CN115063035A - Customer evaluation method, system, equipment and storage medium based on neural network - Google Patents

Customer evaluation method, system, equipment and storage medium based on neural network Download PDF

Info

Publication number
CN115063035A
CN115063035A CN202210859101.9A CN202210859101A CN115063035A CN 115063035 A CN115063035 A CN 115063035A CN 202210859101 A CN202210859101 A CN 202210859101A CN 115063035 A CN115063035 A CN 115063035A
Authority
CN
China
Prior art keywords
neural network
target
information
client
customer
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202210859101.9A
Other languages
Chinese (zh)
Inventor
王美
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Ping An Health Insurance Company of China Ltd
Original Assignee
Ping An Health Insurance Company of China Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Ping An Health Insurance Company of China Ltd filed Critical Ping An Health Insurance Company of China Ltd
Priority to CN202210859101.9A priority Critical patent/CN115063035A/en
Publication of CN115063035A publication Critical patent/CN115063035A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/21—Design, administration or maintenance of databases
    • G06F16/215—Improving data quality; Data cleansing, e.g. de-duplication, removing invalid entries or correcting typographical errors
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/27—Replication, distribution or synchronisation of data between databases or within a distributed database system; Distributed database system architectures therefor
    • G06F16/275—Synchronous replication
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/28—Databases characterised by their database models, e.g. relational or object models
    • G06F16/283—Multi-dimensional databases or data warehouses, e.g. MOLAP or ROLAP
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/20—Natural language analysis
    • G06F40/279—Recognition of textual entities
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/20—Natural language analysis
    • G06F40/279—Recognition of textual entities
    • G06F40/284—Lexical analysis, e.g. tokenisation or collocates
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G06Q40/08—Insurance

Abstract

The invention provides a customer evaluation method, a system, equipment and a storage medium based on a neural network, wherein the method comprises the following steps: acquiring policy information, historical claim settlement information and communication information of a target client; obtaining static characteristics and dynamic characteristics according to the policy information, the historical claim settlement information and the communication information, wherein the static characteristics represent the fixed basic attributes of the target client, and the dynamic characteristics represent the dynamic attributes of the target client changing along with time; and inputting the static features and the dynamic features into a target evaluation neural network to obtain an evaluation score of the target client, wherein the target evaluation neural network is obtained by training the static features and the dynamic features corresponding to the sample clients and the labels corresponding to the sample clients. The main aim at excavates the potential demand of customer to purchasing insurance and improves the avoidance rate of malicious claim settlement to realize the accurate propelling movement to insurance.

Description

Customer evaluation method, system, equipment and storage medium based on neural network
Technical Field
The invention relates to the technical field of artificial intelligence, in particular to a client evaluation method, a system, equipment and a storage medium based on a neural network.
Background
Listening to the voice of the client is a key way for the insurance company to meet the client requirement and optimize the client service. As an important acquisition mode of the sound of insurance customers, analyzing and extracting clues from customer service conversation data is an effective means for the insurance enterprises to improve the self competitiveness. Through customer service session analysis, insurance companies can more fully and comprehensively know the customer appeal, so that the service quality of the enterprise is improved, and the enterprise is assisted to upgrade products. However, the analysis of the customer service session data in the industry depends on manual intervention to a great extent, and the current situation that the voice of the customer cannot be captured efficiently from a large amount of customer service session data and the feedback cannot be provided timely is faced.
When customer service conversation analysis is carried out, the current analysis method which is relatively universal in various fields mainly focuses on analysis of conversation emotion and conversation purposes, wherein the application value of the conversation emotion analysis in the insurance field is relatively limited, and the conversation purpose analysis needs to use a large number of manually made judgment rules as a basis in a complex insurance scene. However, the existing customer service session analysis method is not closely related to the three core services of sales, underwriting and claim settlement in the insurance field, and the customer service session analysis method suitable for the insurance field is still in a relatively deficient state.
Therefore, there is a need for a client session analysis in the insurance field to analyze whether a client is willing to purchase insurance and whether there is a risk of claim settlement after purchasing insurance, so as to mine potential clients and avoid malicious claim settlement.
Disclosure of Invention
The invention provides a customer evaluation method based on a neural network, which mainly aims to mine the potential requirements of customers on insurance purchase and improve the avoidance rate of malicious claims, thereby realizing accurate insurance pushing.
In a first aspect, an embodiment of the present invention provides a neural network-based customer evaluation method, including:
acquiring policy information, historical claim settlement information and communication information of a target client;
obtaining static characteristics and dynamic characteristics according to the policy information, the historical claim settlement information and the communication information, wherein the static characteristics represent the fixed basic attributes of the target client, and the dynamic characteristics represent the dynamic attributes of the target client changing along with time;
and inputting the static features and the dynamic features into a target evaluation neural network to obtain an evaluation score of the target client, wherein the target evaluation neural network is obtained by training the static features and the dynamic features corresponding to the sample clients and the labels corresponding to the sample clients.
Preferably, the label corresponding to the sample client is obtained by:
screening out sample clients with communication records from a data warehouse;
carrying out data cleaning on policy information, historical claim settlement information and communication information of the sample client to obtain sample data after cleaning;
carrying out exploratory analysis on the cleaned sample data to obtain an observation period;
and acquiring a label corresponding to the client according to whether the sample client generates a target behavior in an observation period.
Preferably, the screening out the sample clients with communication records from the data warehouse comprises:
synchronizing the application logs of the data warehouse to a big data cluster, and performing data cleaning on the application logs;
splitting the cleaned application log to obtain field information corresponding to the communication record;
and screening the sample client from a client database according to the field information corresponding to the communication record.
Preferably, the target behavior represents insurance purchasing behavior, the sample client corresponds to labels of purchasing behavior occurring in the observation period and purchasing behavior not occurring in the observation period, and the target evaluation neural network is a purchase evaluation neural network.
Preferably, the target behavior represents a malicious claim behavior, the labels corresponding to the sample clients are that the malicious claim behavior occurs in the observation period and the malicious claim behavior does not occur in the observation period, and the target evaluation neural network is a malicious evaluation neural network.
Preferably, the communication information is obtained by:
acquiring a communication record text of the target client and the customer service staff;
performing word segmentation processing on the communication record text;
and extracting keyword features of the communication record text after word segmentation processing, and taking the extracted keyword features as the communication information.
Preferably, the static characteristics include gender, age, occupation, marital status, premium for purchasing insurance of the target customer, and the dynamic characteristics include the payment amount of the premium of the target customer during a preset historical period, and whether the target customer is renewed during the preset historical period.
In a second aspect, an embodiment of the present invention provides a customer evaluation system based on a neural network, including:
the information acquisition module is used for acquiring policy information, historical claim settlement information and communication information of a target client;
the characteristic acquisition module is used for acquiring static characteristics and dynamic characteristics according to the policy information, the historical claim settlement information and the communication information, wherein the static characteristics represent the fixed basic attributes of the target client, and the dynamic characteristics represent the dynamic attributes of the target client changing along with time;
and the probability calculation module is used for inputting the static characteristics and the dynamic characteristics into a target evaluation neural network to obtain the purchase probability or the malicious claim settlement probability of the target customer, wherein the target evaluation neural network is obtained by training the static characteristics and the dynamic characteristics corresponding to the sample customer and the label corresponding to the sample customer.
In a third aspect, an embodiment of the present invention provides a computer device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the above-mentioned neural network-based customer evaluation method when executing the computer program.
In a fourth aspect, an embodiment of the present invention provides a computer storage medium storing a computer program, which when executed by a processor implements the steps of the above-mentioned neural network-based customer evaluation method.
According to the customer evaluation method, the customer evaluation system, the customer evaluation equipment and the storage medium based on the neural network, by collecting policy information, historical claim settlement information and communication information of a target customer, extracting static characteristics and dynamic characteristics of the target customer according to information such as insurance purchase and malicious claim settlement contained in the information, inputting the characteristics into the target evaluation neural network to obtain an evaluation score of the target customer, and obtaining whether the customer wants to purchase insurance or has risk of malicious claim settlement according to the evaluation score. According to the embodiment of the invention, the information related to insurance purchase, malicious claims settlement and the like is captured as far as possible in the policy information, the historical claims information and the communication information of the target customer, so that the potential requirements of the customer on insurance purchase are furthest excavated, the avoiding rate of the malicious claims is improved, and the insurance is accurately pushed.
Drawings
Fig. 1 is a schematic scene diagram of a neural network-based client evaluation method according to an embodiment of the present invention;
FIG. 2 is a flow chart of a neural network-based customer evaluation method according to an embodiment of the present invention;
fig. 3 is a schematic structural diagram of a neural network-based customer evaluation system according to an embodiment of the present invention;
fig. 4 is a schematic structural diagram of a computer device provided in an embodiment of the present invention.
The implementation, functional features and advantages of the present invention will be further described with reference to the accompanying drawings.
Detailed Description
It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
Fig. 1 is a schematic view of a scenario of a client evaluation method based on a neural network according to an embodiment of the present invention, and as shown in fig. 1, a user inputs policy information, historical claim settlement information, and communication information on a page provided by a client, and after receiving the policy information, historical claim settlement information, and communication information, the client sends the policy information, historical claim settlement information, and communication information to a server. And the server receives the policy information, the historical claim settlement information and the communication information, and executes the neural network-based client evaluation method to realize the evaluation of the target client.
It should be noted that the server may be implemented by an independent server or a server cluster composed of a plurality of servers. The client may be, but is not limited to, a smart phone, a tablet computer, a laptop computer, a desktop computer, and the like. The client and the server may be connected through bluetooth, USB (Universal Serial Bus), or other communication connection manners, which is not limited in this embodiment of the present invention.
The embodiment of the invention can acquire and process related data based on an artificial intelligence technology. Artificial Intelligence (AI) is a theory, method, technique and application system that simulates, extends and expands human Intelligence using a digital computer or a machine controlled by a digital computer, senses the environment, acquires knowledge and uses the knowledge to obtain the best result.
The artificial intelligence base technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, mechatronics, and the like. The artificial intelligence software technology mainly comprises a computer vision technology, a robot technology, a biological recognition technology, a voice processing technology, a natural language processing technology, machine learning, deep learning and the like.
Fig. 2 is a flowchart of a neural network-based customer evaluation method according to an embodiment of the present invention, and as shown in fig. 2, the method includes:
s210, acquiring policy information, historical claim settlement information and communication information of a target client;
the method includes the steps of firstly obtaining policy information, historical claim settlement information and communication information of target clients, generally, insurance companies all have related client data recording systems, basic attribute information of the clients and information related to insurance are recorded on the client data recording systems, the basic attribute information includes basic information such as names, sexes, ages, professions, marital conditions, heights, weights and physical examination records, the information related to insurance includes whether insurance has been purchased before, what insurance has been purchased, the insurance year of insurance purchase, insurance premium, claim settlement amount, historical claim settlement information and the like, each insurance company can record client data according to own needs, specific contained information can be determined according to actual conditions, and the embodiment of the invention is not limited specifically. The target client in the embodiment of the invention is the client recorded in the client data recording system, and the target client is evaluated and analyzed by extracting the policy information, the historical claim settlement information and the communication information of the target client to see whether the target client belongs to a potential client of an insurance company. In the embodiment of the invention, the policy information of the target client comprises information such as insurance type, insurance amount, insurance cost, insurance purchasing age, insurance period, whether the policy is modified or not, and if so, what is modified, the insurance type comprises medical insurance, endowment insurance, accidental insurance and the like, the client analyzes the degree of importance of the client on the insurance from the purchased insurance type, the insurance amount represents the insurance maximum claim amount, the insurance cost represents the cost which needs to be paid by purchasing the insurance target client every year, the insurance purchasing age refers to the age of purchasing the insurance, the insurance period refers to the guarantee time of each insurance, whether the policy is modified or not refers to whether the target client modifies the content of the policy terms when purchasing the insurance, if so, the modified specific content is recorded, the importance degree of the target client to insurance and the importance degree of insurance clauses can be seen through the policy information, the specific policy information can be determined according to the actual situation, and the embodiment of the invention is not specifically limited herein; the historical claim settlement information refers to insurance claim settlement information of the target client in the past time period, and includes information such as claim number, claim settlement time, claim reason, claim limit, beneficiary and the like, the claim number refers to the number of times of insurance claim settlement of the target client in the past time period, the claim settlement time refers to time when insurance claim settlement is performed each time, the claim reason refers to reason for performing insurance claim settlement, if the insurance is medical insurance, diseases for performing insurance claim settlement can be recorded, if the insurance is accidental insurance, accidental events for performing insurance claim settlement can be recorded, the claim limit refers to actual amount of money for performing claim settlement, the beneficiary refers to a final beneficial object of the insurance, the determination can be specifically made according to actual conditions, and the embodiment of the invention is not particularly limited herein; the communication information generally refers to information records for communication between a target customer and the insurance company, if voice communication records exist, the voice communication records are converted into character records, then all the character records are extracted, key sensitive words in the character records are extracted, the extracted key sensitive words are analyzed, and the meaning of the customer on insurance is extracted.
In the era of information explosion, valuable key information is extracted from massive text data, which is very important, keywords are words capable of expressing the center content of a document, and keyword extraction is a branch of the field of text mining. Supervised and unsupervised approaches. In the field of machine learning, "supervised" is defined as that an algorithm needs to label data manually, a supervised keyword extraction algorithm is mainly used for training existing text data and keywords thereof by using a machine learning algorithm to generate a model which can be used for detecting the text keywords, then the model is used for processing the new text data and detecting the keywords therein, and the supervised text keyword extraction algorithm needs high labor cost at present, namely, the existing data set needs to be labeled, for example, the text is labeled as positive, negative or neutral to evaluate the emotion implied by the text. The unsupervised keyword extraction algorithm only needs to select a method for evaluating keywords, such as the frequency and the position of the occurrence of the keyword, and possible keywords are extracted through the method. Common unsupervised keyword extraction methods comprise a TF-IDF algorithm, an LDA algorithm and a Word2vec algorithm, wherein the TF-IDF algorithm mainly evaluates the importance of words to documents through a statistical method. The basic idea is that the more times a word appears in a document, the more representative the word is, but if the word appears in many documents, the more times the word appears, the less document distinguishing capability is provided; the LDA algorithm is one of the most popular methods in the current keyword detection technology, each document is composed of different words, and simultaneously, a plurality of potential subjects exist, such as sports, entertainment, news, politics, and each subject also has different words belonging to the same, such as "football, basketball, match" may exist belonging to the "sports" subject, and "star, movie, record" and the like may exist belonging to the "entertainment" subject; the Word2vec algorithm mainly studies the relationship between words, converts all non-repeated words appearing in all text data sets into vectors, and the data format contains the similarity between the Word and all other words, so that the words can be classified according to the relationship between the words, central words of a plurality of categories are obtained through a classification algorithm, then the similarity between the Word and the category center in each category is calculated and sequenced, and finally the first words of the closest center are selected as keywords.
According to the embodiment of the invention, the information related to insurance purchase, malicious claims and the like is captured as much as possible in the policy information, the historical claims information and the communication information of the target customer, so that the potential requirements of the customer on insurance purchase are furthest mined, and the avoiding rate of the malicious claims is improved.
S220, obtaining static characteristics and dynamic characteristics according to the policy information, the historical claim settlement information and the communication information, wherein the static characteristics represent fixed basic attributes of the target client, and the dynamic characteristics represent dynamic attributes of the target client changing along with time;
and then, acquiring static characteristics and dynamic characteristics according to policy information, historical claim settlement information and communication information, wherein the policy information, the historical claim settlement information and the communication information are recorded through keywords, and the static characteristics and the dynamic characteristics are respectively extracted by extracting the characteristics of the information. When processing text data, a great deal of effort is used for feature extraction of a data set, and therefore a commonly used text feature extraction method needs to be recorded. In natural language processing, text data is converted into vector data, and many language features from the text data can be obtained in the vector data, which is called text representation or text feature construction. In the embodiment of the invention, static characteristics and dynamic characteristics are extracted according to policy information, historical claim settlement information and communication information, the static characteristics refer to inherent stability characteristics of a target client and generally do not change randomly along with time change or insurance type change, the static characteristics can describe some basic conditions and basic attitudes of the target client for insurance purchase, such as gender, occupation, marital conditions, physical conditions and the like, the attitudes of different occupations for insurance are different, such as higher insurance purchase willingness of doctors and higher insurance purchase willingness of main family labor force after marriage, so that the basic conditions of the target client can be determined through the static characteristics, the dynamic characteristics are used for further analyzing the willingness purchase conditions of the target client on the basis of the static characteristics, and the dynamic characteristics generally comprise payment conditions of the target client for existing insurance, The condition of renewal of the insurance per year is full payment or installments payment, and if the insurance is installments payment, the insurance is determined according to the actual condition, and the embodiment of the invention is not limited specifically.
According to the embodiment of the invention, the target customer is comprehensively analyzed by combining the static characteristics and the dynamic characteristics, so that the accuracy of analysis of the target customer is ensured.
And S230, inputting the static features and the dynamic features into a target evaluation neural network to obtain the evaluation score of the target client, wherein the target evaluation neural network is obtained by training the static features and the dynamic features corresponding to the sample clients and the labels corresponding to the sample clients.
And finally, inputting the static characteristics and the dynamic characteristics into a target evaluation neural network to obtain the evaluation score of the target client, wherein the target evaluation neural network is obtained by training the static characteristics and the dynamic characteristics corresponding to the sample client and the label corresponding to the sample client. In the embodiment of the invention, the target evaluation neural network outputs the evaluation score of the target customer, in the embodiment of the invention, the target customer is analyzed to determine whether the target customer has insurance purchase desire, if so, the desire is analyzed to determine how strong the desire is, and the target customer also has probability of cheating, if so, the probability of cheating is analyzed to determine how much, cheating is also malicious claim, for example, a certain customer has already had relevant symptoms of thyroid before purchasing insurance, and some people have intentionally go to thyroid to purchase relevant medical insurance after having suffered relevant diseases for cheating, so as to realize malicious claim, in the embodiment of the invention, the probability that the target customer has malicious claim is determined through policy information, historical claim information and communication information, and corresponding evaluation score is determined according to the probability, that is the evaluation score of possible insurance purchase of the target customer includes two evaluation scores, continuing to purchase the evaluation scores of the insurance and the evaluation scores of the malicious claims, wherein the higher the evaluation score corresponding to the purchase of the insurance is, the greater the potential of the target customer to the insurance company is, and the lower the evaluation score of the malicious claims is, the greater the potential of the target customer to the insurance company is; the lower the evaluation score corresponding to the purchase of insurance, the less the insurance company should distribute the energy to the target customer, and the higher the evaluation score of malicious claims, the more the insurance company should avoid the target customer.
It should be noted that the target evaluation neural network in the embodiment of the present invention belongs to one of neural networks, and before using the target evaluation neural network, it is also necessary to train or update the training, and train the auto-regressive speech synthesis model through the obtained samples and labels. The training process of the target evaluation neural network can be divided into three steps: defining the structure of a target evaluation neural network and an output result of forward propagation; defining a loss function and a back propagation optimization algorithm according to the process described above; finally, a session is generated and a back propagation optimization algorithm is run repeatedly on the training data.
The neuron is the minimum unit forming the neural network, one neuron can have a plurality of inputs and one output, and the input of each neuron can be the output of other neurons or the input of the whole neural network. The output of the neural network is the weighted sum of the inputs of all the neurons, the weight of different inputs is the neuron parameter, and the optimization process of the neural network is the process of optimizing the value of the neuron parameter.
The effect and optimization goal of the neural network are defined by a loss function, the loss function gives a calculation formula of the difference between the output result of the neural network and the real label, and supervised learning is a way of training the neural network, and the idea is that on a labeled data set of known answers, the result given by the neural network is as close as possible to the real answer (namely, the label). The training data is fitted by adjusting parameters in the neural network so that the neural network provides predictive power to unknown samples.
The back propagation algorithm realizes an iterative process, when each iteration starts, a part of training data is taken first, and the prediction result of the neural network is obtained through the forward propagation algorithm. Because the training data all have correct answers, the difference between the predicted result and the correct answer can be calculated. Based on the difference, the back propagation algorithm can correspondingly update the value of the neural network parameter, so that the neural network parameter is closer to the real answer.
After the training process is completed by the method, the auto-regression speech synthesis model after the training can be used for application.
According to the customer evaluation method based on the neural network, the policy information, the historical claim settlement information and the communication information of a target customer are collected, the static characteristics and the dynamic characteristics of the target customer are extracted according to the information, such as insurance purchase and malicious claim settlement, contained in the information, and the characteristics are input into the target evaluation neural network to obtain the evaluation score of the target customer, so that whether the customer wants to purchase insurance or not or whether the customer has risk of malicious claim settlement or not is obtained according to the evaluation score. According to the embodiment of the invention, the information related to insurance purchase, malicious claims settlement and the like is captured as far as possible in the policy information, the historical claims information and the communication information of the target customer, so that the potential requirements of the customer on insurance purchase are furthest excavated, the avoiding rate of the malicious claims is improved, and the insurance is accurately pushed.
On the basis of the above embodiment, preferably, the label corresponding to the sample client is obtained by:
screening out sample clients with communication records from a data warehouse;
carrying out data cleaning on policy information, historical claim settlement information and communication information of the sample client to obtain sample data after cleaning;
carrying out exploratory analysis on the cleaned sample data to obtain an observation period;
and acquiring a label corresponding to the client according to whether the sample client generates a target behavior in an observation period.
As an embodiment, sample clients with communication records are first screened out from the data warehouse, in the embodiment of the present invention, the data warehouse is a strategic set providing all types of data support for all levels of decision making process of the enterprise, and is a single data storage created for analytic reporting and decision support purposes, and provides guidance for business process improvement, monitoring time, cost, quality and control for the enterprise requiring business intelligence. For an insurance company, a data warehouse stores information such as target customer data, insurance data and claim settlement data, because the data warehouse stores information of a plurality of customers, but communication records of each customer and salespersons are not recorded, only the customers who communicate through specified communication software have the communication records, and other communication software except the specified communication records are communicated through telephone communication and micro communication, because of the limitation of a data interface, the communication records have no way to be recorded by the data warehouse, and therefore sample customers with the communication records need to be screened from the data warehouse. And screening the data warehouse, wherein the specific screening method can be that screening is carried out according to the fields corresponding to the communication records, and if only single screening is carried out, the data warehouse can be directly screened by using database statements to screen out data with corresponding fields, so that the screened clients are used as sample clients.
And then, data cleaning is carried out, wherein the data cleaning comprises four parts of missing data processing, repeated data processing, abnormal data processing and inconsistent data sorting. Data loss is a common situation in a database, but in order to obtain a complete information table for data mining, the data loss situation must be solved, so the following three methods are generally used for processing the data loss. The first is to delete a record with missing information, when main information in the record is missing, especially when more key information is missing, data cannot reflect the information which can be represented by the main information, and then a record can be deleted, but this case is only applicable to the case of large data volume, that is, deletion does not affect the integrity of all information, obviously, when the data volume is small, or the record lacking data is more, it is not feasible to completely delete the whole record, which may cause significant impact on the data quality, and the method for deleting the record has a certain application range. The second method is to perform manual completion on information, and also has its own limitations, and when the data volume is large, especially under the condition of massive data, the method consumes large manpower and has extremely low efficiency. The default value can be used to replace the missing information, the attribute of the missing attribute value is regarded as a special attribute, and a special attribute value is set for the missing information value, so that complete information is obtained. The third method is to use a mathematical formula to carry out statistical analysis on the value of the existing information by the data and to use the statistical value to carry out completion. The vacancy value can be filled by using an average value, or the vacancy value can be filled by using a sample prediction value of the same type, and the filling can also be performed by using an inference-based method such as a Bayes formula and a decision tree, so that the quality of information cannot be influenced. Obviously, if the adopted formula is not appropriate, the next information analysis can be adversely affected.
The duplicated data includes both attribute redundancy and redundant part of attribute data in addition to duplicated data in the true sense. The data with repeated real values or attribute values are simple to process and can be directly deleted. But attribute redundancy and redundancy of attribute data need to be analyzed and then deleted. In a data warehouse, data is collected using different databases, and thus, a case where a plurality of attribute names represent the same attribute may occur; moreover, some data can be obtained from other attributes, the age can be obtained from the birthday, and the data of the repeated part can be directly deleted. The redundancy of the attribute data means that values of some attributes already contain values in some attributes, for example, detailed addresses such as country provinces exist when the domestic user addresses are processed, the part of the country in the information belongs to repeated data, and the analysis of the data is not influenced by removing the country. The elimination of the repeated attributes can not only simplify the related records in the database and reduce the occupation of storage space, but also is beneficial to the improvement of the data analysis efficiency.
Abnormal data is also a common situation, which refers to a situation where a part of data is largely different from other data or is inconsistent with other data in a data set. The distinction is not necessarily an anomaly in the data, and the specific data may reflect the actual situation. At this time, it is necessary to judge whether the data is abnormal data, and if the data bit is abnormal, the data needs to be removed, so as to avoid influencing the accuracy of data analysis. However, some inconsistent data is not necessarily abnormal data, and for such data, attention is paid to the information hidden behind the inconsistent data to find out the reason for the inconsistent data.
Then, exploratory analysis is carried out on the cleaned data to obtain an observation period corresponding to the sample client, wherein the observation period represents that the observation period is a time period after a certain time node, the policy information, the historical claim settlement information and the communication information of the sample client before the time node are subjected to feature extraction to obtain the static feature and the dynamic feature of the sample client, but in order to label the sample client, the labeling is carried out according to whether a target behavior occurs in the observation period after the time node, in the embodiment of the invention, the target behavior comprises a purchase insurance behavior and a malicious claim settlement behavior, if the sample client has the purchase insurance behavior in the observation period, the sample client is labeled as having the purchase insurance label, otherwise, the sample client is labeled as not having the purchase insurance label, if the sample client has the malicious claim behavior in the observation period, and marking the sample client as a malicious claim settlement label, otherwise, marking the sample client as a non-malicious claim settlement label.
The exploratory analysis in the embodiment of the invention is deep and detailed descriptive statistical analysis on the variables, and on the basis of general descriptive statistical indexes, text and graphic description about other characteristics of the data are added, so that the analysis result is more detailed and comprehensive, and the further analysis on the data is facilitated. Exploratory analysis can generate comprehensive statistics and graphs about all cases, or different groups of cases; data screening work can be carried out, such as detection of abnormal values, extreme values, data gaps and the like; hypothesis testing may also be performed. Through exploratory analysis, the method can help people to decide which statistical method to select for data modeling, judge whether data needs to be converted into normal distribution or not, and judge whether non-parameter statistics needs to be carried out or not. Exploratory analysis is suitable for analyzing numerical variables (continuous or ratiometric), which should be categorical variables (for grouping data) taking a finite number of discrete values.
On the basis of the above embodiment, preferably, the screening out the sample clients with communication records from the data warehouse includes:
synchronizing the application logs of the data warehouse to a big data cluster, and performing data cleaning on the application logs;
splitting the cleaned application log to obtain field information corresponding to the communication record;
and screening the sample client from a client database according to the field information corresponding to the communication record.
As an implementation manner, the screening of the sample clients with communication records from the data warehouse in the embodiment of the present invention specifically includes the following steps: the application logs of the data warehouse are synchronized to a large data cluster, the application logs are subjected to data cleaning, the offline data line warehouse is used for synchronizing data of different data sources to the data warehouse and synchronizing the data to the service system at regular time, and for realizing the bidirectional synchronization of the data between the different data warehouses, the data must be converted into a certain intermediate state to unify the data formats. Then splitting the cleaned application log to obtain field information corresponding to the communication record; and screening out sample customers from the customer database according to the field information corresponding to the communication records.
On the basis of the above embodiment, preferably, the target behavior represents insurance purchase behavior, the labels corresponding to the sample customers are that purchase behavior occurs in the observation period and purchase behavior does not occur in the observation period, and the target evaluation neural network is a purchase evaluation neural network.
Specifically, the target behavior in the embodiment of the present invention represents insurance purchasing behavior, and then the labels corresponding to the sample customers are that purchasing behavior occurs in the observation period and purchasing behavior does not occur in the observation period, the target evaluation neural network is a purchase evaluation neural network, and the evaluation score of the target customer is an evaluation score of the insurance purchasing behavior. The insurance information, the historical claim settlement information and the communication information of the target customer are also insurance information, historical claim settlement information and communication information related to insurance purchasing behavior.
On the basis of the foregoing embodiment, preferably, the target behavior represents a malicious claim behavior, the labels corresponding to the sample clients are that the malicious claim behavior occurs in the observation period and the malicious claim behavior does not occur in the observation period, and the target evaluation neural network is a malicious evaluation neural network.
Specifically, the target behavior in the embodiment of the present invention represents a malicious claim behavior, and then the labels corresponding to the sample clients are that the malicious claim behavior occurs in the observation period and the malicious claim behavior does not occur in the observation period, the target evaluation neural network evaluates the neural network for the malicious claim, and the evaluation score of the target client is the evaluation score of the malicious claim behavior. The insurance information, the historical claim settlement information and the communication information of the target client are also insurance information, historical claim settlement information and communication information related to malicious claim settlement behaviors.
It should be noted that, the purchase evaluation neural network and the malicious claim neural network are two different neural networks, and the training sample of each neural network is different.
On the basis of the above embodiment, preferably, the communication information is obtained by:
acquiring a communication record text of the target client and the customer service staff;
performing word segmentation processing on the communication record text;
and extracting keyword features of the communication record text after word segmentation processing, and taking the extracted keyword features as the communication information.
Specifically, the communication information comprises communication record texts of the target client and the customer service staff, the communication record texts are subjected to word segmentation, keyword feature extraction is performed on the communication record texts subjected to word segmentation, and the extracted keyword features are used as the communication information.
Fig. 3 is a schematic structural diagram of a neural network-based customer evaluation system according to an embodiment of the present invention, and as shown in fig. 3, the system includes: an information acquisition module 310, a feature acquisition module 320, and a probability calculation module 330, wherein:
the information acquisition module 310 is used for acquiring policy information, historical claim settlement information and communication information of a target client;
the characteristic obtaining module 320 is configured to obtain a static characteristic and a dynamic characteristic according to the policy information, the historical claim settlement information, and the communication information, where the static characteristic represents a basic attribute fixed by the target customer, and the dynamic characteristic represents a dynamic attribute that changes with time of the target customer;
the probability calculation module 330 is configured to input the static features and the dynamic features into a target evaluation neural network to obtain the purchase probability or the malicious claim settlement probability of the target customer, where the target evaluation neural network is obtained by training the static features and the dynamic features corresponding to the sample customers and the tags corresponding to the sample customers.
The present embodiment is a system embodiment corresponding to the above method embodiment, and the specific implementation is the same as the above method execution process, and the system embodiment is not described herein again.
On the basis of the above embodiment, preferably, the probability calculation module includes a screening unit, a washing unit, an analysis unit, and a labeling unit, wherein:
the screening unit is used for screening out sample clients with communication records from the data warehouse;
the cleaning unit is used for carrying out data cleaning on policy information, historical claim settlement information and communication information of the sample client to obtain sample data after cleaning;
the analysis unit is used for carrying out exploratory analysis on the cleaned sample data to obtain an observation period;
and the label unit is used for acquiring a label corresponding to the customer according to whether the sample customer has a target behavior in an observation period.
On the basis of the foregoing embodiment, preferably, the screening unit includes a synchronization unit, a field unit, and a selection unit, where:
the synchronization unit is used for synchronizing the application logs of the data warehouse to a big data cluster and cleaning the application logs;
the field unit is used for splitting the cleaned application log to obtain field information corresponding to the communication record;
the selection unit is used for screening the sample client from a client database according to the field information corresponding to the communication record.
On the basis of the above embodiment, preferably, the target behavior represents insurance purchase behavior, the labels corresponding to the sample customers are that purchase behavior occurs in the observation period and purchase behavior does not occur in the observation period, and the target evaluation neural network is a purchase evaluation neural network.
On the basis of the foregoing embodiment, preferably, the target behavior represents a malicious claim behavior, the labels corresponding to the sample clients are that the malicious claim behavior occurs in the observation period and the malicious claim behavior does not occur in the observation period, and the target evaluation neural network is a malicious evaluation neural network.
On the basis of the above embodiment, preferably, the information obtaining module includes a text module, a word segmentation module and a keyword module, wherein:
the text module is used for acquiring a communication record text of the target client and the customer service staff;
the word segmentation module is used for carrying out word segmentation processing on the communication record text;
the keyword module is used for extracting keyword features of the communication record text after word segmentation processing, and the extracted keyword features are used as the communication information.
On the basis of the above embodiment, preferably, the static characteristics include gender, age, occupation, marital status, and premium of purchasing insurance of the target customer, and the dynamic characteristics include the payment amount of the premium of the target customer in a preset historical time period, and whether the target customer is renewed in the preset historical time period.
The various modules in the neural network-based customer evaluation system described above may be implemented in whole or in part by software, hardware, and combinations thereof. The modules can be embedded in a hardware form or independent from a processor in the computer device, and can also be stored in a memory in the computer device in a software form, so that the processor can call and execute operations corresponding to the modules.
Fig. 4 is a schematic structural diagram of a computer device provided in an embodiment of the present invention, where the computer device may be a server, and an internal structural diagram of the computer device may be as shown in fig. 4. The computer device includes a processor, a memory, a network interface, and a database connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a computer storage medium and an internal memory. The computer storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of an operating system and computer programs in the computer storage media. The database of the computer device is used to store data generated or obtained during execution of the neural network-based customer assessment method, such as policy information, historical claim settlement information, and communication information. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program is executed by a processor to implement a neural network-based customer evaluation method.
In one embodiment, a computer device is provided, comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the steps of the neural network based customer evaluation method in the above embodiments when executing the computer program. Alternatively, the processor, when executing the computer program, implements the functions of the modules/units in this embodiment of the neural network-based customer evaluation system.
In an embodiment, a computer storage medium is provided, on which a computer program is stored, which, when being executed by a processor, implements the steps of the neural network-based client evaluation method in the above embodiments. Alternatively, the computer program realizes the functions of the modules/units in the embodiment of the neural network-based customer evaluation system described above when executed by the processor.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above may be implemented by hardware instructions of a computer program, which may be stored in a non-volatile computer-readable storage medium, and when executed, may include the processes of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in the embodiments provided herein may include non-volatile and/or volatile memory, among others. Non-volatile memory can include read-only memory (ROM), Programmable ROM (PROM), Electrically Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus Direct RAM (RDRAM), direct bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
It will be apparent to those skilled in the art that, for convenience and brevity of description, only the above-mentioned division of the functional units and modules is illustrated, and in practical applications, the above-mentioned function distribution may be performed by different functional units and modules according to needs, that is, the internal structure of the apparatus is divided into different functional units or modules to perform all or part of the above-mentioned functions.
The above-mentioned embodiments are only used to illustrate the technical solution of the present invention, and not to limit the same; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; such modifications and substitutions do not substantially depart from the spirit and scope of the embodiments of the present invention, and are intended to be included within the scope of the present invention.

Claims (10)

1. A neural network-based customer evaluation method, comprising:
acquiring policy information, historical claim settlement information and communication information of a target client;
obtaining static characteristics and dynamic characteristics according to the policy information, the historical claim settlement information and the communication information, wherein the static characteristics represent the fixed basic attributes of the target client, and the dynamic characteristics represent the dynamic attributes of the target client changing along with time;
and inputting the static features and the dynamic features into a target evaluation neural network to obtain an evaluation score of the target client, wherein the target evaluation neural network is obtained by training the static features and the dynamic features corresponding to the sample clients and the labels corresponding to the sample clients.
2. The neural network-based customer evaluation method according to claim 1, wherein the label corresponding to the sample customer is obtained by:
screening out sample clients with communication records from a data warehouse;
carrying out data cleaning on policy information, historical claim settlement information and communication information of the sample client to obtain sample data after cleaning;
carrying out exploratory analysis on the cleaned sample data to obtain an observation period;
and acquiring a label corresponding to the client according to whether the sample client generates a target behavior in an observation period.
3. The neural network-based customer evaluation method of claim 2, wherein the screening of sample customers having communication records from the data warehouse comprises:
synchronizing the application logs of the data warehouse to a big data cluster, and performing data cleaning on the application logs;
splitting the cleaned application log to obtain field information corresponding to the communication record;
and screening the sample client from a client database according to the field information corresponding to the communication record.
4. The neural network-based customer evaluation method of claim 3, wherein the target behavior represents insurance purchase behavior, the sample customers have corresponding labels of occurrence of purchase behavior during the observation period and non-occurrence of purchase behavior during the observation period, and the target evaluation neural network is a purchase evaluation neural network.
5. The customer evaluation method based on neural network as claimed in claim 3, wherein the target behavior represents malicious claim behavior, the labels corresponding to the sample customers are that malicious claim behavior occurs during the observation period and that malicious claim behavior does not occur during the observation period, and the target evaluation neural network is a malicious evaluation neural network.
6. The neural-network-based customer evaluation method according to claim 1, wherein the communication information is obtained by:
acquiring a communication record text of the target client and the customer service staff;
performing word segmentation processing on the communication record text;
and extracting keyword features of the communication record text after word segmentation processing, and taking the extracted keyword features as the communication information.
7. The neural network-based customer evaluation method according to any one of claims 1 to 5, wherein the static characteristics include gender, age, occupation, marital status, and premium for insurance purchase of the target customer, and the dynamic characteristics include the amount of premium payment by the target customer over a preset historical period, and whether or not to renew the premium over the preset historical period.
8. A neural network-based customer evaluation system, comprising:
the information acquisition module is used for acquiring policy information, historical claim settlement information and communication information of a target client;
the characteristic acquisition module is used for acquiring static characteristics and dynamic characteristics according to the policy information, the historical claim settlement information and the communication information, wherein the static characteristics represent the fixed basic attributes of the target client, and the dynamic characteristics represent the dynamic attributes of the target client changing along with time;
and the probability calculation module is used for inputting the static characteristics and the dynamic characteristics into a target evaluation neural network to obtain the purchase probability or the malicious claim settlement probability of the target customer, wherein the target evaluation neural network is obtained by training the static characteristics and the dynamic characteristics corresponding to the sample customer and the label corresponding to the sample customer.
9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor when executing the computer program realizes the steps of the neural network based customer evaluation method according to any one of claims 1 to 7.
10. A computer storage medium storing a computer program, wherein the computer program when executed by a processor implements the steps of the neural network based customer evaluation method of any one of claims 1 to 7.
CN202210859101.9A 2022-07-21 2022-07-21 Customer evaluation method, system, equipment and storage medium based on neural network Pending CN115063035A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210859101.9A CN115063035A (en) 2022-07-21 2022-07-21 Customer evaluation method, system, equipment and storage medium based on neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210859101.9A CN115063035A (en) 2022-07-21 2022-07-21 Customer evaluation method, system, equipment and storage medium based on neural network

Publications (1)

Publication Number Publication Date
CN115063035A true CN115063035A (en) 2022-09-16

Family

ID=83205656

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210859101.9A Pending CN115063035A (en) 2022-07-21 2022-07-21 Customer evaluation method, system, equipment and storage medium based on neural network

Country Status (1)

Country Link
CN (1) CN115063035A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115238195A (en) * 2022-09-19 2022-10-25 太平金融科技服务(上海)有限公司深圳分公司 Method, apparatus, device, medium and product for determining target object
CN116150341A (en) * 2023-04-23 2023-05-23 之江实验室 Method for detecting claim event, computer device and storage medium

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115238195A (en) * 2022-09-19 2022-10-25 太平金融科技服务(上海)有限公司深圳分公司 Method, apparatus, device, medium and product for determining target object
CN116150341A (en) * 2023-04-23 2023-05-23 之江实验室 Method for detecting claim event, computer device and storage medium

Similar Documents

Publication Publication Date Title
US11334635B2 (en) Domain specific natural language understanding of customer intent in self-help
US10990901B2 (en) Training, validating, and monitoring artificial intelligence and machine learning models
JP7090936B2 (en) ESG-based corporate evaluation execution device and its operation method
EP3483797A1 (en) Training, validating, and monitoring artificial intelligence and machine learning models
US20230289665A1 (en) Failure feedback system for enhancing machine learning accuracy by synthetic data generation
CN115063035A (en) Customer evaluation method, system, equipment and storage medium based on neural network
US10671926B2 (en) Method and system for generating predictive models for scoring and prioritizing opportunities
KR102105319B1 (en) Esg based enterprise assessment device and operating method thereof
US10706359B2 (en) Method and system for generating predictive models for scoring and prioritizing leads
CN113535963A (en) Long text event extraction method and device, computer equipment and storage medium
CN110968664A (en) Document retrieval method, device, equipment and medium
US20230308360A1 (en) Methods and systems for dynamic re-clustering of nodes in computer networks using machine learning models
CN113065329A (en) Data processing method and device
CN111639494A (en) Case affair relation determining method and system
US11880394B2 (en) System and method for machine learning architecture for interdependence detection
CN113961811A (en) Conversational recommendation method, device, equipment and medium based on event map
CN113870998A (en) Interrogation method, device, electronic equipment and storage medium
KR20220099690A (en) Apparatus, method and computer program for summarizing document
CN112950392A (en) Information display method, posterior information determination method and device and related equipment
Shanmugarajah et al. WoKnack–A Professional Social Media Platform for Women Using Machine Learning Approach
CN114757275A (en) Business rule extraction method and device, storage medium and electronic equipment
CN117371562A (en) Method, device, equipment and medium for training ring-and-guest model and method, device, equipment and medium for recommending information
Sing et al. Judgemental Analysis of Data and Prediction Using Ann
Yadav et al. Impact of Double Negation through Majority Voting of Machine Learning Algorithms
Kumar et al. Framework On Text And Predictive Analytics

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination