CN113011703B - Topic-sensitive crowdsourcing task diffusion method - Google Patents
Topic-sensitive crowdsourcing task diffusion method Download PDFInfo
- Publication number
- CN113011703B CN113011703B CN202110180146.9A CN202110180146A CN113011703B CN 113011703 B CN113011703 B CN 113011703B CN 202110180146 A CN202110180146 A CN 202110180146A CN 113011703 B CN113011703 B CN 113011703B
- Authority
- CN
- China
- Prior art keywords
- user
- task
- influence
- probability
- winner
- 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
Links
- 238000009792 diffusion process Methods 0.000 title claims abstract description 52
- 238000000034 method Methods 0.000 claims abstract description 20
- 230000007246 mechanism Effects 0.000 claims abstract description 14
- 239000000126 substance Substances 0.000 claims description 9
- 238000007476 Maximum Likelihood Methods 0.000 claims description 5
- 230000004913 activation Effects 0.000 claims description 5
- 230000007480 spreading Effects 0.000 claims description 5
- 230000003213 activating effect Effects 0.000 claims description 4
- 230000000694 effects Effects 0.000 claims description 3
- 230000007704 transition Effects 0.000 claims description 3
- 238000010586 diagram Methods 0.000 description 1
- 230000006855 networking Effects 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000012358 sourcing Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- 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
- G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
- G06Q10/06311—Scheduling, planning or task assignment for a person or group
-
- 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
- G06Q30/0227—Frequent usage incentive value reconciliation between diverse systems
-
- 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
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/01—Social networking
Landscapes
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- Development Economics (AREA)
- Entrepreneurship & Innovation (AREA)
- Theoretical Computer Science (AREA)
- Marketing (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- Game Theory and Decision Science (AREA)
- Finance (AREA)
- Accounting & Taxation (AREA)
- Tourism & Hospitality (AREA)
- Educational Administration (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a topic-sensitive crowdsourcing task diffusion method which is enhanced on the basis of a classical independent cascade model of influence, so that the influence is closely related to a task topic, the topic-sensitive independent cascade model is provided, the problem of feasible task diffusion of budget is formalized, and a platform utility function is maximized under the budget constraint. The invention provides a parameter estimation method for estimating influence parameters in a subject sensitive independent cascade model. By utilizing the sub-model characteristics of the objective function, a budget feasible incentive mechanism is provided, and the mechanism meets the rational characteristics of computational efficiency, individual rationality, budget feasibility and authenticity so as to encourage task diffusion, improve the participation degree of participants in a large-scale crowdsourcing system and have feasible application value.
Description
Technical Field
The invention relates to crowdsourcing task diffusion, in particular to a topic-sensitive crowdsourcing task diffusion method.
Background
Mobile devices such as cell phones are widely popular, and people can perceive data of surrounding environment and perform diffusion sharing through social networks through sensors embedded in the mobile devices. And the incentive mechanism is crucial to crowdsourcing, privacy disclosure of intelligent device users can be effectively avoided, more people are promoted to crowd sourcing, and high-quality data is encouraged to be provided.
Many online communities have self-developed crowdsourcing systems such as steps created by Facebook, Google Image label and transform Community created by Google +, Q-Crowd and Crowdtesting created by QQ, and hundred-degree encyclopedia created by hundred.
The problem that crowdsourcing cannot overlook is participant shortfall. According to statistical data, amazon's MTurk platform distributes 618 intelligent tasks on average per day, but 92 tasks per day expire without human participation for a long time, more than 180 tasks are published for more than 2 weeks, with 61% of the tasks even exceeding one month.
Disclosure of Invention
The purpose of the invention is as follows: the invention aims to provide a diffusion method of a theme-sensitive crowdsourcing task, which improves the participation degree of the crowdsourcing task.
The technical scheme is as follows: the method comprises the steps that a platform operated by an online community is set to release crowdsourcing tasks, a crowdsourcing task diffusion system is modeled into a reverse auction, firstly, each registered user submits a bidding document to the platform, and the bidding document comprises a task set which the registered user is willing to diffuse and corresponding quotations; secondly, the platform selects a winner and determines payment according to the theme of the estimation crowdsourcing task and the influence of the registered user; the winner then distributes the task to other users in the social network; eventually, each winner will get paid well by the platform and the affected social users perform the crowdsourcing task. The present invention designs a realistic incentive mechanism to maximize the value of the winner task spread under budget constraints.
A topic-sensitive crowdsourcing task diffusion method comprises the following specific steps:
step (1.1), the crowdsourcing platform obtains a registered user set U, wherein the number of registered users is n, and a task set T (T) issued by the crowdsourcing platform is T 1 ，t 2 ，…，t m A social network graph G ═ V, E) and a budget B for diffusion tasks, where V represents all users in the social network, E is an edge between all users in the social network, and if there is an edge between a registered user V and a user G, the user G is said to be a social neighbor of the registered user V; each registered user v belongs to U and submits bidding document theta v ＝(T v ，b v ) Wherein
Is a set of tasks that registered user v is willing to spread, b v Is an offer for the registered user v,c v is the true cost of registered user v, c v The information is private information and is known by registered users v; defining the utility of a registered user v as f v This can be obtained from equation (1):
wherein mu v The reward is the reward of the registered user v, and S is a winner set;
step (1.2), defining a budget feasible task diffusion problem;
step (1.3), establishing a theme-sensitive independent cascade influence diffusion model according to the characteristics of a traditional independent cascade influence diffusion model and a crowdsourcing system;
step (1.4), obtaining a task theme and an influence parameter of a theme-sensitive independent cascade influence diffusion model;
and (1.5) acquiring a winner set from all registered users by adopting a budget feasible mechanism, and calculating winner rewards.
Further, the budget feasible task diffusion problem described in step (1.2) is as follows:
step (1.2.1), build constraint sigma v∈S μ v ≤B，μ v The reward of each registered user is represented, and the reward of all registered users is ensured not to exceed budget B;
step (1.2.2), the problem of budget feasible task diffusion is as follows:
max f(S) (2)
s.t.∑ v∈S μ v ≤B (3)
where f (S) represents the platform utility function when crowdsourcing the platform winner set S, the objective of the incentive mechanism is to maximize the platform utility function f (S) under the condition that the rewards of all registered users do not exceed the budget B constraint.
Further, the step (1.3) of obtaining a topic-sensitive independent cascade influence diffusion model according to the traditional independent cascade influence diffusion model and the crowdsourcing system characteristics comprises the following steps:
and (1).3.1), in the independent cascade influence diffusion model, for a given social network, a part of nodes in the social network are in an activated state in an initial stage; at any time period tau epsilon N + Every node C in active state tries to get p C，D E (0, 1) activates a social neighbor node D thereof; if the attempt is successful, D is changed from the inactive state to the active state in the period of tau + 1; each node C in the active state has only one opportunity to activate its social neighbor node, which is called the active node once node D is activated; assuming that the influence of the node C in the activated state on the social neighbor node D is independent from the historical data, so as to obtain the activity state of the social neighbor node D, and then calculating the influence of the node C in the activated state; in the crowdsourcing system, each user is mapped to one node in the network, the influence among the users is closely related to the spread tasks and the subjects of the tasks, the active users in the initial stage are winners among the registered users participating in the reverse auction, and each registered user v is related to the task t j Probability of activating social neighbor g is p v，g (j)∈(0，1)；p v，g (j) Indicating that registered user v will task t j The influence of spreading to social neighbors g;
step (1.3.2), registered user v successfully activates social neighbor g about task t j The probability of (c) can be calculated by equation (4):
wherein, it is provided with
Is task t j With a subject of k, i.e.
And satisfy
I.e. for task t j The sum of the probabilities of all subjects is 1, where z j Is task t j Z is the set of topics for all tasks; probability of
Is the influence of the registered user v on his neighbour g about the topic k;
step (1.3.3), obtaining a platform utility function sensitive to the subject:
where Path (v, u, j) is for task t j The path with the largest multiplicative activation probability from a registered user v to any non-registered user u,<v ', u' refers to the user-mapped node pair for each edge in the path.
Further, in the step (1.4), task topic distribution and influence parameters of the topic-sensitive independent cascade influence diffusion model are obtained, and the steps are as follows:
step (1.4.1), obtaining likelihood probability of historical data and estimation value of the historical data according to an expectation maximization algorithm;
step (1.4.2), obtaining the theme distribution of the randomly generated task and the influence of each edge, and finishing initialization;
step (1.4.3), an expectation maximization algorithm is expanded by combining a crowdsourcing system, namely, the theme distribution of the task is obtained through maximum likelihood estimation according to the newly-calculated influence value of each edge in the social graph;
step (1.4.4), re-estimating the influence of each edge in the social graph according to the newly calculated theme distribution of the tasks;
and (5) step (1.4.5), and step (1.4.3) and step (1.4.4) are executed in an iterative mode until convergence.
In a further step (1.4.1), the likelihood probability of the historical data and the estimated value of the historical data are obtained according to the expectation maximization method, and the method comprises the following steps:
step (1.4.1.1) according to the historical task data x 1 ，x 2 ，…，x m Which isIn x j ＝{x j (τ)|τ∈N * }，x j (τ) is the time period τ with respect to task t j The active set of users. Assume arbitrary x j Independent of each other, obtaining likelihood probability of historical data, i.e. x j The joint distribution of (A) is:
wherein, p (x) j (ii) a λ) is that the sample is x when the parameter is λ j Probability of time;
and step (1.4.1.2), obtaining the Log likelihood probability according to the formula (6) as follows:
step (1.4.1.3) of obtaining task t j Subject z of j The prior probability distribution of (a) is:
wherein the content of the first and second substances,
is the value of the parameter lambda after a iterations;
step (1.4.1.4), obtaining
The likelihood function of (d) is:
step (1.4.1.5) to obtain the new
Estimated value of (a):
further, in the step (1.4.3), in combination with a crowdsourcing system, the expectation maximization algorithm is extended, and according to the newly calculated influence value of each edge in the social graph, the topic distribution of the task is obtained through maximum likelihood estimation, and the method comprises the following steps:
step (1.4.3.1), based on historical data processing and general knowledge, assume the topic of k task diffusion set x j The possibilities of (a) are:
wherein the content of the first and second substances,
is a set of neighbors with potential impact on any unregistered user u, expressed as:
wherein, the first and the second end of the pipe are connected with each other,
is the time period during which any non-registered user u transitions to the active state,
a time phase of a social neighbor h of a non-registered user u transitioning to an active state;
step (1.4.3.2), diffusion task set x is obtained through conditional probability j And the joint probability distribution of topic k is:
p(x j ，k；λ)＝p(x j |k；λ)p(k|λ) (14)
step (1.4.3.3), set pi k P (k | λ) is the prior probability of k being the subject of any general task;
step (1.4.3.4), obtaining prior probability distribution with the task subject being k:
step (1.4.3.5), obtaining the probability of the topic k of any general task as:
further, the step (1.4.4) of estimating the influence of each edge in the social graph according to the newly calculated topic distribution of the task comprises the following steps:
step (1.4.4.1), obtaining likelihood function according to step (1.4.2):
step (1.4.4.2), because of the likelihood function pairs in equation (17)
Is monotonous and can not maximize the likelihood function, when the subject is redesigned to be k, the task diffusion set x j The possibilities of (a) are:
wherein the content of the first and second substances,
is that the user h successfully activates its neighbor u, i.e. the task t with spreading topic k j The probability of success, expressed as:
and step (1.4.4.3), obtaining a corresponding likelihood function as:
and (1.4.4.4) obtaining an influence estimation value of each edge:
further, the step (1.5) comprises the steps of:
step (1.5.1), initialize winner set S ═ phi and pay vector μ ═ μ (μ) 1 ，μ 2 ，...，μ n ) 0, temporary set S * ＝{v|b v B ≦ B }, i.e. set S * Is the set of all registered users v whose quoted price is less than budget B;
step (1.5.2), selecting S with probability of 2/5 * User v with the greatest value f ({ v }) in * As winner, user v * The reward obtained is equal to the budget;
step (1.5.3), carrying out winner selection and payment according to crowdsourcing rules with the probability of 3/5;
further, the step (1.5.3) of winner selection and reward payment according to crowdsourcing rules with a probability of 3/5 comprises the following steps:
step (1.5.3.1) obtaining a marginal value f according to the winner set S v (S), expressed as:
f v (S)＝f(S∪{v})-f(S) (22)
step (1.5.3.2) finding S * User v with maximum medium marginal density max I.e. by
Step (1.5.3.3) is a step of judging user v max Whether the auction base price exceeds
If so, performing step 536, otherwise, performing step 1.5.3.4;
step (1.5.3.4) presents user v max Adding into the winner set S;
step (1.5.3.5) is from set S * V 'of user with maximum marginal density selected from S' max I.e. by
Performing step (1.5.3.3);
step (1.5.3.6) of obtaining sets S for users i in each set S of winners in a remuneration phase *′ ＝S * \ { i }, initializing a new set of winners S ', let S' ═ phi;
step (1.5.3.7) at S * ' find the user with the greatest marginal density i
Step (1.5.3.8) is a step of judging whether or not the auction reserve price of user i' exceeds
If yes, executing step (542), otherwise, continuing to execute step (1.5.3.9);
step (1.5.3.9) is from set S * 'S' selection of the user with the highest marginal density, i.e.
Step (1.5.4.0) calculates eachReward mu for winner i i ：
Wherein, S' i′-1 Is the set of winners before adding with corpse i 'to S';
a step (1.5.4.1) of adding user i 'to the winner set S';
step 1.5.4.2 determines whether all users in the winner set S have performed step 1.5.3.6, if yes, ends, otherwise, performs step 1.5.3.6.
Has the advantages that: compared with the prior art, the invention has the following remarkable advantages:
(1) the budget feasible method of the present invention is computationally efficient, i.e., the set of winners and payment scheme can be computed in polynomial time.
The budget feasible method runtime of the present invention is primarily determined by the winner's choice and payment. The winner-selected run is determined primarily by the computational cost function f(s). First of all,(s) is computationally efficient, given a social network graph G ═ (V, E), the logarithm logp is taken as the weight of the edge between any two nodes u', V ═ V v′，u′ (j) Due to p v′，u′ Has a value of (0, 1), logp v′，u′ Is negative. Then-logp is taken v′，u′ As new weights. Finding a task t from node v to node u j The problem with the maximum multiplicative probability is equivalent to the classical problem of finding the shortest path with weights, which can be solved by Dijkstra's algorithm with a running time of O (| E | log | V |). Thus, for all node pairs of the network map, for all tasks t j The time complexity of the computation of the cost function f (S) is O (n) 2 m | E | log | V |), and thus f(s) is computationally efficient. The time complexity of calculating the marginal density is O (n) 3 m | E | log | V |) so that the entire winner's chosen runtime is O (n) 4 m | E | log | V |). The time of operation for paying is O (n) 5 m | E | log | V |) so that the budget feasible run time is O (n) 5 m|E|log|V|)And is therefore computationally efficient.
(2) The budget feasible method of the present invention is individual rational, i.e., each registered user is not negatively effective in bidding at cost.
V is to be k In place of registered user v, at S * \ { v } order is at the v-th position. Since user v is a candidate, if
Then register user v k Will not be in the v-th position, therefore
Thereby to obtain
The equation holds only if all users k ≧ v satisfy S ═ S'. This ensures
I.e. the reward is greater than the true cost.
(3) The budget feasible approach of the present invention is budget feasible, i.e. the total payment to the registered user is less than the budget of the platform.
Since the feasible solution of the budget feasible method must satisfy the constraint condition, i.e., formula (3), the sum of the rewards of all the winners is less than the total budget of the platform.
(4) The budget feasible approach of the present invention is real, i.e., no user is able to improve its utility by submitting false offers.
Firstly, the Melson theorem is led out: the auction mechanism is only true if and only if:
the selection rule is monotonic: if user v passes bid b v Winning the auction, then bid b v ′＜b v The auction will still be won;
② each winner will get a critical reward: as long as the user v bid is above the value, the auction fails.
According to the Melson theorem, it is sufficient to prove that the winner-selection rule isAdjusted and a reward p per user v v Are all critical values. The monotonicity of the selection rules is evident because users with lower bids are not inserted in the ranking before registered user v. Next, p is demonstrated v Is a critical value for each registered user v. Higher p bid v Causing registered user v to fall out of the auction. It is noted therein
If registered user v bids b v ≥p v Due to the fact that
Means that
Then the v rank of the registered user will be ranked at v k Thereafter, the registered user v does not win the auction.
Drawings
FIG. 1 is a schematic diagram of a crowdsourcing architecture with task diffusion according to the present invention;
FIG. 2 is a flow chart of the present invention for obtaining task topics and influence parameters in a topic-sensitive independent cascade influence diffusion model;
FIG. 3 is a flow chart of the present invention for obtaining a set of winners from all registered users using a budget feasible mechanism;
FIG. 4 is a flow chart of the present invention for calculating winner rewards using a budget feasible mechanism;
Detailed Description
The technical scheme of the invention is further explained by combining the attached drawings.
A topic-sensitive crowdsourcing task diffusion method works on a crowdsourcing structure with task diffusion shown in figure 1, and specifically comprises the following steps:
(1) the crowdsourcing platform obtains a registered user set U, wherein the number of registered users is n, and a task set T ═ T published by the crowdsourcing platform 1 ，t 2 ，…，t m H, social networking graph G ═ V, E, and budget for diffusion tasks B, where V is tableDisplaying all users in the social network, wherein E is an edge between all users in the social network, and if an edge exists between a registered user v and a user g, the user g is called as a social neighbor of the registered user v; each registered user v belongs to U and submits bidding document theta v ＝(T v ，b v ) Wherein
Is a set of tasks that registered user v is willing to spread, b v Is a quote of registered user v, c v Is the true cost of registered user v, c v The information is private information and is known by registered users v; defining the utility of a registered user v as f v This can be obtained from equation (1):
wherein mu v The reward is the reward of the registered user v, and S is a winner set;
(2) defining a budget feasible task diffusion problem;
(3) establishing a theme-sensitive independent cascade influence diffusion model according to the independent cascade influence diffusion model and the crowdsourcing system characteristics;
(4) obtaining a task theme and an influence parameter of an independent cascade influence diffusion model sensitive to the theme;
(5) and acquiring a winner set from all registered users by adopting a budget feasible mechanism, and calculating winner rewards.
The step (2) comprises the following steps:
(21) build constraint sigma v∈S μ v ≤B，μ v The reward of each registered user is represented, and the reward of all registered users is ensured not to exceed budget B;
(22) the problem of budget feasible task diffusion is as follows:
max f(S) (2)
s.t.∑ v∈S μ v ≤B (3)
where f (S) represents the platform utility function when crowdsourcing the platform winner set S, the objective of the incentive mechanism is to maximize the platform utility function f (S) under the condition that the rewards of all registered users do not exceed the budget B constraint.
The step (3) comprises the following steps:
(31) in the independent cascade influence diffusion model, for a given social network, a part of nodes in the social network are in an activated state in an initial stage; at any time interval tau epsilon N + Every node C in active state tries to get p C，D E (0, 1) activates a social neighbor node D thereof; if the attempt is successful, D is changed from the inactive state to the active state in the period of tau + 1; each node C in the active state has only one opportunity to activate its social neighbor node, and once node D is activated, it is called the active node; the influence of the node C in the activated state on the social neighbor node D is assumed to be independent from historical data, so that the activity state of the social neighbor node D is obtained, and then the influence of the node C in the activated state is calculated; in the crowdsourcing system, each user is mapped to one node in the network, the influence among the users is closely related to the spread tasks and the subjects of the tasks, the active users in the initial stage are winners among the registered users participating in the reverse auction, and each registered user v is related to the task t j The probability of activating a social neighbor g is p v，g (j)∈(0，1)；p v，g (j) Indicating that registered user v will task t j The influence of spreading to social neighbors g;
(32) successful activation of social neighbors g by registered users v with respect to tasks t j Is calculated by equation (4):
wherein, it is provided with
Is task t j With a subject of k, i.e.
And satisfy
I.e. for task t j The sum of the probabilities of all subjects is 1, where z j Is task t j Z is the set of topics for all tasks; probability of occurrence
Is the influence of the registered user v on its neighbors u with respect to the topic k;
(33) obtaining a subject-sensitive platform utility function:
where Path (v, u, j) is for task t j The path with the largest multiplicative activation probability from a registered user v to any non-registered user u,<v′，u′>user-mapped node pairs for each edge in a path
In the step (4), the task topic distribution and the influence parameters of the independent cascade influence diffusion model with topic sensitivity are obtained, and as shown in fig. 2, the method comprises the following steps:
(41) obtaining likelihood probability of historical data and an estimated value of the historical data;
(42) obtaining the theme distribution of the randomly generated task and the influence of each edge, and finishing initialization;
(43) combining a crowdsourcing system, expanding an expectation maximization algorithm, namely obtaining the theme distribution of the task through maximum likelihood estimation according to the newly-calculated influence value of each edge in the social graph;
(44) re-estimating the influence of each edge in the social graph according to the newly calculated theme distribution of the tasks;
(44) and (6) iteratively executing the steps (43) and (44) until convergence.
The likelihood probability of the historical data and the estimated value of the historical data are obtained in the step (41), and the method comprises the following steps:
(411) based on historical task data x 1 ，x 2 ，…，x m Wherein x is j ＝{x j (τ)|τ∈N * }，x j (τ) is the time period τ with respect to task t j The active set of users. Assume arbitrary x j Independent of each other, obtaining likelihood probability of historical data, i.e. x j The joint distribution of (A) is:
wherein, p (x) j (ii) a λ) is that the sample is x when the parameter is λ j Probability of time;
(412) the Log likelihood probability obtained according to equation (6) is:
(413) obtaining a task t j Subject z of j The prior probability distribution of (a) is:
wherein the content of the first and second substances,
is the value of the parameter lambda after a iterations;
(414) to obtain
The likelihood function of (d) is:
(415) obtain new
Estimated value of (a):
step (43) of expanding the expectation-maximization algorithm by combining with a crowdsourcing system, and obtaining the theme distribution of the task through maximum likelihood estimation according to the newly-calculated influence value of each edge in the social graph, wherein the method comprises the following steps:
(431) according to the processing of historical data and general knowledge, assume that the topic is k task diffusion set x j Is that
Wherein the content of the first and second substances,
is a set of neighbors with potential impact on any unregistered user u, expressed as:
wherein the content of the first and second substances,
is the time period during which any non-registered user u transitions to the active state,
a time phase of a social neighbor h of a non-registered user u transitioning to an active state;
(432) obtaining a diffusion task set x through conditional probability j And the joint probability distribution of topic k is:
p(x j ，k；λ)＝p(x j |k；λ)p(k|λ) (14)
(433) set pi k P (k | λ) is the prior probability of k being the subject of any general task;
(434) the prior probability distribution of the task with the topic of k is obtained as follows:
(435) the probability of obtaining a topic k for any general task is:
in step (44), the influence of each edge in the social graph is estimated according to the newly calculated theme distribution of the tasks, and the method comprises the following steps:
(441) obtaining a likelihood function according to step (42):
(442) since the pair of likelihood functions in equation (17)
Is monotonous and can not maximize the likelihood function, when the subject is redesigned to be k, the task diffusion set x j The possibilities of (a) are:
wherein the content of the first and second substances,
is the user h becomesSuccessfully activating its neighbors u, i.e. task t with diffusion topic k j The probability of success, expressed as:
(443) the corresponding likelihood function is obtained as:
(444) obtaining an estimate of the influence of each edge:
the step (5) of obtaining a winner set from all registered users by adopting a budget feasible mechanism and calculating winner remuneration comprises the following steps:
(51) initializing winner set S ═ phi and payout vector μ ═ μ 1 ，μ 2 ，…，μ n ) Temporary set S as 0 * ＝{v|b v B ≦ B }, i.e. set S * Is the set of all registered users v whose quotes are less than budget B;
(52) selecting S with a probability of 2/5 * User v with the greatest value f ({ v }) in * As winner, user v * The reward obtained is equal to the budget;
(53) winner selection and remuneration according to crowdsourcing rules with a probability of 3/5;
in step (53), the winner selection and the payment are performed according to the crowdsourcing rule with the probability of 3/5, as shown in fig. 3 and 4, the method comprises the following steps:
(531) obtaining a margin value f according to the winner set S v (S), expressed as:
f v (S)＝f(S∪{v})-f(S) (22)
(532) find S * User v with maximum medium marginal density max I.e. by
(533) Judging user v max Whether the auction base price exceeds
If yes, go to step 536, otherwise go to step 534;
(534) user v max Adding into the winner set S;
(535) from the set S * \ S selecting user v 'with maximum marginal density' max I.e. by
Performing step (533);
(536) in the compensation stage, a set S is obtained for each user i in the winner set S *′ ＝S * \ { i }, initializing a new set of winners S ', let S' ═ phi;
(537) at S * ' find the user with the greatest marginal density i
(538) Determining whether the auction reserve price of user i' exceeds
If yes, executing step (542), otherwise, continuing to execute step (539);
(539) from the set S * 'S' selection of the user with the highest marginal density, i.e.
(540) Calculate the reward mu of each winner i i ：
Wherein S is′ i′-1 Is to add user i 'to the previous set of winners of S';
(541) adding the user i 'into the winner set S';
(542) and (4) judging whether all the users in the winner set S execute the step (536), if so, ending, otherwise, executing the step (536).
Claims (7)
1. A topic-sensitive crowdsourcing task diffusion method is characterized by comprising the following steps:
step 1: the crowdsourcing platform obtains a registered user set U, wherein the number of registered users is n, and a task set T ═ T published by the crowdsourcing platform 1 ，t 2 ，…，t m A social network graph G ═ V, E) and a budget B for diffusion tasks, where V represents all users in the social network, E is an edge between all users in the social network, and if there is an edge between a registered user V and a user G, the user G is said to be a social neighbor of the registered user V; each registered user v belongs to U and submits bidding document theta v ＝(T v ，b v ) Wherein
Is a set of tasks that registered user v is willing to spread, b v Is a quote of registered user v, c v Is the true cost of registered user v, c v The information is private information and is known by registered users v; defining the utility of a registered user v as f v This can be obtained from equation (1):
wherein mu v The reward is the reward of the registered user v, and S is a winner set;
step 2: defining a budget feasible task diffusion problem;
and 3, step 3: establishing a theme-sensitive independent cascade influence diffusion model according to the independent cascade influence diffusion model and the crowdsourcing system characteristics;
and 4, step 4: obtaining a task theme and an influence parameter of an independent cascade influence diffusion model sensitive to the theme;
and 5: a budget feasible mechanism is adopted to obtain a winner set from all registered users, and winner compensation is calculated,
the step 3 comprises the following steps:
step 3.1: in the independent cascade influence diffusion model, for a given social network, a part of nodes in the social network are in an activated state in an initial stage; at any time period tau epsilon N + Every node C in active state tries to get p C，D E (0, 1) activates a social neighbor node D thereof; if the attempt is successful, D is changed from the inactive state to the active state in the period of tau + 1; each node C in the active state has only one opportunity to activate its social neighbor node, which is called the active node once node D is activated; assuming that the influence of the node C in the activated state on the social neighbor node D is independent from the historical data, so as to obtain the activity state of the social neighbor node D, and then calculating the influence of the node C in the activated state; in the crowdsourcing system, each user is mapped to one node in the network, the influence among the users is closely related to the spread tasks and the subjects of the tasks, the active users in the initial stage are winners among the registered users participating in the reverse auction, and each registered user v is related to the task t j The probability of activating a social neighbor g is p v，g (j)∈(0，1)；p v，g (j) Indicating that registered user v will task t j The influence of spreading to social neighbors g;
step 3.2: successful activation of social neighbors g by registered users v with respect to tasks t j Is calculated by equation (4):
wherein, it is provided with
Is task t j Subject of (1) is a probability of kI.e. by
And satisfy
I.e. for task t j The sum of the probabilities of all subjects is 1, where z j Is task t j Z is the set of topics for all tasks; probability of
Is the influence of the registered user v on his social neighbor g about topic k;
step 3.3: obtaining a subject-sensitive platform utility function:
where Path (v, u, j) is for task t j The path with the largest multiplicative activation probability from a registered user v to any non-registered user u,<v′，u′>refers to the user-mapped node pairs for each edge in the path,
the step 4 comprises the following steps:
step 4.1: obtaining likelihood probability of historical data and an estimated value of the historical data;
step 4.2: obtaining the theme distribution of the randomly generated tasks and the influence of each edge, and finishing initialization;
step 4.3: combining with a crowdsourcing system, expanding an expectation maximization algorithm, namely obtaining the theme distribution of tasks through maximum likelihood estimation according to the newly-calculated influence value of each edge in the social graph;
step 4.4: re-estimating the influence of each edge in the social graph according to the newly calculated theme distribution of the tasks;
step 4.4: step 4.3 and step 4.4 are performed iteratively until convergence.
2. The method of claim 1, wherein: the step 2 comprises the following steps:
step 2.1: build constraint sigma v∈S μ v ≤B，μ v The reward of each registered user is represented, and the reward of all registered users is ensured not to exceed budget B;
step 2.2: the problem of budget feasible task diffusion is as follows:
maxf(S) (2)
s.t.∑ v∈S μ v ≤B (3)
wherein f (S) represents the platform utility function when the platform winner set S is crowd-sourced, and the goal of the incentive mechanism is to maximize the platform utility function f (S) under the condition that the rewards of all registered users do not exceed the budget B constraint.
3. The method of claim 1, wherein the step 4.1 comprises the steps of:
step 4.11: based on historical task data x 1 ，x 2 ，…，x m Wherein x is j ＝{x j (τ)|τ∈N * }，x j (τ) is the time period τ with respect to task t j Active set of users, assuming arbitrary x j Independent of each other, obtaining likelihood probability of historical data, i.e. x j The joint distribution of (a) is:
wherein, p (x) j (ii) a λ) is that the sample is x when the parameter is λ j Probability of time;
step 4.12: the Log likelihood probability obtained according to equation (6) is:
step 4.13: obtaining a task t j Subject z of j The prior probability distribution of (a) is:
wherein the content of the first and second substances,
is the value of the parameter lambda after a iterations;
step 4.14: to obtain
The likelihood function of (d) is:
step 4.15: obtain new
The estimated value of (c):
4. the method of claim 1, wherein the step 4.3 comprises the steps of:
step 4.31: according to the processing of historical data and general knowledge, assume that the topic is k task diffusion set x j The possibilities of (a) are:
wherein, the first and the second end of the pipe are connected with each other,
is a set of neighbors with potential impact on any unregistered user u, expressed as:
wherein, the first and the second end of the pipe are connected with each other,
is the time period during which any non-registered user u transitions to the active state,
a time phase of a social neighbor h of a non-registered user u transitioning to an active state;
step 4.32: obtaining a diffusion task set x through conditional probability j And the joint probability distribution of topic k is:
p(x j ，k；λ)＝p(x j |k；λ)p(k|λ) (14)
step 4.33: set pi k P (k | λ) is the prior probability of the subject of any general task being k;
step 4.34: the prior probability distribution of the task with the topic of k is obtained as follows:
step 4.35: the probability of obtaining a topic k for any general task is:
5. the method of claim 1, wherein the step 4.4 comprises the steps of: :
step 4.41: a likelihood function is obtained according to step 4.2:
step 4.42: since the pair of likelihood functions in equation (17)
Is monotonous and can not maximize the likelihood function, when the subject is redesigned to be k, the task diffusion set x j The possibilities of (a) are:
wherein the content of the first and second substances,
the user h successfully activates its neighbor u, i.e. the task t with spreading topic k j The probability of success, expressed as:
step 4.43: the corresponding likelihood function is obtained as:
step 4.44: obtaining an estimate of the influence of each edge:
6. the method of claim 1, wherein the step 5 comprises the steps of:
step 5.1: initializing winner set S ═ phi and payout vector μ ═ μ 1 ，μ 2 ，…，μ n ) Temporary set S as 0 * ＝{v|b v B ≦ B }, i.e. set S * Is the set of all registered users v whose quoted price is less than budget B;
step 5.2: selecting S with a probability of 2/5 * User v with the greatest value f ({ v }) in * As winner, user v * The reward obtained is equal to the budget;
step 5.3: winner selection and remuneration are made according to crowdsourcing rules with a probability of 3/5.
7. The method of claim 1, wherein the step 5.3 comprises the steps of:
step 5.31: obtaining a margin value f according to the winner set S v (S), expressed as:
f v (S)＝f(S∪{v})-f(S) (22)
step 5.32: find S * User v with maximum medium marginal density max I.e. by
Step 5.33: judging user v max Whether the auction base price exceeds
If yes, executing step 5.36, otherwise, executing step 5.34;
step 5.34: will user v max Adding into the winner set S;
step 5.35: from the set S * V 'of user with maximum marginal density selected from S' max I.e. by
Step 5.33 is executed;
step 5.36: in the compensation stage, a set S is obtained for each user i in the winner set S *′ ＝S * \ { i }, initializing a new set of winners S ', let S' ═ phi;
step 5.37: at S *′ To find the user i' with the maximum marginal density, i.e. the user
Step 5.38: determining whether the auction reserve price of user i' exceeds
If yes, executing step 5.42, otherwise, continuing to execute step 5.39;
step 5.39: from the set S *′ Selection of the user with the greatest marginal density in \ S', i.e.
Step 5.40: calculate the reward mu for each winner i i ：
Wherein, S' i′-1 Is to add user i 'to the previous set of winners of S';
step 5.41: adding the user i 'into the winner set S';
step 5.42: and (5) judging whether all the users in the winner set S execute the step 5.36, if so, ending, otherwise, executing the step 5.36.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110180146.9A CN113011703B (en) | 2021-02-08 | 2021-02-08 | Topic-sensitive crowdsourcing task diffusion method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110180146.9A CN113011703B (en) | 2021-02-08 | 2021-02-08 | Topic-sensitive crowdsourcing task diffusion method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113011703A CN113011703A (en) | 2021-06-22 |
| CN113011703B true CN113011703B (en) | 2022-08-26 |
Family
ID=76384119
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110180146.9A Active CN113011703B (en) | 2021-02-08 | 2021-02-08 | Topic-sensitive crowdsourcing task diffusion method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113011703B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114118472B (en) * | 2021-11-26 | 2024-06-18 | 武汉理工大学 | Shared equipment maintenance system and method based on task distribution and user pricing mechanism |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107688964A (en) * | 2017-08-07 | 2018-02-13 | 南京邮电大学 | A kind of motivational techniques for the diffusion of extensive mass-rent task |
-
2021
- 2021-02-08 CN CN202110180146.9A patent/CN113011703B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107688964A (en) * | 2017-08-07 | 2018-02-13 | 南京邮电大学 | A kind of motivational techniques for the diffusion of extensive mass-rent task |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113011703A (en) | 2021-06-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111754000B (en) | Quality-aware edge intelligent federal learning method and system | |
| Cai et al. | Real-time bidding by reinforcement learning in display advertising | |
| CN108133330B (en) | Social crowdsourcing task allocation method and system | |
| US20090307073A1 (en) | Social marketing | |
| US20100250332A1 (en) | System and Method for Adaptive Bidding for Display Advertising | |
| Avadhanula et al. | Stochastic bandits for multi-platform budget optimization in online advertising | |
| JP7219228B2 (en) | Strategic exploration in strategic dialogue between parties | |
| Malik et al. | Why bitcoin will fail to scale? | |
| US11468521B2 (en) | Social media account filtering method and apparatus | |
| Lu et al. | Designing socially-optimal rating protocols for crowdsourcing contest dilemma | |
| CN111626563B (en) | Dual-target robust mobile crowd sensing system and excitation method thereof | |
| CN113011703B (en) | Topic-sensitive crowdsourcing task diffusion method | |
| Jin et al. | DPDA: A differentially private double auction scheme for mobile crowd sensing | |
| Ali et al. | A novel nested q-learning method to tackle time-constrained competitive influence maximization | |
| Shen et al. | Multi-winner contests for strategic diffusion in social networks | |
| Du et al. | Adversarial deep learning for online resource allocation | |
| Bimpikis et al. | Managing market thickness in online B2B markets | |
| CN112292699A (en) | Determining action selection guidelines for an execution device | |
| CN107330754B (en) | Mobile crowd sensing excitation method for cooperative task | |
| CN108337263B (en) | Sybil attack prevention incentive method based on mobile crowd sensing system | |
| CN115099535B (en) | Dual-target crowd sensing excitation mechanism method | |
| Fan et al. | Truthful incentive mechanisms for dynamic and heterogeneous tasks in mobile crowdsourcing | |
| Jiang et al. | Combinatorial double auction for resource allocation with differential privacy in edge computing | |
| Tang et al. | Multi-session budget optimization for forward auction-based federated learning | |
| CN114116740A (en) | Method and device for determining contribution degree of participants in joint learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |