CN111784411A - Customer prediction method and device, electronic equipment and storage medium - Google Patents

Customer prediction method and device, electronic equipment and storage medium Download PDF

Info

Publication number
CN111784411A
CN111784411A CN202010677606.4A CN202010677606A CN111784411A CN 111784411 A CN111784411 A CN 111784411A CN 202010677606 A CN202010677606 A CN 202010677606A CN 111784411 A CN111784411 A CN 111784411A
Authority
CN
China
Prior art keywords
search
customer
sample
training
client
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
CN202010677606.4A
Other languages
Chinese (zh)
Inventor
牟树根
毛小平
徐博
徐海泉
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.)
Industrial and Commercial Bank of China Ltd ICBC
Original Assignee
Industrial and Commercial Bank of China Ltd ICBC
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 Industrial and Commercial Bank of China Ltd ICBC filed Critical Industrial and Commercial Bank of China Ltd ICBC
Priority to CN202010677606.4A priority Critical patent/CN111784411A/en
Publication of CN111784411A publication Critical patent/CN111784411A/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
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0201—Market modelling; Market analysis; Collecting market data
    • G06Q30/0202—Market predictions or forecasting for commercial activities
    • 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
    • G06N3/045—Combinations of networks
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Strategic Management (AREA)
  • Finance (AREA)
  • Development Economics (AREA)
  • Accounting & Taxation (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Linguistics (AREA)
  • Biomedical Technology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • Artificial Intelligence (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Game Theory and Decision Science (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The embodiment of the disclosure provides a customer prediction method, a customer prediction device, electronic equipment and a storage medium. The method comprises the following steps: acquiring attribute information of a client to be predicted; and processing the attribute information of the customer to be predicted by using a customer prediction model to obtain a prediction result aiming at the customer to be predicted, wherein the customer prediction model is obtained by training based on the attribute information of the sample customer and the telemarketing information corresponding to the sample customer.

Description

Customer prediction method and device, electronic equipment and storage medium
Technical Field
The embodiment of the disclosure relates to the technical field of computers, and more particularly, to a customer prediction method, a customer prediction device, an electronic device and a storage medium.
Background
The artificial intelligence technology is developed rapidly, the recognition capability of the artificial intelligence in the aspect of voice recognition is almost close to that of a real person scene, and the artificial intelligence is more highly anthropomorphic in the aspects of voice synthesis and video synthesis. All large Internet enterprises around the world step into the intelligent era, and intelligent transformation is realized by constructing intelligent service platforms with full channels, full media and full flows. Meanwhile, all large financial enterprises are transformed into intelligent types in a dispute.
In the related art, the product marketing mode is gradually changed from human marketing to intelligent marketing of robots.
In implementing the disclosed concept, the inventors found that there are at least the following problems in the related art: it is difficult to determine the target client more accurately using the related art.
Disclosure of Invention
In view of this, the disclosed embodiments provide a client prediction method, device, electronic device and storage medium.
One aspect of the embodiments of the present disclosure provides a customer prediction method, including:
acquiring attribute information of a client to be predicted; and
and processing the attribute information of the customer to be predicted by using a customer prediction model to obtain a prediction result aiming at the customer to be predicted, wherein the customer prediction model is obtained by training based on the attribute information of the sample customer and the telemarketing information corresponding to the sample customer.
According to the embodiment of the disclosure, the customer prediction model is trained based on the attribute information of the sample customer and the telemarketing information corresponding to the sample customer, and comprises:
obtaining a historical sample set, wherein each sample in the historical sample set comprises attribute information of a sample customer and telemarketing information corresponding to the sample customer;
dividing the historical sample set into a search sample set and a training sample set by utilizing a differentiable structure search algorithm;
training an initial search module by using the search sample set to obtain a target search module;
generating a convolutional neural network model based on the target search module; and
and training the convolutional neural network model by using the training sample set to obtain the customer prediction model.
According to the embodiment of the present disclosure, the target search module includes a plurality;
the generating a convolutional neural network model based on the target search module includes:
and obtaining a convolutional neural network model based on the series connection of the target search modules.
According to an embodiment of the present disclosure, the search sample set includes a search training set and a search validation set;
the training of the initial search module by using the search sample set to obtain the target search module comprises the following steps:
determining an initial search module by using the differentiable structure search algorithm, wherein the initial search module comprises at least two search units, and each search unit is a directed acyclic graph consisting of at least two nodes and at least one directed edge;
determining a set of candidate operations corresponding to each of the directed edges, wherein the set of candidate operations includes at least two candidate operations;
inputting the search training set into an initial search module using a first candidate operation to output a first predictive identifier of each sample client in the search training set and a first operation weight of the first candidate operation, and inputting the search validation set into an initial search module using the first candidate operation to output a second predictive identifier of each sample client in the search validation set and a second operation weight of the first candidate operation;
determining a first loss function according to the first operation weight, the real identifier of each sample client in the search training set and the first prediction identifier, and determining a second loss function according to the second operation weight, the real identifier of each sample client in the search verification set and the second prediction identifier;
adjusting structure parameters and network parameters until the first loss function and the second loss function meet preset conditions, and determining target operation corresponding to each directed edge; and
and determining a target searching module according to the target operation.
According to an embodiment of the present disclosure, the telemarketing information includes at least one of: marketing product type, dial-out type, dial-up or not, call duration and call-out time.
Another aspect of the disclosed embodiments provides a prediction apparatus, including:
the acquisition module is used for acquiring attribute information of a client to be predicted; and
and the prediction module is used for processing the attribute information of the customer to be predicted by utilizing a customer prediction model to obtain a prediction result aiming at the customer to be predicted, wherein the customer prediction model is obtained by training based on the attribute information of the sample customer and the telemarketing information corresponding to the sample customer.
According to an embodiment of the present disclosure, the prediction module includes:
the acquisition submodule is used for acquiring a historical sample set, wherein each sample in the historical sample set comprises attribute information of a sample client and telemarketing information corresponding to the sample client;
the dividing submodule is used for dividing the historical sample set into a search sample set and a training sample set by utilizing a differentiable structure search algorithm;
the first training submodule is used for training the initial search module by utilizing the search sample set to obtain a target search module;
a generation submodule for generating a convolutional neural network model based on the target search module; and
and the second training submodule is used for training the convolutional neural network model by using the training sample set to obtain the client prediction model.
According to an embodiment of the present disclosure, the search unit search module includes a plurality;
the generation submodule includes:
and the generating unit is used for obtaining a convolutional neural network model based on the serial connection of the plurality of searching modules.
Another aspect of the disclosed embodiments provides an electronic device, including:
one or more processors;
a memory for storing one or more programs,
wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the method as described above.
Another aspect of the embodiments of the present disclosure provides a computer-readable storage medium having stored thereon executable instructions, which when executed by a processor, cause the processor to implement the method as described above.
Another aspect of embodiments of the present disclosure provides a computer program comprising computer executable instructions for implementing the method as described above when executed.
According to the embodiment of the disclosure, the client prediction model obtained by training based on the attribute information of the sample client and the telemarketing information is adopted to predict the client to be predicted, wherein the client prediction model establishes the relationship between the client and the product in the intelligent marketing scene, so that the technical problem that the target client is difficult to accurately determine in the related technology is at least partially overcome, the target client is accurately determined, and the intelligent marketing efficiency is improved.
Drawings
The above and other objects, features and advantages of the present disclosure will become more apparent from the following description of embodiments of the present disclosure with reference to the accompanying drawings, in which:
FIG. 1 schematically illustrates an exemplary system architecture to which the customer forecasting method of the present disclosure may be applied;
FIG. 2 schematically illustrates a flow chart of a customer prediction method according to an embodiment of the present disclosure;
FIG. 3 schematically illustrates a block diagram of a convolutional neural network model, in accordance with an embodiment of the present disclosure;
FIG. 4 schematically shows a schematic diagram of a search unit according to an embodiment of the present disclosure;
FIG. 5 schematically shows a schematic diagram of another search unit according to an embodiment of the present disclosure;
FIG. 6 schematically shows a schematic diagram of yet another search unit according to an embodiment of the present disclosure;
FIG. 7 schematically shows a schematic diagram of yet another search unit according to an embodiment of the present disclosure;
FIG. 8 schematically illustrates a flow chart of a customer prediction method according to another embodiment of the present disclosure;
FIG. 9 discloses schematically a block diagram of a customer testing device according to an embodiment of the present disclosure; and
fig. 10 discloses schematically a block diagram of an electronic device according to an embodiment of the disclosure.
Detailed Description
Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. It should be understood that the description is illustrative only and is not intended to limit the scope of the present disclosure. In the following detailed description, for purposes of explanation, numerous specific details are set forth in order to provide a thorough understanding of the embodiments of the disclosure. It may be evident, however, that one or more embodiments may be practiced without these specific details. Moreover, in the following description, descriptions of well-known structures and techniques are omitted so as to not unnecessarily obscure the concepts of the present disclosure.
The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the disclosure. The terms "comprises," "comprising," and the like, as used herein, specify the presence of stated features, steps, operations, and/or components, but do not preclude the presence or addition of one or more other features, steps, operations, or components.
All terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art unless otherwise defined. It is noted that the terms used herein should be interpreted as having a meaning that is consistent with the context of this specification and should not be interpreted in an idealized or overly formal sense.
Where a convention analogous to "at least one of A, B and C, etc." is used, in general such a construction is intended in the sense one having skill in the art would understand the convention (e.g., "a system having at least one of A, B and C" would include but not be limited to systems that have a alone, B alone, C alone, a and B together, a and C together, B and C together, and/or A, B, C together, etc.). Where a convention analogous to "A, B or at least one of C, etc." is used, in general such a construction is intended in the sense one having skill in the art would understand the convention (e.g., "a system having at least one of A, B or C" would include but not be limited to systems that have a alone, B alone, C alone, a and B together, a and C together, B and C together, and/or A, B, C together, etc.).
The intelligent marketing can be understood as that the robot is used for marketing the product to be marketed to the client in a telephone mode instead of a manual mode according to the information of the client and the information of the product to be marketed, which are acquired in advance. In implementing the concepts of the present disclosure, the inventors found that at least the following problems exist in the related art of intelligent marketing: although information of a large number of customers can be acquired, due to the lack of the relationship between the customers and the products to be marketed, in the marketing process, the customers are selected to be marketed blindly, target customers are difficult to be determined accurately, and the efficiency of intelligent marketing is further influenced. Wherein the targeted customer may refer to a customer who is interested in purchasing the product to be marketed.
It can be seen that the reason why it is difficult to accurately determine the target client is as follows: there is a lack of relationship between the customer and the product being marketed. Accordingly, in order to solve the above problems, a relationship between the customer and the product to be marketed needs to be established. Since the robot usually markets the product to be marketed to the customer by telephone and generates the telemarketing information corresponding to the customer in the telephone communication with the customer, the telemarketing information can reflect the attitude of the customer to the product to be marketed, i.e. whether the customer intends to purchase or does not intend to purchase, and therefore, the telemarketing information can be used as a basis for establishing the relationship. Meanwhile, since the attribute information of the customer can usually reflect whether the customer has the ability to purchase the corresponding product to be marketed from one side, if a customer has the ability to purchase, the probability of purchasing the customer is high, and if a customer has no ability to purchase, the probability of purchasing the customer is low, the attribute information of the customer can be used as the basis for establishing the relationship.
Based on the above, the relationship between the customer and the product can be established according to the attribute information of the customer and the telemarketing information corresponding to the customer, so as to realize accurate target customer determination. The relationship may be established based on the obtained attribute information of a large number of customers and telemarketing information corresponding to the customers.
In smart marketing, in order to establish the above-described relationship, a relationship between a customer and a product may be established based on attribute information of a sample customer and telemarketing information corresponding to the sample customer. Wherein, the relationship between the customer and the product can be understood as a customer prediction model. In other words, the customer prediction model is generated based on the attribute information of the sample customer and the telemarketing information corresponding to the sample customer. And subsequently, after the attribute information of the client to be predicted is obtained, processing the attribute information of the client to be predicted by adopting a client prediction model to obtain a prediction result, wherein the prediction result represents whether the client to be predicted is a target client or not. The following description will be made in conjunction with examples.
The embodiment of the disclosure provides a customer prediction method and device for an intelligent marketing scene and electronic equipment capable of applying the method. The method includes a prediction process and a training process. In the prediction process, obtaining attribute information of a client to be predicted, and processing the attribute information of the client to be predicted by using a client prediction model obtained based on the attribute information of the sample client and telemarketing information corresponding to the sample client to obtain a prediction result aiming at the client to be predicted. In the training process, a historical sample set is obtained, the historical sample set is divided into a search sample set and a training sample set, the search samples in the search sample set are trained by utilizing a differentiable structure search algorithm to obtain a search module, a convolutional neural network model is generated based on the search module, and the convolutional neural network model is trained by utilizing the training sample set to obtain a client prediction model.
Fig. 1 schematically illustrates an exemplary system architecture 100 to which a customer prediction method may be applied, according to an embodiment of the disclosure. It should be noted that fig. 1 is only an example of a system architecture to which the embodiments of the present disclosure may be applied to help those skilled in the art understand the technical content of the present disclosure, and does not mean that the embodiments of the present disclosure may not be applied to other devices, systems, environments or scenarios.
As shown in fig. 1, a system architecture 100 according to this embodiment may include a server 101, a network 102, and terminal devices 103, 104. The network 102 serves as a medium for providing communication links between the terminal devices 103, 104 and the server 101. Network 102 may include various connection types, such as wired and/or wireless communication links, and so forth.
The client may use the terminal devices 103, 104 to interact with the server 101 over the network 102 to receive or send messages or the like. The terminal devices 103, 104 may have installed thereon various messaging client applications, such as shopping-like applications, web browser applications, search-like applications, instant messaging tools, mailbox clients, and/or social platform software, etc. (by way of example only).
The terminal devices 103, 104 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, and the like.
The server 101 may be a server that provides various services, such as a back-office management server (for example only) that initiates smart marketing to the terminal devices 103, 104.
It should be noted that the client prediction method provided by the embodiment of the present disclosure may be generally executed by the server 101. Accordingly, the electronic device provided by the embodiment of the present disclosure may be generally disposed in the server 101. The client prediction method provided by the embodiments of the present disclosure may also be performed by a server or a server cluster that is different from the server 101 and is capable of communicating with the terminal devices 103 and 104 and/or the server 101. Accordingly, the electronic device provided by the embodiment of the present disclosure may also be disposed in a server or a server cluster different from the server 101 and capable of communicating with the terminal devices 103 and 104 and/or the server 101.
It should be understood that the number of terminal devices, networks, and servers in fig. 1 is merely illustrative. There may be any number of terminal devices, networks, and servers, as desired for implementation.
Fig. 2 schematically illustrates a flow chart of a customer prediction method according to an embodiment of the present disclosure.
As shown in fig. 2, the method includes operations S210 to S220.
In operation S210, attribute information of a customer to be predicted is acquired.
In an embodiment of the present disclosure, a customer to be predicted may refer to a customer who predicts whether or not it is a target customer. The attribute information may include at least one of a customer name, a customer gender, a customer age, a location, an education level, a unit property, a marital status, a occupation, an administration level, an annual income, and whether or not a customer is a gold card.
It should be noted that the attribute information may be embodied in the form of a vector. Vectorization of attribute information may be achieved using one-hot encoding. Illustratively, the attribute information may include, for example, the customer's gender and whether it is a gold card customer. The gender of the customer comprises a man and a woman, and whether the money card customer corresponds to yes or no. "male" may be set to 1 and "female" to 0. "yes" is set to 1, and "no" is 0. For a certain client to be tested, the gender of the client to be tested is male, and the client to be tested is a gold card client, the attribute information is processed after the unique hot coding is adopted, and the obtained corresponding vectorized attribute information can be represented as [1, 0|1, 0 ].
In operation S220, the attribute information of the customer to be predicted is processed using a customer prediction model, which is trained based on the attribute information of the sample customer and the telemarketing information corresponding to the sample customer, to obtain a prediction result for the customer to be predicted.
In the embodiment of the disclosure, in order to achieve accurate determination of the target customer, a customer prediction model generated based on the attribute information of the sample customer and the telemarketing information corresponding to the sample customer may be adopted, and the attribute information of the customer to be predicted is processed, wherein the sample customer may refer to a customer participating in generation of the customer prediction model. The telemarketing message may include at least one of a marketing product type, an outbound type, whether to dial, a call duration, and an outbound time. Wherein an outbound call represents a telephone call initiated by the service center to the customer on its own initiative. The outbound time may indicate when to place a call to the customer. For example, if a call is placed to the customer at 15:00 every three to five weeks, the outbound time is 15:00 every three to five weeks.
And inputting the attribute information of the client to be predicted into a client prediction model, and outputting a prediction result corresponding to the client to be predicted. The prediction result may be whether the customer to be predicted is the target customer. The specific form of the prediction result can be a customer identifier of the customer to be predicted and a prediction identifier corresponding to the customer identifier. The predicted identity may be a first identity or a second identity, wherein the first identity may be indicative of the target customer and the second identity may be indicative of the target customer being absent.
According to the embodiment of the disclosure, the attribute information of the client to be predicted is input into the client prediction model, and if the prediction identifier corresponding to the client identifier is the first identifier, the client to be predicted can be indicated as the target client. If the prediction identifier corresponding to the client identifier is the second identifier, it may be stated that the client to be tested is not the target identifier. In addition, since the target customer refers to a customer who has an intention to purchase a product, the prediction result may be whether the user to be predicted has an intention to purchase a product. The predictive identifier may be a third identifier that may indicate an intent to purchase the product or a fourth identifier that may indicate an intent to purchase the product. Based on the above, the attribute information of the customer to be forecasted is input into the customer forecasting model, and if the forecasting identification of the original customer is the third identification, it can be stated that the customer to be forecasted intentionally purchases the product, which is the target customer. If the forecast identification of the original customer is the fourth identification, it can be stated that the customer to be forecasted does not intend to purchase the product, which is not the target customer.
It should be noted that the customer prediction model may be generated by training the attribute information of the sample customer and the telephone attribute information corresponding to the sample customer by using a machine learning algorithm. The machine learning algorithm may be automatic machine learning (AutoML). Compared with other machine learning algorithms, the automatic machine learning enables steps such as model construction and the like to be automated and can be applied without participation of excessive professionals.
According to the technical scheme of the embodiment of the disclosure, the client to be predicted is predicted by the client prediction model obtained by training based on the attribute information of the sample client and the telemarketing information, wherein the client prediction model establishes the relation between the client and the product in the intelligent marketing scene, so that the technical problem that the target client cannot be accurately determined in the related technology is at least partially overcome, the target client is accurately determined, and the intelligent marketing efficiency is improved.
Optionally, on the basis of the foregoing technical solution, the customer prediction model is trained based on attribute information of the sample customer and telemarketing information corresponding to the sample customer, and may include: a set of historical samples is obtained, wherein each sample in the set of historical samples includes attribute information of a sample customer and telemarketing information corresponding to the sample customer. The historical sample set is divided into a search sample set and a training sample set by utilizing a differentiable structure search algorithm. And training the initial search module by utilizing the search sample set to obtain the target search module. A convolutional neural network model is generated based on the target search module. And training the convolutional neural network model by utilizing the training sample set to obtain a customer prediction model.
In an embodiment of the present disclosure, in order to obtain the customer prediction model, a differentiated ArchiTecture Search (dart) algorithm may be used to train the historical sample set. The differentiable means that the candidate module or the search space is not discrete but continuous, so that iterative training optimization of the continuous space can be performed by a gradient descent method. Based on the optimization of gradient, can effectively shorten the time of training search module, and then save hardware resources, it is specific:
the acquired historical sample set may be divided into a search sample set and a training sample set using a differentiable structure search algorithm. Wherein the historical sample set may include at least two samples, each sample may include attribute information of a sample customer and telemarketing information corresponding to the sample customer. The set of search samples may be used to participate in training an initial search module, and the set of training samples may be used to participate in training a convolutional neural network model generated based on the initial search module. The number ratio of the search sample set and the training sample set may be set according to actual situations, and is not particularly limited herein. Alternatively, 20% of the historical sample set is determined as the search sample set, and 80% of the historical sample set is determined as the training sample set. The attribute information of the sample customer may include at least one of a customer name, a customer gender, a customer age, a location, an education level, a unit property, a marital status, a occupation, an administrative level, an annual income, and whether or not the customer is a gold card customer. The telemarketing information of the sample customer may include at least one of a marketing product type, an outbound type, whether to dial, a duration of the call, and an outbound time.
An initial search module may be determined, and the initial search module is trained using a search sample set to obtain a target search module. Wherein, the initial search module may include two or more search units. Each search unit may include nodes and directed edges, the number of the nodes being at least two, and the number of the directed edges being at least one. A set of candidate operations corresponding to each directed edge is provided, which may include at least two candidate operations, which may include convolution operations, pooling operations, identities, and join operations. The convolution operation may include a separable convolution operation and a hole convolution operation. The pooling operations may include a maximum pooling operation and an average pooling operation. According to an embodiment of the present disclosure, the process of training the initial search module to obtain the target search module is a process of determining a target operation for each directed edge, where the target operation is one candidate operation in a candidate operation set.
After determining the target search module, a convolutional neural network model may be generated from the target search module. Wherein the convolutional neural network model may include at least two target search modules. A plurality of target searching modules can be connected in series to obtain a convolutional neural network model. It should be noted that the number of the target search modules included in the convolutional neural network model and the number of channels in the target search modules may be set according to actual situations, and are not limited specifically herein. The actual situation described herein may refer to the size of the training sample set. I.e. the number of target search modules and the number of channels of the target search modules are matched to the size of the training sample set. Specifically, the method comprises the following steps: if the scale of the training sample set is larger, the number of target search modules and/or the number of channels in the target search modules can be set to be larger, so as to improve the model accuracy. If the scale of the training sample set is smaller, the number of target search modules and/or the number of channels of the target search modules can be set smaller to improve the model training speed. Illustratively, the number of target search modules is 20, and the number of channels of the target search modules is 36.
After the convolutional neural network model is determined, the convolutional neural network can be trained by adopting a training sample set to obtain a client prediction model. The training sample set is input into a convolutional neural network model to obtain the prediction identification of each sample client. And determining a loss function of the convolutional neural network model according to the real identification and the prediction identification of each sample client. And adjusting the network parameters of the convolutional neural network model until the output value of the loss function of the convolutional neural network model is less than or equal to a preset threshold value, and taking the trained convolutional neural network model as a client prediction model. It should be noted that, in the process of training the convolutional neural network model by using the training sample set to obtain the client prediction model, the training sample set may be divided into a training set, a verification set and a test set, where the verification set may be used to participate in training the convolutional neural network model to obtain the client prediction model. The validation set may be used to avoid overfitting during the training process. The overfitting phenomenon means that if the accuracy of the training set is improved, but the accuracy of the verification set is kept unchanged or reduced, the overfitting phenomenon can be indicated, and the training should be stopped. The test set may be used to test the performance of the trained convolutional neural network model, i.e., the performance of the customer prediction model.
It should be noted that the history sample set may be vectorized by using one-hot encoding. Specifically, the method comprises the following steps: the historical sample set is converted into a matrix arranged according to a time dimension through one-hot coding. Illustratively, as each column represents a historical sample having the same time.
The customer prediction model is built by adopting the differentiable structure search algorithm, so that the automatic model building is realized, and compared with the manual model building in the related technology, the time and the energy are saved. In addition, the number of the target search modules and/or the number of channels of the target search modules are adjusted according to the scale of the training sample set, so that a customer prediction model matched with the scale of the training sample set is established, and further the model precision and/or the model training speed are improved.
Optionally, on the basis of the above technical solution, the target search module includes a plurality of modules. Generating a convolutional neural network model based on a target search module may include: and obtaining a convolutional neural network model based on the serial connection of a plurality of target search modules.
In an embodiment of the present disclosure, the number of target search modules may be at least two. The plurality of target search modules can be connected in series in sequence to obtain a convolutional neural network model.
Illustratively, fig. 3 schematically illustrates a block diagram of a convolutional neural network model in accordance with an embodiment of the present disclosure. The convolutional neural network model in fig. 3 includes 8 object search modules, which are a first object search module, a second object search module, a third object search module, a fourth object search module, a fifth object search module, a sixth object search module, a seventh object search module, and an eighth object search module, respectively.
Optionally, on the basis of the above technical solution, the search sample set includes a search training set and a search validation set. Training the initial search module by using the search sample set to obtain a target search module, which may include: determining an initial search module by utilizing a differentiable structure search algorithm, wherein the initial search module comprises at least two search units, and each search unit is a directed acyclic graph formed by at least two nodes and at least one directed edge. A set of candidate operations corresponding to each directed edge is determined, wherein the set of candidate operations includes at least two candidate operations. The method includes inputting a search training set to an initial search module using a first candidate operation to output a first predictive identifier for each sample client in the search training set and a first operation weight for the first candidate operation, and inputting a search validation set to the initial search module using the first candidate operation to output a second predictive identifier for each sample client in the search validation set and a second operation weight for the first candidate operation. And determining a first loss function according to the first operation weight, the real identifier and the first prediction identifier of each sample client in the search training set, and determining a second loss function according to the second operation weight, the real identifier and the second prediction identifier of each sample client in the search verification set. And adjusting the structural parameters and the network parameters until the first loss function and the second loss function meet preset conditions, and determining target operation corresponding to each directed edge. And determining a target searching module according to the target operation.
In the embodiment of the present disclosure, in order to obtain the target search module, a differentiable structure search algorithm may be adopted, specifically:
the initial search module may be determined using a differentiable structure search algorithm, wherein the initial search module may be composed of a plurality of search units. Each search unit is a directed acyclic graph formed by at least two nodes and at least one directed edge. The nodes in the search unit may be divided into input nodes, intermediate nodes, and output nodes. The number of input nodes is two, and the number of output nodes is one. For each search cell, the input node is the output of the first two levels of the search cell. The intermediate node is obtained by re-summing its predecessor nodes through candidate operation transitions. The output nodes are obtained by splicing the channels of all the intermediate nodes.
A set of candidate operations corresponding to each directed edge may be determined, where the set of candidate operations may include at least two candidate operations. The candidate operations may include convolution operations, pooling operations, identities, and join operations. The convolution operation may include a separable convolution operation and a hole convolution operation. The pooling operations may include a maximum pooling operation and an average pooling operation. Alternatively, the separable convolution operations may include a 3 × 3 separable convolution and a 5 × 5 separable convolution. The hole convolution may include a 3 × 3 hole convolution and a 5 × 5 hole convolution. The maximum pooling operation may be 3 x 3 maximum pooling. The average pooling operation may be 3 x 3 average pooling.
Fig. 4 schematically shows a schematic diagram of a search unit according to an embodiment of the present disclosure. Fig. 4 does not show the input node and the output node of the search unit, and node 0, node 1, node 2, and node 3 are all intermediate nodes. The question mark on each directed edge indicates that the candidate operation set corresponding to the directed edge is not determined.
Fig. 5 schematically shows a schematic diagram of another search unit according to an embodiment of the present disclosure. As in fig. 4, fig. 5 does not show the input node and the output node of the search unit, and node 0, node 1, node 2, and node 3 are all intermediate nodes. In fig. 5, a corresponding set of candidate operations is set for each directed edge. Different lines represent different candidate operations.
The search sample set may be divided into a search training set and a search validation set. The search training set and the search verification set can be used for completing training of the initial search module in a matched mode. The ratio of the search training set to the search validation set may be set according to actual conditions, and is not particularly limited herein. Optionally, the search sample set is divided into a search training set and a search validation set in a ratio of 1: 1.
The search training set may be input to an initial search module that uses a first candidate operation, and a first predictive identifier for each sample customer in the search training set may be output, along with a first operation weight for the first candidate operation. Wherein the first candidate operation is any one of the candidate operations in the candidate operation set. That is, for each candidate operation, the search training set may be input to the initial search module that used the candidate operation, resulting in a first operation weight for the candidate operation. The first predictive identity may be the first identity or the second identity. The first identification may indicate a target customer and the second identification may indicate a non-target customer.
The search validation set may be input to an initial search module that uses the first candidate operation, and the second predictive identity for each sample client in the search validation set may be output, along with the second operation weight for the first candidate operation. Wherein the first candidate operation is any one of the candidate operations in the candidate operation set. That is, for each candidate operation, the search validation set may be input to the initial search module that used the candidate operation, resulting in a second operation weight for the candidate operation. Likewise, the second predictive flag may be the first flag or the second flag. The first identification may indicate a target customer and the second identification may indicate a non-target customer.
It should be noted that, according to the embodiments of the present disclosure, in order to achieve continuity of the search space, the normalized exponential function may be used to reuse the normalized exponential probability weighted representation of different candidate operations for the first operation weight of the first candidate operation. Likewise, the normalized exponential function may be used to reuse the normalized exponential probability weighted representation of a different candidate operation for the second operational weight of the first candidate operation.
A first loss function may be derived based on the first operational weight, the true identity and the first predicted identity of each sample customer in the search training set. The first loss function is a function of a network parameter and a structure parameter. Wherein the network parameters may represent weights of the convolutional neural network.
Likewise, a second loss function may be derived based on the second operational weight, the true identity of each sample customer in the search validation set, and the second predicted identity. The second loss function is a function of the network parameter and the configuration parameter.
After the first loss function and the second loss function are obtained, the structure parameters and the network parameters can be subjected to bidirectional optimization, so that the first loss function and the second loss function meet preset conditions. The bidirectional optimization is as follows: an optimal configuration parameter is determined such that the output value of the second loss function is minimized, and an optimal network parameter is determined such that the output value of the first loss function is minimized. In the bidirectional optimization process, under the condition that the network parameters are not changed, the structural parameters are adjusted, so that the output value of the second loss function is minimum. And under the condition that the structure parameter is not changed, the network parameter is adjusted to enable the output value of the first loss function to be minimum. Based on the above, in the bidirectional optimization process, since the network parameter is adjusted under the condition that the structure parameter is not changed, so that the output value of the first loss function is the minimum, at this time, the first loss function is the function of the network parameter. Since the structure parameter is adjusted under the condition that the network parameter is not changed, so that the output value of the second loss function is minimum, at this time, the second loss function is the function of the structure parameter.
According to embodiments of the present disclosure, the search training set may be used to adjust network parameters, while the search validation set may be used to adjust structural parameters.
It should be noted that, in order to implement bidirectional optimization, an approximate iterative optimization strategy may be adopted: namely, the structure parameter and the network parameter are mutually optimized by a gradient descent method.
After determining the structural parameters, a target operation may be determined from the set of candidate operations for each directed edge based on the structural parameters. The target operation is the candidate operation with the highest probability in the candidate operation set. According to the embodiment of the present disclosure, since the first candidate operation is any one of the candidate operations in the candidate operation set, the target operation is the one first candidate operation.
According to embodiments of the present disclosure, after determining the target operation for each directed edge, the search unit is trained. Since the initial search module includes a plurality of identical search units, the initial search module is trained, and the trained initial search module may be referred to as a target search module. On the basis, in order to improve the precision of the search unit, multiple times of training can be performed randomly by adopting the random seed, and then one search unit which is optimal in performance is selected from the multiple times of training to be used as the search unit which forms the target search module. Optionally, 4 times of training are performed randomly based on the random seed, each training iteration is performed 100 times, and one search unit which shows the best performance is selected from the 4 times of training as the search unit which forms the target search module.
Fig. 6 schematically shows a schematic diagram of yet another search unit according to an embodiment of the present disclosure. As in fig. 4 and 5, fig. 6 does not show the input node and the output node of the search unit, and node 0, node 1, node 2, and node 3 are all intermediate nodes. The process of jointly optimizing the structural parameters and the network parameters by solving a two-layer optimization problem is shown in fig. 6.
Fig. 7 schematically shows a schematic diagram of yet another search unit according to an embodiment of the present disclosure. Like fig. 4, 5 and 6, fig. 7 does not show the input node and the output node of the search unit, and node 0, node 1, node 2 and node 3 are all intermediate nodes. In fig. 7, the target operation of each directed edge is determined, and fig. 7 is the search unit after training.
For better understanding of the generation process of the target search module in the embodiment of the present disclosure, the following description is made in conjunction with a formula, specifically as follows:
the setting search unit includes N nodes x(i)Each node x(i)Representing a feature map in a convolutional neural network. Each directed edge (i,) is to node x(i)Candidate operation o(i，j)。
The output result of each node is expressed by the following formula:
Figure BDA0002583025770000161
where, j represents a sequence number smaller than the sequence number, which represents the sequence number of the current node.
To make the search space continuous, the selection of the classification of each candidate operation is continued as a normalized exponential function of all possible operations:
Figure BDA0002583025770000162
wherein the content of the first and second substances,
Figure BDA0002583025770000163
the operation weight of the candidate operation of the directed edge (i, j) is represented. O represents the set of candidate operations for each directed edge.
The process of selecting different candidate operations and their weights by the process of equation (2) is a continuous process. After the discrete space is serialized, the task of determining the target search module is simplified as follows: learning structural parameters
Figure BDA0002583025770000171
And learning the network weight w.
After establishing the structure of the search unit and the continuous space to be searched, the parameters of the structure are followed
Figure BDA0002583025770000172
And the network weight w, bidirectional optimization can be adopted, namely:
Figure BDA0002583025770000173
Figure BDA0002583025770000174
wherein L istrainRepresenting a first loss function, LvalRepresenting a second loss function. The formula (3) and the formula (4) show that an optimal structure parameter is determined
Figure BDA0002583025770000175
So that the second loss function LvalIs minimized and an optimal network parameter w is determined*So that the first loss function LtrainThe output value of (c) is minimum.
To solve the above two-way optimization problem, an approximate iterative optimization strategy, i.e., a gradient descent method, may be used to implement the structural parameters
Figure BDA0002583025770000176
And mutual optimization of the network weights w. The following formula (5), formula (6), formula (7) and formula (8).
Setting the current structure parameters when the k step is reached
Figure BDA0002583025770000177
Weighting w the network by heading in a direction that reduces training lossk-1Is adjusted to wk. Then, the network weight w is maintainedkUnchanged, updating the structural parameters
Figure BDA00025830257700001714
So that it can make the second loss function LvalHas the smallest output value, i.e.
Figure BDA0002583025770000178
Where ξ represents the learning rate with the gradient decreasing.
From equation (5), the structural parameters can be obtained
Figure BDA0002583025770000179
Namely:
Figure BDA00025830257700001710
where w' represents the network weights of a model for a single step iteration.
Figure BDA00025830257700001711
Since the calculation of equation (6) is costly, the approximate calculation can be performed by the following equations (7) and (8):
Figure BDA00025830257700001712
Figure BDA00025830257700001713
in which a scalar parameter is represented, the value of which is small.
The search unit of the continuous structure can be determined. In order to improve the precision of the search unit, multiple times of training can be performed randomly by adopting the random seed, and one search unit which is optimal in performance is selected from the multiple times of training to be used as the search unit which forms the target search module. Optionally, 4 times of training are performed randomly based on the random seed, each training iteration is performed 100 times, and one search unit which shows the best performance is selected from the 4 times of training as the search unit which forms the target search module.
In obtaining structural parameters of continuous structure
Figure BDA0002583025770000181
Thereafter, the search unit of the discrete structure may retain k candidate operations with the highest probability for each directed edge, and the probability of each candidate operation may be determined by the following formula (9):
Figure BDA0002583025770000182
for each directed edge, the candidate operation with the highest probability is used as the target operation of the directed edge, and the target operation may be set for each directed edge by using an argmax function.
Optionally, on the basis of the above technical solution, the telemarketing message includes at least one of: marketing product type, dial-out type, dial-up or not, call duration and call-out time.
Fig. 8 schematically illustrates a flow chart of a customer prediction method according to another embodiment of the present disclosure.
As shown in fig. 8, the method includes operations S801 to S814.
In operation S801, a history sample set is obtained, wherein each sample in the history sample set includes attribute information of a sample customer and telemarketing information corresponding to the sample customer.
In operation S802, a history sample set is divided into a search sample set and a training sample set using a differentiable structure search algorithm, wherein the search sample set includes a search training set and a search validation set.
In operation S803, an initial search module is determined, wherein the initial search module includes at least two search units, wherein each search unit is a directed acyclic graph composed of at least two nodes and at least one directed edge.
In operation S804, a set of candidate operations corresponding to each directed edge is determined, where the set of candidate operations includes at least two candidate operations.
In operation S805, the search training set is input to the initial search module using the first candidate operation to output a first predictive identifier of each sample client in the search training set and a first operation weight of the first candidate operation.
In operation S806, the search validation set is input to the initial search module using the first candidate operation to output a second predictive identifier of each sample client in the search validation set and a second operation weight of the first candidate operation.
In operation S807, a first loss function is determined according to the first operation weight, the true identity of each sample client in the search training set, and the first predicted identity.
In operation S808, a second loss function is determined according to the second operation weight, the true identity of each sample client in the search validation set, and the second predicted identity.
In operation S809, the structure parameter and the network parameter are adjusted until the first loss function and the second loss function satisfy a preset condition, and then a target operation corresponding to each directed edge is determined.
In operation S810, a target search module is determined according to the target operation.
In operation S811, a convolutional neural network model is obtained based on a plurality of target search modules connected in series.
In operation S812, the convolutional neural network model is trained using the training sample set to obtain a customer prediction model.
In operation S813, attribute information of the customer to be predicted is acquired.
In operation S814, the attribute information of the customer to be predicted is processed using the customer prediction model, and a prediction result for the customer to be predicted is obtained.
According to the technical scheme disclosed by the embodiment, the client prediction model obtained based on the attribute information of the sample client and the telemarketing information training is adopted to predict the client to be predicted, wherein the client prediction model establishes the relation between the client and the product in the intelligent marketing scene, so that the technical problem that the target client is difficult to accurately determine in the related technology is at least partially overcome, the target client is accurately determined, and the intelligent marketing efficiency is improved. In addition, the customer prediction model is built by adopting a differentiable structure search algorithm, so that the automatic model building is realized, and compared with the manual model building in the related technology, the time and the energy are saved. Meanwhile, the number of the target search modules and/or the number of channels of the target search modules are/is adjusted according to the scale of the training sample set, so that a customer prediction model matched with the scale of the training sample set is established, and further the model precision and/or the model training speed are/is improved.
FIG. 9 schematically shows a block diagram of a customer testing device according to an embodiment of the disclosure.
As shown in fig. 9, the client prediction apparatus 900 may include an acquisition module 910 and a prediction module 920. The acquisition module 910 is communicatively coupled to the prediction module 920.
An obtaining module 910, configured to obtain attribute information of a client to be predicted.
And the prediction module 920 is configured to process the attribute information of the customer to be predicted by using a customer prediction model to obtain a prediction result for the customer to be predicted, where the customer prediction model is trained based on the attribute information of the sample customer and the telemarketing information corresponding to the sample customer.
According to the technical scheme of the embodiment of the disclosure, the client prediction model obtained based on the attribute information of the sample client and the telemarketing information training is adopted to predict the client to be predicted, wherein the client prediction model establishes the relation between the client and the product in the intelligent marketing scene, so that the technical problem that the target client is difficult to accurately determine in the related technology is at least partially overcome, the target client is accurately determined, and the intelligent marketing efficiency is improved.
Optionally, on the basis of the foregoing technical solution, the prediction module 920 may include an obtaining sub-module, a dividing sub-module, a first training sub-module, a generating sub-module, and a second training sub-module.
And the obtaining submodule is used for obtaining a historical sample set, wherein each sample in the historical sample set comprises the attribute information of the sample client and the telemarketing information corresponding to the sample client.
And the dividing submodule is used for dividing the historical sample set into a search sample set and a training sample set by utilizing a differentiable structure search algorithm.
And the first training submodule is used for training the initial search module by utilizing the search sample set to obtain the target search module.
And the generation submodule is used for generating a convolutional neural network model based on the target search module.
And the second training submodule is used for training the convolutional neural network model by utilizing the training sample set to obtain a client prediction model.
Optionally, on the basis of the above technical solution, the target search module includes a plurality of modules. The generation submodule may include a generation unit.
And the generating unit is used for obtaining a convolutional neural network model based on the serial connection of a plurality of target searching modules.
Optionally, on the basis of the above technical solution, the search sample set includes a search training set and a search validation set. The first training submodule may include a first determination unit, a second determination unit, an output unit, a third determination unit, a fourth determination unit, and a fifth determination unit.
The device comprises a first determining unit and a second determining unit, wherein the first determining unit is used for determining an initial searching module by utilizing a differentiable structure searching algorithm, the initial searching module comprises at least two searching units, and each searching unit is a directed acyclic graph formed by at least two nodes and at least one directed edge.
And a second determining unit, configured to determine a candidate operation set corresponding to each directed edge, where the candidate operation set includes at least two candidate operations.
An output unit configured to input the search training set to the initial search module using the first candidate operation to output the first prediction identification of each sample client in the search training set and the first operation weight of the first candidate operation, and input the search validation set to the initial search module using the first candidate operation to output the second prediction identification of each sample client in the search validation set and the second operation weight of the first candidate operation.
And the third determining unit is used for determining the first loss function according to the first operation weight, the real identifier and the first prediction identifier of each sample client in the search training set, and determining the second loss function according to the second operation weight, the real identifier and the second prediction identifier of each sample client in the search verification set.
And the fourth determining unit is used for adjusting the structural parameters and the network parameters until the first loss function and the second loss function meet preset conditions, and then determining the target operation corresponding to each directed edge.
And the fifth determining unit is used for determining the target searching module according to the target operation.
Optionally, on the basis of the above technical solution, the telemarketing message includes at least one of: marketing product type, dial-out type, dial-up or not, call duration and call-out time.
Any number of modules, sub-modules, units, or at least part of the functionality of any number thereof according to embodiments of the present disclosure may be implemented in one module. Any one or more of the modules, sub-modules, units according to the embodiments of the present disclosure may be implemented by being split into a plurality of modules. Any one or more of the modules, sub-modules, units according to the embodiments of the present disclosure may be implemented at least partially as a hardware Circuit, such as a Field Programmable Gate Array (FPGA), a Programmable Logic Array (PLA), a system on a chip, a system on a substrate, a system on a package, an Application Specific Integrated Circuit (ASIC), or may be implemented by hardware or firmware in any other reasonable manner of integrating or packaging a Circuit, or implemented by any one of three implementations of software, hardware, and firmware, or any suitable combination of any of them. Alternatively, one or more of the modules, sub-modules, units according to embodiments of the disclosure may be implemented at least partly as computer program modules, which, when executed, may perform corresponding functions.
For example, any number of the obtaining module 910 and the predicting module 920 may be combined and implemented in one module/sub-module/unit, or any one of the modules/sub-modules/units may be split into a plurality of modules/sub-modules/units. Alternatively, at least part of the functionality of one or more of these modules/sub-modules/units may be combined with at least part of the functionality of other modules/sub-modules/units and implemented in one module/sub-module/unit. According to an embodiment of the disclosure, at least one of the obtaining module 910 and the predicting module 920 may be implemented at least partially as a hardware Circuit, such as a Field Programmable Gate Array (FPGA), a Programmable Logic Array (PLA), a system on a chip, a system on a substrate, a system on a package, an Application Specific Integrated Circuit (ASIC), or any other reasonable manner of integrating or packaging a Circuit. Alternatively, at least one of the obtaining module 910 and the predicting module 920 may be at least partially implemented as a computer program module, which when executed may perform the corresponding functions.
It should be noted that the client prediction device part in the embodiment of the present disclosure corresponds to the client prediction method part in the embodiment of the present disclosure, and the description of the client prediction device part specifically refers to the client prediction method part, and is not repeated herein.
Fig. 10 schematically shows a block diagram of an electronic device adapted to implement the above described method according to an embodiment of the present disclosure. The electronic device shown in fig. 10 is only an example, and should not bring any limitation to the functions and the scope of use of the embodiments of the present disclosure.
As shown in fig. 10, an electronic device 1000 according to an embodiment of the present disclosure includes a processor 1001 that can perform various appropriate actions and processes according to a program stored in a Read-Only Memory (ROM) 1002 or a program loaded from a storage section 1008 into a Random Access Memory (RAM) 1003. The processor 1001 may include, for example, a general purpose microprocessor (e.g., a CPU), an instruction set processor and/or related chipset, and/or a special purpose microprocessor (e.g., an Application Specific Integrated Circuit (ASIC)), among others. The processor 1001 may also include onboard memory for caching purposes. The processor 1001 may include a single processing unit or multiple processing units for performing different actions of a method flow according to embodiments of the present disclosure.
In the RAM1003, various programs and data necessary for the operation of the electronic apparatus 1000 are stored. The processor 1001, ROM 1002, and RAM1003 are connected to each other by a bus 1004. The processor 1001 performs various operations of the method flow according to the embodiments of the present disclosure by executing programs in the ROM 1002 and/or the RAM 1003. Note that the programs may also be stored in one or more memories other than the ROM 1002 and the RAM 1003. The processor 1001 may also perform various operations of the method flows according to embodiments of the present disclosure by executing programs stored in the one or more memories.
Electronic device 1000 may also include an input/output (I/O) interface 1005, the input/output (I/O) interface 1005 also being connected to bus 1004, according to an embodiment of the present disclosure. Electronic device 1000 may also include one or more of the following components connected to I/O interface 1005: an input section 1006 including a keyboard, a mouse, and the like; an output section 1007 including a Display such as a Cathode Ray Tube (CRT), a Liquid Crystal Display (LCD), and a speaker; a storage portion 1008 including a hard disk and the like; and a communication section 1009 including a network interface card such as a LAN card, a modem, or the like. The communication section 1009 performs communication processing via a network such as the internet. The driver 1010 is also connected to the I/O interface 1005 as necessary. A removable medium 1011 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like is mounted on the drive 1010 as necessary, so that a computer program read out therefrom is mounted into the storage section 1008 as necessary.
According to embodiments of the present disclosure, method flows according to embodiments of the present disclosure may be implemented as computer software programs. For example, embodiments of the present disclosure include a computer program product comprising a computer program embodied on a computer readable storage medium, the computer program containing program code for performing the method illustrated by the flow chart. In such an embodiment, the computer program may be downloaded and installed from a network through the communication part 1009 and/or installed from the removable medium 1011. The computer program performs the above-described functions defined in the system of the embodiment of the present disclosure when executed by the processor 1001. The systems, devices, apparatuses, modules, units, etc. described above may be implemented by computer program modules according to embodiments of the present disclosure.
The present disclosure also provides a computer-readable storage medium, which may be contained in the apparatus/device/system described in the above embodiments; or may exist separately and not be assembled into the device/apparatus/system. The computer-readable storage medium carries one or more programs which, when executed, implement the method according to an embodiment of the disclosure.
According to an embodiment of the present disclosure, the computer-readable storage medium may be a non-volatile computer-readable storage medium. Examples may include, but are not limited to: a portable Computer diskette, a hard disk, a Random Access Memory (RAM), a Read-Only Memory (ROM), an erasable Programmable Read-Only Memory (EPROM) (erasable Programmable Read-Only Memory) or flash Memory), a portable compact Disc Read-Only Memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the preceding. In the present disclosure, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device.
For example, according to embodiments of the present disclosure, a computer-readable storage medium may include the ROM 1002 and/or the RAM1003 described above and/or one or more memories other than the ROM 1002 and the RAM 1003.
The flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams or flowchart illustration, and combinations of blocks in the block diagrams or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions. Those skilled in the art will appreciate that various combinations and/or combinations of features recited in the various embodiments and/or claims of the present disclosure can be made, even if such combinations or combinations are not expressly recited in the present disclosure. In particular, various combinations and/or combinations of the features recited in the various embodiments and/or claims of the present disclosure may be made without departing from the spirit or teaching of the present disclosure. All such combinations and/or associations are within the scope of the present disclosure.
The embodiments of the present disclosure have been described above. However, these examples are for illustrative purposes only and are not intended to limit the scope of the present disclosure. Although the embodiments are described separately above, this does not mean that the measures in the embodiments cannot be used in advantageous combination. The scope of the disclosure is defined by the appended claims and equivalents thereof. Various alternatives and modifications can be devised by those skilled in the art without departing from the scope of the present disclosure, and such alternatives and modifications are intended to be within the scope of the present disclosure.

Claims (10)

1. A customer prediction method, comprising:
acquiring attribute information of a client to be predicted; and
and processing the attribute information of the customer to be predicted by using a customer prediction model to obtain a prediction result aiming at the customer to be predicted, wherein the customer prediction model is obtained by training based on the attribute information of the sample customer and the telemarketing information corresponding to the sample customer.
2. The method of claim 1, wherein the customer prediction model is trained based on attribute information of a sample customer and telemarketing information corresponding to the sample customer, comprising:
obtaining a historical sample set, wherein each sample in the historical sample set comprises attribute information of a sample customer and telemarketing information corresponding to the sample customer;
dividing the historical sample set into a search sample set and a training sample set by utilizing a differentiable structure search algorithm;
training an initial search module by using the search sample set to obtain a target search module;
generating a convolutional neural network model based on the target search module; and
and training the convolutional neural network model by using the training sample set to obtain the customer prediction model.
3. The method of claim 2, wherein the target search module comprises a plurality;
the generating a convolutional neural network model based on the target search module includes:
and obtaining a convolutional neural network model based on the series connection of the target search modules.
4. The method of claim 2 or 3, wherein the search sample set comprises a search training set and a search validation set;
the training of the initial search module by using the search sample set to obtain the target search module comprises the following steps:
determining an initial search module by using the differentiable structure search algorithm, wherein the initial search module comprises at least two search units, and each search unit is a directed acyclic graph consisting of at least two nodes and at least one directed edge;
determining a set of candidate operations corresponding to each of the directed edges, wherein the set of candidate operations includes at least two candidate operations;
inputting the search training set into an initial search module using a first candidate operation to output a first predictive identifier of each sample client in the search training set and a first operation weight of the first candidate operation, and inputting the search validation set into an initial search module using the first candidate operation to output a second predictive identifier of each sample client in the search validation set and a second operation weight of the first candidate operation;
determining a first loss function according to the first operation weight, the real identifier of each sample client in the search training set and the first prediction identifier, and determining a second loss function according to the second operation weight, the real identifier of each sample client in the search verification set and the second prediction identifier;
adjusting structure parameters and network parameters until the first loss function and the second loss function meet preset conditions, and determining target operation corresponding to each directed edge; and
and determining a target searching module according to the target operation.
5. The method of any one of claims 1-3, wherein the telemarketing information includes at least one of: marketing product type, dial-out type, dial-up or not, call duration and call-out time.
6. A customer prediction apparatus comprising:
the acquisition module is used for acquiring attribute information of a client to be predicted; and
and the prediction module is used for processing the attribute information of the customer to be predicted by utilizing a customer prediction model to obtain a prediction result aiming at the customer to be predicted, wherein the customer prediction model is obtained by training based on the attribute information of the sample customer and the telemarketing information corresponding to the sample customer.
7. The apparatus of claim 6, wherein the prediction module comprises:
the acquisition submodule is used for acquiring a historical sample set, wherein each sample in the historical sample set comprises attribute information of a sample client and telemarketing information corresponding to the sample client;
the dividing submodule is used for dividing the historical sample set into a search sample set and a training sample set by utilizing a differentiable structure search algorithm;
the first training submodule is used for training the initial search module by utilizing the search sample set to obtain a target search module;
a generation submodule for generating a convolutional neural network model based on the target search module; and
and the second training submodule is used for training the convolutional neural network model by using the training sample set to obtain the client prediction model.
8. The apparatus of claim 7, wherein the search unit search module comprises a plurality;
the generation submodule includes:
and the generating unit is used for obtaining a convolutional neural network model based on the serial connection of the plurality of searching modules.
9. An electronic device, comprising:
one or more processors;
a memory for storing one or more programs,
wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the method of any of claims 1-5.
10. A computer readable storage medium having stored thereon executable instructions which, when executed by a processor, cause the processor to carry out the method of any one of claims 1 to 5.
CN202010677606.4A 2020-07-14 2020-07-14 Customer prediction method and device, electronic equipment and storage medium Pending CN111784411A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010677606.4A CN111784411A (en) 2020-07-14 2020-07-14 Customer prediction method and device, electronic equipment and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010677606.4A CN111784411A (en) 2020-07-14 2020-07-14 Customer prediction method and device, electronic equipment and storage medium

Publications (1)

Publication Number Publication Date
CN111784411A true CN111784411A (en) 2020-10-16

Family

ID=72768639

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010677606.4A Pending CN111784411A (en) 2020-07-14 2020-07-14 Customer prediction method and device, electronic equipment and storage medium

Country Status (1)

Country Link
CN (1) CN111784411A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112686448A (en) * 2020-12-31 2021-04-20 重庆富民银行股份有限公司 Loss early warning method and system based on attribute data

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107688967A (en) * 2017-08-24 2018-02-13 平安科技（深圳）有限公司 The Forecasting Methodology and terminal device of client's purchase intention
CN108898429A (en) * 2018-06-19 2018-11-27 平安科技（深圳）有限公司 Electronic device, preference tendency prediction technique and computer readable storage medium
CN110415002A (en) * 2019-07-31 2019-11-05 中国工商银行股份有限公司 Customer behavior prediction method and system
CN111105265A (en) * 2019-10-12 2020-05-05 平安银行股份有限公司 Prediction method and device based on customer information, computer equipment and storage medium

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107688967A (en) * 2017-08-24 2018-02-13 平安科技（深圳）有限公司 The Forecasting Methodology and terminal device of client's purchase intention
CN108898429A (en) * 2018-06-19 2018-11-27 平安科技（深圳）有限公司 Electronic device, preference tendency prediction technique and computer readable storage medium
CN110415002A (en) * 2019-07-31 2019-11-05 中国工商银行股份有限公司 Customer behavior prediction method and system
CN111105265A (en) * 2019-10-12 2020-05-05 平安银行股份有限公司 Prediction method and device based on customer information, computer equipment and storage medium

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
AI异构: "神经网络构建搜索——可微分搜索（DARTS）", pages 1 - 6, Retrieved from the Internet <URL:https://blog.csdn.net/u012347027/article/details/106175020?ops_request_misc=%257B%2522request%255Fid%2522%253A%2522168612426216800215052026%2522%252C%2522scm%2522%253A%252220140713.130102334.pc%255Fall.%2522%257D&request_id=168612426216800215052026&biz_id=0&utm_medium=distribute.pc_search_result.none-task-blog-2~all~first_rank_ecpm_v1~rank_v31_ecpm-2-106175020-null-null.142^v88^control,239^v2^insert_chatgpt&utm_term=%E5%8F%AF%E5%BE%AE%E5%88%86%E7%BB%93%E6%9E%84%E6%90%9C%E7%B4%A2DARTS&spm=1018.2226.3001.4187> *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112686448A (en) * 2020-12-31 2021-04-20 重庆富民银行股份有限公司 Loss early warning method and system based on attribute data
CN112686448B (en) * 2020-12-31 2024-02-13 重庆富民银行股份有限公司 Loss early warning method and system based on attribute data

Similar Documents

Publication Publication Date Title
US20230252327A1 (en) Neural architecture search for convolutional neural networks
CN109947919B (en) Method and apparatus for generating text matching model
US11861474B2 (en) Dynamic placement of computation sub-graphs
WO2020108605A1 (en) Recommendation method, apparatus, and storage medium
US20190171935A1 (en) Robust gradient weight compression schemes for deep learning applications
US11157316B1 (en) Determining action selection policies of an execution device
US20150356703A1 (en) Arranging a transport service based on computed vectors associated with service providers
CN112288572B (en) Service data processing method and computer equipment
US11204803B2 (en) Determining action selection policies of an execution device
US20200111027A1 (en) Systems and methods for providing recommendations based on seeded supervised learning
CN112785157A (en) Risk identification system updating method and device and risk identification method and device
CN113348472A (en) Convolutional neural network with soft kernel selection
CN110276387B (en) Model generation method and device
CN113283680B (en) Address selection method, device, equipment and storage medium thereof
CN110263250B (en) Recommendation model generation method and device
CN113689270B (en) Method for determining black product device, electronic device, storage medium, and program product
CN111784411A (en) Customer prediction method and device, electronic equipment and storage medium
CN116910373B (en) House source recommendation method and device, electronic equipment and storage medium
CN114092162B (en) Recommendation quality determination method, and training method and device of recommendation quality determination model
US20170160730A1 (en) Production system having constraint evaluator generating constrained sequences
CN110689117A (en) Information processing method and device based on neural network
CN112818235B (en) Method and device for identifying illegal user based on association characteristics and computer equipment
Hu et al. ROUTERBENCH: A Benchmark for Multi-LLM Routing System
CN113128677A (en) Model generation method and device
CN115204386B (en) Training method of prediction model and method for recommending broker

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