CN106339852A - Crowdsourcing task allocation method considering preferences - Google Patents

Abstract

The invention discloses a crowdsourcing task allocation method considering preferences. The objective of the invention is to solve a value maximized allocation problem. Two kinds of crowdsourcing models, namely, a same-value task crowdsourcing model and a different-value task crowdsourcing model, are provided, and the two models both consider competitive relationships between requesters and the preferences of the task requesters to crowdsourcing workers; and a matching-based task allocation mechanism is put forwards respectively for the same-value task crowdsourcing model and the different-value task crowdsourcing model. With the method adopted, a task total value maximized allocation problem can be solved. The task allocation method provided by the invention is advantageous in calculation effectiveness, work load feasibility, preference authenticity and constant factor approximation ratio.

Description

There is the mass-rent method for allocating tasks of preference

Technical field

The present invention relates in mass-rent system task distribution method, belong to the Internet and the game theoretic crossing domain of algorithm.

Background technology

Mass-rent completes, by integrating unknown subscriber on intelligent movable mobile phone and the Internet, the task that machine is difficult to complete, Become the main realization mechanism that many Internet resources produce.At present mass-rent have been widely used for information retrieval, artificial intelligence, The fields such as video analysis, knowledge excavation, smart city, man-machine interaction study, image quality measure.

Mechanism of Task Allocation in mass-rent system is particularly significant, and existing research often swashs by task distribution with to user Encourage and combine, form the incentive mechanism in mass-rent system.At present, the such incentive mechanism of emphasized in design is come for many researchs Encouragement workers participate in mass-rent.However, existing incentive mechanism assumes that mass-rent task belongs to the task requests of a monopolization Person, and do not account for the preference issues to worker for the requestor.Therefore work on hand the distribution of mass-rent task is only considered right The excitation of mass-rent workman, and have ignored the excitation to task requester.

In fact, in many mass-rent markets, being to there is competition between the requestor of task, they are likely to make plan Slightly behavior is maximizing the interests of itself.This requestor competitive market may be by mass-rent workman, especially high-quality mass-rent work People's rare numbers lead to.

Another in esse phenomenon is that the requestor of task often has preference to mass-rent workman.Such as, requestor and There is the conflict of interest between mass-rent workman, certain task need to meet the mass-rent workman of certain condition to complete, the work of mass-rent workman Make the reasons such as load restriction, requestor all may be led to have preference to mass-rent workman.

The present invention devise the task with preference in competitive mass-rent system distribution method maximize complete appoint The aggregate value of business.

Content of the invention

It is an object of the invention to provide the method in competitive mass-rent system, the task with preference effectively distributed, Solve Maximum Value assignment problem.Present invention firstly provides two kinds of mass-rent models: identical value task mass-rent model and difference Value task mass-rent model.Two models all consider competitive relation between requestor and the preference of requestor.Next is directed to Identical value task mass-rent model and different value task mass-rent model propose a kind of Mechanism of Task Allocation respectively, solve task total Value maximization assignment problem, and prove this method have calculating effectiveness, live load feasibility, preference verity and The characteristic of invariant approximation ratio.The present invention, with respect to the motivational techniques that presently, there are, solves in mass-rent system and has competition Property and the exciting torque problem of tendentiousness requestor.

The technical solution of the present invention is:

A kind of task mass-rent process with preference: the information of mass-rent workman is distributed to all of task requests by mass-rent platform Person.Each task requester returns one and asks platform, and this request bag contains a set of preferences, and that is, one to this task requester For compatible mass-rent workman set.Mass-rent platform calculates a distribution, and this distribution is task and the set of mass-rent workman couple, and Notify to task requester and mass-rent workman.The mass-rent workman execution task of being assigned to, and implementing result is returned to mass-rent put down Platform.Finally, mass-rent platform provides service to task requester.

The mass-rent method for allocating tasks with preference of the present invention, comprises an identical value task mass-rent model.Should Model includes a mass-rent platform, task requester set r={ 1,2 ..., n } and mass-rent workman set w={ 1,2 ..., m }. Mass-rent platform issues mass-rent workman set w to all of task requester.Each task requester i submits a request b to_i= (t_i,p_i), wherein t_iRepresent the task that task requester is submitted to.Each task t_iOne sets of preferences of associationThis Bright consideration sets of preferences q_iIt is private information,It is the preference of task requester i statement.Expression task t_iValue, this The all of value of identical value task mass-rent model hypothesisIt is equal.B={ b₁,b₂,…,b_nRepresent all task requests Set.Mass-rent platform calculates distributing so that having completed the aggregate value of task between a task and mass-rent workman Greatly, each of which task is at most given to a mass-rent workman, and each mass-rent workman at most can execute a task.

Identical value task mass-rent model of the present invention, step is as follows:

Step 201: platform one mass-rent workman set w=of issue 1,2 ..., and m } give all of task requester；

Step 202: set task requester collection and be combined into r={ 1,2 ..., n }, each task requester i submits one to platform Individual request b_i=(t_i,p_i), wherein t_iIt is the task that task requester i submits to.It is task requester i statement to crowd The sets of preferences of job contract people；

Step 203: the distribution a between mass-rent platform calculating task and mass-rent workman, that is, task-mass-rent workman is to (i, k) Set, i ∈ r, k ∈ p_i；

Step 204: mass-rent workman executes allocated task and result is fed back to mass-rent platform；

Step 205: mass-rent platform is supplied to task requester service.

In step 203, the assignment problem formalization representation between mass-rent platform calculating task and mass-rent workman is

max_aV (a)=∑_k∈w|a^k|=σ_i∈r|a_i|

$\begin{array}{cc}s.t.& \left(1\right)---\left|a^k\right|\≤1,\∀k\&element;w\end{array}$

$\left(2\right)---\left|a_i\right|\≤1,\∀i\&element;r$

(3)a∈{(i,k)|,i∈r,k∈p_i}

Wherein, a^k, k ∈ w represents the task of distributing to mass-rent workman k, a_i, i ∈ r represents the mass-rent workman of execution task i, v A () represents the total value of task in distribution a.

The essence of above-mentioned Formalization Problems is: the distribution set between one task of searching, mass-rent workman is so that distribute to The aggregate value of the set of tasks of mass-rent workman is maximum, and each task is at most given to a mass-rent workman, each Mass-rent workman at most can execute a task.

In step 203, the step of the distribution a between mass-rent platform calculating task and mass-rent workman is as follows:

Step 301: initialization task distribution a is sky；

Step 302: all of mass-rent workman-task requester is ranked into (1,1) to (i, k) according to lexicographic order, (1,2) ..., (2,1), (2,2) ..., (m, n), wherein i ∈ r, k ∈ p_i.H represents the sequence through sequence.

Step 303: by sequences h be assigned to sequences h '；

Step 304: by the maximum Bipartite Matching algorithm sequence of calculationMaximum match size n；

Step 305: for each mass-rent workman-task requester to j, j ∈ h, by execution step 306- step 308；

Step 306: by j from sequences h ' delete, by the maximum match of maximum Bipartite Matching algorithm sequence of calculation h' Size n'；

Step 307: check that whether n' is more than or equal to n, if greater than being equal to, execution step 308, otherwise, execution step 305

Step 308: remove j from h'；

Step 309: by sequences h ' be assigned to distribute a；

Step 310: return distribution a.

The mass-rent method for allocating tasks with preference of the present invention, comprises a different value task mass-rent model.Its InImplication with identical in identical value task mass-rent model.Each mass-rent workman k ∈ w has one Individual effort indicator i_k>=0, represent that mass-rent workman k executes the level of effort of task.Two tuples (w, i) issued by mass-rent platform To all of task requester, wherein i=(i₁,i₂,…,i_m).Each task requester i submits a request b to_i=(t_i, a_i,p_i), wherein a_iIt is task t_iType,It is the sets of preferences of task requester i statement.For each task t_i There is degree-of-difficulty factor d_i,d_i=f (a_i).Task t_iValue be defined asB={ b₁,b₂,…,b_nRepresent The set of all requests.Mass-rent platform calculates the distribution between a task and mass-rent workman, and each of which task is at most It is given to a worker, each worker at most can execute a task.

Difference value task mass-rent model of the present invention, step is as follows:

Step 401: mass-rent platform issues two tuples (w, i) to all of task requester；

Step 402: set task requester collection and be combined into r={ 1,2 ..., n }, each task requester i carries to mass-rent platform Hand over a request b_i=(t_i,a_i,p_i), wherein t_iIt is the task that task requester i submits to, each task t_iThere is one therewith Related task type a_i,It is the sets of preferences of task requester i statement；

Step 403: the distribution a between mass-rent platform calculating task and mass-rent workman, that is, task-mass-rent workman is to (i, k) Set, i ∈ r, k ∈ p；

Step 404: mass-rent workman executes allocated task and result is fed back to mass-rent platform；

Step 405: mass-rent platform is supplied to task requester service.

In step 403, the assignment problem formalization representation between platform calculating task and worker is

${\max }_{a}v\left(a\right)={\σ}_{(i,k)\&element;a}{v}_i^k$

$\begin{array}{cc}s.t.& \left(1\right)---\left|a^k\right|\≤1,\∀k\&element;w\end{array}$

$\left(2\right)---\left|a_i\right|\≤1,\∀i\&element;r$

(3)a∈{(i,k)|,i∈r,k∈p_i}

Wherein, a^k, k ∈ w represents the task of distributing to mass-rent workman k, a_i, i ∈ r represents the i mass-rent workman of execution task, v A () represents the total value of task in distribution a.

The essence of above-mentioned Formalization Problems is: the distribution set between one task of searching, mass-rent workman is so that distribute to The aggregate value of the set of tasks of mass-rent workman is maximum, and each task is at most given to a mass-rent workman, each Mass-rent workman at most can execute a task.

In step 403, the step of the distribution a between mass-rent platform calculating task and mass-rent workman is as follows:

Step 501: for each task requester i ∈ r and each worker k ∈ p_i, order

Step 502: according toNon-increasing order all of task-mass-rent workman is ranked up to (i, k), sequence use J represents；

Step 503: initialization task distribution a is sky；

Step 504: check whether each of j task-mass-rent workman was all considered to j ∈ j, if it is, execution Step 507, otherwise, execution step 505；

Step 505: judge whether a ∪ { j } is a coupling in g (r, w, j), if it is, execution step 506, otherwise hold Row step 504；

Step 506: make a=a ∪ { j }；

Step 507: return distribution a.

Beneficial effect

There is the mass-rent method for allocating tasks of preference, can be used for task requester in task mass-rent system and mass-rent workman is had Task during preference is had to distribute.The present invention has a following significant advantage:

Calculating time complexity is low, and the task for identical value task mass-rent model distributes computational methods time complexity For o (n²m²(n+m)), the task distribution computational methods time complexity for different value task mass-rent models is o (nm max (log (nm), min (n, m))), wherein n is business requestor's quantity, and m is mass-rent number of workers.It is a complete multinomial time Method, has the value of practical application.

Task for identical value task mass-rent model distributes computational methods and for different value task mass-rent models Task distribution computational methods all there is live load feasibility, that is, each mass-rent workman is at most assigned to 1 task.

Task for identical value task mass-rent model distributes computational methods and for different value task mass-rent models Task distribution computational methods all there is preference verity, that is, no matter other task requester submit what preference, neither one to Task requester can improve the value of oneself acquisition by submitting a sets of preferences being different from actual preferences collection to.Change sentence Talk about, for all task requester, one actual preferences set of report is a dominating stragegy.Therefore task requester inclines To in the report actual preferences set of itself.Preference verity is for preventing corner on the market or gang up with important function.

The present invention adopts matching process to solve total value maximization problems, wherein, for identical value task mass-rent model Task distribution computational methods be optimal allocation method, the task for different value task mass-rent models distributes computational methods Approximation ratio is 2.

Brief description

Fig. 1 is a kind of task mass-rent process with preference；

Fig. 2 is identical value task mass-rent model；

Fig. 3 is that the task for identical value task mass-rent model distributes computational methods；

Fig. 4 is different value task mass-rent model；

Fig. 5 is that the task for different value task mass-rent models distributes computational methods.

Specific embodiment

Describe the preferred embodiments of the present invention below in conjunction with the accompanying drawings in detail.

A kind of task mass-rent process with preference is as shown in Figure 1: the information of mass-rent workman is distributed to all by mass-rent platform Task requester.Each task requester returns one and asks platform, and this request bag contains a set of preferences, and that is, one to this Compatible mass-rent workman set for task requester.Mass-rent platform calculates a distribution, and this distribution is task and mass-rent workman To set, and notify to task requester and mass-rent workman.The task that mass-rent workman execution is assigned to, and implementing result is returned Back to mass-rent platform.Finally, mass-rent platform provides service to task requester.

The mass-rent method for allocating tasks with preference of the present invention, comprises an identical value task mass-rent model.Should Model includes a mass-rent platform, task requester set r={ 1,2 ..., n } and mass-rent workman set w={ 1,2 ..., m }. Mass-rent platform issues mass-rent workman set w to all of task requester.Each task requester i submits a request b to_i= (t_i,p_i), wherein t_iRepresent the task that task requester is submitted to.Each task t_iOne sets of preferences of associationThis Bright consideration sets of preferences q_iIt is private information,It is the preference of task requester i statement.Expression task t_iValue, this The all of value of identical value task mass-rent model hypothesisIt is equal.B={ b₁,b₂,…,b_nRepresent all task requests Set.Mass-rent platform calculates distributing so that having completed the aggregate value of task between a task and mass-rent workman Greatly, each of which task is at most given to a mass-rent workman, and each mass-rent workman at most can execute a task.

Identical value task mass-rent model of the present invention, flow process is as shown in Fig. 2 specifically comprise the following steps that

Step 201: platform one mass-rent workman set w=of issue 1,2 ..., and m } give all of task requester；

Step 202: set task requester collection and be combined into r={ 1,2 ..., n }, each task requester i submits one to platform Individual request b_i=(t_i,p_i), wherein t_iIt is the task that task requester i submits to.It is task requester i statement to crowd The sets of preferences of job contract people；

Step 203: the distribution a between mass-rent platform calculating task and mass-rent workman, that is, task-mass-rent workman is to (i, k) Set, i ∈ r, k ∈ p_i；

Step 204: mass-rent workman executes allocated task and result is fed back to mass-rent platform；

Step 205: mass-rent platform is supplied to task requester service.

In step 203, the assignment problem formalization representation between mass-rent platform calculating task and mass-rent workman is

max_aV (a)=∑_k∈w|a^k|=σ_i∈r|a_i|

$\begin{array}{cc}s.t.& \left(1\right)---\left|a^k\right|\≤1,\∀k\&element;w\end{array}$

$\left(2\right)---\left|a_i\right|\≤1,\∀i\&element;r$

(3)a∈{(i,k)|,i∈r,k∈p_i}

Wherein, a^k, k ∈ w represents the k task distributing to mass-rent workman, a_i, i ∈ r represents the mass-rent workman of execution task i, v A () represents the total value of task in distribution a.

The essence of above-mentioned Formalization Problems is: the distribution set between one task of searching, mass-rent workman is so that distribute to The aggregate value of the set of tasks of mass-rent workman is maximum, and each task is at most given to a mass-rent workman, each Mass-rent workman at most can execute a task.

In step 203, the distribution a between mass-rent platform calculating task and mass-rent workman, flow process is as shown in figure 3, concrete Step is as follows:

Step 301: initialization task distribution a is sky；

Step 302: all of mass-rent workman-task requester is ranked into (1,1) to (i, k) according to lexicographic order, ..., (1,2) (2,1), (2,2) ..., (m, n) wherein i ∈ r, k ∈ p_i.H represents the sequence through sequence.

Step 303: by sequences h be assigned to sequences h '；

Step 304: by size n of the maximum match of maximum Bipartite Matching algorithm sequence of calculation h；

Step 305: for each mass-rent workman-task requester to j, j ∈ h, by execution step 306- step 308；

Step 306: by j from sequences h ' delete, by the maximum match of maximum Bipartite Matching algorithm sequence of calculation h' Size n'；

Step 307: check that whether n' is more than or equal to n, if greater than being equal to, execution step 308, otherwise, execution step 305；

Step 308: remove j from h'；

Step 309: by sequences h ' be assigned to distribute a；

Step 310: return distribution a.

The mass-rent method for allocating tasks with preference of the present invention, comprises a different value task mass-rent model.Its InImplication with identical in identical value task mass-rent model.Each mass-rent workman k ∈ w has one Individual effort indicator i_k>=0, represent that mass-rent workman k executes the level of effort of task.Two tuples (w, i) issued by mass-rent platform To all of task requester, wherein i=(i₁,i₂,…,i_m).Each task requester i submits a request b to_i=(t_i, a_i,p_i), wherein a_iIt is task t_iType,It is the sets of preferences of task requester i statement.For each task t_i There is degree-of-difficulty factor d_i, d_i=f (a_i).Task t_iValue be defined asB={ b₁,b₂,…,b_nRepresent The set of all requests.Mass-rent platform calculates the distribution between a task and mass-rent workman, and each of which task is at most It is given to a worker, each worker at most can execute a task.

Difference value task mass-rent model of the present invention, flow process is as shown in figure 4, specifically comprise the following steps that

Step 401: mass-rent platform issues two tuples (w, i) to all of task requester；

Step 402: set task requester collection and be combined into r={ 1,2 ..., n }, each task requester i carries to mass-rent platform Hand over a request b_i=(t_i,a_i,p_i), wherein t_iIt is the task that task requester i submits to, each task t_iThere is one therewith Related task type a_i,It is the sets of preferences of task requester i statement；

Step 403: the distribution a between mass-rent platform calculating task and mass-rent workman, that is, task-mass-rent workman is to (i, k) Set, i ∈ r, k ∈ p_i；

Step 404: mass-rent workman executes allocated task and result is fed back to mass-rent platform；

Step 405: mass-rent platform is supplied to task requester service.

In step 403, the assignment problem formalization representation between platform calculating task and worker is

${\max }_{a}v\left(a\right)={\σ}_{(i,k)\&element;a}{v}_i^k$

$\begin{array}{cc}s.t.& \left(1\right)---\left|a^k\right|\≤1,\∀k\&element;w\end{array}$

$\left(2\right)---\left|a_i\right|\≤1,\∀i\&element;r$

(3)a∈{(i,k)|,i∈r,k∈p_i}

Wherein, a^k, k ∈ w represents the task of distributing to mass-rent workman k, a_i, i ∈ r represents the mass-rent workman of execution task i, v A () represents the total value of task in distribution a.

The essence of above-mentioned Formalization Problems is: the distribution set between one task of searching, mass-rent workman is so that distribute to The aggregate value of the set of tasks of mass-rent workman is maximum, and each task is at most given to a mass-rent workman, each Mass-rent workman at most can execute a task.

In step 403, the distribution a between mass-rent platform calculating task and mass-rent workman, flow process is as shown in figure 5, concrete Step is as follows:

Step 501: for each task requester i ∈ r and each worker k ∈ p_i, order

Step 502: according toNon-increasing order all of task-mass-rent workman is ranked up to (i, k), sequence use J represents；

Step 503: initialization task distribution a is sky；

Step 504: check whether each of j task-mass-rent workman was all considered to j ∈ j, if it is, execution Step 507, otherwise, execution step 505；

Step 505: judge whether a ∪ { j } is a coupling in g (r, w, j), if it is, execution step 506, otherwise hold Row step 504；

Step 506: make a=a ∪ { j }；

Step 507: return distribution a.

Claims (6)

1. there is the mass-rent method for allocating tasks of preference it is characterised in that comprising an identical value task mass-rent model, step As follows:

Step 201: platform one mass-rent workman set w=of issue 1,2 ..., and m } give all of task requester；

Step 202: set task requester collection and be combined into r={ 1,2 ..., n }, each task requester i submits to one to ask to platform Seek b_i=(t_i,p_i), wherein t_iIt is the task that task requester i submits to.It is task requester i statement to mass-rent work The sets of preferences of people；

Step 203: the distribution a between mass-rent platform calculating task and mass-rent workman, i.e. task-collection to (i, k) for the mass-rent workman Close, i ∈ r, k ∈ p_i；

Step 204: mass-rent workman executes allocated task and result is fed back to mass-rent platform；

Step 205: mass-rent platform is supplied to task requester service.

2. the method for claim 1 is it is characterised in that in step 203, mass-rent platform calculating task and mass-rent workman Between assignment problem formalization representation be

max_aV (a)=∑_k∈w|a^k|=∑_i∈r|a|_i

$s.t.\left(1\right)\left|a^k\right|\≤1,\∀k\&element;w$

$\left(2\right)\left|a_i\right|\≤1,\∀i\&element;r$

(3)a∈{(i,k)|,i∈r,k∈p_i}

Wherein, a^k, k ∈ w represents the task of distributing to mass-rent workman k, a_i, i ∈ r represents the mass-rent workman of execution task i, v (a) Represent the total value of task in distribution a；

The essence of above-mentioned Formalization Problems is: the distribution set between one task of searching, mass-rent workman is so that distribute to mass-rent The aggregate value of the set of tasks of workman is maximum, and each task is at most given to a mass-rent workman, each mass-rent Workman at most can execute a task.

3. the method for claim 1 is it is characterised in that in step 203, mass-rent platform calculating task and mass-rent workman Between distribution a step as follows:

Step 301: initialization task distribution a is sky；

Step 302: all of mass-rent workman-task requester is ranked into (1,1) to (i, k) according to lexicographic order, (1, 2) ..., (2,1), (2,2) ..., (m, n), wherein i ∈ r, k ∈ p_i.H represents the sequence through sequence；

Step 303: by sequences h be assigned to sequences h '；

Step 304: by size n of the maximum match of maximum Bipartite Matching algorithm sequence of calculation h；

Step 305: for each mass-rent workman-task requester to j, j ∈ h, by execution step 306- step 308；

Step 306: by j from sequences h ' delete, big by the maximum match of maximum Bipartite Matching algorithm sequence of calculation h' Little n'；

Step 307: check that whether n' is more than or equal to n, if greater than being equal to, execution step 308, otherwise, execution step 305

Step 308: remove j from h'；

Step 309: by sequences h ' be assigned to distribute a；

Step 310: return distribution a.

4. there is the mass-rent method for allocating tasks of preference it is characterised in that comprising a different value task mass-rent model, step As follows:

Step 401: mass-rent platform issues two tuples (w, i) to all of task requester；

Step 402: set task requester collection and be combined into r={ 1,2 ..., n }, each task requester i submits one to mass-rent platform Individual request b_i=(t_i,a_i,p_i), wherein t_iIt is the task that task requester i submits to, each task t_iHave one associated Task type a_i,It is the sets of preferences of task requester i statement；

Step 403: the distribution a between mass-rent platform calculating task and mass-rent workman, i.e. task-collection to (i, k) for the mass-rent workman Close, i ∈ r, k ∈ p_i；

Step 404: mass-rent workman executes allocated task and result is fed back to mass-rent platform；

Step 405: mass-rent platform is supplied to task requester service.

5. it is characterised in that in step 403, platform calculates appoints difference value task mass-rent model as claimed in claim 4 Business and worker between assignment problem formalization representation be

${\max }_{a}v\left(a\right)={\mathrm{\σ}}_{(i,k)\&element;a}{v}_i^k$

$s.t.\left(1\right)\left|a^k\right|\≤1,\∀k\&element;w$

$\left(2\right)\left|a_i\right|\≤1,\∀i\&element;r$

(3)a∈{(i,k)|,i∈r,k∈p_i}

Wherein, a^k, k ∈ w represents the task of distributing to mass-rent workman k, a_i, i ∈ r represents the mass-rent workman of execution task i, v (a) Represent the total value of task in distribution a；

The essence of above-mentioned Formalization Problems is: the distribution set between one task of searching, mass-rent workman is so that distribute to mass-rent The aggregate value of the set of tasks of workman is maximum, and each task is at most given to a mass-rent workman, each mass-rent Workman at most can execute a task.

6. as claimed in claim 4 difference value task mass-rent model it is characterised in that in step 403, mass-rent platform meter The step of the distribution a between calculation task and mass-rent workman is as follows:

Step 501: for each task requester i ∈ r and each worker k ∈ p_i, order

Step 502: according toNon-increasing order all of task-mass-rent workman is ranked up to (i, k), sequence j table Show；

Step 503: initialization task distribution a is sky；

Step 504: check whether each of j task-mass-rent workman was all considered to j ∈ j, if it is, execution step 507, otherwise, execution step 505；

Step 505: judge whether a ∪ { j } is a coupling in g (r, w, j), if it is, execution step 506, otherwise execute step Rapid 504；

Step 506: make a=a ∪ { j }；

Step 507: return distribution a.