CN110232517B - Mobile crowd sensing user profit selection method - Google Patents

Mobile crowd sensing user profit selection method Download PDF

Info

Publication number
CN110232517B
CN110232517B CN201910493082.0A CN201910493082A CN110232517B CN 110232517 B CN110232517 B CN 110232517B CN 201910493082 A CN201910493082 A CN 201910493082A CN 110232517 B CN110232517 B CN 110232517B
Authority
CN
China
Prior art keywords
user
data
representing
data quality
publisher
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
CN201910493082.0A
Other languages
Chinese (zh)
Other versions
CN110232517A (en
Inventor
王慧强
邵子豪
冯光升
吕宏武
郭方方
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.)
Harbin Engineering University
Original Assignee
Harbin Engineering University
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 Harbin Engineering University filed Critical Harbin Engineering University
Priority to CN201910493082.0A priority Critical patent/CN110232517B/en
Publication of CN110232517A publication Critical patent/CN110232517A/en
Application granted granted Critical
Publication of CN110232517B publication Critical patent/CN110232517B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/11—Complex mathematical operations for solving equations, e.g. nonlinear equations, general mathematical optimization problems
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • G06Q10/06395—Quality analysis or management
    • 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
    • G06Q30/0207—Discounts or incentives, e.g. coupons or rebates
    • G06Q30/0226—Incentive systems for frequent usage, e.g. frequent flyer miles programs or point systems
    • 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)
  • Business, Economics & Management (AREA)
  • Physics & Mathematics (AREA)
  • Human Resources & Organizations (AREA)
  • General Physics & Mathematics (AREA)
  • Development Economics (AREA)
  • Theoretical Computer Science (AREA)
  • Strategic Management (AREA)
  • Mathematical Physics (AREA)
  • Economics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • General Business, Economics & Management (AREA)
  • Mathematical Analysis (AREA)
  • Data Mining & Analysis (AREA)
  • Operations Research (AREA)
  • Marketing (AREA)
  • Game Theory and Decision Science (AREA)
  • Accounting & Taxation (AREA)
  • Finance (AREA)
  • Computational Mathematics (AREA)
  • Pure & Applied Mathematics (AREA)
  • Mathematical Optimization (AREA)
  • Educational Administration (AREA)
  • Tourism & Hospitality (AREA)
  • Algebra (AREA)
  • Quality & Reliability (AREA)
  • Databases & Information Systems (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Complex Calculations (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention provides a mobile crowd sensing user profit selecting method, firstly modeling the choice of the profits of two parties as a publisher-user evolution game model; calculating the data quality deviation according to the related attribute of the uploaded data of the user, and removing the low-quality user data; and solving evolution strategy solutions of the screened users and analyzing stability to obtain the optimal user profit strategies under different conditions. Compared with the prior art, the invention screens out all reasonable users in the user data evaluation stage, has the advantages of high abnormal data identification accuracy and data classification accuracy compared with the performance of two common abnormal detection methods (namely BP method and SVM method), and simultaneously selects high-quality user data as far as possible under the condition of user data quality confirmation, thereby ensuring the maximum income of winning users and ensuring the overall utility of a perception platform.

Description

Mobile crowd sensing user profit selection method
Technical Field
The invention belongs to the field of mobile crowd sensing, and particularly relates to a method for selecting benefits of a mobile crowd sensing user.
Background
In recent years, with the development of intelligent terminal equipment, wireless networks and crowd sensing, mobile crowd sensing (Mobile Crowdsensing, MCS) has become a leading edge research problem of cross-space and large-scale data sensing, and it depends on a large number of mobile devices of ordinary users and various sensors of the mobile devices to complete a large number of sensing tasks. Although research on mobile crowd sensing has been developed in a long way, along with popularization of smart phones, diversification of user data and existence of a large number of malicious users, how to ensure rationalized benefits of users has become a problem to be solved.
Mobile crowd sensing is sensing with existing devices of users, and the perceived data quality is difficult to grasp. If the data quality of the user is not evaluated, the earnings of honest and high-quality users are affected, and the effectiveness of mobile crowd sensing is reduced. It falls into a vicious circle of "low quality user data-low task rewards-low perceived platform efficiency". Therefore, the evaluation of the data quality not only can ensure the benefit of the user, but also can enhance the effectiveness of intelligent perception of the mobile group.
Game theory is an important means of formalizing the user's revenue maximization when the user's data quality is ensured. At present, research on games in mobile crowd sensing is also mostly based on assumption of complete rationality of individuals, but such assumption is not generally consistent with practical situations. Because of the constantly changing topology of mobile crowd sensing, users' behavior is unlikely to be a completely rational behavior, but rather a limited rational behavior, based on the fact that people in real society often have limited rational constraints. Neglecting the limitation of finite rationality, modeling and analyzing the user behavior game can generate deviation from the actual situation, and the accuracy and the actual guiding value of the finally obtained optimal profit strategy selection method are reduced.
Disclosure of Invention
The invention aims to provide a mobile crowd sensing user income selecting method, which selects high-quality user data as far as possible under the condition of user data quality confirmation, thereby ensuring the maximum income of winning users and ensuring the overall utility of a sensing platform.
The purpose of the invention is realized in the following way:
a mobile crowd sensing user profit selection method comprises the following specific implementation steps:
step 1, constructing a publisher and user evolution game model, and defining the model as a quadruple;
step 2, calculating the data quality deviation according to the related attribute of the uploaded data of the user, removing low-quality data, and screening a user data quality strategy set meeting the requirements;
and 3, combining the reasonable user data quality strategy set obtained in the step 2 with the task excitation selectable strategy set of the publisher in the step 1, solving an evolution strategy solution and performing stability analysis to obtain user optimal benefit strategies under different conditions.
The quadruple of the step 1 is as follows:
N＝(N u ,N p ) Representing participants of an evolving game, where N u Representing the user, N p Representing a publisher;
w= (uw, pw) represents the choice of both gaming strategies, where uw= { uw 1 ,uw 2 ,...,uw n The policy set of the user, i.e. the set of user data quality, pw= { pw } represents 1 ,pw 2 ,...,pw m -represents a set of policies for the publisher, i.e. a set of task incentives;
r= (d, q) represents a set of game two-party policy selection probabilities, where d i Representing a publisher's selection of a publication task incentive policy pw i Probability, q of j Indicating user selection of data quality policy uw j Probability of (2);
B＝(B u ,B p ) Representing a set of game two-party revenue functions, where B u Representing the game benefits of the user, B p Representing the game revenue of the publisher.
The specific process of the step 2 is as follows:
step 2.1. Data quality policy set uw from each user i In the method, m data quality evaluation indexes are quantized to be C= { C 1 ,c 2 ,...,c m Dividing the evaluation index of the data quality into a cost index C cost And benefit index C profit WhereinC cost ∪C profit ＝C，/>Error value t of user data i,j Expressed as:
where i=1, 2, n, j=1, 2, m,representing the expected lower limit of the data quality evaluation index, < + >>Indicating the expected upper limit g of the data quality evaluation index i,j Representing user data u i In the data quality evaluation index c j The following values, error values are represented by a constant ρ, ρ=1+ε, ε is a positive infinity fraction, and each user data u is calculated i The sequence of error values under each data quality evaluation index is denoted as { t } i,1 ,t i,2 ,...,t i,m }；
Step 2.2. Search for user data u i Maximum error value, expressed as When->When the user data does not meet the condition, removing the user;
step 2.3. Calculating limit loss values for each data quality evaluation index expressed asIn calculating->In the process, the data quality evaluation index is normalized,
calculating limit loss value of each data quality evaluation index
Step 2.4, calculating a user data error loss sequence; introducing a weight omega i Can realize the change of the attention degree of different task demands and the error loss sequence of the user dataExpressed as:
where i=1, 2, n, j=1, 2, m,
step 2.5. Calculate the user data quality deviation, denoted R i The secondary screening of the user data is realized, and the calculation formula of the data quality deviation degree is as follows:
step 2.6. Calculating abnormal data identification accuracy and data classification accuracy, wherein the data identification accuracy is as followsAnd data classification accuracy->Sum in jad Representing the total number of real abnormal data judged as abnormal data, sum ad Representing the total number of actual abnormal data, sum jd Representing the total number of data judged to be normal, sum d Representing the total amount of data;
the specific process of the step 3 is as follows: in mobile crowd sensing, the publisher is known as p, the user is u, { pw l ,pw h Sum { uw } l ,uw h The method is characterized in that the method comprises the steps that two sides adopt simple strategies, the two sides select strategies in respective strategy sets with different probabilities and generate different benefits, and the benefits of the two sides are defined as that in a publisher and user evolution game model, the benefits of the two sides meet the following conditions: a, a hh ＞a lh ＞a ll ＞a hl And b hh ＞b hl ＞b ll ＞b lh ；
Based on the above conditions, expected benefits of users under cooperative and non-cooperative policies are calculated
Calculating average benefit of users
The number of people who choose these policies will vary over time, represented by q (t) and 1-q (t), for user policy uw l The proportion of the number of people selecting the strategy is selected according to a time function, and the replication dynamic equation is expressed as follows:
similarly, the expected benefits, average benefits and dynamic equation construction are performed for the publishers, expressed as:
combining the strategy selection and duplication dynamic equations of the two parties to construct a publisher-user game evolution equation set, and enablingFive sets of balance points were finally obtained:
and analyzing the established replication dynamic publisher-user evolution equation set by adopting a system dynamics method, performing stability analysis on all evolution balance points by using a jacobian matrix local stability method, and solving a stable equilibrium solution to obtain different optimal profit strategies of users under different conditions.
The invention has the beneficial effects that: compared with the prior art, the invention screens out all reasonable users in the user data evaluation stage, has the advantages of high abnormal data identification accuracy and data classification accuracy compared with the performance of two common abnormal detection methods (namely BP method and SVM method), and simultaneously selects high-quality user data as far as possible under the condition of user data quality confirmation, thereby ensuring the maximum income of winning users and ensuring the overall utility of a perception platform.
Drawings
FIG. 1 is a flow chart of the present invention.
FIG. 2 is a graph showing the comparison of the recognition accuracy of abnormal data in three methods according to the present invention at different abnormal scales.
FIG. 3 is a graph showing the comparison of data classification accuracy of three methods at different anomaly scales according to the present invention.
FIG. 4 is a strategy selection chart of the optimal benefits of the user under the condition that the user selects different strategies according to the embodiment of the invention.
Detailed Description
The invention is further described below with reference to the accompanying drawings:
example 1
The aim of the invention can be achieved by the following technical scheme:
a mobile crowd-sourced user revenue picking method based on data quality assessment and evolutionary game, the method comprising the steps of:
the first step, a publisher-user evolution game model is built and defined as a quadruple, and the method specifically comprises the following steps:
1)N＝(N u ,N p ) Representing participants of an evolving game, where N u Representing the user, N p Representing a publisher;
2) W= (uw, pw) represents the choice of both gaming strategies, where uw= { uw 1 ,uw 2 ,...,uw n The policy set of the user, i.e. the set of user data quality, pw= { pw } represents 1 ,pw 2 ,...,pw m -represents a set of policies for the publisher, i.e. a set of task incentives;
3) R= (d, q) represents a set of game two-party policy selection probabilities, where d i Representing a publisher's selection of a publication task incentive policy pw i Probability, q of j Indicating user selection of data quality policy uw j Probability of (2);
4)B＝(B u ,B p ) Representing a set of game two-party revenue functions, where B u Representing the game benefits of the user, B p Representing game benefits of the publisher;
the second step, calculate the data quality deviation degree according to the relevant attribute of the user uploading data, remove the low-quality data, screen the user data quality tactics set meeting the requirements, its concrete step is as follows:
1) From the data quality policy set uw for each user i In the quantized m data quality evaluation indexes, the quantized m data quality evaluation indexes are expressed as c= { C 1 ,c 2 ,...,c m The evaluation index of the data quality is divided into a cost type index and a benefit type index, wherein the cost type index is that the higher the data quality of a user is, the smaller the value of the evaluation index is, otherwise, the benefit type index is respectively used as C cost And C profit Representing, whereinC cost ∪C profit ＝C，/>Calculating an error value t of user data i,j ；
2) Searching the user data u according to the error values of different indexes of the user data obtained in the last step i Maximum error value, expressed asJudging rationality, and realizing one-time screening of user data, namely removing abnormal user data;
3) Calculating limit loss value of each data quality evaluation index based on the user data retained in the previous step, expressed as
4) Calculating a user data error loss sequence according to each data quality evaluation index limit loss value obtained in the last step, which is expressed as
5) According to the previous stepThe obtained error loss sequence of the user data, calculating the quality deviation degree of the user data, and representing the quality deviation degree as R i Secondary screening of user data is realized;
6) Calculating abnormal data identification accuracy and data classification accuracy, which are respectively expressed as X da And X dp Performance comparison is carried out with two common anomaly detection methods (namely a BP method and an SVM method), so that the method has the advantages of high anomaly data identification accuracy and data classification accuracy in the evaluation of user data quality;
thirdly, combining the reasonable user data quality strategy set obtained in the second step with the task excitation selectable strategy set of the publisher in the first step, solving an evolution strategy solution and performing stability analysis to obtain user optimal benefit strategies under different conditions, and guaranteeing maximum benefit of the winning bid user.
The example provides a mobile crowd sensing user benefit selection method based on data quality assessment and evolution game, wherein a flow chart of the method is shown in fig. 1, and the method comprises the following steps:
the first step, a publisher-user evolution game model is constructed and defined as a quadruple.
For example, 1) n= (N u ,N p ) Representing participants of an evolving game, where N u Representing the user, N p Representing a publisher;
2) W= (uw, pw) represents the choice of both gaming strategies, where uw= { uw 1 ,uw 2 ,...,uw n The policy set of the user, i.e. the set of user data quality, pw= { pw } represents 1 ,pw 2 ,...,pw m -represents a set of policies for the publisher, i.e. a set of task incentives;
3) R= (d, q) represents a set of game two-party policy selection probabilities, where d i Representing a publisher's selection of a publication task incentive policy pw i Probability, q of j Indicating user selection of data quality policy uw j Probability of (2);
4)B＝(B u ,B p ) Representing a set of game two-party revenue functions, where B u Representing the game benefits of the user, B p Representing game benefits of the publisher;
the second step, calculate the data quality deviation degree according to the relevant attribute of the user uploading data, remove the low-quality data, screen the user data quality tactics set meeting the requirements, its concrete step is as follows:
1) From the data quality policy set uw for each user i In the quantized m data quality evaluation indexes, the quantized m data quality evaluation indexes are expressed as c= { C 1 ,c 2 ,...,c m The evaluation index of the data quality is divided into a cost type index and a benefit type index, wherein the cost type index is that the higher the data quality of a user is, the smaller the value of the evaluation index is, otherwise, the benefit type index is respectively used as C cost And C profit Representing, whereinC cost ∪C profit ＝C，/>Error value t of user data i,j Expressed as:
where i=1, 2, n, j=1, 2, m,and g i,j Respectively represent the expected lower limit, the expected upper limit and the user data u of the data quality evaluation index i In the data quality evaluation index c j The following numerical values, equation (1) indicates that when the user data quality evaluation index is a cost index (c) j ∈C cost ) Error value calculation method of user data, at this time, attribute value g i,j Smaller than the meshSign->Or greater than target->When the cost index of the user does not meet the actual requirement, the error value is represented by a constant rho, so that the order of magnitude is the same, rho=1+epsilon, and epsilon is positive infinity decimal; when attribute value g i,j Greater than or equal to->And less than or equal to->When the attribute value is larger, the deviation is larger, and the value range is [0,1]Similarly, the expression (2) indicates that when the data quality evaluation index is a benefit type index (c j ∈C profit ) Error value calculation method of user data, each user data u can be obtained by the formula (1) and the formula (2) i The sequence of error values under each data quality evaluation index is denoted as { t } i,1 ,t i,2 ,...,t i,m }。
2) Searching the user data u according to the error values of different indexes of the user data obtained in the last step i Maximum error value, expressed as
The step is used for judging the rationality of the user and realizing one-time screening of the user data, namely removing the abnormal user data; when (when)And if the user data does not meet the condition, the user data is error data, and the user is removed.
3) Calculating limit loss value of each data quality evaluation index based on the user data retained in the previous step, expressed as
In the calculationWhen the method is used, the data quality evaluation index is normalized, namely:
then, the limit loss value of each data quality evaluation index is calculated and expressed as
4) Calculating a user data error loss sequence according to each data quality evaluation index limit loss value obtained in the previous step;
in mobile crowd sensing, the degree of interest in data is often different for different tasks, for example, in rescue vehicle fastest path search traffic monitoring, the required data is more focused on accuracy and effectiveness. And the integrity and accuracy of the data are more concerned in the environmental pollution detection. Thus introducing a weight ω i Changes in the degree of attention to different task demands can be achieved. Error loss sequence for user dataExpressed as:
where i=1, 2, n, j=1, 2, m,
5) Calculating the quality deviation degree of the user data according to the error loss sequence of the user data obtained in the last step, and representing the quality deviation degree as R i The secondary screening of the user data is realized, and the calculation formula of the data quality deviation degree is as follows:
6) Calculating abnormal data identification accuracy and data classification accuracy, which are respectively expressed as X da And X dp Performance comparison is carried out with two common anomaly detection methods (namely a BP method and an SVM method), so that the method has the advantages of high anomaly data identification accuracy and data classification accuracy in the evaluation of user data quality;
wherein, abnormal data identification accuracy X da And data classification accuracy X dp The expression of (2) is:
sum in jad Representing the total number of real abnormal data judged as abnormal data, sum ad Representing the total number of actual abnormal data, sum jd Representing the total number of data judged to be normal, sum d Representing the total amount of data. Fig. 2 and 3 show graphs of the accuracy of identifying abnormal data and the accuracy of classifying data of three methods at different abnormal scales in the quality evaluation of user data, respectively. As can be seen from fig. 2, as the specific gravity of the abnormal data increases, the accuracy of the BP method and the SVM method will increase,this is because both methods require a large number of training samples to achieve more accurate recognition. However, the accuracy of the method is better than that of the BP method and the SVM method, and the accuracy is always 1, because the method has the starting point of removing error abnormal data, and the method is very sensitive when the numerical value deviates, namely the method is not in a reasonable data range. This demonstrates that the method of the present invention has high accuracy in identifying anomalous data, and that the training sample requirement is small, only in relation to the limits of the quality of each type of data. As can be seen from fig. 3, along with the increase of the specific gravity of the abnormal data, the accuracy of the identification data of the method and the SVM method is basically maintained above 95%, which illustrates that the method also has the advantages of high accuracy and anti-interference in data classification.
Thirdly, combining the reasonable user data quality strategy set obtained in the second step with the task excitation selectable strategy set of the publisher in the first step, solving an evolution strategy solution and performing stability analysis to obtain user optimal benefit strategies under different conditions, and guaranteeing maximum benefit of the winning bid user.
In mobile crowd sensing, the publisher is known as p, the user is u, { pw l ,pw h Sum { uw } l ,uw h Respectively, are simple policies adopted by both parties. I.e., the publisher's incentive policy is low and high, and the user-provided data quality policy is uncooperative and cooperative. Both sides select the strategies in the respective strategy set with different probabilities, and different benefits can be generated. At the same time, the benefits of both parties can be maximized only if the publishers adopt a high incentive strategy and the users adopt cooperation. The benefits of both parties are defined as follows:
definition 1: in the publisher-user evolution game model, the strategy benefits of the two parties should meet the following conditions: a, a hh ＞a lh ＞a ll ＞a hl And b hh ＞b hl ＞b ll ＞b lh 。
Based on the above conditions, the expected benefits of the user using the collaborative and non-collaborative strategies are calculated as shown in equation (10). The average benefit of the user is shown in equation (11).
The number of people who choose these strategies will vary over time, represented by q (t) and 1-q (t). For user policy uw l The proportion of the number of people selecting the strategy is selected according to a time function, and the replication dynamic equation is expressed as follows:
similarly, the expected benefits, average benefits and dynamic equation construction are performed for the publishers, expressed as:
combining the strategy selection and duplication dynamic equations of the two parties to construct a publisher-user game evolution equation set, and enablingFive sets of balance points were finally obtained: />
Aiming at the established replication dynamic publisher-user evolution equation set, analyzing the system dynamic publisher-user evolution equation set by adopting a system dynamics method, performing stability analysis on all evolution balance points by using a jacobian matrix local stability method, and solving a stable equilibrium solution to obtain a final stable state F 1 ，F 4 And F 5 I.e. the user can get different optimal profit strategies under different conditions.
FIG. 4 shows the final stable equilibrium solution values in different initial states, wherein a value of 1 indicates that the gaming parties tend to stabilize the equilibrium solution F 5 I.e. the user selects the uncooperative strategy uw l The optimal income can be obtained, and when the value is 0, the two parties in the game tend to stabilize the equilibrium solution F 1 I.e. the user selects the collaboration policy uw h The optimal income can be obtained, and when the value is 0.5, the two parties of the game tend to stabilize the equilibrium solution F 4 I.e. the user chooses the strategy uw with the mixed probability 0.6,0.4 l ,uw h Optimal yields may be obtained. As can be seen from fig. 4, in the same publisher policy probability selection, the higher the user data quality, the more the stable equilibrium solution will tend to cooperate with the policy, and the greater the likelihood that the user will obtain the optimal benefit. Therefore, by adopting the invention, not only can the high-accuracy identification of the user data be realized, but also the overall benefit of the user can be improved.
Example 2
The invention is suitable for the mobile crowd sensing income selection method which fully considers the user data quality. The invention models the whole mobile crowd sensing process as a publisher-user evolution game model and evaluates the quality of user data in the model. The existing design of various mobile crowd sensing income selection methods ignores the data quality submitted by users, and the research on income games is mostly based on the complete rationality of individuals, so that the income of users cannot reach expectations. Based on the method, firstly, the selection of the benefits of the two parties is modeled as a publisher-user evolution game model; then, calculating the data quality deviation according to the related attribute of the uploaded data of the user, and removing the low-quality user data; and finally, solving evolution strategy solutions of the screened users and performing stability analysis to obtain the optimal user profit strategies under different conditions. By adopting the invention, not only can the high-accuracy identification of the user data be realized, but also the overall benefit of the user can be improved.
1. A mobile crowd sensing user gain selection method based on data quality assessment and evolution game comprises the following steps:
firstly, constructing a publisher-user evolution game model, and defining the model as a quadruple;
1)N＝(N u ,N p ) Representing participants of an evolving game, where N u Representing the user, N p Representing a publisher;
2) W= (uw, pw) represents the choice of both gaming strategies, where uw= { uw 1 ,uw 2 ,...,uw n The policy set of the user, i.e. the set of user data quality, pw= { pw } represents 1 ,pw 2 ,...,pw m -represents a set of policies for the publisher, i.e. a set of task incentives;
3) R= (d, q) represents a set of game two-party policy selection probabilities, where d i Representing a publisher's selection of a publication task incentive policy pw i Probability, q of j Indicating user selection of data quality policy uw j Probability of (2);
4)B＝(B u ,B p ) Representing a set of game two-party revenue functions, where B u Representing the game benefits of the user, B p Representing game benefits of the publisher;
calculating the data quality deviation according to the related attribute of the uploaded data of the user, removing low-quality data, and screening a user data quality strategy set meeting the requirements;
1) From the data quality policy set uw for each user i In the quantized m data quality evaluation indexes, the quantized m data quality evaluation indexes are expressed as c= { C 1 ,c 2 ,...,c m Dividing the evaluation index of the data quality into a cost type index and a benefit type index, and using C respectively cost And C profit Representing, whereinC cost ∪C profit ＝C，/>Error value t of user data i,j Expressed as:
where i=1, 2, n, j=1, 2, m,and g i,j Respectively represent the expected lower limit, the expected upper limit and the user data u of the data quality evaluation index i In the data quality evaluation index c j The values below are represented by the constant ρ, where ρ=1+ε, ε is a positive infinity fraction to ensure that the order of magnitude is the same, and each user data u can be obtained by equations (1) and (2) i The sequence of error values under each data quality evaluation index is denoted as { t } i,1 ,t i,2 ,...,t i,m }；
2) Searching the user data u according to the error values of different indexes of the user data obtained in the last step i Maximum error value, expressed as
When (when)The number of usersRemoving the user if the condition is not satisfied, namely the error data;
3) Calculating limit loss value of each data quality evaluation index based on the user data retained in the previous step, expressed as
In the calculationWhen the method is used, the data quality evaluation index is normalized, namely:
then, the limit loss value of each data quality evaluation index is calculated and expressed as
4) Calculating a user data error loss sequence according to each data quality evaluation index limit loss value obtained in the previous step;
introducing a weight omega i Can realize the change of the attention degree of different task demands and the error loss sequence of the user dataExpressed as:
where i=1, 2, n, j=1, 2, m,
5) Calculating the quality deviation degree of the user data according to the error loss sequence of the user data obtained in the last step, and representing the quality deviation degree as R i The secondary screening of the user data is realized, and the calculation formula of the data quality deviation degree is as follows:
6) Calculating abnormal data identification accuracy and data classification accuracy, which are respectively expressed as X da And X dp Wherein abnormal data identification accuracy X da And data classification accuracy X dp The expression of (2) is:
sum in jad Representing the total number of real abnormal data judged as abnormal data, sum ad Representing the total number of actual abnormal data, sum jd Representing the total number of data judged to be normal, sum d Representing the total amount of data;
thirdly, combining the reasonable user data quality strategy set obtained in the second step with the task excitation selectable strategy set of the publisher in the first step, solving an evolution strategy solution and performing stability analysis to obtain user optimal benefit strategies under different conditions, and guaranteeing maximum benefit of the winning bid user;
in mobile crowd sensing, the publisher is known as p, the user is u, { pw l ,pw h Sum { uw } l ,uw h The method is characterized in that the method comprises the steps that two sides adopt simple strategies, the two sides select the strategies in respective strategy sets with different probabilities, different benefits are generated, and benefits of the two sides are defined as follows:
definition 1: at the publisher-subscriberIn the evolution game model, the strategy yields of the two parties meet the following conditions: a, a hh ＞a lh ＞a ll ＞a hl And b hh ＞b hl ＞b ll ＞b lh ；
Based on the above conditions, the expected benefits of the user using the collaborative and non-collaborative strategies are calculated, as shown in equation (10), the average benefits of the user are shown in equation (11),
the number of people who choose these policies will vary over time, represented by q (t) and 1-q (t), for user policy uw l The proportion of the number of people selecting the strategy is selected according to a time function, and the replication dynamic equation is expressed as follows:
similarly, the expected benefits, average benefits and dynamic equation construction are performed for the publishers, expressed as:
combining the strategy selection and duplication dynamic equations of the two parties to construct the game evolution of the publisher and the userEquation set, letFive sets of balance points were finally obtained: />
And aiming at the established replication dynamic publisher-user evolution equation set, analyzing the system dynamic system by adopting a system dynamics method, performing stability analysis on all evolution balance points by using a jacobian matrix local stability method, and solving a stable equilibrium solution to obtain different optimal profit strategies of users under different conditions.

Claims (1)

1. A mobile crowd sensing user profit selection method is characterized by comprising the following specific implementation steps:
step 1: constructing a publisher and user evolution game model, and defining the model as a quadruple;
N＝(N u ,N p ) Representing participants of an evolving game, where N u Representing the user, N p Representing a publisher;
w= (uw, pw) represents the choice of both gaming strategies, where uw= { uw 1 ,uw 2 ,...,uw n The policy set of the user, i.e. the set of user data quality, pw= { pw } represents 1 ,pw 2 ,...,pw m -represents a set of policies for the publisher, i.e. a set of task incentives;
r= (d, q) represents a set of game two-party policy selection probabilities, where d i Representing a publisher's selection of a publication task incentive policy pw i Probability, q of j Indicating user selection of data quality policy uw j Probability of (2);
B＝(B u ,B p ) Representing a set of game two-party revenue functions, where B u Representing the game benefits of the user, B p Representing game benefits of the publisher;
step 2: calculating the data quality deviation according to the related attribute of the uploaded data of the user, removing low-quality data, and screening a user data quality strategy set meeting the requirements;
step 2.1: from the data quality policy set uw for each user i In the method, m data quality evaluation indexes are quantized to be C= { C 1 ,c 2 ,...,c m Dividing the evaluation index of the data quality into a cost index C cost And benefit index C profit WhereinC cost ∪C profit ＝C，/>Error value t of user data i,j Expressed as:
where i=1, 2, n, j=1, 2, m;and g i,j Respectively represent the expected lower limit, the expected upper limit and the user data u of the data quality evaluation index i In the data quality evaluation index c j The following values;
equation (1) shows that when the user data quality evaluation index is a cost index, i.e., c j ∈C cost Error value calculation method of user data, at this time, attribute value g i,j Less than the target valueOr is largeAt target value +.>When the cost index of the user does not meet the actual requirement, the error value is represented by a constant rho, so that the order of magnitude is the same, rho=1+epsilon, and epsilon is positive infinity decimal; when attribute value g i,j Greater than or equal to->And less than or equal to->When the attribute value is larger, the deviation is larger, and the value range is [0,1]；
Equation (2) shows that when the data quality evaluation index is a benefit type index, i.e., c j ∈C profit A method for calculating error value of user data; each user data u is obtained by the formula (1) and the formula (2) i The sequence of error values under each data quality evaluation index is denoted as { t } i,1 ,t i,2 ,...,t i,m }；
Step 2.2: searching for user data u i Maximum error value, expressed as When->When the user data does not meet the condition, removing the user;
step 2.3: calculating the limit loss value of each data quality evaluation index, expressed asIn calculating->When the data quality evaluation index is processed, normalization processing is carried out;
calculating limit loss value of each data quality evaluation index
Step 2.4: calculating a user data error loss sequence; introducing a weight omega i Can realize the change of the attention degree of different task demands and the error loss sequence of the user dataExpressed as:
where i=1, 2, n, j=1, 2, m,
step 2.5: calculating the quality deviation of user data, denoted as R i The secondary screening of the user data is realized, and the calculation formula of the data quality deviation degree is as follows:
step 2.6: calculating abnormal data identification accuracy and data divisionClass accuracy, data identification accuracyAnd data classification accuracy->Sum in jad Representing the total number of real abnormal data judged as abnormal data, sum ad Representing the total number of actual abnormal data, sum jd Representing the total number of data judged to be normal, sum d Representing the total amount of data;
step 3: combining the reasonable user data quality strategy set obtained in the step 2 with the task excitation selectable strategy set of the publisher in the step 1, solving an evolution strategy solution and performing stability analysis to obtain user optimal income strategies under different conditions;
in mobile crowd sensing, the publisher is known as p, the user is u, { pw l ,pw h Sum { uw } l ,uw h The method is characterized in that the method comprises the steps that two sides adopt simple strategies, the two sides select strategies in respective strategy sets with different probabilities and generate different benefits, and the benefits of the two sides are defined as that in a publisher and user evolution game model, the benefits of the two sides meet the following conditions: a, a hh ＞a lh ＞a ll ＞a hl And b hh ＞b hl ＞b ll ＞b lh ；
Based on the above conditions, expected benefits of users under cooperative and non-cooperative policies are calculated
Calculating average benefit of users
Selecting these policiesThe number of people in (a) will change with time, and is expressed by q (t) and 1-q (t), and is aimed at a user policy uw l The proportion of the number of people selecting the strategy is selected according to a time function, and the replication dynamic equation is expressed as follows:
similarly, the expected benefits, average benefits and dynamic equation construction are performed for the publishers, expressed as:
combining the strategy selection and duplication dynamic equations of the two parties to construct a publisher-user game evolution equation set, and enablingFive sets of balance points were finally obtained:
and analyzing the established replication dynamic publisher-user evolution equation set by adopting a system dynamics method, performing stability analysis on all evolution balance points by using a jacobian matrix local stability method, and solving a stable equilibrium solution to obtain different optimal profit strategies of users under different conditions.
CN201910493082.0A 2019-06-06 2019-06-06 Mobile crowd sensing user profit selection method Active CN110232517B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910493082.0A CN110232517B (en) 2019-06-06 2019-06-06 Mobile crowd sensing user profit selection method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910493082.0A CN110232517B (en) 2019-06-06 2019-06-06 Mobile crowd sensing user profit selection method

Publications (2)

Publication Number Publication Date
CN110232517A CN110232517A (en) 2019-09-13
CN110232517B true CN110232517B (en) 2023-08-01

Family

ID=67859324

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910493082.0A Active CN110232517B (en) 2019-06-06 2019-06-06 Mobile crowd sensing user profit selection method

Country Status (1)

Country Link
CN (1) CN110232517B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112396102B (en) * 2020-11-17 2022-08-02 哈尔滨工程大学 Mobile crowd sensing user alliance clustering method based on space-time mobile feature distribution

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108337656A (en) * 2018-01-16 2018-07-27 武汉工程大学 A kind of mobile intelligent perception motivational techniques
CN109544261A (en) * 2018-09-29 2019-03-29 南京邮电大学 A kind of intelligent perception motivational techniques based on diffusion and the quality of data

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP4386973B2 (en) * 1996-10-23 2009-12-16 株式会社野村総合研究所 Hierarchical prediction model construction apparatus and method
US9767137B2 (en) * 2010-04-26 2017-09-19 Apple Inc. Method and system for distributed data verification
US20120284090A1 (en) * 2011-05-02 2012-11-08 Sergejs Marins System and method for accumulation and verification of trust for participating users in a crowd sourcing activity
CN108668253A (en) * 2018-04-09 2018-10-16 南京邮电大学 A kind of gunz cooperative sensing motivational techniques based on evolutionary Game
CN109492887B (en) * 2018-10-25 2021-10-26 浙江工商大学 Mobile crowd sensing and exciting method based on potential game theory

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108337656A (en) * 2018-01-16 2018-07-27 武汉工程大学 A kind of mobile intelligent perception motivational techniques
CN109544261A (en) * 2018-09-29 2019-03-29 南京邮电大学 A kind of intelligent perception motivational techniques based on diffusion and the quality of data

Also Published As

Publication number Publication date
CN110232517A (en) 2019-09-13

Similar Documents

Publication Publication Date Title
CN106097043B (en) The processing method and server of a kind of credit data
Lee et al. Can Twitter data be used to validate travel demand models
WO2018157808A1 (en) Method and system for evaluating internet credit
CN111325619A (en) Credit card fraud detection model updating method and device based on joint learning
CN114418109A (en) Node selection and aggregation optimization system and method for federal learning under micro-service architecture
CN103577876A (en) Credible and incredible user recognizing method based on feedforward neural network
Xiao et al. Incentive mechanism design for federated learning: A two-stage stackelberg game approach
CN110796485A (en) Method and device for improving prediction precision of prediction model
CN114372803A (en) Quick anti-money laundering detection method based on transaction map
Wang et al. TVD-RA: A truthful data value discovery based reverse auction incentive system for mobile crowd sensing
CN110163378A (en) Characteristic processing method, apparatus, computer readable storage medium and computer equipment
CN110232517B (en) Mobile crowd sensing user profit selection method
CN110310199B (en) Method and system for constructing loan risk prediction model and loan risk prediction method
CN116582568A (en) Method for exciting federal learning participation equipment for time-efficiency perception in industrial Internet of things
CN115309985A (en) Fairness evaluation method and AI model selection method of recommendation algorithm
Kang et al. DTD: An intelligent data and bid dual truth discovery scheme for MCS in IIoT
CN117592819B (en) Dynamic supervision method and device for land-sea complex oriented to ocean pasture
Xiao et al. AoI-aware incentive mechanism for mobile crowdsensing using stackelberg game
CN116091106A (en) Multifunctional financial cost evaluation system
CN116627781A (en) Target model verification method and device
CN115860417A (en) Agricultural water irrigation distribution method and system and terminal equipment
CN109873836A (en) A kind of methods of risk assessment and device of data
CN114707664A (en) Case scale prediction method and system based on graph evolution game model
CN103942278A (en) Method for conducting friend recommendation through analysis of user active friends making will
CN114756891A (en) Data transaction method for individualized privacy protection requirements of terminal equipment

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