CN117057756A - Client relationship management method and device based on RPA technology and related medium - Google Patents
Client relationship management method and device based on RPA technology and related medium Download PDFInfo
- Publication number
- CN117057756A CN117057756A CN202311312217.1A CN202311312217A CN117057756A CN 117057756 A CN117057756 A CN 117057756A CN 202311312217 A CN202311312217 A CN 202311312217A CN 117057756 A CN117057756 A CN 117057756A
- Authority
- CN
- China
- Prior art keywords
- data
- rpa
- client
- relationship management
- 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
Links
- 238000007726 management method Methods 0.000 title claims abstract description 97
- 238000005516 engineering process Methods 0.000 title claims abstract description 45
- 238000000034 method Methods 0.000 claims abstract description 84
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 44
- 238000013528 artificial neural network Methods 0.000 claims abstract description 40
- 230000002787 reinforcement Effects 0.000 claims abstract description 17
- 238000007781 pre-processing Methods 0.000 claims abstract description 10
- 238000012545 processing Methods 0.000 claims description 32
- 230000006870 function Effects 0.000 claims description 24
- 238000012549 training Methods 0.000 claims description 23
- 210000002569 neuron Anatomy 0.000 claims description 17
- 238000005457 optimization Methods 0.000 claims description 13
- 238000004590 computer program Methods 0.000 claims description 10
- 238000002922 simulated annealing Methods 0.000 claims description 10
- 238000000605 extraction Methods 0.000 claims description 8
- 230000004913 activation Effects 0.000 claims description 6
- 230000009466 transformation Effects 0.000 claims description 6
- 230000002159 abnormal effect Effects 0.000 claims description 5
- 238000010606 normalization Methods 0.000 claims description 4
- 230000008569 process Effects 0.000 abstract description 45
- 238000010801 machine learning Methods 0.000 abstract description 5
- 238000003058 natural language processing Methods 0.000 abstract description 4
- 230000009471 action Effects 0.000 description 11
- 230000007704 transition Effects 0.000 description 8
- 230000006399 behavior Effects 0.000 description 7
- 239000007787 solid Substances 0.000 description 7
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 230000008901 benefit Effects 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 4
- 230000003993 interaction Effects 0.000 description 4
- 239000011159 matrix material Substances 0.000 description 4
- 238000005309 stochastic process Methods 0.000 description 4
- 238000000137 annealing Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 3
- 239000002245 particle Substances 0.000 description 3
- 230000006978 adaptation Effects 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000001816 cooling Methods 0.000 description 2
- 230000005284 excitation Effects 0.000 description 2
- 210000002364 input neuron Anatomy 0.000 description 2
- 230000007774 longterm Effects 0.000 description 2
- 238000012423 maintenance Methods 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 210000004205 output neuron Anatomy 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 230000009849 deactivation Effects 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000007636 ensemble learning method Methods 0.000 description 1
- 230000008713 feedback mechanism Effects 0.000 description 1
- 230000005283 ground state Effects 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000009191 jumping Effects 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000010355 oscillation Effects 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 238000011426 transformation method Methods 0.000 description 1
- 230000001131 transforming effect Effects 0.000 description 1
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
- G06Q10/10—Office automation; Time management
- G06Q10/103—Workflow collaboration or project management
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- 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
- G06N3/092—Reinforcement learning
Abstract
The application discloses a client relationship management method, a device and a related medium based on RPA technology, wherein the method comprises the following steps: acquiring customer data and preprocessing the customer data; extracting target features from the client data through a deep neural network based on an RPA technology; combining reinforcement learning algorithm to make decision selection on target characteristics to obtain final decision strategy, so as to construct RPA customer relationship management model; and carrying out strategy prediction output on the appointed client data by utilizing the RPA client relation management model. The RPA customer relationship management model provided by the embodiment of the application adopts advanced machine learning and natural language processing technology, and can automatically understand and process a large amount of customer data, thereby greatly improving the efficiency of customer relationship management. Meanwhile, the RPA customer relation management model also comprises a rule engine, so that not only can a large amount of data be processed, but also a complex business process can be automated while business rules are followed.
Description
Technical Field
The present application relates to the field of computer software technologies, and in particular, to a method and an apparatus for managing a client relationship based on RPA technology, and a related medium.
Background
Customer relationship management (Customer Relationship Management, abbreviated as CRM) refers to a process of coordinating the interaction between enterprises and customers in sales, marketing and service by using corresponding information technology and internet technology to improve core competitiveness of the enterprises, thereby improving management modes and providing innovative personalized customer interaction and service for the customers. Its final goal is to attract new customers, retain old customers, turn existing customers into faithful customers, and increase the market.
In modern business environments, enterprises face increasingly complex customer relationship management challenges. The traditional manual mode for managing the customer relationship is inefficient and easy to make mistakes, and is difficult to deal with a large amount of customer data and complicated business processes. In some client relationship management methods based on big data, a corresponding service topology relationship is established by utilizing big data resources, and each topology chain is assigned; then extracting features of the client login information to obtain corresponding task information and client information; and comparing the client information with the service topological relation, performing similarity calculation on the task information and the client information after all the client information is contained in the service topological relation, generating a corresponding service link, and tracking, recording and feeding back the processing condition of the task information based on the service link. Although the efficiency of customer relationship management can be improved by the above manner or permission, the accuracy of the customer relationship management cannot be adjusted, for example, since the data volume is gradually increased, the initial service topology relationship is difficult to expire after the time, and therefore, a new service topology relationship is required to be continuously established to meet the subsequent needs, but because of different front-back development and different initial data volumes, a larger management accuracy problem exists in the newly established service topology relationship.
How to improve the efficiency and quality of customer relationship management is a matter that one skilled in the art would need to solve.
Disclosure of Invention
The embodiment of the application provides a client relationship management method, a client relationship management device, a client relationship management computer device and a client relationship management storage medium based on an RPA technology, which aim to improve the client relationship management efficiency and realize the accuracy and consistency of a business flow of client relationship management.
In a first aspect, an embodiment of the present application provides a client relationship management method based on RPA technology, including:
acquiring customer data and preprocessing the customer data; wherein the client data comprises client historical purchase data and client historical browsing data;
extracting target features from the client data through a deep neural network based on an RPA technology;
combining reinforcement learning algorithm to make decision selection on target characteristics to obtain final decision strategy, so as to construct RPA customer relationship management model;
and carrying out strategy prediction output on the appointed client data by utilizing the RPA client relation management model.
In a second aspect, an embodiment of the present application provides a client relationship management apparatus based on RPA technology, including:
the data acquisition unit is used for acquiring client data and preprocessing the client data; wherein the client data comprises client historical purchase data and client historical browsing data;
the feature extraction unit is used for extracting target features from the client data through a deep neural network based on an RPA technology;
the decision selection unit is used for carrying out decision selection on the target characteristics by combining with the reinforcement learning algorithm to obtain a final decision strategy so as to construct an RPA client relationship management model;
and the prediction output unit is used for carrying out strategy prediction output on the appointed client data by utilizing the RPA client relation management model.
In a third aspect, an embodiment of the present application provides a computer device, including a memory, a processor, and a computer program stored on the memory and capable of running on the processor, where the processor implements the RPA technology-based client relationship management method according to the first aspect when executing the computer program.
In a fourth aspect, an embodiment of the present application is a computer readable storage medium, where a computer program is stored, where the computer program is executed by a processor to implement the client relationship management method based on RPA technology according to the first aspect.
The embodiment of the application provides a client relationship management method, a device, computer equipment and a storage medium based on RPA technology, wherein the method comprises the following steps: acquiring customer data and preprocessing the customer data; wherein the client data comprises client historical purchase data and client historical browsing data; extracting target features from the client data through a deep neural network based on an RPA technology; combining reinforcement learning algorithm to make decision selection on target characteristics to obtain final decision strategy, so as to construct RPA customer relationship management model; and carrying out strategy prediction output on the appointed client data by utilizing the RPA client relation management model. The RPA customer relationship management model provided by the embodiment of the application adopts advanced machine learning and natural language processing technology, and can automatically understand and process a large amount of customer data, thereby greatly improving the efficiency of customer relationship management. At the same time, the RPA customer relationship management model also contains a powerful rules engine that can define rules according to a series of conditions and actions, and then automate the decision process according to these rules. This enables the RPA customer relationship management model to not only handle large amounts of data, but also to automate complex business processes while following business rules. Furthermore, the RPA customer relationship management model also uses an optimization algorithm, so that an optimization decision can be made when complex business processes are processed, the business efficiency is further improved, the decision process can be automated according to a predefined rule, and the accuracy and consistency of the business processes are ensured, so that the efficiency is maximized or the cost is minimized.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for the description of the embodiments will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present application, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic flow chart of a client relationship management method based on RPA technology according to an embodiment of the present application;
FIG. 2 is a schematic sub-flowchart of a client relationship management method based on the RPA technology according to an embodiment of the present application;
FIG. 3 is a schematic block diagram of a client relationship management device based on RPA technology according to an embodiment of the present application;
FIG. 4 is a sub-schematic block diagram of a client relationship management device based on RPA technology according to an embodiment of the present application;
FIG. 5 is a schematic diagram of a neuron structure in a client relationship management method based on the RPA technology according to an embodiment of the present application;
fig. 6 is a schematic diagram of a deep neural network structure in a client relationship management method based on RPA technology according to an embodiment of the present application.
Detailed Description
The following description of the embodiments of the present application will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are some, but not all embodiments of the application. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.
It should be understood that the terms "comprises" and "comprising," when used in this specification and the appended claims, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof.
It is also to be understood that the terminology used in the description of the application herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. As used in this specification and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise.
It should be further understood that the term "and/or" as used in the present specification and the appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes such combinations.
Referring to fig. 1, fig. 1 is a flow chart of a client relationship management method based on RPA technology according to an embodiment of the present application, which specifically includes: steps S101 to S104.
S101, acquiring client data and preprocessing the client data; wherein the client data comprises client historical purchase data and client historical browsing data;
s102, extracting target features from the client data through a deep neural network based on an RPA technology;
s103, combining a reinforcement learning algorithm to perform decision selection on the target features to obtain a final decision strategy, so as to construct an RPA client relationship management model;
s104, performing strategy prediction output on the appointed client data by utilizing the RPA client relation management model.
In this embodiment, customer data including, but not limited to, customer purchase history, behavior patterns, browsing preferences, and the like, is first obtained. These customer data come from various sources, such as databases, social media, email, and the like. The customer data is then cleaned and formatted for subsequent processing. And then, on the basis of the RPA technology, carrying out feature extraction on the client data through a deep neural network to obtain corresponding target features, and then, carrying out decision selection on the target features by utilizing a reinforcement learning algorithm so as to output a corresponding decision strategy. Through the process, an RPA customer relationship management model can be constructed, and the RPA customer relationship management model can be used for carrying out policy prediction on input customer data.
The RPA customer relationship management model provided by the embodiment adopts advanced machine learning and natural language processing technology, and can automatically understand and process a large amount of customer data, thereby greatly improving the efficiency of customer relationship management. At the same time, the RPA customer relationship management model also contains a powerful rules engine that can define rules according to a series of conditions and actions, and then automate the decision process according to these rules. This enables the RPA customer relationship management model to not only handle large amounts of data, but also to automate complex business processes while following business rules. Furthermore, the RPA customer relationship management model also uses an optimization algorithm, so that an optimization decision can be made when complex business processes are processed, the business efficiency is further improved, the decision process can be automated according to a predefined rule, and the accuracy and consistency of the business processes are ensured, so that the efficiency is maximized or the cost is minimized. Compared with the prior art, for example, compared with the traditional manual mode, the method can obviously improve the efficiency and the accuracy of customer relationship management. Compared with the mode of adopting big data and the like, the method and the device can ensure the efficiency and the precision of the customer relationship management and avoid the defect that the prior art only can improve one aspect (for example, only can improve the efficiency but can not ensure the precision, or can improve the precision but reduce the efficiency). In addition, the RPA customer relationship management model provided by the embodiment can directly or indirectly accumulate customer resources for users, realize long-term culture of potential customers, and bring long-term economic benefits for users. The RPA customer relationship management model provided by the embodiment can also improve the accuracy of sales decision, thereby ensuring the success rate of sales transactions and improving sales profits.
It should be noted that, the RPA client relationship management model provided in this embodiment can adapt to a plurality of different scenarios, for example:
retail industry: retailers may use the RPA customer relationship management model provided by the present embodiments to automate customer service flows, such as processing customer shopping queries, providing personalized shopping recommendations, or automatically processing return requests, etc.;
financial services: banks and other financial institutions may use the RPA customer relationship management model provided by the present embodiments to automate customer relationship management processes, such as processing a customer's loan application, providing personalized financial product recommendations, or automatically handling complaints and disputes.
The telecommunications industry: the RPA customer relationship management model provided by the present embodiments may be used by a telecommunications carrier to automate customer service flows, such as processing customer service queries, providing personalized package recommendations, or automatically handling billing issues.
Medical care: the RPA customer relationship management model provided by the present embodiments may be used by medical institutions to automate their customer relationship management processes, such as processing patient reservation requests, providing personalized health consultation, or automatically processing insurance claims.
Travel and hotel industries: travel agencies and hotels may use the RPA customer relationship management model provided by the present embodiments to automate customer service flows, such as processing customer reservation queries, providing personalized travel recommendations, or automatically processing unsubscribe requests.
It should be further noted that the reinforcement learning adopted in this embodiment may have the following advantages:
based on the evaluation: the goal of reinforcement learning is to maximize the jackpot, so it relies on the rewards function to evaluate the behavior of the intelligent agent. The intelligent agent maximizes the jackpot by attempting different actions and adjusts the behavior strategy based on the feedback of the rewards.
Interactivity: reinforcement learning is the process of learning an optimal strategy through the interaction of an intelligent agent with an environment. The intelligent agent observes the current state at each time step and selects an action to execute based on the current state. The environment will return to the next state and corresponding rewards based on the actions of the smart agent.
Sequence decision process: reinforcement learning is a sequential decision process in which the intelligent agent needs to select actions based on the current state and previous behavior history at each time step. The intelligent agent needs to take future rewards and risks into account to make optimal decisions.
Bonus hysteresis: in reinforcement learning, the behavior of the smart agent affects future rewards, and thus rewards are often hysteresis. That is, a reward for an action may be presented in several time steps in the future. This is a challenge for the intelligent agent because it needs to make decisions in cases where future cannot be predicted.
Based on sample evaluation: in reinforcement learning, the intelligent agent needs to learn the optimal strategy through interactions with the environment. Since the state space of an environment is typically very large, the intelligent agent needs to evaluate different strategies by sampling. The intelligent agent needs to constantly try for errors to find the optimal strategy.
Discrete random process: a discrete random process refers to a set of random variable sequences at discrete points in time, typically used to describe the evolution of random events over time.
In addition, the RPA technique used in this embodiment includes a powerful rule engine. The rules engine may define and integrate rules from a series of conditions and actions, and then automate the decision process from these rules, which enables the present embodiment to not only handle large amounts of data when employing RPA techniques, but also to automate complex business processes while following business rules. Specifically, the rule engine may specifically include: naming conventions, code annotations, log records, configuration information, folder structures, and rules for exception capture. The naming specification refers to naming according to an internally defined rule, and includes naming modes such as variable, parameter, flow name, file name and the like, and can follow the coding specification of software development; code annotation refers to annotations containing flow, each activity, and business logic; the log in the log record comprises two types, namely a system log and a service log, and the system log in the perfect frame has complete functions and does not need to be recorded again under the general condition; the configuration information refers to configuration information required by the project, the configuration information is required to be stored in a configuration file, a user account and a password are required to be stored in a server, and information required to be frequently modified can be stored in the server; the folder structure refers to a structure capable of clearly defining a project folder; the exception capturing means that a perfect exception capturing mechanism is required, including system exception and business exception, and exception information and screen capturing are recorded. The rules can be adjusted differently in different scenes to better adapt to scene requirements, for example, the naming convention rules are adjusted according to specific scenes so as to be more fit with the specific scenes. Of course, when the RPA technology is utilized, maintenance of the RPA technology is also required to be considered, so that the smoothness and reliability of the whole model are guaranteed. When RPA maintenance is performed, feedback of an operation stage needs to be collected in time, corresponding measures are taken based on the feedback, and meanwhile, a problem log, a perfect problem feedback mechanism and the like can be built.
In one embodiment, the obtaining the client data and preprocessing the client data includes:
carrying out missing value processing and abnormal value processing on the client data in sequence;
interpolation filling is carried out based on the missing value processing result and the abnormal value processing;
and carrying out normalization processing on the customer data after interpolation filling.
In this embodiment, for the acquired client data, first, missing value processing and outlier processing are performed on the acquired client data, and then interpolation filling is performed. In the practical application process, the central tendency measurement (mean/median/mode) of the attribute can be adopted for interpolation. As for continuous data, average and median values are typically used, while for discrete data mode is typically used. The measurement value well bears overall information, and certain characteristics of the original sample cannot be reproduced (or can be wiped off), but the overall fitting effect is not affected, and the method is a robust filling method. Taking mean interpolation as an example, considering that samples with homogeneity may exhibit similar characteristics, it is obviously simpler and less efficient to interpolate with the average of sub-samples with some same property therein, which is also called "conditional average filling", if the variable average of the whole samples is taken. The same holds for other central tendency metrics.
Meanwhile, since the customer data contains different dimensions, such as purchase history, behavior pattern, browsing preference and the like of the customer, the customer data is normalized so as to be conveniently and uniformly input into the deep neural network.
In an embodiment, as shown in fig. 2, the extracting, by the deep neural network, the target feature from the client data includes: steps S201 to S203.
S201, inputting the client data into an input layer of the deep neural network, and inputting the client data into a plurality of hidden layers through the input layer;
s202, performing linear transformation on the client data by utilizing neurons in each hidden layer to obtain intermediate characteristics of hidden layer output;
and S203, extracting information from the intermediate features through a nonlinear activation function of an output layer in the deep neural network, and outputting the information to obtain the target features.
In this embodiment, the client data is input into the deep neural network, and is used as an input neuron of the deep neural network, and after multi-layer information extraction, final output information (i.e., target information) is obtained. Specifically, in one layer of information extraction process, as shown in fig. 5, each neuron can receive input information from a plurality of other neurons (x 1-xn), extract the information, and then transmit the output to the next neuron. Wherein x 1-xn are output information of other neurons, and each neuron learns a weight (w 1-wn) for each input information after receiving information of other neurons; then carrying out weighted summation on the input information according to the weight, and adding offset information to obtain summarized information; finally, the information is further extracted by a nonlinear activation function (the excitation function is a dynamics rule with a short time scale of most neural network models, and is used for defining how the neurons change their own excitation values according to the activities of other neurons), so as to obtain final output information. The deep neural network is formed by connecting a large number of basic neurons, nodes are connected by weights, and the weights are obtained by algorithm training. The output of the deep neural network is different according to different connection modes, weights and activation functions.
As shown in fig. 6, the deep neural network is composed of multiple layers of neurons, each layer may include multiple neurons, the connection mode between the neurons and the activation function may be manually specified, so that the complexity of the deep neural network can approximate to any complex function.
In an embodiment, the combining reinforcement learning algorithm performs decision selection on the target feature to obtain a final decision strategy, so as to construct an RPA client relationship management model, including:
and carrying out decision selection on the target features by using a Markov chain to obtain the decision strategy.
In this embodiment, the Markov chain is a random process of discrete time, discrete states, characterized in that the current state is related to only the previous state and not to the past state. Thus, the Markov chain has the feature of "memoryless", i.e., its future state is only related to the current state, and not to the previous historical state. The markov chain was originally applied to describe the problem of walk, i.e. in a graph of nodes, the probability of a point jumping from a current position to a next position is only dependent on the current position and not on previous paths. This point may jump between different nodes over time.
Markov chains can be used to model and analyze such a process, and the probability distribution of points at different nodes at different time steps can be calculated so that future states can be predicted. In addition to the walk problem, markov chains are also used to simulate and predict various stochastic processes such as weather changes, stock price changes, credit ratings, etc. It is also used to solve various practical problems such as a language model in natural language processing, a probability map model in machine learning, and the like. While the Markov decision process is much like the Markov chain name, it is a two-concept solution and the problem solved and the application scenario are not very similar.
A markov chain is a mathematical model describing a stochastic process that assumes that the current state is related only to the previous state and not to the previous historical state. Markov chains are mainly used to analyze and predict discrete time, random processes of discrete states, such as weather changes, stock price changes, language models, etc. A markov decision process is a mathematical framework for describing sequential decision problems that formalizes the problem that a decision maker needs to make in an uncertain environment into a state space, a decision space, a state transition probability, and a reward function. The markov decision process is mainly used for solving the problem that a decision maker needs to make a decision in an uncertain environment, such as robot navigation, automatic driving, resource management, financial investment and the like.
In short, markov chains are mainly used to predict and simulate stochastic processes, whereas Markov decision processes are used to make optimal decisions in an uncertain environment. In some cases, a Markov chain may be used to describe the state transition probabilities in a Markov decision process, but it does not provide an optimal decision strategy, and therefore does not completely replace the Markov decision process in a sequence decision problem.
When we describe a discrete-time markov process, a state transition matrix can be used to represent transition probabilities between states. Assuming n states in our state space, the state transition matrix P is an n matrix, where P ij Representing the probability of transitioning from state (i) to state (j). The general form of the state transition matrix can be expressed as:
；
wherein each element Pij must fulfil the following condition:
，/>，i=1,2,……,n。
these conditions ensure that the transition probability of each state is between 0 and 1, and that the sum of the probabilities of each state transitioning from one state to another state is 1. State transition matrices are a very important concept in markov processes, which can be used to analyze and predict the behavior and properties of stochastic processes.
The markov process may be expressed in terms of a simple formula:
；
wherein X is t Representing states at time t, i and j represent two states in the state space respectively,representing the probability of transitioning from state i to state j.
In an embodiment, the client relationship management method based on RPA technology further includes:
acquiring a training data set with labels, and training the deep neural network by using the training data set;
and carrying out iterative updating on the deep neural network by adopting a gradient descent algorithm until the preset iterative times or the preset error range is reached.
In this embodiment, the gradient descent algorithm is one of the most commonly used optimization algorithms in machine learning, which continuously adjusts model parameters in an iterative manner so that the loss function is minimized. In practical applications, the gradient descent algorithm has various variants and improvements, including batch gradient descent, random gradient descent, small batch gradient descent, momentum gradient descent, adaptive gradient descent, etc. For example, the batch gradient descent algorithm is the most basic gradient descent algorithm that uses all training data in each iteration to calculate the gradient and then updates the model parameters. The bulk gradient descent algorithm has the advantage of faster convergence speed, but has the disadvantage of higher temporal and spatial complexity in computing gradients when processing large-scale data sets. The stochastic gradient descent algorithm is an online learning algorithm that randomly selects one sample in each iteration to calculate the gradient and then updates the model parameters. The random gradient descent algorithm has the advantages of low time and space complexity of calculating gradients, suitability for processing large-scale data sets, but low convergence speed and possible oscillation phenomenon.
Furthermore, instead of the gradient descent algorithm described above, the optimization may be performed in other manners, for example, using an Adam optimizer or the like. Adam optimizers are faster than SGD (random gradient descent), mainly derived from the calculation of its adaptive learning rate. Specifically, in the aspect of self-adaptive learning rate, the Adam optimizer utilizes the first moment and the second moment estimation value of the gradient to independently adjust the learning rate of each weight, and the self-adaptive learning rate method can bring more efficient updating and faster convergence; in terms of efficient gradient descent, unlike SGDs that require the same size update for all parameters, adam optimizers achieve efficient gradient descent by making smaller updates for frequently updated parameters and larger updates for infrequently updated parameters. Also, adam optimizers have less dependence on the initial learning rate, i.e., adam optimizers have less sensitivity to the initial learning rate, thereby reducing the time for the super parameter adjustment.
In an embodiment, the obtaining the training data set with the label and training the deep neural network by using the training data set includes:
performing model generalization treatment on the deep neural network by adopting a regularization method; the regularization method is any one of L1 regularization, L2 regularization and Dropout method.
In this embodiment, consider that when there is insufficient training data, overfitting (overfitting) is often caused, and the regularization method is a generic term for a class of methods that introduces additional information into the deep neural network at this time so as to prevent overfitting and improve the generalization performance of the deep neural network. In practical application, regularization methods such as an L1 regularization method, an L2 regularization method and a Dropout method can be adopted. Wherein, L2 regularization is to make the weight closer to the origin by adding a regularization term to the objective function. The L2 regularization method is also called weight decay, sometimes also called ridge regression (ridge regression). L2 weight decay is the most common form of weight decay, and other methods may be used to limit the scale of model parameters. For example, L1 parameter regularization may also be used. Similar to the L2 weight decay, the intensity of the L1 weight decay may also be controlled by scaling the positive hyper-parameters of the penalty term.
The Dropout method may also be referred to as a "discard method," or "random deactivation. The value of Dropout is mainly reflected in solving the over fitting problem of a model, and although the method is not the only means for solving the over fitting problem, the method is the best means for achieving the two-point combination of light weight and high efficiency. Similar to the Bagging ensemble learning method, the Dropout training process has a three-layer neural network (two input neurons, two hidden neurons, and one output neuron), and non-output neurons are randomly deleted from the basic network to construct subsets. Each training is thus like randomly selecting a different network model among the subsets for training, and finally selecting a best model by means of "voting" or averaging etc.
In other embodiments, indices such as mean square error or cross entropy loss are used to measure the prediction error of the deep neural network.
In one embodiment, the outputting the policy prediction for the specified client data by using the RPA client relationship management model includes:
and adopting a simulated annealing algorithm to perform decision optimization on the RPA customer relationship management model, and performing strategy prediction output on the appointed customer data by using the optimized RPA customer management model.
In this embodiment, the simulated annealing algorithm is derived from the solid annealing principle, heats the solid to a sufficiently high temperature, then slowly cools the solid, when the solid is heated, the internal particles of the solid become disordered with the temperature rise, the internal energy increases, and when the solid is slowly cooled, the particles gradually become ordered, the temperature reaches an equilibrium state at each temperature, and finally reaches a ground state at normal temperature, and the internal energy is reduced to the minimum. According to the Metropolis criterion, the probability that the particles will tend to equilibrate at temperature T is E (- ΔE/(kT)), where E is the internal energy at temperature T, ΔE is the amount of change it changes, and k is the Boltzmann constant. Simulating the combination optimization problem by using solid annealing, simulating the internal energy E as an objective function value f, and evolving the temperature T into a control parameter T to obtain a simulated annealing algorithm for solving the combination optimization problem: starting from the initial solution i and the initial value t of the control parameter, repeating the iteration of generating a new solution, calculating the objective function difference, accepting or rejecting the objective function difference and gradually attenuating the value t, wherein the current solution when the algorithm is terminated is the obtained approximate optimal solution, which is a heuristic random search process based on the Monte Carlo iteration solution. The annealing process is controlled by a Cooling Schedule comprising initial values t of control parameters and their decay factors Δt, the number of iterations L at each value of t and stop conditions S.
The simulated annealing algorithm can be decomposed into three parts, namely a solution space, an objective function and an initial solution. The basic idea is as follows:
(1) Initializing: initial temperature T (sufficiently large), initial solution state S (which is the starting point of algorithm iteration), number of iterations L for each T value
(2) For k=1, …, L is taken as steps (3) to 6:
(3) Generating a new solution S'
(4) Calculating an increment Δt=c (S') -C (S), wherein C (S) is an evaluation function
(5) If DeltaT <0 then accept S 'as the new current solution, otherwise accept S' as the new current solution with probability exp (-DeltaT/T).
(6) And if the termination condition is met, outputting the current solution as an optimal solution, and ending the program.
The termination condition is typically taken as terminating the algorithm when none of a number of new solutions in succession are accepted.
(7) T gradually decreases, and T- >0, and then go to step 2.
The generation and acceptance of new solutions for the simulated annealing algorithm can be divided into four steps:
the first step is to generate a new solution from the current solution in solution space by a generating function; in order to facilitate subsequent calculation and acceptance, the algorithm time consumption is reduced, and a method for generating a new solution by simply transforming the current new solution is generally selected, for example, all or part of elements constituting the new solution are replaced, interchanged, etc., and it is noted that the transformation method for generating the new solution determines the neighborhood structure of the current new solution, so that the selection of the cooling schedule is affected to a certain extent.
The second step is to calculate the objective function difference corresponding to the new solution. Since the objective function difference is generated by the transformation section only, the calculation of the objective function difference is preferably calculated in increments. It has been shown that for most applications this is the fastest way to calculate the difference in objective functions.
The third step is to judge whether the new solution is accepted or not, the judging basis is an acceptance criterion, the most common acceptance criterion is Metropolis criterion, if DeltaT <0, the acceptance S' is used as the new current solution S, otherwise, the probability exp (-DeltaT/T) is used as the new current solution S.
The fourth step is to replace the current solution with the new solution when the new solution is determined to be accepted, which is only required to implement the transformation part corresponding to the time when the new solution is generated in the current solution, and to correct the objective function value. At this point, the current solution achieves one iteration. The next round of testing can be started on this basis. And when the new solution is judged to be abandoned, continuing the next round of test on the basis of the original current solution.
The simulated annealing algorithm is irrelevant to an initial value, and the solution obtained by the algorithm is irrelevant to an initial solution state S (which is the starting point of algorithm iteration); the simulated annealing algorithm has asymptotic convergence, and has been proved to be a global optimization algorithm converging on a global optimal solution with probability l in theory; the simulated annealing algorithm has parallelism.
Fig. 3 is a schematic block diagram of a client relationship management apparatus 300 based on RPA technology according to an embodiment of the present application, where the apparatus 300 includes:
a data acquisition unit 301, configured to acquire client data and perform preprocessing on the client data; wherein the client data comprises client historical purchase data and client historical browsing data;
a feature extraction unit 302, configured to extract target features from the client data through a deep neural network based on an RPA technique;
the decision selection unit 303 is configured to combine the reinforcement learning algorithm to perform decision selection on the target feature, so as to obtain a final decision strategy, thereby constructing an RPA client relationship management model;
and the prediction output unit 304 is configured to perform policy prediction output on the specified client data by using the RPA client relationship management model.
In an embodiment, the data acquisition unit 301 includes:
the missing exception processing unit is used for sequentially carrying out missing value processing and exception value processing on the client data;
the interpolation filling unit is used for carrying out interpolation filling based on the missing value processing result and the abnormal value processing;
and the normalization unit is used for carrying out normalization processing on the customer data after interpolation filling.
In one embodiment, as shown in fig. 4, the feature extraction unit 302 includes:
a data input unit 401, configured to input the client data into an input layer of the deep neural network, and input the client data into a plurality of hidden layers through the input layer;
a linear transformation unit 402, configured to perform linear transformation on the client data by using neurons in each hidden layer, so as to obtain an intermediate feature of hidden layer output;
and the information extraction unit 403 is configured to extract information from the intermediate feature by using a nonlinear activation function of an output layer in the deep neural network, and output the extracted information to obtain the target feature.
In an embodiment, the decision selecting unit 303 comprises:
and the strategy acquisition unit is used for carrying out decision selection on the target characteristics by using a Markov chain to obtain the decision strategy.
In one embodiment, the client relationship management apparatus 300 based on RPA technology further includes:
the network training unit is used for acquiring a training data set with labels and training the deep neural network by utilizing the training data set;
and the iteration updating unit is used for carrying out iteration updating on the deep neural network by adopting a gradient descent algorithm until the preset iteration times or the preset error range is reached.
In an embodiment, the network training unit comprises:
the generalization processing unit is used for carrying out model generalization processing on the deep neural network by adopting a regularization method; the regularization method is any one of L1 regularization, L2 regularization and Dropout method.
In an embodiment, the prediction output unit 304 includes:
and the decision optimization unit is used for carrying out decision optimization on the RPA client relationship management model by adopting a simulated annealing algorithm, and carrying out strategy prediction output on the appointed client data by utilizing the optimized RPA client management model.
The embodiment of the application also provides a computer device, which can comprise a memory and a processor, wherein the memory stores a computer program, and the processor can realize the steps provided by the embodiment when calling the computer program in the memory. Of course, the computer device may also include various network interfaces, power supplies, and the like.
In the description, each embodiment is described in a progressive manner, and each embodiment is mainly described by the differences from other embodiments, so that the same similar parts among the embodiments are mutually referred. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant points refer to the description of the method section. It should be noted that it will be apparent to those skilled in the art that various modifications and adaptations of the application can be made without departing from the principles of the application and these modifications and adaptations are intended to be within the scope of the application as defined in the following claims.
It should also be noted that in this specification, relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
Claims (10)
1. A customer relationship management method based on RPA technology, comprising:
acquiring customer data and preprocessing the customer data; wherein the client data comprises client historical purchase data and client historical browsing data;
extracting target features from the client data through a deep neural network based on an RPA technology;
combining reinforcement learning algorithm to make decision selection on target characteristics to obtain final decision strategy, so as to construct RPA customer relationship management model;
and carrying out strategy prediction output on the appointed client data by utilizing the RPA client relation management model.
2. The RPA technology-based customer relationship management method according to claim 1, wherein the acquiring customer data and preprocessing the customer data includes:
carrying out missing value processing and abnormal value processing on the client data in sequence;
interpolation filling is carried out based on the missing value processing result and the abnormal value processing;
and carrying out normalization processing on the customer data after interpolation filling.
3. The RPA technology-based customer relationship management method according to claim 1, wherein the extracting target features for the customer data through the deep neural network comprises:
inputting the client data into an input layer of the deep neural network, and inputting the client data into a plurality of hidden layers through the input layer;
performing linear transformation on the client data by using neurons in each hidden layer to obtain intermediate characteristics of hidden layer output;
and extracting information from the intermediate features through a nonlinear activation function of an output layer in the deep neural network, and outputting to obtain the target features.
4. The RPA technology-based customer relationship management method according to claim 1, wherein the combining the reinforcement learning algorithm performs decision selection on the target feature to obtain a final decision strategy, thereby constructing an RPA customer relationship management model, and the method comprises:
and carrying out decision selection on the target features by using a Markov chain to obtain the decision strategy.
5. The RPA technology-based customer relationship management method of claim 3, further comprising:
acquiring a training data set with labels, and training the deep neural network by using the training data set;
and carrying out iterative updating on the deep neural network by adopting a gradient descent algorithm until the preset iterative times or the preset error range is reached.
6. The RPA technology-based customer relationship management method of claim 5, wherein the obtaining a training dataset with labels and training the deep neural network using the training dataset comprises:
performing model generalization treatment on the deep neural network by adopting a regularization method; the regularization method is any one of L1 regularization, L2 regularization and Dropout method.
7. The RPA technology-based customer relationship management method according to claim 1, wherein the performing policy prediction output on specified customer data using the RPA customer relationship management model includes:
and adopting a simulated annealing algorithm to perform decision optimization on the RPA customer relationship management model, and performing strategy prediction output on the appointed customer data by using the optimized RPA customer management model.
8. A customer relationship management apparatus based on RPA technology, comprising:
the data acquisition unit is used for acquiring client data and preprocessing the client data; wherein the client data comprises client historical purchase data and client historical browsing data;
the feature extraction unit is used for extracting target features from the client data through a deep neural network based on an RPA technology;
the decision selection unit is used for carrying out decision selection on the target characteristics by combining with the reinforcement learning algorithm to obtain a final decision strategy so as to construct an RPA client relationship management model;
and the prediction output unit is used for carrying out strategy prediction output on the appointed client data by utilizing the RPA client relation management model.
9. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the RPA technology-based customer relationship management method of any one of claims 1 to 7 when the computer program is executed.
10. A computer readable storage medium, wherein a computer program is stored on the computer readable storage medium, the computer program, when executed by a processor, implementing the RPA technology-based customer relationship management method according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311312217.1A CN117057756A (en) | 2023-10-11 | 2023-10-11 | Client relationship management method and device based on RPA technology and related medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311312217.1A CN117057756A (en) | 2023-10-11 | 2023-10-11 | Client relationship management method and device based on RPA technology and related medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117057756A true CN117057756A (en) | 2023-11-14 |
Family
ID=88659463
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311312217.1A Pending CN117057756A (en) | 2023-10-11 | 2023-10-11 | Client relationship management method and device based on RPA technology and related medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117057756A (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20040015386A1 (en) * | 2002-07-19 | 2004-01-22 | International Business Machines Corporation | System and method for sequential decision making for customer relationship management |
| CN114926151A (en) * | 2022-06-21 | 2022-08-19 | 中关村科学城城市大脑股份有限公司 | RPA flow automatic generation method and device based on reinforcement learning |
| CN115081538A (en) * | 2022-07-19 | 2022-09-20 | 平安银行股份有限公司 | Customer relationship identification method, device, equipment and medium based on machine learning |
-
2023
- 2023-10-11 CN CN202311312217.1A patent/CN117057756A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20040015386A1 (en) * | 2002-07-19 | 2004-01-22 | International Business Machines Corporation | System and method for sequential decision making for customer relationship management |
| CN114926151A (en) * | 2022-06-21 | 2022-08-19 | 中关村科学城城市大脑股份有限公司 | RPA flow automatic generation method and device based on reinforcement learning |
| CN115081538A (en) * | 2022-07-19 | 2022-09-20 | 平安银行股份有限公司 | Customer relationship identification method, device, equipment and medium based on machine learning |
Non-Patent Citations (1)
| Title |
|---|
| 李元章 等著: "马尔科夫过程与实用随机模型", 华南理工大学出版社, pages: 140 - 142 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Li et al. | Risk prediction in financial management of listed companies based on optimized BP neural network under digital economy | |
| US11663493B2 (en) | Method and system of dynamic model selection for time series forecasting | |
| US10002322B1 (en) | Systems and methods for predicting transactions | |
| KR20230070272A (en) | Computer-based systems, computing components, and computing objects configured to implement dynamic outlier bias reduction in machine learning models | |
| US11599740B2 (en) | Computer-based systems, computing components and computing objects configured to implement dynamic outlier bias reduction in machine learning models | |
| JP7365840B2 (en) | Automatic assessment of project acceleration | |
| CN109767032A (en) | A kind of business finance operation digital management optimization system based on data analysis | |
| Desirena et al. | Maximizing customer lifetime value using stacked neural networks: An insurance industry application | |
| Relich | A computational intelligence approach to predicting new product success | |
| Chen et al. | Session-based recommendation: Learning multi-dimension interests via a multi-head attention graph neural network | |
| Branchi et al. | Learning to act: a reinforcement learning approach to recommend the best next activities | |
| CN113657990A (en) | Ant-lion algorithm optimized NARX neural network risk prediction system and method | |
| US11468352B2 (en) | Method and system for predictive modeling of geographic income distribution | |
| Jadli et al. | A Novel LSTM-GRU-Based Hybrid Approach for Electrical Products Demand Forecasting. | |
| Ge et al. | Single stock trading with deep reinforcement learning: A comparative study | |
| Li et al. | A multiple long short-term model for product sales forecasting based on stage future vision with prior knowledge | |
| CN117057756A (en) | Client relationship management method and device based on RPA technology and related medium | |
| CN113435927B (en) | User willingness prediction method, device, equipment and storage medium | |
| CN115600009A (en) | Deep reinforcement learning-based recommendation method considering future preference of user | |
| Lian et al. | A fuel sales forecast method based on variational Bayesian structural time series | |
| CN114626898A (en) | Sales forecasting method, tool, system, equipment and storage medium | |
| US11004156B2 (en) | Method and system for predicting and indexing probability of financial stress | |
| Dash et al. | Designing an efficient predictor model using PSNN and crow search based optimization technique for gold price prediction | |
| Esfandiari et al. | Explicit preference elicitation for task completion time | |
| KR102596740B1 (en) | Method for predicting macroeconomic factors and stock returns in the context of economic uncertainty news sentiment using machine learning |
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 |