related in the embodiments of the present invention may be implemented by software, or may be implemented by hardware, for example, the units and modules may be located in a processor.

In order to enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, some technical terms related to the embodiments of the present invention are briefly described below.

1. Support Vector Machine (SVM): the support vector machine (Support Vector Machine, abbreviated as SVM) is a supervised learning model widely used in the field of machine learning, and is particularly suitable for classification problems. The purpose of the support vector machine is to find the best interface (also called hyperplane) in the feature space, separating the different classes of data as accurately and robustly as possible. The machine learning model of the SVM is excellent in high-dimensional space and small sample data, and is effective even when the number of features in the data is larger than the number of samples.

2. Chinese word segmentation library Jieba: jieba is a popular library of chinese segmentation Python. It supports three word segmentation modes: the method mainly provides the functions of part-of-speech tagging, keyword extraction and the like.

① Accurate mode: attempting to cut the sentence most accurately, suitable for text analysis;

② Full mode: all possible words in the sentence are scanned out, the speed is very high, but ambiguity cannot be resolved;

③ Search engine mode: based on the accurate mode, the long word is segmented again, the recall rate is improved, and the method is suitable for word segmentation of a search engine.

Jieba has good performance in Chinese word segmentation accuracy, adopts a high-efficiency prefix dictionary realized based on a Trie structure to carry out word graph scanning, constructs a Directed Acyclic Graph (DAG), adopts dynamic programming to search a maximum probability path, and realizes high-efficiency word frequency statistics. Generally, jieba is particularly well suited for rapid development and early stages of chinese text preprocessing due to its simplicity and ease of use and relatively good performance, especially in environments with limited resources and computing power.

3. Work level platform Airflow: airflow is an open source workflow management platform for orchestrating, planning, and monitoring complex workflows of data pipelines. It provides a flexible and extensible framework for defining tasks and workflows, commonly referred to as "DAGs" (DIRECTED ACYCLIC GRAPHS ).

Airflow's powerful orchestration capability makes it very popular among data engineers because it can orchestrate complex large data workflows, ensuring efficient and reliable operation of data pipes. Using Airflow, a team can automate and monitor the lifecycle of complex data processing tasks, thereby improving the predictability and transparency of data operations.

Example 1:

The embodiment provides a method for grading potential value of government enterprise customers, as shown in fig. 1, which comprises the following steps:

Step S101: enterprise portrait data, client history subscription data and client industry history subscription data corresponding to government enterprise clients are obtained.

In this embodiment, the enterprise portrayal data includes enterprise public information, call records, and associated business opportunities.

Optionally, the obtaining enterprise portrait data corresponding to the government enterprise clients specifically includes:

Acquiring enterprise public information of all known government and/or enterprise clients within a preset range through a third party service interface;

access records and associated business opportunities entered when the government and/or enterprise clients are visited in the field are obtained.

In this embodiment, the enterprise image data includes enterprise public information such as an operation range, a financial situation, an industrial structure, etc. acquired through a third party service interface, and a visit record and an associated business opportunity, etc. which are recorded by a sales manager to visit a customer in the field.

Specifically, by analyzing a network request of an enterprise information service site or a financial service site, such as an industrial and commercial registration information network, a tax bureau, a stock exchange (for a marketing company), and a third party professional information service company, the data interface of the site is requested to be accessed through a python construction after the structure of the request is analyzed, so that public information of all known enterprises/governments within a preset range is obtained.

Specifically, sales personnel may be assigned as customer managers of some enterprises/governments in a certain industry, and it is responsible for visiting the enterprises/governments in real time and filling out visit records, if the opposite customer is intentional for a certain product, the customer manager may reenter the associated business, for example, the filled-out visit records may include: the associated business machine input by the visiting type, the client care person, the visiting level, the visiting mode, the visiting matter or the intentional of the cooperation, the visiting person and the like can comprise information such as a client name, a client manager, a marketing unit, a business suit, a home service, a supporting person, creation time, work order duration, uploading amount, backfilling amount, effective state or not and the like.

Optionally, the enterprise public information includes basic information, business activities that the company can engage in, registered capital, stakeholder information, industry scope, financial conditions, intellectual property rights, administrative permissions and penalties, legal litigation and arbitration, market share, corporate governance, and the like.

Step S102: and processing the enterprise portrait data, the client historical contract signing data and the client industry historical contract signing data based on Airflow workflow management platform to obtain client portrait characteristic information.

In this embodiment, the enterprise portrait data, the customer history subscription data and the customer industry history subscription data are processed at the DAG timing designed by the Airflow workflow management platform, so that customer portrait feature information as a dataset can be obtained.

Optionally, the workflow management platform based on Airflow processes the enterprise portrait data, the client history subscription data and the client industry history subscription data to obtain client portrait feature information, which specifically includes:

Defining a directed acyclic graph DAG based on Airflow workflow management platforms;

defining a plurality of tasks and execution sequences and dependency relationships among the plurality of tasks in the DAG, wherein the plurality of tasks comprise a data extraction task, a client portrait generation task and a text analysis processing task;

and executing the tasks according to the DAG to obtain the customer portrait characteristic information.

In the present embodiment, a Airflow DAG is used primarily to represent a data processing flow, where each task represents a step in the flow, such as data extraction, conversion, generation, etc. Edges between tasks represent the order of execution, ensuring that a particular task will not be executed until the task it depends on is successfully completed.

Optionally, the data extraction task is configured to extract an enterprise information table, a visit record table, an associated business table, a client history subscription table and a history subscription table of the client industry according to the enterprise portrait data, the client history subscription data and the client industry history subscription data;

The client portrait generation task is used for aggregating the enterprise information table, the visit record table, the associated business table, the client history subscription table and the history subscription table of the client industry, and selecting characteristics related to the client potential value prediction from the aggregated data to generate a client portrait;

The text analysis processing task is used for calling a Chinese word segmentation library jieba to segment the features in the customer portrait to extract text contents, and converting the text contents into word vectors to obtain the feature information of the customer portrait.

In this embodiment, the data extraction tasks include an enterprise information table, a visit record table, an associated business table, a customer history subscription information table, and a customer industry history subscription information table, and specifically operate to specify a task number, a PG connection configuration, an sql script file, and a dag number by using an execution operator (PostgresOperator) of the PG database in Airflow.

In this embodiment, the processing steps of the client portrait creation task may include:

① Data aggregation: creating a broad table, combining the enterprise information table, the visit record table, the associated business table, the client history subscription table and the client industry history subscription table into a single view, wherein each record represents an all-round image of an enterprise.

② Feature selection: the method comprises the steps of selecting useful characteristics related to the prediction of potential value of a client from original data (namely a broad table), wherein the selected characteristics comprise total subscription amount, average subscription amount, visit frequency, business opportunity conversion rate, industry subscription trend, financial situation, financial report grade, legal dispute grade, business opportunity distribution, client intention grade, client demand grade and the like of an enterprise.

In this embodiment, the processing steps of the text analysis processing task may include:

① Adding a custom word stock: the word stock sources are enterprise business scope word stock, financial word stock and industry general word stock which are obtained from the Internet.

② Data cleaning: the method comprises the steps of lowercase removal, special character removal, messy code removal, and calling of a third party library to correct spelling errors and text formatting.

③ Long text keyword extraction processing: the python calls jieba a library to use an accurate mode to segment long text, and matches the word library with a pre-obtained word library, if the business scope of a certain enterprise relates to electronic equipment or retail industry, after the matching with the word library is successful, the text content is converted into word vectors, namely, the feature conversion in an SVM algorithm is preprocessed, and tag coding is adopted.

Step S103: and training the SVM model by adopting the customer portrait characteristic information to obtain a customer potential value prediction model.

In this embodiment, based on the customer portrait feature information processed through text analysis, the customer portrait feature information is trained by combining with a Support Vector Machine (SVM) model, so as to obtain a customer potential value prediction model.

Optionally, training the support vector machine SVM model by using the customer portrait characteristic information to obtain a customer potential value prediction model, which specifically includes:

taking the customer portrait characteristic information as a data set, and dividing the data set into a training set, a verification set and a test set;

Training the SVM model by using a cross verification method based on the training set, wherein different kernel function verification results are respectively applied to find out optimal configuration parameters of the SVM model in the training process;

Evaluating and optimizing the SVM model by using a preset evaluation index based on the verification set;

and testing the SVM model by using the test set to finally obtain the customer potential value prediction model.

In this embodiment, the step of model training may include:

① Data set partitioning: the three data sets are divided according to time, namely a training set, a verification set and a test set, wherein the training set is client information and historical subscription data within 3 months to 2 years, and the verification set and the test set are randomly divided into data within 3 months. The total data set after such division is roughly divided into 70%, 15%.

② Selecting an SVM kernel function: the selection of the kernel functions is critical to performance, and the invention uses a cross verification method, wherein a subset in the data set is selected for training each time, different kernel functions are respectively applied, and the quality of the results is verified, wherein the kernel functions comprise a linear kernel, a polynomial kernel, an RBF (Radial Basis Function radial basis function) kernel and a sigmoid kernel. And in combination with grid search, find the best configuration parameters of the model.

③ Model evaluation and tuning:

a) And (3) selecting an evaluation index: accuracy, precision, recall, F1 score, confusion matrix.

B) Adjusting parameters using a separate validation set provides generalization ability of the model, such as kernel parameters.

④ Model test: the model was tested using a separate test set. And testing on the test set by using the final model to obtain the final efficacy evaluation of the model.

Step S104: and grading the potential value of the government enterprise client based on the client potential value prediction model.

Specifically, customer portrait characteristic information corresponding to the government enterprise customers to be predicted is input into the customer potential value prediction model, and potential value grading labels of the government enterprise customers to be predicted are obtained. The potential value grading labels can be classified into high, medium and low 3 grades, or can be classified into 5 grades, and the highest grade is 5 grades, and the potential value grading labels can be specifically set according to actual conditions.

In a specific embodiment, the method for grading potential values of government and enterprise customers is applied to a system for grading potential values of government and enterprise customers as shown in fig. 2, an algorithm relates to a data source acquisition system, the data source of the algorithm consists of enterprise portrait data, customer history subscription data and customer industry history subscription data, the enterprise portrait data comprises enterprise public information such as operation range, financial situation, industrial structure and the like acquired by a third party service interface, and key information extracted by a jieba Chinese word segmentation library from text contents such as intention surveys, demand books, conference records and the like collected by a sales manager visiting customers. As shown in fig. 2, the government enterprise customer potential value grading system includes 4 modules: the system comprises an enterprise information acquisition module, a client visit management module, a text analysis processing module and a core algorithm module. Wherein, each module is described as follows:

(1) The enterprise information acquisition module:

The function of the module is mainly to analyze network requests of enterprise information service sites or financial service sites, such as business registration information networks, tax authorities, stock exchanges (aiming at marketing companies) and third party professional information service companies, and after the request body structure is analyzed, the request is constructed by python to access the data interface of the sites, so that enterprise public information of all known enterprises/governments in a certain jurisdiction of the city is obtained, and the enterprise public information is stored in an enterprise information table in a database through structural processing, and the data flows to a text analysis processing module, wherein the specific collected contents are as follows:

a) Basic information: including company name, established date, registration number, legal representative, registration address, contact, business scope, etc.;

b) Business activities that a company may engage in: a service range requiring special permissions or qualification;

c) Register capital: a registered capital amount for the enterprise;

d) Stakeholder information: the method comprises the steps of stakeholder name, a stakeholder right structure, a stakeholder right change record and the like;

e) Industry range: including the category to which the business belongs, the assets or licenses associated with a particular industry, etc.;

f) Financial situation: including annual financial reports (including asset liabilities sheets, profit sheets, etc.), audit reports, credit records (e.g., credit ratings, liability information), etc.;

g) Intellectual property rights: registration of intellectual property rights such as patents, trademarks, copyrights and the like;

h) Administrative permissions and administrative penalties: government issued licenses and license records, known administrative penalty records;

i) Legal litigation and arbitration: litigation and arbitration case information related to the enterprise;

j) Market share: the share of an enterprise in its service or product market (typically obtained through industry analysis reports);

k) Corporate governance: including board and prison member lists, important decisions, corporate laws, etc.

(2) Customer visit management module:

The module has the main functions of realizing the application of visiting and recovering intention information of a client, wherein sales personnel can be distributed as client managers of partial enterprises/governments of a certain industry, the responsibility is to visit the enterprises/governments in real time and fill out visit records, if opposite clients have intention on a certain product, the clients can be managed to enter relevant business machines again, for example, the filled-out visit records can comprise: the associated business machine input by the visiting type, the client care person, the visiting level, the visiting mode, the visiting matter or the intentional of the cooperation, the visiting person and the like can comprise information such as a client name, a client manager, a marketing unit, a business suit, a home service, a supporting person, creation time, work order duration, uploading amount, backfilling amount, effective state or not and the like. The visit records and associated business opportunities are also stored in a database, and the data flows to a text analysis processing module.

(3) A customer portrait generation module:

The data source of the module contains 5 tables altogether, one part of the 3 tables generated by the enterprise information acquisition module and the visit management module are respectively an enterprise information table, a visit record table and an associated business table, and the other part of the data source is a client history subscription table and a history subscription table of the client industry (belonging to an intermediate result table generated by aggregation operation). The enterprise information table, the visit record table and the association business table are associated through client names, and then the history subscription table of the client Industry is associated with other tables through Industry fields.

The government clients include national enterprises and institutions, and also have enterprise information tables.

The customer history subscription table may include the following fields:

subscription ID (ContractID): an ID uniquely identifying each subscription.

Customer ID (CustomerID): an ID identifying the subscribing client is associated with the client profile.

Customer name (CustomerName): the name of the customer is convenient for visual display and searching.

Belongs to Industry: industry of customers

Contact (Contact): the principal contacts of the customer.

Contact phone (ContactNumber): telephone number of customer contact.

Subscription date (StartDate): the start date of the contract.

Expiration date (EndDate): the expiration date of the contract.

Contract amount (ContractValue): the contract amount is used for recording and analyzing the contract value.

Contract type (ContractType): such as sales contracts, service contracts, etc.

Payment condition (PAYMENTTERMS): such as prepaid, pay-for-goods, installment, and the like.

Lead time (DELIVERYTIME): expected delivery time of goods or services.

Service scope (ServiceScope): the service scope or the product supply scope agreed in the contract.

Case (RenewalStatus): indicating whether to continue, and the status of the continuation.

Contract status (ContractStatus): such as executing, completed, cancelled, etc.

Remarks (Notes): any additional information about the subscription.

Subscription responsible person (ResponsiblePerson): responsible for the corporate staff who signs up this time.

Customer satisfaction evaluation (CustomerSatisfaction) the customer gives a satisfaction evaluation after the contract is completed.

Subsequent service feedback (FollowUpFeedback): customer feedback for subsequent service conditions after contract execution.

Update time (UpdateTime): the time of the last update of the contract information is recorded for tracking the latest state.

Service Level Agreements (SLAs): recording service level agreement details with the customer, which is a very important part of the telecommunications industry, such as availability, fault response time, etc.

Network coverage (NetworkCoverage): network coverage, representing service or contract coverage, may involve network services at the urban, regional, or even national level.

Data traffic packet (DATAPACKAGE): the details of the data traffic package purchased or presented by the customer include information such as total amount, expiration date, etc.

Device information (EquipmentDetails): details of network devices purchased or rented by the customer, such as routers, switches, etc., including model numbers, quantities, configurations, etc.

Technical support level (TechSupportLevel): the technical support service level purchased by the customer may include 24/7 phone support, field support, etc.

Bandwidth requirements (BandwidthRequirements): there may be different bandwidth requirements depending on the type of service, and the bandwidth requirements of the client pool are recorded.

Custom service (CustomServices): customized service content for specific needs that a telecommunications large customer may have.

Access point information (AccessPointDetails): for access services, details of the client access point, such as location, access technology, etc., are recorded.

Priority code (PriorityCode): priority levels of customer traffic during peak network traffic periods.

Security service (SecurityServices): additional network security service details purchased by the customer, such as firewalls, intrusion detection services, etc.

Contract renewal condition (RenewalConditions): especially for contracts in the telecommunications industry, renewal terms may be relevant for price adjustment, service upgrades, etc.

Compliance requirements (ComplianceRequirements): the customer's requirements for compliance with particular industry standards or regulations, such as data protection, privacy laws, etc., are recorded.

Project implementation plan (ProjectImplementationPlan): large telecommunication service contracts may require detailed implementation plans, including staged goals, key milestones, etc.

The client industry history subscription table is an intermediate table generated by Airflow at regular time, and comprises the following fields: industry name, industry classification, number of contracts, total amount of contracts, average contract amount, maximum contract amount, minimum contract amount, contract type distribution, contract status distribution, renewal rate, average customer satisfaction, average service deadline, technical support requirements, security service requirements, value added service requirements, common contract renewal conditions, bid replacement frequency, and the like.

The core process of this module includes designing a DAG based on Airflow workflow management platform, one Airflow DAG primarily to represent data processing flows, where each task represents a step in the flow, such as data extraction, conversion, loading, etc. Edges between tasks represent the order of execution, ensuring that a particular task will not be executed until the task it depends on is successfully completed.

The DAG is divided into a plurality of tasks, including customer information data extraction, visit record extraction, associated business extraction, data fusion, data conversion, text analysis and output target data, wherein the text analysis processing task calls the python Chinese word stock jieba. Each task node is equivalent to executing a specific function and plays a role in decoupling the whole final target task. The DAG aims to generate a data set in a core algorithm SVM generated by text analysis processing based on five tables at 3-point timing in the morning, and the specific process is as follows:

step one: defining a Airflow DAG:

Specifying default parameter configuration initialization DAG: execution time, timing execution rules, number of failed retries, failed retry interval, etc.

Step two: defining a data extraction task:

Five tasks are defined in the DAG, namely, an enterprise information table, a visit record table, an associated business machine table, a client history subscription information table and a data extraction task of the client industry history subscription information table, wherein the specific operations are that a task number, a PG connection configuration, an sql script file and a DAG number are specified by using an execution operator (PostgresOperator) of a PG database in Airflow.

Step three: defining a client image generation task:

the task is based on the data extracted in the previous step, and the following processing steps are carried out:

① Data aggregation: creating a broad table combining step two 5 different sources of data into a single view, each record representing an omnidirectional image of an enterprise.

② Feature selection: the method comprises the steps of selecting useful characteristics related to the prediction of potential value of a client from original data (namely a broad table), wherein the selected characteristics comprise total subscription amount, average subscription amount, visit frequency, business opportunity conversion rate, industry subscription trend, financial situation, financial report grade, legal dispute grade, business opportunity distribution, client intention grade, client demand grade and the like of an enterprise.

Step four: defining text analysis processing tasks:

The task is divided into the following steps:

① Adding a custom word stock: the word stock sources are enterprise business scope word stock, financial word stock and industry general word stock which are obtained from the Internet.

② Data cleaning: the method comprises the steps of lowercase removal, special character removal, messy code removal, and calling of a third party library to correct spelling errors and text formatting.

③ Long text keyword extraction processing: the python calls jieba a library to use an accurate mode to segment long text, and matches the word library with a pre-obtained word library, if the business scope of a certain enterprise relates to electronic equipment or retail industry, after the matching with the word library is successful, the text content is converted into word vectors, namely, the feature conversion in an SVM algorithm is preprocessed, and tag coding is adopted.

It should be noted that, the data extraction task mainly extracts data of tables distributed among different libraries and summarizes the data into 5 tables: enterprise information table, visit record table, association business table, customer history subscription information table, customer industry history subscription information table, because these data are distributed on different platforms, text analysis mainly vectorizes text content.

(4) The core algorithm module:

The module trains the customer representation data in conjunction with a Support Vector Machine (SVM) model based on pre-processed (i.e., text analysis processed) customer representation feature information. The method comprises the following steps:

① Data set partitioning: the three data sets are divided according to time, namely a training set, a verification set and a test set, wherein the training set is client information and historical subscription data within 3 months to 2 years, and the verification set and the test set are randomly divided into data within 3 months. The total data set after such division is roughly divided into 70%, 15%.

② Selecting an SVM kernel function: the selection of the kernel functions is critical to performance, and the invention uses a cross verification method, wherein a subset in the data set is selected for training each time, different kernel functions are respectively applied, and the quality of the results is verified, including a linear kernel, a polynomial kernel, an RBF kernel and a sigmoid kernel. And in combination with grid search, find the best configuration parameters of the model.

③ Model evaluation and tuning:

a) And (3) selecting an evaluation index: accuracy, precision, recall, F1 score, confusion matrix.

B) Adjusting parameters using a separate validation set provides generalization ability of the model, such as kernel parameters.

④ Model test: the model was tested using a separate test set. And testing on the test set by using the final model to obtain the final efficacy evaluation of the model.

It should be noted that, the input of the prediction stage SVM model is the customer portrait characteristic information after text analysis processing, and the output is the customer potential value grading label, for example, 5 grades, and the value is 5 at the highest.

The invention designs a closed-loop controlled customer data recovery and generation for government and enterprise customer groups of telecom operators, designs a classification algorithm for the potential value degree of customers based on the combination of customer portrait and historical subscription data and a classification algorithm support vector machine commonly used in machine learning, can effectively extract important features in multidimensional data, and has higher robustness and accuracy compared with other methods.

The potential value grading method for the government enterprise customers comprises the steps of firstly, obtaining enterprise portrait data, customer history subscription data and customer industry history subscription data corresponding to the government enterprise customers; then processing the enterprise portrait data, the client historical contract signing data and the client industry historical contract signing data based on Airflow workflow management platform to obtain client portrait characteristic information; training a Support Vector Machine (SVM) model by adopting the customer portrait characteristic information to obtain a customer potential value prediction model; finally, grading the potential value of the government enterprise customers based on the customer potential value prediction model, and the method can identify the customers with high potential value from a huge government enterprise customer group through grading prediction of the potential value of the government enterprise customers, thereby solving the problem of how to identify the customers with high potential value from the huge government enterprise customer group in the prior art, and the problem of no related scheme at present.

Example 2:

As shown in fig. 3, the present embodiment provides a grading apparatus for potential value of a government enterprise customer, which is configured to execute the grading method for potential value of a government enterprise customer, including:

The client data acquisition module 11 is used for acquiring enterprise portrait data, client historical subscription data and client industry historical subscription data corresponding to government enterprise clients;

The characteristic information acquisition module 12 is connected with the client data acquisition module 11 and is used for processing the enterprise portrait data, the client historical contract signing data and the client industry historical contract signing data based on the Airflow workflow management platform to obtain client portrait characteristic information;

The prediction model training module 13 is connected with the feature information acquisition module 12 and is used for training a Support Vector Machine (SVM) model by adopting the customer portrait feature information to obtain a customer potential value prediction model;

And the potential value grading module 14 is connected with the prediction model training module 13 and is used for grading the potential value of the government enterprise customers based on the customer potential value prediction model.

Optionally, the enterprise portrait data includes enterprise public information, visit records and associated business opportunities, and the client data acquisition module 11 includes:

The public information acquisition unit is used for acquiring enterprise public information of all known governments and/or enterprise clients within a preset range through a third party service interface;

And the visit and business machine acquisition unit is used for acquiring visit records and associated business machines which are recorded when the government and/or enterprise clients are visited in the field.

Optionally, the enterprise public information includes basic information, business activities that the company can engage in, registered capital, stakeholder information, industry wide, financial conditions, intellectual property rights, administrative permissions and penalties, legal litigation and arbitration, market share, and corporate governance.

Optionally, the feature information obtaining module 12 includes:

A first defining unit for defining a directed acyclic graph, DAG, based on Airflow workflow management platform;

A second definition unit, configured to define a plurality of tasks and execution sequences and dependency relationships among the plurality of tasks in the DAG, where the plurality of tasks include a data extraction task, a client portrait generation task, and a text analysis processing task;

And the task execution unit is used for executing the tasks according to the DAG to obtain the customer portrait characteristic information.

Optionally, the data extraction task is configured to extract an enterprise information table, a visit record table, an associated business table, a client history subscription table and a history subscription table of the client industry according to the enterprise portrait data, the client history subscription data and the client industry history subscription data;

The client portrait generation task is used for aggregating the enterprise information table, the visit record table, the associated business table, the client history subscription table and the history subscription table of the client industry, and selecting characteristics related to the client potential value prediction from the aggregated data to generate a client portrait;

The text analysis processing task is used for calling a Chinese word segmentation library jieba to segment the features in the customer portrait to extract text contents, and converting the text contents into word vectors to obtain the feature information of the customer portrait.

Optionally, the prediction model training module 13 includes:

The data set dividing unit is used for taking the customer portrait characteristic information as a data set and dividing the data set into a training set, a verification set and a test set;

The training unit is used for training the SVM model by using a cross verification method based on the training set, and different kernel function verification results are respectively applied to find the optimal configuration parameters of the SVM model in the training process;

A verification unit for evaluating and tuning the SVM model by using a preset evaluation index based on the verification set;

And the testing unit is used for testing the SVM model by using the testing set to finally obtain the customer potential value prediction model.

Optionally, the potential value ranking module 14 is specifically configured to:

And inputting customer portrait characteristic information corresponding to the government enterprise customers to be predicted into the customer potential value prediction model to obtain potential value grading labels of the government enterprise customers to be predicted.

Example 3:

referring to fig. 4, the present embodiment provides a device for grading potential value of a government-enterprise customer, comprising a memory 21 and a processor 22, the memory 21 storing a computer program, the processor 22 being arranged to run the computer program to perform the grading method of potential value of a government-enterprise customer of embodiment 1.

The memory 21 is connected to the processor 22, the memory 21 may be a flash memory, a read-only memory, or other memories, and the processor 22 may be a central processing unit or a single chip microcomputer.

Example 4:

the present embodiment provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements the method of grading potential value of a government-enterprise customer in the above-described embodiment 1.

Computer-readable storage media include volatile or nonvolatile, removable or non-removable media implemented in any method or technology for storage of information such as computer-readable instructions, data structures, computer program modules or other data. Computer-readable storage media includes, but is not limited to, RAM (Random Access Memory ), ROM (Read-Only Memory), EEPROM (ELECTRICALLY ERASABLE PROGRAMMABLE READ ONLY MEMORY, charged erasable programmable Read-Only Memory), flash Memory or other Memory technology, CD-ROM (Compact Disc Read-Only Memory), digital Versatile Disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store the desired information and which can be accessed by a computer.

In summary, the method, the device and the medium for grading potential values of government enterprise customers provided by the embodiment of the application firstly acquire enterprise portrait data, customer history subscription data and customer industry history subscription data corresponding to the government enterprise customers; then processing the enterprise portrait data, the client historical contract signing data and the client industry historical contract signing data based on Airflow workflow management platform to obtain client portrait characteristic information; training a Support Vector Machine (SVM) model by adopting the customer portrait characteristic information to obtain a customer potential value prediction model; finally, grading the potential value of the government enterprise customers based on the customer potential value prediction model, and the method can identify the customers with high potential value from a huge government enterprise customer group through grading prediction of the potential value of the government enterprise customers, thereby solving the problem of how to identify the customers with high potential value from the huge government enterprise customer group in the prior art, and the problem of no related scheme at present.

It is to be understood that the above embodiments are merely illustrative of the application of the principles of the present invention, but not in limitation thereof. Various modifications and improvements may be made by those skilled in the art without departing from the spirit and substance of the invention, and are also considered to be within the scope of the invention.

Claims (10)

1. A method of ranking potential value of a government enterprise customer, the method comprising:

Acquiring enterprise portrait data, client history subscription data and client industry history subscription data corresponding to government enterprise clients;

Processing the enterprise portrait data, the client history subscription data and the client industry history subscription data based on Airflow workflow management platform to obtain client portrait feature information;

training a Support Vector Machine (SVM) model by adopting the customer portrait characteristic information to obtain a customer potential value prediction model;

And grading the potential value of the government enterprise client based on the client potential value prediction model.

2. The method according to claim 1, wherein the enterprise portrait data includes enterprise public information, visit records and associated business opportunities, and the acquiring enterprise portrait data corresponding to government enterprise customers specifically includes:

Acquiring enterprise public information of all known government and/or enterprise clients within a preset range through a third party service interface;

access records and associated business opportunities entered when the government and/or enterprise clients are visited in the field are obtained.

3. The method of claim 2, wherein the enterprise public information includes basic information, business activities that a company can engage in, registered capital, stakeholder information, industry wide, financial aspects, intellectual property, administrative permissions and penalties, legal litigation and arbitration, market share, and corporate governance.

4. The method according to claim 1, wherein the processing the enterprise portrayal data, the client history subscription data and the client industry history subscription data based on Airflow workflow management platform to obtain client portrayal feature information specifically includes:

Defining a directed acyclic graph DAG based on Airflow workflow management platforms;

defining a plurality of tasks and execution sequences and dependency relationships among the plurality of tasks in the DAG, wherein the plurality of tasks comprise a data extraction task, a client portrait generation task and a text analysis processing task;

and executing the tasks according to the DAG to obtain the customer portrait characteristic information.

5. The method of claim 4, wherein the data extraction task is configured to extract an enterprise information table, a visit record table, an associated business table, a customer history subscription table, and a customer industry history subscription table based on the enterprise portrait data, the customer history subscription data, and the customer industry history subscription data;

The client portrait generation task is used for aggregating the enterprise information table, the visit record table, the associated business table, the client history subscription table and the history subscription table of the client industry, and selecting characteristics related to the client potential value prediction from the aggregated data to generate a client portrait;

The text analysis processing task is used for calling a Chinese word segmentation library jieba to segment the features in the customer portrait to extract text contents, and converting the text contents into word vectors to obtain the feature information of the customer portrait.

6. The method of claim 4, wherein training the support vector machine SVM model using the customer representation feature information to obtain the customer potential value prediction model, comprises:

taking the customer portrait characteristic information as a data set, and dividing the data set into a training set, a verification set and a test set;

Training the SVM model by using a cross verification method based on the training set, wherein different kernel function verification results are respectively applied to find out optimal configuration parameters of the SVM model in the training process;

Evaluating and optimizing the SVM model by using a preset evaluation index based on the verification set;

and testing the SVM model by using the test set to finally obtain the customer potential value prediction model.

7. The method according to claim 4, wherein the ranking of potential value of government enterprise customers based on the customer potential value prediction model is implemented, specifically comprising:

And inputting customer portrait characteristic information corresponding to the government enterprise customers to be predicted into the customer potential value prediction model to obtain potential value grading labels of the government enterprise customers to be predicted.

8. A device for ranking potential value of government and enterprise customers, said device comprising:

The client data acquisition module is used for acquiring enterprise portrait data, client historical contract signing data and client industry historical contract signing data corresponding to government enterprise clients;

The characteristic information acquisition module is connected with the client data acquisition module and is used for processing the enterprise portrait data, the client historical contract signing data and the client industry historical contract signing data based on the Airflow workflow management platform to obtain client portrait characteristic information;

The prediction model training module is connected with the characteristic information acquisition module and is used for training a Support Vector Machine (SVM) model by adopting the customer portrait characteristic information to obtain a customer potential value prediction model;

And the potential value grading module is connected with the prediction model training module and is used for grading the potential value of the government enterprise client based on the client potential value prediction model.

9. A device for ranking potential values of government and enterprise customers, comprising a memory and a processor, the memory having stored therein a computer program, the processor being arranged to run the computer program to implement a method of ranking potential values of government and enterprise customers as claimed in any of claims 1 to 7.

10. A computer readable storage medium having stored thereon a computer program which when executed by a processor implements a method of ranking potential value of a political party as defined in any one of claims 1 to 7.