CN112214668B - Personalized financial service recommendation device and method based on big data - Google Patents

Personalized financial service recommendation device and method based on big data Download PDF

Info

Publication number
CN112214668B
CN112214668B CN202011040637.5A CN202011040637A CN112214668B CN 112214668 B CN112214668 B CN 112214668B CN 202011040637 A CN202011040637 A CN 202011040637A CN 112214668 B CN112214668 B CN 112214668B
Authority
CN
China
Prior art keywords
service
user
matrix
hidden
behavior
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.)
Active
Application number
CN202011040637.5A
Other languages
Chinese (zh)
Other versions
CN112214668A (en
Inventor
许明
周玥
罗辛
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.)
Shenzhen Wanjiaan Interconnected Technology Co ltd
Chongqing Institute of Green and Intelligent Technology of CAS
Original Assignee
Shenzhen Wanjiaan Interconnected Technology Co ltd
Chongqing Institute of Green and Intelligent Technology of CAS
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 Shenzhen Wanjiaan Interconnected Technology Co ltd, Chongqing Institute of Green and Intelligent Technology of CAS filed Critical Shenzhen Wanjiaan Interconnected Technology Co ltd
Priority to CN202011040637.5A priority Critical patent/CN112214668B/en
Publication of CN112214668A publication Critical patent/CN112214668A/en
Application granted granted Critical
Publication of CN112214668B publication Critical patent/CN112214668B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24—Querying
    • G06F16/245—Query processing
    • G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
    • G06F16/2462—Approximate or statistical queries
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/958—Organisation or management of web site content, e.g. publishing, maintaining pages or automatic linking
    • 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
    • G06Q40/06—Asset management; Financial planning or analysis
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Databases & Information Systems (AREA)
  • General Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Probability & Statistics with Applications (AREA)
  • Finance (AREA)
  • Development Economics (AREA)
  • Accounting & Taxation (AREA)
  • Computational Linguistics (AREA)
  • Operations Research (AREA)
  • Mathematical Physics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Fuzzy Systems (AREA)
  • Game Theory and Decision Science (AREA)
  • Human Resources & Organizations (AREA)
  • Software Systems (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • Strategic Management (AREA)
  • Technology Law (AREA)
  • General Business, Economics & Management (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The invention discloses a personalized financial service recommendation device and method based on big data, comprising S1, obtaining user-service behavior statistical data and establishing a user-service behavior statistical matrix T; s2: receiving an instruction and setting initialization parameters; s3: constructing a unified loss function phi, and carrying out non-negative constraint on the unified loss function; s4: constructing a Nesterov acceleration user-service behavior recommendation model by adopting a Nesterov acceleration gradient method and training; s5: and outputting the user hidden characteristic matrix and the service hidden characteristic matrix, and restoring the unknown behavior data in the user-service behavior statistical matrix. By adopting the Nesterov acceleration gradient method, the internal statistical rule of the known user-service behavior statistical data is analyzed with smaller computational complexity, so that the personalized financial service unknown behavior data based on big data is accurately restored, and personalized and accurate financial service is provided for the user.

Description

Personalized financial service recommendation device and method based on big data
Technical Field
The invention relates to the technical field of data processing, in particular to a personalized financial service recommendation device and method based on big data.
Background
Along with the continuous development of technologies such as big data and artificial intelligence, traditional financial services with industrial era structures gradually move towards scenerization, individualization and intellectualization personalized financial services, such as a large number of financial services such as 'Living treasures', 'profit treasures', 'balance treasures', and the like, and the consumption experience of users is comprehensively satisfied. However, with the continuous expansion of financial services, users have difficulty in selecting services which are required by themselves from massive services, so that the financial service industry hopes to actively capture the consumption demands of users through big data technology and machine learning, and provide more accurate services for users.
According to the history of a financial service system, a user-service behavior statistical matrix is a commonly used data description structure, wherein each row pair uses a user, each column corresponds to a service, and one element of each matrix corresponds to the history behavior data of a user to a service. In the financial service, the user faces a huge amount of services, and cannot operate all services, and also one service cannot be operated by all users, so that the user-service behavior statistical matrix is often extremely sparse, and the unknown behavior of the user to the service is far more than the known behavior.
In a financial service system, behavior data of user services often have a positive natural law, so the financial service system recommends services for users by adopting the existing non-negative matrix hidden characteristic analysis method, but the non-negative matrix hidden characteristic analysis method has a low convergence speed, and low data recovery accuracy, so that the user cannot be rapidly and effectively recommended for the required services, and therefore, how to perform efficient non-negative hidden characteristic analysis aiming at an extremely sparse user-service behavior statistical matrix in the financial service system, thereby rapidly and accurately recommending the required services for the users, improving user experience, and being a key problem for achieving personalized finance.
Disclosure of Invention
Aiming at the problem of low matching degree between users and services in the prior art, the invention provides a personalized financial service recommendation device and method based on big data.
In order to achieve the above object, the present invention provides the following technical solutions:
a personalized financial service recommendation device based on big data comprises a data receiving module, a data storage module, an initializing module, a Nesterov training module and a service recommendation module; wherein,,
the data receiving module is used for acquiring user-service behavior statistical data from the financial service system, establishing a user-service behavior statistical matrix and storing the user-service behavior statistical matrix into the data storage module;
the initialization module is used for preprocessing the user-service behavior statistical matrix and initializing parameters;
the Nesterov training module is used for constructing a Nesterov acceleration gradient user-service behavior statistical model according to the user-service behavior statistical data and the initialized parameters so as to accelerate recommendation of the user-service behavior statistical data;
the service recommending module is used for restoring the unknown behavior data of the user-service through the constructed Nesterov acceleration gradient user-service behavior statistical model and recommending corresponding service to the user according to the unknown behavior data.
Preferably, the user-service behavior statistical matrix is a matrix of |U| rows and |S| columns, U represents a user set, and S represents a service set.
Preferably, the Nesterov training module comprises a Nesterov updating unit and a training unit; wherein,,
the Nesterov updating unit is used for carrying out parameter updating on the user hidden characteristic matrix X, the service hidden characteristic matrix Y, the user linear deviation vector B and the service linear deviation vector C by adopting a Nesterov acceleration gradient method;
the training unit adopts non-negative matrix hidden characteristic analysis to train the user hidden characteristic matrix X, the service hidden characteristic matrix Y, the user linear deviation vector B and the service linear deviation vector C.
Preferably, the service recommendation module comprises an output hidden characteristic unit and a recommendation unknown interestingness unit; wherein,,
the output hidden characteristic unit outputs a user hidden characteristic matrix X and a service hidden characteristic matrix Y in the user-service behavior statistical data;
recommending an unknown interestingness unit, and restoring the user-service unknown behavior data according to the predicted user-service behavior statistical matrix T obtained by the user hidden characteristic matrix X and the service hidden characteristic matrix Y in the output user-service behavior statistical data.
The invention also provides a personalized financial service recommendation method based on big data, which comprises the following steps:
s1, acquiring user-service behavior statistical data and establishing a user-service behavior statistical matrix T;
s2: receiving an instruction and setting initialization parameters;
s3: constructing a unified loss function, and carrying out non-negative constraint on the unified loss function;
s4: constructing a Nesterov acceleration user-service behavior recommendation model by adopting a Nesterov acceleration gradient method and training;
s5: outputting a user hidden characteristic matrix and a service hidden characteristic matrix, predicting scores of unknown behavior data in a user-service behavior statistical matrix, and recommending services corresponding to the unknown behavior data of the front top k to the user.
Preferably, the S1 includes:
the method comprises the steps that a user set is U, a service set is S, a matrix of I U I row and I S I column is established as a user-service behavior statistical matrix T, and then a rule matrix factorization is used for decomposing the T to respectively obtain a user hidden characteristic matrix X and a service hidden characteristic matrix Y, wherein X is a matrix of I U I row and F column, and each row vector in X corresponds to a user and is a hidden characteristic vector of the user; y is a matrix of |S| rows and f columns, and each row vector in Y corresponds to a service and is a hidden feature vector of the service; f is the dimension of the user implicit feature space and the service implicit feature space.
Preferably, the initialization parameters include hidden feature matrices X and Y; hidden feature space dimension f; a Nesterov momentum control parameter mu, a regularization factor lambda; the maximum iteration round number R; the iteration round number in the training process controls variable r; the linear bias conceals feature vectors B and C.
Preferably, in the step S3, the unified loss function Φ (X, Y, B, C) has the expression:
Figure BDA0002706526870000041
in the formula (1), t u,s Historical behavior data representing the entity relationship between the user u and the service s, namely the user u on the service s; Λ represents a set of known behavior data of the user for the financial service in the user-service behavior statistics matrix T;
Figure BDA0002706526870000042
representing a reduction value of an unknown behavior in the user-service behavior statistics matrix; b u A linear deviation (U) th row characteristic value of the user set is represented; c s Representing the linear deviation S-th line characteristic value of the service set S; x is x u,k A kth characteristic value of a ith row and a kth column of the user hidden characteristic matrix X is represented; y is s,k The kth row and the kth column eigenvalue of the service hidden eigenvalue matrix Y are represented; λ represents the canonical control parameter and f represents the dimension of the implicit feature space.
Preferably, the step S4 includes the steps of:
s4-1: in the first iteration, training a user hidden characteristic matrix X, a service hidden characteristic matrix Y, a user linear deviation vector B and a service linear deviation vector C by adopting a non-negative matrix hidden characteristic analysis method, wherein the training formula is as follows:
Figure BDA0002706526870000051
in the formula (2), Λ (u) represents a set of services in the user-service known behavior data related to the user u, and Λ(s) represents a set of users in the user-service known behavior data related to the service s;
s4-2: and updating the user hidden characteristic matrix X, the service hidden characteristic matrix Y, the user linear deviation vector B and the service linear deviation vector C by adopting a Nesterov acceleration gradient method, wherein the updating formula is as follows:
Figure BDA0002706526870000052
in the formula (3),
Figure BDA0002706526870000053
representation b u State at time t->
Figure BDA0002706526870000054
Indicating updated time b at time t+1 u (t+1) μ is a Nesterov control parameter, the acceleration effect is controlled, and the Nesterov acceleration gradient method is only added for the second time in the training process, namely t is more than or equal to 2, and the iteration times are represented;
s4-3: according to a non-negative matrix hidden characteristic analysis method, the updated matrix hidden characteristic is used for updating
Figure BDA0002706526870000055
Training is carried out, wherein the training formula is as follows: />
Figure BDA0002706526870000061
Parameters after training
Figure BDA0002706526870000062
Indicating the use of intermediate states +.>
Figure BDA0002706526870000063
The state value at the time t+1 after updating;
s4-4: judging whether the training reaches the iteration termination condition, if so, ending the iteration training, and if not, continuing the training and updating until the iteration termination condition is reached.
Preferably, the step S5 includes the steps of:
s5-1: acquiring hidden features corresponding to a user u, a service s, a user linear deviation b and a service linear deviation c from an output hidden feature unit;
s5-2: the inner product of the vector in the user hidden characteristic matrix X and the vector in the service hidden characteristic matrix Y corresponding to the service s and the user linear deviation b and the service linear deviation c is taken as the unknown behavior data value of the user u to the service s
Figure BDA0002706526870000064
I.e.
Figure BDA0002706526870000065
In summary, due to the adoption of the technical scheme, compared with the prior art, the invention has at least the following beneficial effects:
according to the invention, the Nesterov acceleration gradient method is adopted, and the internal statistical rule of the known user-service behavior statistical data is analyzed with smaller calculation complexity, so that the personalized financial service unknown behavior data based on big data is accurately restored, and the corresponding personalized and accurate financial service is provided for the user; the invention also realizes that: (1) The non-negative matrix hidden characteristic analysis method is used for carrying out non-negative limitation on the user-service behavior statistical matrix, so that the recommendation accuracy is improved, and the user characteristics are better ensured; (2) The recommendation precision is improved by a Nesterov acceleration gradient method; (3) By using the Nesterov acceleration gradient method, the recommendation speed is increased, the time cost is saved, and the matching degree of the user and the service is improved.
Description of the drawings:
fig. 1 is a schematic view of a personalized financial service recommendation device based on big data according to an exemplary embodiment of the present invention.
Fig. 2 is a schematic diagram of a personalized financial service recommendation method based on big data according to an exemplary embodiment of the present invention.
Fig. 3 is a diagram illustrating a user-service behavior statistics analysis time versus intent in accordance with an exemplary embodiment of the present invention.
Fig. 4 is a comparative schematic diagram of RMSE according to an exemplary embodiment of the invention.
Detailed Description
The present invention will be described in further detail with reference to examples and embodiments. It should not be construed that the scope of the above subject matter of the present invention is limited to the following embodiments, and all techniques realized based on the present invention are within the scope of the present invention.
In the description of the present invention, it should be understood that the terms "longitudinal," "transverse," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," and the like indicate orientations or positional relationships based on the orientation or positional relationships shown in the drawings, merely to facilitate describing the present invention and simplify the description, and do not indicate or imply that the devices or elements referred to must have a specific orientation, be configured and operated in a specific orientation, and therefore should not be construed as limiting the present invention.
As shown in fig. 1, the present invention proposes a personalized financial service recommendation device based on big data, which includes a data receiving module 10, a data storage module 20, an initializing module 30, a nestrov training module 40 and a service recommendation module 50. The output end of the data receiving module 10 is connected with the input end of the data storage module 20, the output end of the data storage module 20 is connected with the input end of the initializing module 30, the output end of the initializing module 30 is connected with the input end of the Nesterov training module 40, and the output end of the Nesterov training module 40 is connected with the input end of the service recommending module 50. Wherein,,
the data receiving module 10 is configured to obtain user-service behavior statistics data from the financial service system, establish a user-service behavior statistics matrix, and store the user-service behavior statistics matrix in the data storage module 20.
In the obtained user-service behavior statistics data, the user set is denoted as U, the service set is denoted as S, a matrix of |u| rows and |s| columns is established as the user-service behavior statistics matrix T, and the user-service behavior statistics matrix T is stored in the data storage module 20. Decomposing T by using a protocol matrix factorization to respectively obtain a user hidden characteristic matrix X and a service hidden characteristic matrix Y, wherein X is a matrix of I U I row and f column, and each row vector in X corresponds to a user and is a hidden characteristic vector of the user; y is a matrix of |S| rows and f columns, and each row vector in Y corresponds to a service and is a hidden feature vector of the service; f is the dimension of the user implicit feature space and the service implicit feature space; and simultaneously storing the user hidden characteristic matrix X and the service hidden characteristic matrix Y in the data storage module 20.
An initialization module 30 is configured to initialize parameters in the recommendation process of the user-service behavior characteristics.
The initialized parameters comprise hidden feature matrixes X and Y of users and services; hidden feature space dimension f; a Nesterov momentum control parameter mu, a regularization factor lambda; the maximum iteration round number R; the iteration round number in the training process controls variable r; the linear bias conceals feature vectors B and C. F determines the feature space dimension of each hidden feature matrix and initializes the feature space dimension to a positive integer; in the hidden feature matrices X and Y: the hidden feature matrix with X being |U| row and f column and the hidden feature matrix with Y being |S| row and f column are initialized by using random smaller positive numbers respectively; linear bias hidden feature vectors B and C: b is a vector of |u| row and C is a vector of |s| row, and are initialized by using random smaller positive numbers respectively; the maximum training iteration round number R is a variable for controlling the upper limit of training times and is initialized to be a larger positive integer; initializing an iteration round number control variable r to 0; regularization factor λ is a measure of L 2 The limiting effect of the regularization term on the recommendation model is initialized to be smaller positive number; and the Nesterov momentum control parameter mu is used for controlling the acceleration effect of the Nesterov momentum on the model, so that the data recovery accuracy of the recommended model on unknown behavior data is highest, and the data recovery accuracy is initialized to be a positive number close to 1.
The Nesterov training module 40 is configured to construct a Nesterov acceleration gradient user-service behavior statistical model according to the user-service behavior statistical data and the initialized parameters so as to accelerate recommendation of the user-service behavior statistical data.
In this embodiment, the Nesterov training module 40 includes a Nesterov updating unit and a training unit.
And the Nesterov updating unit is used for carrying out parameter updating on the user hidden characteristic matrix X, the service hidden characteristic matrix Y, the user linear deviation vector B and the service linear deviation vector C by adopting a Nesterov acceleration gradient method, and updating corresponding hidden characteristics in a single element mode.
The training unit adopts non-negative matrix hidden characteristic analysis to train the hidden characteristic matrix X, the hidden characteristic matrix Y, the linear deviation vector B and the linear deviation vector C of the user, and updates the corresponding hidden characteristic in a single element mode.
The service recommendation module 50 is configured to restore the user-service unknown behavior data through the constructed nestrov acceleration gradient user-service behavior statistical model, and output the user-service unknown behavior restoration data, and the hidden feature matrix of the user and the service.
In this embodiment, the service recommendation module 50 includes an output hidden feature unit and a recommendation unknown interestingness unit.
And outputting the hidden characteristic unit, namely outputting a user hidden characteristic matrix X and a service hidden characteristic matrix Y in the user-service behavior statistical data.
Recommending an unknown interestingness unit, and restoring the unknown behavior data of the user-service according to the predicted user-service behavior statistical matrix T-A obtained by outputting the user hidden characteristic matrix X and the service hidden characteristic matrix Y in the user-service behavior statistical data.
Based on the above device, as shown in fig. 2, the invention further provides a personalized financial service recommendation method based on big data, which specifically comprises the following steps:
s1, acquiring user-service behavior statistical data, and establishing a user-service behavior statistical matrix T, wherein for each element in the matrix, rows represent corresponding users, and columns represent corresponding services;
in the obtained user-service behavior statistics data, the user set is denoted as U, the service set is denoted as S, a matrix of |u| rows and |s| columns is established as the user-service behavior statistics matrix T, and the user-service behavior statistics matrix T is stored in the data storage module 20. Decomposing T by using a protocol matrix factorization to respectively obtain a user hidden characteristic matrix X and a service hidden characteristic matrix Y, wherein X is a matrix of I U I row and f column, and each row vector in X corresponds to a user and is a hidden characteristic vector of the user; y is a matrix of |S| rows and f columns, and each row vector in Y corresponds to a service and is a hidden feature vector of the service; f is the dimension of the user implicit feature space and the service implicit feature space.
S2: receiving an instruction and setting initialization parameters;
in this embodiment, the initialized parameters include:
user-service known behavior data t u,s The entity relation between the user u and the service s is represented as historical behavior data of the user u on the service s;
a linear deviation hidden feature vector B which represents a linear deviation feature vector of the user set U;
a linear bias hidden feature vector C representing a linear bias feature vector of the service set S;
the hidden characteristic matrix X represents a user hidden characteristic matrix after the user-service behavior statistical data matrix is factorized by a protocol matrix;
the hidden characteristic matrix Y represents a service hidden characteristic matrix after the user-service behavior statistical data matrix is factorized by the protocol matrix;
the regular control parameter lambda is used for measuring the limiting effect of the regular term on the model, so that the robustness of the model is enhanced;
and the Nesterov momentum control parameter mu is used for controlling the acceleration effect of the Nesterov momentum on the model, so that the data recovery accuracy of the recommended model on unknown behavior data is highest.
S3: constructing a unified loss function phi (X, Y, B and C), and carrying out non-negative constraint on the unified loss function to ensure non-negativity of the user and service hidden characteristic matrix in the training process;
Figure BDA0002706526870000111
in the formula (1), t u,s Historical behavior data representing the entity relationship between the user u and the service s, namely the user u on the service s; Λ represents a set of known behavior data of the user for the financial service in the user-service behavior statistics matrix T;
Figure BDA0002706526870000113
representing a reduction value of an unknown behavior in the user-service behavior statistics matrix; b u A linear deviation (U) th row characteristic value of the user set is represented; c s Representing the linear deviation S-th line characteristic value of the service set S; x is x u,k Representing user hidden featuresThe ith row and the kth column eigenvalue of the matrix X; y is s,k The kth row and the kth column eigenvalue of the service hidden eigenvalue matrix Y are represented; λ represents the canonical control parameter and f represents the dimension of the implicit feature space.
S4: and constructing a Nesterov acceleration user-service behavior recommendation model by adopting a Nesterov acceleration gradient method and training.
In this embodiment, the expression for constructing the Nesterov acceleration user-service behavior recommendation model is represented by a unified loss function Φ (X, Y, B, C).
S4-1: in the first iteration, i.e. t=1, training the user hidden feature matrix X, the service hidden feature matrix Y, the user linear deviation vector B and the service linear deviation vector C by adopting a non-negative matrix hidden feature analysis method, wherein the training formula is as follows:
Figure BDA0002706526870000112
in the formula (2), Λ (u) | represents a set of services in the user-service known behavior data related to the user u, and Λ(s) | represents a set of users in the user-service known behavior data related to the service s.
S4-2: the Nesterov acceleration gradient method is added to accelerate the Nesterov acceleration user-service behavior recommendation model, and the Nesterov acceleration gradient method is adopted to update the user hidden characteristic matrix X, the service hidden characteristic matrix Y, the user linear deviation vector B and the service linear deviation vector C, wherein the updating formula is as follows:
Figure BDA0002706526870000121
in the formula (3),
Figure BDA0002706526870000122
representation b u State at time t->
Figure BDA0002706526870000123
Representing the representation warpPost-update at time b of t+1 u (t+1) μ is a Nesterov control parameter, and the acceleration effect is controlled. The Nesterov acceleration gradient method is only added for the second time in the training process, namely t is more than or equal to 2, and the iteration times are represented.
S4-3: according to a non-negative matrix hidden characteristic analysis method, the updated matrix hidden characteristic is used for updating
Figure BDA0002706526870000124
Training is carried out, wherein the training formula is as follows:
Figure BDA0002706526870000125
parameters after training
Figure BDA0002706526870000126
Indicating the use of intermediate states +.>
Figure BDA0002706526870000127
The state value at time t+1 after updating.
S4-4: judging whether the training reaches the iteration termination condition, if so, ending the iteration training, and if not, continuing the training and updating until the iteration termination condition is reached.
Iteration termination condition: if the iteration round number control variable R exceeds the training iteration round number upper limit R, ending the iterative training; if the iteration round number control variable R is smaller than the training iteration round number upper limit R (R can take the value of 1000), training and updating are continued until the iteration termination condition is reached.
S5: outputting a user hidden characteristic matrix and a service hidden characteristic matrix, predicting unknown behavior data in a user-service behavior statistical matrix, and according to the predicted unknown behavior data
Figure BDA0002706526870000133
Recommended services, i.e. unknown behavioural data->
Figure BDA0002706526870000134
The larger the value of (2) is, the tableThe greater the probability that the user will be shown to use the service, the service will thus be provided (depending on the unknown behavioural data +.>
Figure BDA0002706526870000135
Is a value of (c) is recommended to the interested user.
S5-1: and obtaining hidden features corresponding to the user u, the service s, the user linear deviation b and the service linear deviation c from the output hidden feature unit.
S5-2: the inner product of the vector in the user hidden characteristic matrix X and the vector in the service hidden characteristic matrix Y corresponding to the service s and the user linear deviation b and the service linear deviation c is taken as the unknown behavior data value of the user u to the service s
Figure BDA0002706526870000131
I.e.
Figure BDA0002706526870000132
b u A linear deviation (U) th row characteristic value of the user set is represented; c s Representing the linear deviation S-th line characteristic value of the service set S; x is x u,k A kth characteristic value of a ith row and a kth column of the user hidden characteristic matrix X is represented; y is s,k And the kth eigenvalue of the s-th row of the service hidden eigenvector matrix Y is represented.
In this embodiment, fig. 3 is a comparison chart of data analysis time of a recommendation model before and after the recommendation device and the method of the present invention are applied. As can be seen from fig. 3, the recommendation device and method of the present invention are applied, and the execution time is far less for the user-service behavior statistics than in the case where the recommendation device and method of the present invention is not applied. Specifically, as can be seen from fig. 3, the model for which the recommendation apparatus and method of the present invention was not used was executed about 7 times longer than the recommendation apparatus and method of the present invention. The execution time of the model is improved by 7 times compared with that of the original model by using the recommendation device and the recommendation method, and the operation efficiency of the model is greatly improved.
Fig. 4 is a graph of RMSE comparison during data analysis before and after application of the recommendation apparatus and method of the present invention, where RMSE (root mean square error ) is a measure of recommended error, and the smaller the RMSE, the higher the accuracy. As can be seen from fig. 4, the recommendation device and method of the present invention greatly improves the accuracy of personalized service selection for users in financial services. In practical application, the service for improving the personalized requirements for the user can be better realized.
According to the personalized financial service recommendation device and method based on the big data, the Nesterov acceleration gradient method is adopted, and the internal statistical rule of the statistical data of the known user-service behaviors is analyzed with smaller calculation complexity, so that the unknown behavior data of the personalized financial service based on the big data is accurately restored, and personalized and accurate financial service is provided for the user; the invention also realizes that: (1) The non-negative matrix hidden characteristic analysis method is used for carrying out non-negative limitation on the user-service behavior statistical matrix, so that the recommendation accuracy is ensured, and the user characteristics are better ensured; (2) The recommendation precision of the recommendation model is improved by a Nesterov acceleration gradient method; (3) The calculation speed of the recommended model is accelerated by a Nesterov acceleration gradient method.
It will be understood by those of ordinary skill in the art that the foregoing embodiments are specific examples of carrying out the invention and that various changes in form and details may be made therein without departing from the spirit and scope of the invention.

Claims (9)

1. The personalized financial service recommending device based on big data is characterized by comprising a data receiving module, a data storage module, an initializing module, a Nesterov training module and a service recommending module; wherein,,
the data receiving module is used for acquiring user-service behavior statistical data from the financial service system, establishing a user-service behavior statistical matrix and storing the user-service behavior statistical matrix into the data storage module;
the initialization module is used for preprocessing the user-service behavior statistical matrix and initializing parameters;
the Nesterov training module is used for constructing a Nesterov acceleration gradient user-service behavior statistical model according to the user-service behavior statistical data and the initialized parameters so as to accelerate recommendation of the user-service behavior statistical data;
the construction and training method of the Nesterov acceleration gradient user-service behavior statistical model comprises the following steps:
s4-1: in the first iteration, training a user hidden characteristic matrix X, a service hidden characteristic matrix Y, a user linear deviation vector B and a service linear deviation vector C by adopting a non-negative matrix hidden characteristic analysis method, wherein the training formula is as follows:
Figure FDA0004201521610000011
in the formula (1), b u A linear deviation (U) th row characteristic value of the user set is represented; Λ represents a set of known behavior data of the user for the financial service in the user-service behavior statistics matrix T; t is t u,s Historical behavior data representing the entity relationship between the user u and the service s, namely the user u on the service s;
Figure FDA0004201521610000012
representing a reduction value of an unknown behavior in the user-service behavior statistics matrix; λ represents a canonical control parameter; the Λ (u) represents the set of services in the user-service known behavior data related to user u;
c s representing the linear deviation S-th line characteristic value of the service set S; Λ(s) represents the set of users in the user-service known behavior data related to service s;
x u,k a kth characteristic value of a ith row and a kth column of the user hidden characteristic matrix X is represented; y is s,k The kth row and the kth column eigenvalue of the service hidden eigenvalue matrix Y are represented;
s4-2: and updating the user hidden characteristic matrix X, the service hidden characteristic matrix Y, the user linear deviation vector B and the service linear deviation vector C by adopting a Nesterov acceleration gradient method, wherein the updating formula is as follows:
Figure FDA0004201521610000021
/>
in the formula (2),
Figure FDA0004201521610000022
representation b u State at time t->
Figure FDA0004201521610000023
Indicating updated time b at time t+1 u (t+1) Intermediate state values of (2); />
Figure FDA0004201521610000024
Representation b u A state at time t-1; mu is a Nesterov control parameter, the acceleration effect is controlled, and the Nesterov acceleration gradient method is only added for the second time in the training process, namely t is more than or equal to 2, and the iteration times are represented;
Figure FDA0004201521610000025
indicating updated +.at time t+1>
Figure FDA0004201521610000026
Intermediate state value of->
Figure FDA0004201521610000027
Representation c s The state at the time point of t,
Figure FDA0004201521610000028
representation c s A state at time t-1;
Figure FDA0004201521610000029
indicating updated +.at time t+1>
Figure FDA00042015216100000210
Intermediate state value of->
Figure FDA00042015216100000211
Represents x u,k State at time t->
Figure FDA00042015216100000212
Represents x u,k A state at time t-1;
Figure FDA00042015216100000213
indicating updated +.at time t+1>
Figure FDA00042015216100000214
Intermediate state value of->
Figure FDA00042015216100000215
Representing y s,k State at time t->
Figure FDA00042015216100000216
Representing y s,k A state at time t-1;
s4-3: according to a non-negative matrix hidden characteristic analysis method, the updated matrix hidden characteristic is used for updating
Figure FDA0004201521610000031
Figure FDA0004201521610000032
Training is carried out, wherein the training formula is as follows:
Figure FDA0004201521610000033
in the formula (3),
Figure FDA0004201521610000034
indicating the use of intermediate states +.>
Figure FDA0004201521610000035
The state value at the time t+1 after updating; />
Figure FDA0004201521610000036
Indicating the use of intermediate states +.>
Figure FDA0004201521610000037
The state value at the time t+1 after updating; />
Figure FDA0004201521610000038
Indicating the use of intermediate states +.>
Figure FDA0004201521610000039
The state value at the time t+1 after updating; />
Figure FDA00042015216100000310
Indicating the use of intermediate states +.>
Figure FDA00042015216100000311
The state value at the time t+1 after updating; r is R k Representing a set of known behavior data of the user for the financial service in the user-service behavior statistics matrix T; r is% k (u) | represents the set of services in the user-service known behavior data related to user u; r is% k (s) | represents the set of users in the user-service known behavior data related to service s;
s4-4: judging whether the training reaches an iteration termination condition, if so, ending the iteration training, and if not, continuing the training and updating until the iteration termination condition is reached;
the service recommending module is used for restoring the unknown behavior data of the user-service through the constructed Nesterov acceleration gradient user-service behavior statistical model and recommending corresponding service to the user according to the unknown behavior data.
2. The personalized financial service recommendation device based on big data according to claim 1, wherein the user-service behavior statistics matrix is a matrix of |u| rows, |s| columns, U represents a user set, and S represents a service set.
3. The personalized financial service recommendation device based on big data according to claim 1, wherein the nestrov training module comprises a nestrov updating unit and a training unit; wherein,,
the Nesterov updating unit is used for carrying out parameter updating on the user hidden characteristic matrix X, the service hidden characteristic matrix Y, the user linear deviation vector B and the service linear deviation vector C by adopting a Nesterov acceleration gradient method;
the training unit adopts non-negative matrix hidden characteristic analysis to train the user hidden characteristic matrix X, the service hidden characteristic matrix Y, the user linear deviation vector B and the service linear deviation vector C.
4. The personalized financial service recommendation device based on big data according to claim 1, wherein the service recommendation module comprises an output hidden feature unit and a recommendation unknown interestingness unit; wherein,,
the output hidden characteristic unit outputs a user hidden characteristic matrix X and a service hidden characteristic matrix Y in the user-service behavior statistical data;
recommending an unknown interestingness unit, and restoring the user-service unknown behavior data according to the predicted user-service behavior statistical matrix T obtained by the user hidden characteristic matrix X and the service hidden characteristic matrix Y in the output user-service behavior statistical data.
5. The personalized financial service recommending method based on the big data is characterized by comprising the following steps of:
s1, acquiring user-service behavior statistical data and establishing a user-service behavior statistical matrix T;
s2: receiving an instruction and setting initialization parameters;
s3: constructing a unified loss function, and carrying out non-negative constraint on the unified loss function;
s4: constructing a Nesterov acceleration user-service behavior recommendation model by adopting a Nesterov acceleration gradient method and training, wherein the method comprises the following steps of:
s4-1: in the first iteration, training a user hidden characteristic matrix X, a service hidden characteristic matrix Y, a user linear deviation vector B and a service linear deviation vector C by adopting a non-negative matrix hidden characteristic analysis method, wherein the training formula is as follows:
Figure FDA0004201521610000051
in the formula (4), b u A linear deviation (U) th row characteristic value of the user set is represented; Λ represents a set of known behavior data of the user for the financial service in the user-service behavior statistics matrix T; t is t u,s Historical behavior data representing the entity relationship between the user u and the service s, namely the user u on the service s;
Figure FDA0004201521610000052
representing a reduction value of an unknown behavior in the user-service behavior statistics matrix; λ represents a canonical control parameter; the Λ (u) represents the set of services in the user-service known behavior data related to user u;
c s representing the linear deviation S-th line characteristic value of the service set S; Λ(s) represents the set of users in the user-service known behavior data related to service s;
x u,k a kth characteristic value of a ith row and a kth column of the user hidden characteristic matrix X is represented; y is s,k The kth row and the kth column eigenvalue of the service hidden eigenvalue matrix Y are represented;
s4-2: and updating the user hidden characteristic matrix X, the service hidden characteristic matrix Y, the user linear deviation vector B and the service linear deviation vector C by adopting a Nesterov acceleration gradient method, wherein the updating formula is as follows:
Figure FDA0004201521610000053
in the formula (5) of the present invention,
Figure FDA0004201521610000054
representation b u State at time t->
Figure FDA0004201521610000055
Indicating updated time b at time t+1 u (t+1) Intermediate state values of (2); />
Figure FDA0004201521610000061
Representation b u A state at time t-1; mu is a Nesterov control parameter, the acceleration effect is controlled, and the Nesterov acceleration gradient method is only added for the second time in the training process, namely t is more than or equal to 2, and the iteration times are represented;
Figure FDA0004201521610000062
indicating updated +.at time t+1>
Figure FDA0004201521610000063
Intermediate state value of->
Figure FDA0004201521610000064
Representation c s The state at the time point of t,
Figure FDA0004201521610000065
representation c s A state at time t-1;
Figure FDA0004201521610000066
indicating updated +.at time t+1>
Figure FDA0004201521610000067
Intermediate state value of->
Figure FDA0004201521610000068
Represents x u,k State at time t->
Figure FDA0004201521610000069
Represents x u,k A state at time t-1;
Figure FDA00042015216100000610
indicating updated +.at time t+1>
Figure FDA00042015216100000611
Intermediate state value of->
Figure FDA00042015216100000612
Representing y s,k State at time t->
Figure FDA00042015216100000613
Representing y s,k A state at time t-1;
s4-3: according to a non-negative matrix hidden characteristic analysis method, the updated matrix hidden characteristic is used for updating
Figure FDA00042015216100000614
Figure FDA00042015216100000615
Training is carried out, wherein the training formula is as follows: />
Figure FDA00042015216100000616
In the formula (6) of the present invention,
Figure FDA00042015216100000617
indicating the use of intermediate states +.>
Figure FDA00042015216100000618
The state value at the time t+1 after updating; />
Figure FDA00042015216100000619
Indicating the use of intermediate states +.>
Figure FDA00042015216100000620
The state value at the time t+1 after updating; />
Figure FDA00042015216100000621
Indicating the use of intermediate states +.>
Figure FDA00042015216100000622
The state value at the time t+1 after updating; />
Figure FDA00042015216100000623
Indicating the use of intermediate states +.>
Figure FDA00042015216100000624
The state value at the time t+1 after updating; r is R k Representing a set of known behavior data of the user for the financial service in the user-service behavior statistics matrix T; r is% k (u) | represents the set of services in the user-service known behavior data related to user u; r is% k (s) | represents the set of users in the user-service known behavior data related to service s;
s4-4: judging whether the training reaches an iteration termination condition, if so, ending the iteration training, and if not, continuing the training and updating until the iteration termination condition is reached; s5: outputting a user hidden characteristic matrix and a service hidden characteristic matrix, predicting scores of unknown behavior data in a user-service behavior statistical matrix, and recommending services corresponding to the unknown behavior data of the front top k to the user.
6. The personalized financial service recommendation method based on big data according to claim 5, wherein S1 comprises:
the method comprises the steps that a user set is U, a service set is S, a matrix of I U I row and I S I column is established as a user-service behavior statistical matrix T, and then a rule matrix factorization is used for decomposing the T to respectively obtain a user hidden characteristic matrix X and a service hidden characteristic matrix Y, wherein X is a matrix of I U I row and F column, and each row vector in X corresponds to a user and is a hidden characteristic vector of the user; y is a matrix of |S| rows and f columns, and each row vector in Y corresponds to a service and is a hidden feature vector of the service; f is the dimension of the user implicit feature space and the service implicit feature space.
7. The personalized financial service recommendation method based on big data according to claim 5, wherein the initialization parameters comprise hidden feature matrices X and Y; hidden feature space dimension f; a Nesterov momentum control parameter mu, a regularization factor lambda; the maximum iteration round number R; the iteration round number in the training process controls variable r; the linear bias conceals feature vectors B and C.
8. The personalized financial service recommendation method according to claim 5, wherein in S3, the unified loss function Φ (X, Y, B, C) has the expression:
Figure FDA0004201521610000071
in the formula (7), t u,s Historical behavior data representing the entity relationship between the user u and the service s, namely the user u on the service s; Λ represents a set of known behavior data of the user for the financial service in the user-service behavior statistics matrix T;
Figure FDA0004201521610000081
representing a reduction value of an unknown behavior in the user-service behavior statistics matrix; b u A linear deviation (U) th row characteristic value of the user set is represented; c s Linear deviation representing service set Ss rows of characteristic values; x is x u,k A kth characteristic value of a ith row and a kth column of the user hidden characteristic matrix X is represented; y is s,k The kth row and the kth column eigenvalue of the service hidden eigenvalue matrix Y are represented; λ represents the canonical control parameter and f represents the dimension of the implicit feature space.
9. The personalized financial service recommendation method based on big data according to claim 5, wherein S5 comprises the steps of:
s5-1: acquiring hidden features corresponding to a user u, a service s, a user linear deviation b and a service linear deviation c from an output hidden feature unit;
s5-2: the inner product of the vector in the user hidden characteristic matrix X and the vector in the service hidden characteristic matrix Y corresponding to the service s and the user linear deviation b and the service linear deviation c is taken as the unknown behavior data value of the user u to the service s
Figure FDA0004201521610000082
I.e.
Figure FDA0004201521610000083
/>
CN202011040637.5A 2020-09-28 2020-09-28 Personalized financial service recommendation device and method based on big data Active CN112214668B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011040637.5A CN112214668B (en) 2020-09-28 2020-09-28 Personalized financial service recommendation device and method based on big data

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011040637.5A CN112214668B (en) 2020-09-28 2020-09-28 Personalized financial service recommendation device and method based on big data

Publications (2)

Publication Number Publication Date
CN112214668A CN112214668A (en) 2021-01-12
CN112214668B true CN112214668B (en) 2023-06-02

Family

ID=74051477

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011040637.5A Active CN112214668B (en) 2020-09-28 2020-09-28 Personalized financial service recommendation device and method based on big data

Country Status (1)

Country Link
CN (1) CN112214668B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115599598B (en) * 2022-10-08 2023-08-15 国网江苏省电力有限公司南通供电分公司 Power load sensing data recovery method and device

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110008404A (en) * 2019-03-22 2019-07-12 成都理工大学 Enigmatic language justice model optimization method based on the optimization of NAG momentum
CN111563203A (en) * 2020-05-08 2020-08-21 深圳市万佳安人工智能数据技术有限公司 Intelligent household user-service interest degree personalized prediction device and method based on rapid non-negative implicit characteristic analysis

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170228433A1 (en) * 2016-02-04 2017-08-10 Microsoft Technology Licensing, Llc Method and system for diverse set recommendations

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110008404A (en) * 2019-03-22 2019-07-12 成都理工大学 Enigmatic language justice model optimization method based on the optimization of NAG momentum
CN111563203A (en) * 2020-05-08 2020-08-21 深圳市万佳安人工智能数据技术有限公司 Intelligent household user-service interest degree personalized prediction device and method based on rapid non-negative implicit characteristic analysis

Also Published As

Publication number Publication date
CN112214668A (en) 2021-01-12

Similar Documents

Publication Publication Date Title
Cai et al. Proxylessnas: Direct neural architecture search on target task and hardware
US10304008B2 (en) Fast distributed nonnegative matrix factorization and completion for big data analytics
CN107729999A (en) Consider the deep neural network compression method of matrix correlation
US11669716B2 (en) System and method for implementing modular universal reparameterization for deep multi-task learning across diverse domains
CN113361685B (en) Knowledge tracking method and system based on learner knowledge state evolution expression
US11704604B2 (en) Optimization method, apparatus, computer device and storage medium for engine model
CN113705793B (en) Decision variable determination method and device, electronic equipment and medium
CN108009635A (en) A kind of depth convolutional calculation model for supporting incremental update
CN111311324B (en) User-commodity preference prediction system and method based on stable neural collaborative filtering
EP3525136A1 (en) Distributed machine learning system
CN110633417B (en) Web service recommendation method and system based on service quality
CN112214668B (en) Personalized financial service recommendation device and method based on big data
CN111563203A (en) Intelligent household user-service interest degree personalized prediction device and method based on rapid non-negative implicit characteristic analysis
CN114334013A (en) Single cell clustering method, device, equipment and readable storage medium
Gabrijel et al. On-line identification and reconstruction of finite automata with generalized recurrent neural networks
Chatterjee et al. SSFN--Self Size-estimating Feed-forward Network with Low Complexity, Limited Need for Human Intervention, and Consistent Behaviour across Trials
CN113836174B (en) Asynchronous SQL (structured query language) connection query optimization method based on reinforcement learning DQN (direct-to-inverse) algorithm
US20140006321A1 (en) Method for improving an autocorrector using auto-differentiation
Giffon et al. PSM-nets: Compressing neural networks with product of sparse matrices
CN112037850B (en) Momentum acceleration-based device and method for predicting interaction between missing proteins
CN112037849B (en) Device and method for predicting protein-protein interaction based on alternative direction multiplier method
CN113485107B (en) Reinforced learning robot control method and system based on consistency constraint modeling
CN113591383B (en) Digital twin modeling method and prediction system for multi-fidelity data
CN117439901A (en) Dynamic weighted directional network weight prediction method and device
Awoga et al. Using Deep Q-Networks to Train an Agent to Navigate the Unity ML-Agents Banana Environment

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