CN110852785A - User grading method, device and computer readable storage medium - Google Patents

User grading method, device and computer readable storage medium Download PDF

Info

Publication number
CN110852785A
CN110852785A CN201910976136.9A CN201910976136A CN110852785A CN 110852785 A CN110852785 A CN 110852785A CN 201910976136 A CN201910976136 A CN 201910976136A CN 110852785 A CN110852785 A CN 110852785A
Authority
CN
China
Prior art keywords
user
grading
product
behavior record
training
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.)
Granted
Application number
CN201910976136.9A
Other languages
Chinese (zh)
Other versions
CN110852785B (en
Inventor
张浩然
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Ping An Life Insurance Company of China Ltd
Original Assignee
Ping An Life Insurance Company of China Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Ping An Life Insurance Company of China Ltd filed Critical Ping An Life Insurance Company of China Ltd
Priority to CN201910976136.9A priority Critical patent/CN110852785B/en
Publication of CN110852785A publication Critical patent/CN110852785A/en
Application granted granted Critical
Publication of CN110852785B publication Critical patent/CN110852785B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G06Q40/08—Insurance
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management

Abstract

The invention relates to an artificial intelligence technology, and discloses a user grading method, which comprises the following steps: the method comprises the steps of obtaining an original user tag set, an original user behavior record set and a product grading set, processing the original user tag set, the original user behavior record set and the product grading set to obtain a primary user tag set and a primary user behavior record set, conducting grading operation on the primary user tag set to obtain a common user tag set and a core user tag set, preprocessing the primary user behavior record set to obtain a training set, inputting the training set, the common user tag set and the core user tag set into a grading model for training, receiving a user behavior record, inputting the behavior record into the grading model to obtain a user grading set, establishing a user product corresponding model according to the user grading set and the product grading set, and recommending products corresponding to users according to the user product corresponding model. The invention also provides a user grading device and a computer readable storage medium. The invention can realize accurate and efficient user grading function.

Description

User grading method, device and computer readable storage medium
Technical Field
The present invention relates to the field of artificial intelligence technology, and in particular, to a method and apparatus for ranking users according to their behavior, and a computer-readable storage medium.
Background
The user rating is an operation means for rating the user according to the characteristics of the user, such as how to rate the customer according to the customer information for the life insurance company, thereby providing a suitable product more accurately and automatically. At present, based on user classification, the user classification is mostly based on naive Bayes or a support vector machine, for example, the naive Bayes establishes a conditional probability model according to a user, and solves the user classification according to the conditional probability model, and the support vector machine classifies the user by constructing a hyperplane.
Disclosure of Invention
The invention provides a user grading method, a user grading device and a computer readable storage medium, and mainly aims to provide a method which can combine analysis between users and products to achieve the purpose of accurate user grading.
In order to achieve the above object, the present invention provides a user ranking method, including:
acquiring an original user tag set, an original user behavior record set and a product grading set, and performing exception removal processing on the original user tag set and the original user behavior record set to obtain a primary user tag set and a primary user behavior record set with corresponding relations;
performing hierarchical operation on the primary user tag set according to user attributes to obtain a common user tag set and a core user tag set;
preprocessing the primary user behavior record set to obtain a training set, inputting the training set, the common user tag set and the core user tag set into a pre-constructed hierarchical model for training to obtain a loss value, judging the size relation between the loss value and a preset loss value, continuing training internal parameters by the hierarchical model if the loss value is larger than the preset loss value, and finishing training by the hierarchical model if the loss value is smaller than or equal to the preset loss value;
receiving a behavior record of a user, inputting the behavior record of the user into the trained hierarchical model, and grading the user to obtain a user grading set;
and establishing a user-product corresponding model according to the user grading set and the product grading set, and recommending products corresponding to the user according to the user-product corresponding model.
Optionally, the hierarchical model comprises an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer; and
the training comprises the following steps:
receiving the primary user behavior record set by using the input layer, training according to the primary user behavior record set by using the convolutional layer, the pooling layer and the full-link layer in combination with an activation function to obtain a training value, and outputting a predicted value through an output layer;
and performing loss calculation on the predicted value, the common user tag set and the core user tag set to obtain a loss value.
Optionally, the activation function includes:
Figure BDA0002231401420000021
wherein, OjRepresents the output value, I, of the jth neuron of the fully-connected layerjRepresenting an input value of a jth neuron of the output layer, t representing a total amount of neurons of the output layer, e being an infinite acyclic fraction;
the loss calculation includes:
Figure BDA0002231401420000022
wherein s is the predicted value, k is the number of the primary user tag sets, yiIs the common user tag set, y'iA set of labels for the core user.
Optionally, the establishing a user-product correspondence model includes:
calculating preference similarity of different users to different products according to the user grading set and the product grading set;
calculating the similarity of the user behavior habits in the user hierarchical set;
and constructing the user-product corresponding model based on the preference similarity and the user behavior habit similarity.
Optionally, the preference similarity includes:
Figure BDA0002231401420000031
Figure BDA0002231401420000032
wherein ps (m, n) is preference similarity of different users to different products, where m, n represents different user level data in the user hierarchy set, m is user level data of a core user in the user hierarchy set, n is user level data of a general user in the user hierarchy set, and p issRepresenting said set of graded products, s being different graded products, rmi、rniRepresenting the number of users in one level of said hierarchical product set at the same level of said user hierarchical set, rm、rnAnd a and b are correlation coefficients of the preference similarity.
In addition, to achieve the above object, the present invention further provides a user rating device, including a memory and a processor, wherein the memory stores a user rating program operable on the processor, and the user rating program, when executed by the processor, implements the steps of:
acquiring an original user tag set, an original user behavior record set and a product grading set, and performing exception removal processing on the original user tag set and the original user behavior record set to obtain a primary user tag set and a primary user behavior record set with corresponding relations;
performing hierarchical operation on the primary user tag set according to user attributes to obtain a common user tag set and a core user tag set;
preprocessing the primary user behavior record set to obtain a training set, inputting the training set, the common user tag set and the core user tag set into a pre-constructed hierarchical model for training to obtain a loss value, judging the size relation between the loss value and a preset loss value, continuing training internal parameters by the hierarchical model if the loss value is larger than the preset loss value, and finishing training by the hierarchical model if the loss value is smaller than or equal to the preset loss value;
receiving a behavior record of a user, inputting the behavior record of the user into the trained hierarchical model, and grading the user to obtain a user grading set;
and establishing a user-product corresponding model according to the user grading set and the product grading set, and recommending products corresponding to the user according to the user-product corresponding model.
Optionally, the hierarchical model comprises an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer; and
the training comprises the following steps:
receiving the primary user behavior record set by using the input layer, training according to the primary user behavior record set by using the convolutional layer, the pooling layer and the full-link layer in combination with an activation function to obtain a training value, and outputting a predicted value through an output layer;
and performing loss calculation on the predicted value, the common user tag set and the core user tag set to obtain a loss value.
Optionally, the activation function includes:
Figure BDA0002231401420000041
wherein, OjRepresents the output value, I, of the jth neuron of the fully-connected layerjRepresents the input value of the j-th neuron of the output layer, and t representsE is an infinite acyclic decimal number;
the loss calculation includes:
Figure BDA0002231401420000042
wherein s is the predicted value, k is the number of the primary user tag sets, yiIs the common user tag set, y'iA set of labels for the core user.
Optionally, the establishing a user-product correspondence model includes:
calculating preference similarity of different users to different products according to the user grading set and the product grading set;
calculating the similarity of the user behavior habits in the user hierarchical set;
and constructing the user-product corresponding model based on the preference similarity and the user behavior habit similarity.
Further, to achieve the above object, the present invention also provides a computer-readable storage medium having a user rating program stored thereon, the user rating program being executable by one or more processors to implement the steps of the user rating method as described above.
The invention carries out exception removal processing on the original user label set, the original user behavior record set and the product grading set, obtains a common user label set and a core user label set through grading operation, can divide users through the first grading operation, trains a pre-constructed grading model at the same time, judges user behavior by utilizing the trained grading model, establishes a user-product corresponding model according to the user grading set and the product grading set, achieves final user product recommendation, and divides and recommends through three steps in total, thereby effectively improving the grading efficiency. Therefore, the user grading method, the user grading device and the computer readable storage medium provided by the invention can realize accurate and efficient user grading.
Drawings
Fig. 1 is a schematic flow chart of a user ranking method according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of an internal structure of a user classifying device according to an embodiment of the present invention;
fig. 3 is a block diagram illustrating a user classification program in a user classification device according to an embodiment of the present invention.
The implementation, functional features and advantages of the objects of the present invention will be further explained with reference to the accompanying drawings.
Detailed Description
It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
The invention provides a user grading method. Fig. 1 is a schematic flow chart of a user ranking method according to an embodiment of the present invention. The method may be performed by an apparatus, which may be implemented by software and/or hardware.
In this embodiment, the user ranking method includes:
s1, obtaining an original user label set, an original user behavior record set and a product grading set, and performing exception removal processing on the original user label set and the original user behavior record set to obtain a primary user label set and a primary user behavior record set with corresponding relations.
Preferably, the user grading method of the present invention performs grading operation on the user according to the preset attribute. For example, the telecommunication company divides the user into light telephone charge consumption and heavy telephone charge consumption according to the monthly telephone charge consumption record of the user, the monthly province times of the user, the working type of the user and other attributes; the insurance company divides the users into non-insurance users, light insurance users, heavy insurance users and the like according to the attributes of the users such as the insurance application expense, the academic history, the number of family members and the like on insurance each year, therefore, when a new user is received, the invention grades the user according to the preset attribute of the new user.
In the preferred embodiment of the present invention, the original user behavior record set is the above-mentioned predetermined attribute. For example, the user may be classified by the longevity company, and the original user behavior record set generally includes: the total value and the total times of the life insurance products purchased by the user in the life insurance company, the types of the life insurance products collected by the user on the webpage of the life insurance company, the times of the life insurance products browsed by the user on the webpage of the life insurance company and the like.
Further, the original user tag set is the numbers of different users, such as a user a001, a user a002, a user B001, and the like in a manner of adding numbers and letters.
Preferably, in the preferred embodiment of the present invention, the original product is classified according to the attribute of the product to obtain the product classification set, and generally, different products can be divided according to different product attributes. For example, the life insurance company has the risk of children, health, old age, security, two total risks, accident, additional risk, divination risk, etc., so the life insurance can be divided into the product grading sets of popular products, long-tailed products, good-base products, etc. according to the sales condition of each risk.
Preferably, the exception removing process is to perform one-to-one correspondence on the relationship between the original user tag set and the original user behavior record set, and if the one-to-one correspondence is not satisfied, the redundant data in the original user tag set and the original user behavior record set is removed, or the data lacking in the original user tag set and the original user behavior record set is added.
Preferably, the deincoding process of the present invention includes, for example: the life insurance company carries out hierarchical division on users, the original user label set records a user A001, the original user behavior record set records that the total value of life insurance products purchased by the user A001 in the life insurance company is 13000 yuan, the total times of purchasing the life insurance products is 9 times, the types of the life insurance products collected by the user A001 on a webpage of the life insurance company are 18, the times of browsing the life insurance products on the webpage of the life insurance company are 621 times and the like, and therefore the one-to-one correspondence relationship is met; if the original user tag set records a user B001 and any behavior record of the user B001 is not recorded in the original user behavior record set, removing the user B001 from the original user tag set; similarly, if the original user behavior record set records the behavior record of the user C001, but the original user tag set does not have the user C001, the user C001 is added to the original user tag set.
In summary, the low classification accuracy caused by incomplete data or unequal data in the early stage can be effectively prevented according to the exception removal processing.
And S2, carrying out hierarchical operation on the primary user label set according to the user attributes to obtain a common user label set and a core user label set.
Preferably, as the above-mentioned life insurance company performs hierarchical division on users, the primary user tag set may be divided into a general user tag set and a core user tag set according to whether a user purchases a general product and a core product of the life insurance company, the general user tag set is a general product that the life insurance company has purchased, and the core user tag set is a core product that the life insurance company has purchased, without concern about whether the general product has been purchased. Therefore, when the prediction of the new user is received, the invention predicts whether the new user is a core user or a common user.
S3, preprocessing the primary user behavior record set to obtain a training set, inputting the training set, the common user label set and the core user label set into a pre-constructed hierarchical model for training to obtain a loss value, judging the size relation between the loss value and a preset loss value, if the loss value is larger than the preset loss value, continuing training internal parameters by the hierarchical model, and if the loss value is smaller than or equal to the preset loss value, finishing training by the hierarchical model.
Preferably, the preprocessing operation of the present invention includes qualitative and quantitative conversion, digital extraction, digital normalization, and the like.
The qualitative and quantitative transformation is to transform non-numeric attributes in the set of primary user behavior records to numeric attributes. If the online store A carries out consumption grading on the consumption capacity of the user, if the user collects the products a, b, e, s and the like of the online store A, the qualitative and quantitative conversion is to count the number of all collected products and output the result.
The digital extraction is to extract digital attributes, such as product collection number, browsing number and the like.
The numeric normalization is to map the numbers in the primary user behavior record set into the range of [0,1], and is generally necessary to narrow the gap caused by the magnitude, because the magnitude of each number generally varies greatly, for example, the total amount of life insurance purchases must be much larger than the number of purchases. Preferably, the normalization method is:
Figure BDA0002231401420000071
wherein, XnormalIs the number after normalization, X is the number before normalization, XmaxThe number, X, of the primary user behavior record setminRecording the lowest numerical digit in the set for the primary user behavior.
In a preferred embodiment of the present invention, the hierarchical model includes an input layer, a convolutional layer, a pooling layer, a fully-connected layer, and an output layer.
Preferably, the training process specifically comprises: receiving the primary user behavior record set by using the input layer, training by using the convolutional layer, the pooling layer and the full link layer in combination with an activation function to obtain a training value, outputting a prediction value set through an output layer, performing loss calculation on the prediction value, the common user label set and the core user label set to obtain a loss value, judging the size of the loss value and a preset loss value, continuing training if the loss value is greater than the preset loss value, and quitting training if the loss value is less than the preset loss value.
In a preferred embodiment of the present invention, the activation function comprises a Softmax function. The Softmax function is:
Figure BDA0002231401420000081
wherein, OjRepresents the output value, I, of the jth neuron of the fully-connected layerjRepresenting an input value of a jth neuron of the output layer, t representing a total amount of neurons of the output layer, e being an infinite acyclic fraction;
the loss calculation includes:
Figure BDA0002231401420000082
wherein s is the predicted value, k is the number of the primary user tag sets, yiIs the common user tag set, y'iA set of labels for the core user.
S4, receiving the behavior record of the user, inputting the behavior record of the user into the trained hierarchical model, and grading the user to obtain a user grading set.
Preferably, for example, the life insurance client grading model is a life insurance company grading users, and when receiving all the behavior records of the users a to Z, the life insurance client grading model grades the users into core users, common users and the like according to all the behavior records of the users a to Z.
S5, establishing a user-product corresponding model according to the user hierarchical set and the product hierarchical set, and recommending products corresponding to the user according to the user-product corresponding model.
As described above, the product taxonomy may include popular products, long-tailed products, niche products, and the like. Wherein the popular product is the product with the most popular purchasing number; the long-tail product is a product with low demand or poor sales volume; the product is a product which is sold in a common quantity, obviously shows different unique benefits from other products and can be identified by users.
Preferably, the purpose of the user-product correspondence model is to correspond users at different levels in the user hierarchical set to products with different product attributes in the product hierarchical set, so as to perform association recommendation on the users at different levels and the products with different product attributes, thereby achieving the purpose of more accurate intelligent recommendation.
Further, the user-product corresponding model comprises the steps of calculating preference similarity of different users to different products, calculating user behavior habit similarity, and constructing the user-product corresponding model based on the preference similarity and the user behavior habit similarity.
Preferably, the similarity of the user's preferences for different products is:
Figure BDA0002231401420000091
Figure BDA0002231401420000092
wherein ps (m, n) is the preference similarity of different users to different products, where m, n represents different levels in the user hierarchy, e.g. m may represent core users, n may represent ordinary users, psRepresents the hierarchical product set, s is different hierarchical products, such as the popular product, long-tailed product, and liji product, so when i ∈ psWhen i is 1, represents a product of one grade of the graded product set, such as the popular product, and the like, rmi、rniRepresenting the number of users in one level of said hierarchical product set at the same level of said user hierarchical set, rm、rnAnd a and b are correlation coefficients of the preference similarity.
Further, the similarity of the user behavior habits is as described in step S1 above, for example, the record of the user behavior of the life insurance company on the user includes the total value and the total times of the life insurance products purchased by the user in the life insurance company, the number of types of the life insurance products collected by the user on the web page of the life insurance company, the number of times of the life insurance products browsed by the user on the web page of the life insurance company, and the like, so that the user behavior habits can be obtained according to the calculation of the user similarity at different levels in the user classification set.
Preferably, the user-product correspondence model is finally:
sim(x，[m，n])＝α*ps(x，[m，n])+β*hs(x，[m，n])
wherein sim (x, [ m, n ]) is the user-product corresponding model, hs (x, [ m, n ]) is the user behavior habit similarity, α is the weight of the user-product corresponding model respectively, x is an unknown user, when the unknown user x is received, the user-product corresponding model sequentially performs similarity calculation with users [ m, n ] of different levels, and sequentially traverses the level user corresponding to the highest similarity, so that the corresponding product of the same level is recommended according to the level user, and the purpose of product recommendation is achieved.
The invention also provides a user grading device. Fig. 2 is a schematic diagram illustrating an internal structure of a user classifying device according to an embodiment of the present invention.
In this embodiment, the user classifying device 1 may be a PC (Personal Computer), a terminal device such as a smart phone, a tablet Computer, or a mobile Computer, or may be a server. The user rating device 1 comprises at least a memory 11, a processor 12, a communication bus 13, and a network interface 14.
The memory 11 includes at least one type of readable storage medium, which includes a flash memory, a hard disk, a multimedia card, a card type memory (e.g., SD or DX memory, etc.), a magnetic memory, a magnetic disk, an optical disk, and the like. The memory 11 may in some embodiments be an internal storage unit of the user rating device 1, such as a hard disk of the user rating device 1. The memory 11 may also be an external storage device of the user rating device 1 in other embodiments, such as a plug-in hard disk provided on the user rating device 1, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card), or the like. Further, the memory 11 may also include both an internal storage unit of the user rating apparatus 1 and an external storage device. The memory 11 may be used not only to store application software installed in the user classifying device 1 and various types of data such as a code of the user classifying program 01, but also to temporarily store data that has been output or is to be output.
Processor 12, which in some embodiments may be a Central Processing Unit (CPU), controller, microcontroller, microprocessor or other data Processing chip, executes program code or processes data stored in memory 11, such as executing user rating program 01.
The communication bus 13 is used to realize connection communication between these components.
The network interface 14 may optionally include a standard wired interface, a wireless interface (e.g., WI-FI interface), typically used to establish a communication link between the apparatus 1 and other electronic devices.
Optionally, the apparatus 1 may further comprise a user interface, which may comprise a Display (Display), an input unit such as a Keyboard (Keyboard), and optionally a standard wired interface, a wireless interface. Alternatively, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode) touch device, or the like. The display, which may also be referred to as a display screen or display unit, is suitable for displaying information processed in the user rating device 1 and for displaying a visualized user interface.
While FIG. 2 only shows the user rating device 1 with components 11-14 and a user rating program 01, those skilled in the art will appreciate that the configuration shown in FIG. 1 does not constitute a limitation of the user rating device 1 and may include fewer or more components than shown, or some components may be combined, or a different arrangement of components.
In the embodiment of the apparatus 1 shown in fig. 2, a user rating program 01 is stored in the memory 11; the processor 12, when executing the user rating program 01 stored in the memory 11, implements the following steps:
the method comprises the steps of firstly, obtaining an original user tag set, an original user behavior record set and a product grading set, and conducting exception removal processing on the original user tag set and the original user behavior record set to obtain a primary user tag set and a primary user behavior record set which have corresponding relations.
Preferably, the user grading method of the present invention performs grading operation on the user according to the preset attribute. For example, the telecommunication company divides the user into light telephone charge consumption and heavy telephone charge consumption according to the monthly telephone charge consumption record of the user, the monthly province times of the user, the working type of the user and other attributes; the insurance company divides the users into non-insurance users, light insurance users, heavy insurance users and the like according to the attributes of the users such as the insurance application expense, the academic history, the number of family members and the like on insurance each year, therefore, when a new user is received, the invention grades the user according to the preset attribute of the new user.
In the preferred embodiment of the present invention, the original user behavior record set is the above-mentioned predetermined attribute. For example, the user may be classified by the longevity company, and the original user behavior record set generally includes: the total value and the total times of the life insurance products purchased by the user in the life insurance company, the types of the life insurance products collected by the user on the webpage of the life insurance company, the times of the life insurance products browsed by the user on the webpage of the life insurance company and the like.
Further, the original user tag set is the numbers of different users, such as a user a001, a user a002, a user B001, and the like in a manner of adding numbers and letters.
Preferably, in the preferred embodiment of the present invention, the original product is classified according to the attribute of the product to obtain the product classification set, and generally, different products can be divided according to different product attributes. For example, the life insurance company has the risk of children, health, old age, security, two total risks, accident, additional risk, divination risk, etc., so the life insurance can be divided into the product grading sets of popular products, long-tailed products, good-base products, etc. according to the sales condition of each risk.
Preferably, the exception removing process is to perform one-to-one correspondence on the relationship between the original user tag set and the original user behavior record set, and if the one-to-one correspondence is not satisfied, the redundant data in the original user tag set and the original user behavior record set is removed, or the data lacking in the original user tag set and the original user behavior record set is added.
Preferably, the deincoding process of the present invention includes, for example: the life insurance company carries out hierarchical division on users, the original user label set records a user A001, the original user behavior record set records that the total value of life insurance products purchased by the user A001 in the life insurance company is 13000 yuan, the total times of purchasing the life insurance products is 9 times, the types of the life insurance products collected by the user A001 on a webpage of the life insurance company are 18, the times of browsing the life insurance products on the webpage of the life insurance company are 621 times and the like, and therefore the one-to-one correspondence relationship is met; if the original user tag set records a user B001 and any behavior record of the user B001 is not recorded in the original user behavior record set, removing the user B001 from the original user tag set; similarly, if the original user behavior record set records the behavior record of the user C001, but the original user tag set does not have the user C001, the user C001 is added to the original user tag set.
In summary, the low classification accuracy caused by incomplete data or unequal data in the early stage can be effectively prevented according to the exception removal processing.
And step two, carrying out hierarchical operation on the primary user tag set according to the user attributes to obtain a common user tag set and a core user tag set.
Preferably, as the above-mentioned life insurance company performs hierarchical division on users, the primary user tag set may be divided into a general user tag set and a core user tag set according to whether a user purchases a general product and a core product of the life insurance company, the general user tag set is a general product that the life insurance company has purchased, and the core user tag set is a core product that the life insurance company has purchased, without concern about whether the general product has been purchased. Therefore, when the prediction of the new user is received, the invention predicts whether the new user is a core user or a common user.
And step three, preprocessing the primary user behavior record set to obtain a training set, inputting the training set, the common user tag set and the core user tag set into a pre-constructed hierarchical model for training to obtain a loss value, judging the size relation between the loss value and a preset loss value, if the loss value is larger than the preset loss value, continuing training internal parameters by the hierarchical model, and if the loss value is smaller than or equal to the preset loss value, finishing training by the hierarchical model.
Preferably, the preprocessing operation of the present invention includes qualitative and quantitative conversion, digital extraction, digital normalization, and the like.
The qualitative and quantitative transformation is to transform non-numeric attributes in the set of primary user behavior records to numeric attributes. If the online store A carries out consumption grading on the consumption capacity of the user, if the user collects the products a, b, e, s and the like of the online store A, the qualitative and quantitative conversion is to count the number of all collected products and output the result.
The digital extraction is to extract digital attributes, such as product collection number, browsing number and the like.
The numeric normalization is to map the numbers in the primary user behavior record set into the range of [0,1], and is generally necessary to narrow the gap caused by the magnitude, because the magnitude of each number generally varies greatly, for example, the total amount of life insurance purchases must be much larger than the number of purchases. Preferably, the normalization method is:
wherein, XnormalIs the number after normalization, X is the number before normalization, XmaxThe above-mentionedMaximum number of primary user behavior record set, XminRecording the lowest numerical digit in the set for the primary user behavior.
In a preferred embodiment of the present invention, the hierarchical model includes an input layer, a convolutional layer, a pooling layer, a fully-connected layer, and an output layer.
Preferably, the training process specifically comprises: receiving the primary user behavior record set by using the input layer, training by using the convolutional layer, the pooling layer and the full link layer in combination with an activation function to obtain a training value, outputting a prediction value set through an output layer, performing loss calculation on the prediction value, the common user label set and the core user label set to obtain a loss value, judging the size of the loss value and a preset loss value, continuing training if the loss value is greater than the preset loss value, and quitting training if the loss value is less than the preset loss value.
In a preferred embodiment of the present invention, the activation function comprises a Softmax function. The Softmax function is:
Figure BDA0002231401420000132
wherein, OjRepresents the output value, I, of the jth neuron of the fully-connected layerjRepresenting an input value of a jth neuron of the output layer, t representing a total amount of neurons of the output layer, e being an infinite acyclic fraction;
the loss calculation includes:
Figure BDA0002231401420000133
wherein s is the predicted value, k is the number of the primary user tag sets, yiIs the common user tag set, y'iA set of labels for the core user.
And step four, receiving a behavior record of a user, inputting the behavior record of the user into the trained hierarchical model, and grading the user to obtain a user grading set.
Preferably, for example, the life insurance client grading model is a life insurance company grading users, and when receiving all the behavior records of the users a to Z, the life insurance client grading model grades the users into core users, common users and the like according to all the behavior records of the users a to Z.
And fifthly, establishing a user-product corresponding model according to the user grading set and the product grading set, and recommending products corresponding to the user according to the user-product corresponding model.
As described above, the product taxonomy may include popular products, long-tailed products, niche products, and the like. Wherein the popular product is the product with the most popular purchasing number; the long-tail product is a product with low demand or poor sales volume; the product is a product which is sold in a common quantity, obviously shows different unique benefits from other products and can be identified by users.
Preferably, the purpose of the user-product correspondence model is to correspond users at different levels in the user hierarchical set to products with different product attributes in the product hierarchical set, so as to perform association recommendation on the users at different levels and the products with different product attributes, thereby achieving the purpose of more accurate intelligent recommendation.
Further, the user-product corresponding model comprises the steps of calculating preference similarity of different users to different products, calculating user behavior habit similarity, and constructing the user-product corresponding model based on the preference similarity and the user behavior habit similarity.
Preferably, the similarity of the user's preferences for different products is:
Figure BDA0002231401420000141
Figure BDA0002231401420000142
where ps (m, n) is different for different pairs of usersPreference similarity of products, where m, n represent different levels in the user hierarchy, e.g. m may represent core users, n may represent normal users, psRepresents the hierarchical product set, s is different hierarchical products, such as the popular product, long-tailed product, and liji product, so when i ∈ psWhen i is 1, represents a product of one grade of the graded product set, such as the popular product, and the like, rmi、rniRepresenting the number of users in one level of said hierarchical product set at the same level of said user hierarchical set, rm、rnAnd a and b are correlation coefficients of the preference similarity.
Further, the similarity of the user behavior habits is as described in the first step, for example, the record of the user behavior of the life insurance company on the user includes the total value and the total times of the life insurance products purchased by the user in the life insurance company, the number of the types of the life insurance products collected by the user on the web page of the life insurance company, the number of times of the life insurance products browsed by the user on the web page of the life insurance company, and the like, so that the user behavior habits can be obtained by calculating the similarity of the users at different levels in the user classification set.
Preferably, the user-product correspondence model is finally:
sim(x，[m，n])＝α*ps(x，[m，n])+β*hs(x，[m，n])
wherein sim (x, [ m, n ]) is the user-product corresponding model, hs (x, [ m, n ]) is the user behavior habit similarity, α is the weight of the user-product corresponding model respectively, x is an unknown user, when the unknown user x is received, the user-product corresponding model sequentially performs similarity calculation with users [ m, n ] of different levels, and sequentially traverses the level user corresponding to the highest similarity, so that the corresponding product of the same level is recommended according to the level user, and the purpose of product recommendation is achieved.
Alternatively, in other embodiments, the user rating program may be divided into one or more modules, and the one or more modules are stored in the memory 11 and executed by one or more processors (in this embodiment, the processor 12) to implement the present invention, where the module referred to in the present invention refers to a series of computer program instruction segments capable of performing a specific function for describing the execution process of the user rating program in the user rating device.
For example, referring to fig. 3, a schematic diagram of program modules of a user rating program in an embodiment of the user rating apparatus of the present invention is shown, in this embodiment, the user rating program may be divided into a data receiving and processing module 10, a rating operation module 20, a model training module 30, and a user rating and product recommending module 40, which exemplarily:
the data receiving and processing module 10 is configured to: the method comprises the steps of obtaining an original user tag set, an original user behavior record set and a product grading set, and conducting exception removal processing on the original user tag set and the original user behavior record set to obtain a primary user tag set and a primary user behavior record set with corresponding relations.
The grading operation module 20 is used for; and carrying out hierarchical operation on the primary user tag set according to the user attributes to obtain a common user tag set and a core user tag set.
The model training module 30 is configured to: preprocessing the primary user behavior record set to obtain a training set, inputting the training set, the common user tag set and the core user tag set into a pre-constructed hierarchical model for training to obtain a loss value, judging the size relation between the loss value and a preset loss value, continuing training internal parameters by the hierarchical model if the loss value is larger than the preset loss value, and finishing training by the hierarchical model if the loss value is smaller than or equal to the preset loss value.
The user rating and product recommendation module 40 is configured to: receiving a behavior record of a user, inputting the behavior record of the user into the trained hierarchical model, classifying the user to obtain a user hierarchical set, establishing a user-product corresponding model according to the user hierarchical set and the product hierarchical set, and recommending a product corresponding to the user according to the user-product corresponding model.
The functions or operation steps implemented by the data receiving and processing module 10, the ranking module 20, the model training module 30, the user ranking and product recommending module 40 and other program modules are substantially the same as those of the above embodiments, and are not repeated herein.
Furthermore, an embodiment of the present invention further provides a computer-readable storage medium, on which a user rating program is stored, where the user rating program is executable by one or more processors to implement the following operations:
the method comprises the steps of obtaining an original user tag set, an original user behavior record set and a product grading set, and conducting exception removal processing on the original user tag set and the original user behavior record set to obtain a primary user tag set and a primary user behavior record set with corresponding relations.
And carrying out hierarchical operation on the primary user tag set according to the user attributes to obtain a common user tag set and a core user tag set.
Preprocessing the primary user behavior record set to obtain a training set, inputting the training set, the common user tag set and the core user tag set into a pre-constructed hierarchical model for training to obtain a loss value, judging the size relation between the loss value and a preset loss value, continuing training internal parameters by the hierarchical model if the loss value is larger than the preset loss value, and finishing training by the hierarchical model if the loss value is smaller than or equal to the preset loss value.
Receiving a behavior record of a user, inputting the behavior record of the user into the trained hierarchical model, classifying the user to obtain a user hierarchical set, establishing a user-product corresponding model according to the user hierarchical set and the product hierarchical set, and recommending a product corresponding to the user according to the user-product corresponding model.
It should be noted that the above-mentioned numbers of the embodiments of the present invention are merely for description, and do not represent the merits of the embodiments. And the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, apparatus, article, or method 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, apparatus, article, or method. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, apparatus, article, or method that includes the element.
Through the above description of the embodiments, those skilled in the art will clearly understand that the method of the above embodiments can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware, but in many cases, the former is a better implementation manner. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium (e.g., ROM/RAM, magnetic disk, optical disk) as described above and includes instructions for enabling a terminal device (e.g., a mobile phone, a computer, a server, or a network device) to execute the method according to the embodiments of the present invention.
The above description is only a preferred embodiment of the present invention, and not intended to limit the scope of the present invention, and all modifications of equivalent structures and equivalent processes, which are made by using the contents of the present specification and the accompanying drawings, or directly or indirectly applied to other related technical fields, are included in the scope of the present invention.

Claims (10)

1. A method for user ranking, the method comprising:
acquiring an original user tag set, an original user behavior record set and a product grading set, and performing exception removal processing on the original user tag set and the original user behavior record set to obtain a primary user tag set and a primary user behavior record set with corresponding relations;
performing hierarchical operation on the primary user tag set according to user attributes to obtain a common user tag set and a core user tag set:
preprocessing the primary user behavior record set to obtain a training set, inputting the training set, the common user tag set and the core user tag set into a pre-constructed hierarchical model for training to obtain a loss value, judging the size relation between the loss value and a preset loss value, continuing training internal parameters by the hierarchical model if the loss value is larger than the preset loss value, and finishing training by the hierarchical model if the loss value is smaller than or equal to the preset loss value;
receiving a behavior record of a user, inputting the behavior record of the user into the trained hierarchical model, and grading the user to obtain a user grading set;
and establishing a user-product corresponding model according to the user grading set and the product grading set, and recommending products corresponding to the user according to the user-product corresponding model.
2. The user ranking method of claim 1 wherein the ranking model comprises an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer; and
the training comprises the following steps:
receiving the primary user behavior record set by using the input layer, training according to the primary user behavior record set by using the convolutional layer, the pooling layer and the full-link layer in combination with an activation function to obtain a training value, and outputting a predicted value through an output layer;
and performing loss calculation on the predicted value, the common user tag set and the core user tag set to obtain a loss value.
3. The user rating method of claim 2, wherein the activation function comprises:
Figure FDA0002231401410000011
wherein, OjPresentation instrumentThe output value of the jth neuron of the full connection layer, IjRepresenting an input value of a jth neuron of the output layer, t representing a total amount of neurons of the output layer, e being an infinite acyclic fraction;
the loss calculation includes:
wherein s is the predicted value, k is the number of the primary user tag sets, yiIs the common user tag set, y'iA set of labels for the core user.
4. The user rating method of any of claims 1 to 3, wherein the establishing a user-product correspondence model comprises:
calculating preference similarity of different users to different products according to the user grading set and the product grading set;
calculating the similarity of the user behavior habits in the user hierarchical set;
and constructing the user-product corresponding model based on the preference similarity and the user behavior habit similarity.
5. The user ranking method of claim 4 wherein the preference similarity comprises:
Figure FDA0002231401410000022
wherein ps (m, n) is preference similarity of different users to different products, where m, n represents different user level data in the user hierarchy set, m is user level data of a core user in the user hierarchy set, n is user level data of a general user in the user hierarchy set, and p issRepresenting said set of graded products, s being different graded products, rmi、rniRepresenting said user at one of said hierarchical product setsThe number of users in the same level, r, of the hierarchical setm、rnAnd a and b are correlation coefficients of the preference similarity.
6. A user rating device, the device comprising a memory and a processor, the memory having stored thereon a user rating program executable on the processor, the user rating program when executed by the processor implementing the steps of:
acquiring an original user tag set, an original user behavior record set and a product grading set, and performing exception removal processing on the original user tag set and the original user behavior record set to obtain a primary user tag set and a primary user behavior record set with corresponding relations;
performing hierarchical operation on the primary user tag set according to user attributes to obtain a common user tag set and a core user tag set;
preprocessing the primary user behavior record set to obtain a training set, inputting the training set, the common user tag set and the core user tag set into a pre-constructed hierarchical model for training to obtain a loss value, judging the size relation between the loss value and a preset loss value, continuing training internal parameters by the hierarchical model if the loss value is larger than the preset loss value, and finishing training by the hierarchical model if the loss value is smaller than or equal to the preset loss value;
receiving a behavior record of a user, inputting the behavior record of the user into the trained hierarchical model, and grading the user to obtain a user grading set;
and establishing a user-product corresponding model according to the user grading set and the product grading set, and recommending products corresponding to the user according to the user-product corresponding model.
7. The user rating apparatus of claim 6, wherein the rating model comprises an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer; and
the training comprises the following steps:
receiving the primary user behavior record set by using the input layer, training according to the primary user behavior record set by using the convolutional layer, the pooling layer and the full-link layer in combination with an activation function to obtain a training value, and outputting a predicted value through an output layer;
and performing loss calculation on the predicted value, the common user tag set and the core user tag set to obtain a loss value.
8. The user rating apparatus of claim 7, wherein the activation function comprises:
wherein, OjRepresents the output value, I, of the jth neuron of the fully-connected layerjRepresenting an input value of a jth neuron of the output layer, t representing a total amount of neurons of the output layer, e being an infinite acyclic fraction;
the loss calculation includes:
Figure FDA0002231401410000032
wherein s is the predicted value, k is the number of the primary user tag sets, yiIs the common user tag set, y'iA set of labels for the core user.
9. The user rating apparatus of any of claims 6 to 8, wherein the establishing the user-product correspondence model comprises:
calculating preference similarity of different users to different products according to the user grading set and the product grading set;
calculating the similarity of the user behavior habits in the user hierarchical set;
and constructing the user-product corresponding model based on the preference similarity and the user behavior habit similarity.
10. A computer-readable storage medium having a user rating program stored thereon, the user rating program being executable by one or more processors to implement the steps of the user rating method as claimed in any one of claims 1 to 5.
CN201910976136.9A 2019-10-12 2019-10-12 User grading method, device and computer readable storage medium Active CN110852785B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910976136.9A CN110852785B (en) 2019-10-12 2019-10-12 User grading method, device and computer readable storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910976136.9A CN110852785B (en) 2019-10-12 2019-10-12 User grading method, device and computer readable storage medium

Publications (2)

Publication Number Publication Date
CN110852785A true CN110852785A (en) 2020-02-28
CN110852785B CN110852785B (en) 2023-11-21

Family

ID=69596616

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910976136.9A Active CN110852785B (en) 2019-10-12 2019-10-12 User grading method, device and computer readable storage medium

Country Status (1)

Country Link
CN (1) CN110852785B (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2021139432A1 (en) * 2020-10-13 2021-07-15 平安科技（深圳）有限公司 Artificial intelligence-based user rating prediction method and apparatus, terminal, and medium
CN114860788A (en) * 2022-04-22 2022-08-05 北京猎人纪信息科技有限公司 Technology promotion information service system and method
CN115511124A (en) * 2022-09-27 2022-12-23 上海网商电子商务有限公司 Customer grading method based on after-sale maintenance records

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2009057103A2 (en) * 2007-10-31 2009-05-07 Musicgenome.Com Inc. Content-based recommendations across domains
WO2010037286A1 (en) * 2008-09-27 2010-04-08 华为技术有限公司 Collaborative filtering-based recommendation method and system
CN107820619A (en) * 2017-09-21 2018-03-20 达闼科技（北京）有限公司 One kind classification interactive decision making method, interactive terminal and cloud server
CN110246012A (en) * 2019-06-14 2019-09-17 哈尔滨哈银消费金融有限责任公司 Consumer finance Products Show method, apparatus and equipment based on social data
CN110264306A (en) * 2019-05-21 2019-09-20 平安银行股份有限公司 Products Show method, apparatus, server and medium based on big data
CN110298574A (en) * 2019-06-21 2019-10-01 国网辽宁省电力有限公司鞍山供电公司 A kind of electricity consumption subscriber payment risk rating method based on convolutional neural networks

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2009057103A2 (en) * 2007-10-31 2009-05-07 Musicgenome.Com Inc. Content-based recommendations across domains
WO2010037286A1 (en) * 2008-09-27 2010-04-08 华为技术有限公司 Collaborative filtering-based recommendation method and system
CN107820619A (en) * 2017-09-21 2018-03-20 达闼科技（北京）有限公司 One kind classification interactive decision making method, interactive terminal and cloud server
CN110264306A (en) * 2019-05-21 2019-09-20 平安银行股份有限公司 Products Show method, apparatus, server and medium based on big data
CN110246012A (en) * 2019-06-14 2019-09-17 哈尔滨哈银消费金融有限责任公司 Consumer finance Products Show method, apparatus and equipment based on social data
CN110298574A (en) * 2019-06-21 2019-10-01 国网辽宁省电力有限公司鞍山供电公司 A kind of electricity consumption subscriber payment risk rating method based on convolutional neural networks

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2021139432A1 (en) * 2020-10-13 2021-07-15 平安科技（深圳）有限公司 Artificial intelligence-based user rating prediction method and apparatus, terminal, and medium
CN114860788A (en) * 2022-04-22 2022-08-05 北京猎人纪信息科技有限公司 Technology promotion information service system and method
CN115511124A (en) * 2022-09-27 2022-12-23 上海网商电子商务有限公司 Customer grading method based on after-sale maintenance records
CN115511124B (en) * 2022-09-27 2023-04-18 上海网商电子商务有限公司 Customer grading method based on after-sale maintenance records

Also Published As

Publication number Publication date
CN110852785B (en) 2023-11-21

Similar Documents

Publication Publication Date Title
CN109493199A (en) Products Show method, apparatus, computer equipment and storage medium
CN112785397A (en) Product recommendation method, device and storage medium
CN110060090A (en) Method, apparatus, electronic equipment and the readable storage medium storing program for executing of Recommendations combination
CN111523976A (en) Commodity recommendation method and device, electronic equipment and storage medium
CN105446970A (en) Item recommendation method and device
CN106844407B (en) Tag network generation method and system based on data set correlation
CN110135942A (en) Products Show method, apparatus and computer readable storage medium
CN110852785A (en) User grading method, device and computer readable storage medium
CN112070577A (en) Commodity recommendation method, system, equipment and medium
CN104517216A (en) Enhanced recommender system and method
CN109740642A (en) Invoice category recognition methods, device, electronic equipment and readable storage medium storing program for executing
CN111400613A (en) Article recommendation method, device, medium and computer equipment
CN110276382A (en) Listener clustering method, apparatus and medium based on spectral clustering
CN113157752A (en) Scientific and technological resource recommendation method and system based on user portrait and situation
CN113592605A (en) Product recommendation method, device, equipment and storage medium based on similar products
CN111582932A (en) Inter-scene information pushing method and device, computer equipment and storage medium
CN113505273B (en) Data sorting method, device, equipment and medium based on repeated data screening
CN113850654A (en) Training method of item recommendation model, item screening method, device and equipment
US20210090105A1 (en) Technology opportunity mapping
CN116308683A (en) Knowledge-graph-based clothing brand positioning recommendation method, equipment and storage medium
CN116127184A (en) Product recommendation method and device, nonvolatile storage medium and electronic equipment
CN113656690B (en) Product recommendation method and device, electronic equipment and readable storage medium
CN114429384A (en) Intelligent product recommendation method and system based on e-commerce platform
CN113837847A (en) Knowledge-intensive service recommendation method based on heterogeneous multivariate relation fusion
Agarwal et al. Binarized spiking neural networks optimized with Nomadic People Optimization-based sentiment analysis for social product recommendation

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
GR01 Patent grant
GR01 Patent grant