CN106203893A - A kind of method for allocating tasks of based on genetic algorithm under mass-rent environment - Google Patents

Abstract

The present invention relates to the method for allocating tasks based on genetic algorithm under a kind of mass-rent environment.The present invention receives the multiple subtasks after a task submitter submits task to and decomposes, the workman's information participating in this task competitive bidding provided by mass-rent platform is provided, build the technical ability hierarchical tree of relation between each participant's technical ability, show that each workman can be provided by the value of each subtask, extract key word information, obtain the key word dictionary of each workman, draw the similarity between each workman, use genetic algorithm for solving task allocative decision, mass-rent platform the scheme drawn carries out the distribution of task.Instant invention overcomes and waste time and energy, the defect of poor effect.The present invention distributes in a most suitable team to task, is so possible not only to promote the delivery quality of final task, and workman also can be made to experience consistent workman's preferably cooperative cooperating with oneself exploitation as far as possible when completing this task.

Description

A kind of method for allocating tasks of based on genetic algorithm under mass-rent environment

Technical field

The present invention relates to a kind of each subtask and the competitive bidding participant of a more complicated task on mass-rent platform be carried out The method of distribution, particularly to the method for allocating tasks based on genetic algorithm under a kind of mass-rent environment.

Background technology

In recent years, mass-rent has obtained the extensive concern of industrial quarters and academia.Mass-rent (crowdsourcing) this concept It is to be proposed in June, 2006 by reporter Jeff person of outstanding talent (Jeff Howe) of the U.S.'s " line " magazine. Jeff person of outstanding talent is to " mass-rent " Definition be: " task that a company or mechanism were performed the past by employee is contracted out to non-spy with freely the most voluntary form The way of fixed (and the most large-scale) public network. the task of mass-rent is generally undertaken by individual, if but related to Need the task that multiple person cooperational completes, it is also possible to rely on the individual form produced increased income to occur. "

Nowadays, mass-rent platform has been proven that it supports various software development activities such as exploitation and the ability of test, permissible See several successful mass-rent platform such as TopCoder, uTest and domestic one of the chief characters in "Pilgrimage To The West" who was supposedly incarnated through the spirit of pig, a symbol of man's cupidity's net.But, under mass-rent environment For extensive and complicated software development and test assignment, several optimization problem is still had to need to solve.Such as, the task division of labor is i.e. Search one group of suitable registrant and the Task Allocation Problem to registrant.In mass-rent environment, a task may be divided The registrant of the different majors background of dispensing different regions, the quality that therefore final task is paid is a key issue.As Put forward the high-quality first step, it would be desirable to a systematization and automated method optimize the task distribution of registrant, purpose It is exactly to find one group of most suitable registrant for different tasks.

And the task allocative decision workman often of existing mass-rent platform went from appointing that line search selection desire to participate in Business, is then manually gone distribution by task submitter or platform, both wasted time and energy, and often can not get preferable result.Some Method for allocating tasks under only mass-rent platform is also only typically from the angle of task, focuses simply on task and workman Between contact, and the partnership not accounting for affecting because of the similarity between workman and workman be also final task hand over Pay one of influence factor of quality.Additionally, some existing mass-rent Task Assigned Policies, usually require that craftsmanship and task institute Needing technical ability strictly to mate, this frequently results in and does not distributes failure because having the workman of strict skills match.

Summary of the invention

The purpose of the present invention is that and overcomes drawbacks described above, develops the task based on genetic algorithm under a kind of mass-rent environment Distribution method.

The technical scheme is that

Method for allocating tasks based on genetic algorithm under a kind of mass-rent environment, it is mainly characterized by following step Rapid:

(1) receive after task submitter submits task to and decomposes multiple subtasks, each subtask is wrapped respectively Containing following feature: task names, required by task technical ability and task budget；

(2) receiving the workman's information participating in this task competitive bidding provided by mass-rent platform, each workman comprises following feature: Technical ability and grade of skill, workman's credit worthiness that workman is had, history completes task list and this workman is pre-to each subtask The salary of phase；

(3) on mass-rent platform, excavate the performance information of the competitive bidding participant of this task, build between each participant's technical ability The technical ability hierarchical tree of relation；

(4) the technical ability hierarchical tree constructed by integrating step 3 and each workman are had technical ability and credit worthiness thereof and each Technical ability needed for subtask, show that each workman can be provided by the value of each subtask；

(5) history of participation workman's information that integrating step 2 is provided completes task list and carries out pretreatment and extract pass Keyword information, obtains the key word dictionary of each workman, then is compared by the key word dictionary of each workman, draw each work Similarity between people；

(6) genetic algorithm for solving task allocative decision is used, i.e. corresponding relation between each subtask and many group workmans；

(7) carried out the distribution of task according to the scheme drawn in step 6 by mass-rent platform.

Technical ability hierarchical tree described in described step (1), refers to a kind of tree, and its each node represents a kind of Technical ability, as having technical ability s ' workman can approximate the task of execution demand technical ability s, then s ' is the child node of s.

Workman i in described step 4 is provided that to value Value of subtask j_ijCalculating formula is as follows:

Value_ij=maxs ∈ skill (i) (level (i, s) * (1-d (skill (j), s))

$d(s,s^{\′})=\left\{\begin{array}{cc}0,& if(s=s^{\′})\\ \frac{d_{max}-depth\left(lca\right(s,s^{\′}))}{d_{\max }},& otherwise\end{array}\right.$

Wherein, the technical ability list that skill (i) is had by workman i, skill (j) is the technical ability of subtask j demand, d_max For the depth capacity of technical ability hierarchical tree, lca (s, s ') is the minimum common ancestor of s and s ', and depth (s) is that s is in this sets The degree of depth, d (s, s ') is that technical ability s is away from technical ability s ' distance, (i s) is the grade of skill of technical ability s of workman i to level.

The idiographic flow of the employing genetic algorithm for solving task allocative decision that described step (6) is told is as follows:

(6.1). initialize population；

(6.2). calculate the fitness value of population；

(6.3). perform selection, intersect, the genetic manipulation such as variation, generate a new generation population；

(6.4). judge whether to meet termination condition: be, export optimal solution: no, perform step (6.2).

The population that initializes described in described step (6.1) refers to encode the solution of task distribution, and its genetic coding is used Matrix T (m × n) represents, T_ij=1 represents that workman i is assigned in task j, T_ij=0 represents that workman i is not previously allocated takes office In business j.

Fitness value described in described step (6.2) is calculated by fitness function, and fitness function is chosen for:

$f\left(T\right)=\underset{j=1}{\overset{m}{\Σ}}(\mathrm{\α}\×J_j(T)+\mathrm{\β}\×S_j(T\left)\right)$

$J_j\left(T\right)=\underset{j=1}{\overset{n}{\Σ}}{T}_{ij}{\mathrm{Value}}_{ij}$

$S_j\left(T\right)=\underset{i=1}{\overset{n}{\Σ}}{T}_{ij}{\mathrm{Similarity}}_{ij}$

${\mathrm{Similarity}}_{ij}=\frac{\underset{k=1}{\overset{n}{\Σ}}{T}_{kj}{\mathrm{Sim}}_{ik}}{\underset{k=1}{\overset{n}{\Σ}}{T}_{kj}}$

Wherein, f (T) is fitness, J_iFor the object function of task j, S_jFor being assigned to the total of one group of workman of task j Similarity, similarity_ijFor the similarity of workman i with the one group of workman being assigned to task j, Sim_ikFor workman i and workman Similarity between k, α β is weight coefficient.

Advantages of the present invention and effect are, mainly for the various tasks under mass-rent platform, to assign them to skills match And one group of workman that group's similarity is bigger.When distributing task, the present invention not only allows for the had technical ability of workman and task The matching degree of required technical ability, it is also considered that the history to the team being assigned in same task completes similarity of tasks.This Invent it is also contemplated that in knowledge-intensive software development, required by task technical ability is mated the most completely with the technical ability of workman, Therefore use a kind of method of technical ability hierarchical tree that workman is preferably mated with task.

The present invention, on the premise of meeting the budget required by task, distributes a most suitable team to as far as possible and appoints In business, so it is possible not only to promote the delivery quality of final task, also can make the workman can be as far as possible when completing this task The workman preferably cooperative cooperating consistent with oneself exploitation experience.

Further advantage and the effect of the present invention also reside in:

(1) current mass-rent platform is mostly that the form with workman's competitive bidding obtains alone task, and the present invention can allow and appoint Business is automatically matched to one group of skills match and the high workman of group's similarity.Complete task obtains from the history of this group workman The similarity of this group workman.

(2) in knowledge-intensive software development, required by task technical ability is mated the most completely with the technical ability of workman, therefore The present invention uses a kind of method of technical ability hierarchical tree that workman is preferably mated with task.

(3) in the case of meeting task budget, the present invention not only makes the quality finally paid the highest more good, and this Group workman has similar exploitation experience, thus team collaboration is the best.

Other concrete advantage of the present invention and effect will go on to say below.

Accompanying drawing explanation

Fig. 1 schematic flow sheet of the present invention.

The technical ability hierarchical tree structure schematic diagram of Fig. 2 present invention.

Fig. 3 present invention exports result schematic diagram.

In Fig. 4 present invention, workman's history completes task list schematic diagram.

Workman's key word dictionary schematic diagram in Fig. 5 present invention.

Detailed description of the invention

The present invention is further described below in conjunction with embodiment and accompanying drawing:

Embodiment:

The task of task submitter submission and workman's information of participation are as shown in table 1, table 2.

Table 1 task data:

Subtask title	1	2	3
				Required technical ability	s2	s2	s4
Task budget	100	80	70

Table 2 workman's information (participant):

The history of every workman completes task list as shown in Figure 4.

Method for allocating tasks based on genetic algorithm (overview flow chart such as Fig. 1) under a kind of mass-rent environment, its feature exists In following steps:

Step 1). receive a task submitter submit to task and decompose after multiple subtasks, each subtask is divided Do not comprise following feature: task names, required by task technical ability and task budget；

As shown in table 1, the relevant data of task are read in, including task names, required by task technical ability and task budget.

Step 2). receiving the workman's information participating in this task competitive bidding provided by mass-rent platform, each workman comprises as follows Feature: technical ability and grade of skill, workman's credit worthiness that workman is had, history completes task list and each son is appointed by this workman The intended salary of business；

As shown in table 2, the data of workman are read in.

Step 3). on mass-rent platform, excavate the performance information of the competitive bidding participant of this task, build each participant's skill The technical ability hierarchical tree of relation between energy；

This step is to construct a technical ability hierarchical tree, in order to describe the matching degree between each technical ability, is not having work When people has the technical ability strictly mated with required by task technical ability, it is possible to use the most close technical ability goes to mate this task, with Anti-task is distributed unsuccessfully.The technical ability that specific configuration method is had by the technical ability and workman considering required by task, artificially gives Matching relationship between each technical ability, can be nearly completed the task of demand technical ability B as having the workman of technical ability A, then A is the son joint of B Point.

In the present example it is assumed that workman's information and platform data according to reading in can construct a technical ability hierarchical tree as shown in Figure 2. The figure shows the matching relationship between each technical ability in this sub-distribution, as when nobody has technical ability s2, it may be considered that use The most closely located technical ability s3 goes this technical ability of approximate match.

Step 4). technical ability that the technical ability hierarchical tree constructed by integrating step 3 and each workman are had and credit worthiness thereof with And technical ability needed for each subtask, show that each workman can be provided by the value of each subtask；

Workman i is provided that to value Value of subtask j_ijCalculating formula is as follows:

Value_ij=max s ∈ skill (i) (level (i, s) * (1-d (skill (j), s))

$d(s,s^{\′})=\left\{\begin{array}{cc}0,& if(s=s^{\′})\\ \frac{d_{max}-depth\left(lca\right(s,s^{\′}))}{d_{\max }},& otherwise\end{array}\right.$

The technical ability list that wherein skill (i) is had by workman i, skill (j) is the technical ability of subtask j demand, d_maxFor The depth capacity of technical ability hierarchical tree, lca (s, s ') is the minimum common ancestor of s and s ', and depth (s) is deep in this sets of s Degree, d (s, s ') is that technical ability s is away from technical ability s ' distance, (i s) is the grade of skill of technical ability s of workman i to level.

To calculate Value₁₂As a example by, the technical ability that workman 1 is had is s3 and s6, and the technical ability of task 2 demand is s2, d_max= 3, then

Therefore level (1, s3) * (1-d (s3, s2))=0.6, level (1, s6) * (1-d (s6, s2))=0,

Value₁₂Take maximum, therefore Value₁₂=0.6.

Step 5). the history participating in workman's information that integrating step 2 is provided completes task list and carries out pretreatment and carry Take key word information, obtain the key word dictionary of each workman, then the key word dictionary of each workman is compared, draw each Similarity between individual workman；

Assume that the history of workman completes task list as shown in Figure 4, the figure shows every workman before be complete Task names information list, can reflect the history development Experience of every workman.Preprocessed and extract can after key word information Obtaining workman's key word dictionary as shown in Figure 5, the content that the figure shows is the history development Experience of more specifically every workman, So that the similarity degree quantified between each workman.Can be obtained between each workman by the key word dictionary of each workman of comparison Similarity degree is in case solving use afterwards.The purpose of this step is bigger in order to solve one group of similarity degree afterwards as far as possible Workman data support is provided, and history exploitation is experienced one group of similar workman and obviously can preferably have been cooperated task.

Step 6). use genetic algorithm for solving task allocative decision, i.e. corresponding pass between each subtask and many group workmans System；

Step (6.1). initialize population；

Using matrix T (m × n) presentation code, wherein m is subtask number, and n is workman's number, T_ij=1 represents workman i It is assigned in task j, T_ij=0 represents that workman i is not allocated in task j.Set population quantity as 50, give birth at random Become initial population.

Step (6.2). calculate the fitness value of population；

Fitness value is calculated by fitness function, and fitness function is chosen for:

$f\left(T\right)=\underset{j=1}{\overset{m}{\Σ}}(\mathrm{\α}\×J_j(T)+\mathrm{\β}\×S_j(T\left)\right)$

$J_j\left(T\right)=\underset{i=1}{\overset{n}{\Σ}}{T}_{ij}{\mathrm{Value}}_{ij}$

$S_j\left(T\right)=\underset{i=1}{\overset{n}{\Σ}}{T}_{ij}{\mathrm{Similarity}}_{ij}$

${\mathrm{Similarity}}_{ij}=\frac{\underset{k=1}{\overset{n}{\Σ}}{T}_{kj}{\mathrm{Sim}}_{ik}}{\underset{k=1}{\overset{n}{\Σ}}{T}_{kj}}$

Wherein, f (T) is fitness, J_iFor the object function of task j, S_jFor being assigned to the total of one group of workman of task j Similarity, similarity_ijFor the similarity of workman i with the one group of workman being assigned to task j, Sim_ikFor workman i and workman Similarity between k, α β is weight coefficient.

Weight coefficient therein can be according to the personalized need needing to take different values to meet different task of different task Ask.

Step (6.3). perform selection, intersect, the genetic manipulation such as variation, generate a new generation population；

Individuality is selected by the mode using roulette, the select probability=individual fitness of each individuality/the most individual Body adaptive value summation.

For crossover operator, taking crossover probability is 0.8, first randomly generates a 0/1 matrix P consistent with individual UVR exposure structure (m × n), for two individualities performing intersection operation, if P_ij=1, the most do not carry out the intersection operation on corresponding position；If P_ij =0, then the gene on the corresponding position of exchange.

For mutation operator, take mutation probability 0.1, the individuality in population is carried out mutation operation, will the base of its correspondence Because negating.

Step (6.4). judge whether to meet termination condition: be, export optimal solution: no, perform step (6.2)

Iterations adds 1, judges whether to have reached maximum iteration time defined in system simultaneously, if it has, Perform step 7, without reaching, perform step (6.2).

Step 7). carried out the distribution of task according to the scheme drawn in step 6 by mass-rent platform.

Final result should be the individuality that the fitness in final population is the highest, and its form should be a matrix T (m × n), represents Allocation result, it is assumed that final result T is as it is shown on figure 3, the most corresponding allocative decision is:

Task 1: workman 1 task 2: workman 2 task 3: workman 3, workman 4

Final allocation result transfers to mass-rent platform to perform.

Claims (6)

1. the method for allocating tasks based on genetic algorithm under a mass-rent environment, it is characterised in that following steps:

(1) receive after task submitter submits task to and decomposes multiple subtasks, each subtask comprise respectively as Lower feature: task names, required by task technical ability and task budget；

(2) receiving the workman's information participating in this task competitive bidding provided by mass-rent platform, each workman comprises following feature: workman The technical ability being had and grade of skill thereof, workman's credit worthiness, history complete task list and this workman to expected from each subtask Salary；

(3) on mass-rent platform, excavate the performance information of the competitive bidding participant of this task, build relation between each participant's technical ability Technical ability hierarchical tree；

(4) the technical ability hierarchical tree constructed by integrating step 3 is had with each workman technical ability and credit worthiness and each son thereof are appointed The required technical ability of business, show that each workman can be provided by the value of each subtask；

(5) history of participation workman's information that integrating step 2 is provided completes task list and carries out pretreatment and extract key word Information, obtains the key word dictionary of each workman, then is compared by the key word dictionary of each workman, draw each workman it Between similarity；

(6) genetic algorithm for solving task allocative decision is used, i.e. corresponding relation between each subtask and many group workmans；

(7) carried out the distribution of task according to the scheme drawn in step 6 by mass-rent platform.

Method for allocating tasks based on genetic algorithm under a kind of mass-rent environment the most according to claim 1, its feature exists In: the technical ability hierarchical tree described in described step (1), refer to a kind of tree, its each node represents a kind of technical ability, As having technical ability s ' workman can approximate the task of execution demand technical ability s, then s ' is the child node of s.

Method for allocating tasks based on genetic algorithm under a kind of mass-rent environment the most according to claim 1, its feature exists In: the workman i in described step 4 is provided that to value Value of subtask j_ijCalculating formula is as follows:

Value_ij=max_s∈skill(i)(level (i, s) * (1-d (skill (j), s))

$d(s,s^{\′})=\left\{\begin{array}{c}0,if(s=s^{\′})\\ \frac{d_{\max }-depth\left(lca\right(s,s^{\′}))}{d_{\max }},otherwise\end{array}\right.$

Wherein, the technical ability list that skill (i) is had by workman i, skill (j) is the technical ability of subtask j demand, d_maxFor skill Can the depth capacity of hierarchical tree, lca (s, s ') is the minimum common ancestor of s and s ', and depth (s) is deep in this sets of s Degree, d (s, s ') is that technical ability s is away from technical ability s ' distance, (i s) is the grade of skill of technical ability s of workman i to level.

Method for allocating tasks based on genetic algorithm under a kind of mass-rent environment the most according to claim 1, its feature exists In: the idiographic flow of the employing genetic algorithm for solving task allocative decision that described step (6) is told is as follows:

(6.1). initialize population；

(6.2). calculate the fitness value of population；

(6.3). perform selection, intersect, the genetic manipulation such as variation, generate a new generation population；

(6.4). judge whether to meet termination condition: be, export optimal solution: no, perform step (6.2).

Method for allocating tasks based on genetic algorithm under a kind of mass-rent environment the most according to claim 4, its feature exists In: the population that initializes described in described step (6.1) refers to encode the solution of task distribution, its genetic coding matrix T (m × n) represents, T_ij=1 represents that workman i is assigned in task j, T_ij=0 represents that workman i is not allocated in task j Go.

Method for allocating tasks based on genetic algorithm under a kind of mass-rent environment the most according to claim 4, its feature exists In: the fitness value described in described step (6.2) is calculated by fitness function, and fitness function is chosen for:

$f\left(T\right)=\underset{j=1}{\overset{n}{\Σ}}(\mathrm{\α}\×J_j(T)+\mathrm{\β}\×S_j(T\left)\right)$

$J_j\left(T\right)=\underset{j=1}{\overset{n}{\Σ}}{T}_{ij}{\mathrm{Value}}_{ij}$

$S_j\left(T\right)=\underset{j=1}{\overset{n}{\Σ}}{T}_{ij}{\mathrm{Similarity}}_{ij}$

${\mathrm{Similarity}}_{ij}=\frac{\underset{k=1}{\overset{n}{\Σ}}{T}_{kj}{\mathrm{Sim}}_{ik}}{\underset{k=1}{\overset{n}{\Σ}}{T}_{kj}}$

Wherein, f (T) is fitness, J_iFor the object function of task j, S_jFor being assigned to the most similar of one group of workman of task j Degree, similarity_ijFor the similarity of workman i with the one group of workman being assigned to task j, Sim_ikFor workman i and workman k it Between similarity, α β is weight coefficient.