Crowdsourcing test task allocation method based on Kuhn-Munkres

Abstract

The crowdsourcing test task allocation method based on the tester model provided by the invention has a matching degree calculation mechanism based on the tester model and a task allocation mechanism based on Kuhn-Munkres, and can effectively improve the crowdsourcing test quality. The method is characterized in that 1, according to task requirements, proper tester characteristics are selected, and a complete tester model is constructed; 2, defining a matching degree calculation method based on the characteristics of a tester, and taking the matching degree as an estimated value of the test quality; and 3, constructing a task allocation model, converting a public test task allocation problem into a weighted bipartite graph optimal matching problem, and performing task allocation by using a K-M algorithm. The method has the beneficial effects that the portrait of a tester is constructed by analyzing the dimensions such as ability and willingness, and the matchable dimensions of workers and tasks are quantified; workers with high comprehensive evaluation values are effectively screened out by designing a matching degree calculation method; and based on a K-M crowdsourcing test task allocation mode, realizing reasonable allocation of tasks.

Description

Technical Field

[0001] The present invention belongs to the field of software crowdsourcing testing, relates to Web application technology, and in particular to the field of task allocation for functional testing of a crowdsourcing testing platform. The present invention aims to design an effective task allocation strategy for the crowdsourcing testing platform to solve the problems of low defect detection rate and insufficient coverage of testing requirements during the completion of crowdsourcing testing tasks, and to promote the steady development of the crowdsourcing testing platform. Background Art

[0002] Software crowdsourcing testing (crowd testing) refers to outsourcing software testing tasks to a group of independent testers who can use fragmented time to test software in their unique testing environment. Through the application of this model, task publishers can quickly and cost-effectively discover defects, ensure software quality, and collect insights and feedback in real usage scenarios. Crowd testing aims to discover defects in its own design, corresponding functions, and user use through actual testers' real testing experience of the product. Therefore, the testers in actual crowd testing are non-professional testers from all walks of life and various equipment environments; but with the accumulation of testing experience and the requirements of professional testing, crowd testing platforms usually also include professional testers who can conduct multi-frequency and high-quality testing. They can be assigned some targeted tests by the platform, or they can take the initiative to perform tasks of interest. Crowd testing platforms serve many industries, such as media, software, communications and many other professional fields. When the crowd testing platform is efficiently managed, not only can customers choose from a variety of testers, but testers can also take the initiative to conduct more preference tests. For the software itself, the rich tester types accumulated by the testing platform can better meet the requirements of its device diversity and environmental complexity, not only can it find more defects, but it can even provide imagination related to user use. Based on this, in order to improve the efficiency of software testing and reduce the cost of manual testing, most crowd testing activities are carried out on professional crowd testing platforms.

[0003] With the development of crowd-testing platforms, the types of new software involved in crowd-testing activities are also increasing, and the number of tasks and testers involved is also increasing dramatically. It is common to use test management tools (such as JIRA, TestRail, HPALM, etc.) to track the distribution, progress and results of test tasks. It also includes automated testing tools (such as Selenium, Appium, Postman, etc.) to perform some highly repetitive and easy-to-automate test tasks. However, these methods and technologies fail to comprehensively examine the characteristics of testers. In order to ensure the high-quality completion of the overall task, how to evaluate and identify the tester's ability and arrange suitable executors for the test tasks has become a key point and difficulty. In this regard, the present invention will carry out detailed modeling and task allocation method design for the task allocation problem in crowdsourcing testing. Summary of the invention

[0004] In order to achieve the above objectives, the present invention implements a matching degree calculation mechanism based on the tester model and a crowdsourcing test task allocation mechanism based on Kuhn-Munkres. In the matching degree calculation mechanism, in order to reasonably evaluate and distinguish testers, a tester feature set is selected, a time weight function is introduced to define relevant calculations, and a tester evaluation model is constructed. By combining the preference weights of different tasks for features, a matching degree algorithm for calculating the degree of adaptation of tasks and testers is obtained, which is used to estimate the expected completion quality of each task by the tester and provide a basis for task allocation. In the crowdsourcing test task allocation mechanism based on Kuhn-Munkres, in order to improve the overall task completion quality, a crowd testing task allocation model is established with the goal of maximizing the overall matching degree, and the crowd testing task allocation problem is transformed into a weighted bipartite graph optimal matching problem with matching degree as the weight. Improvements are made on the basis of the traditional algorithm, and a crowd testing task allocation method based on Kuhn-Munkres is obtained, which realizes the effective allocation of tasks.

[0005] 1. Matching calculation mechanism based on tester model

[0006] 1.1 Tester Model

[0007] Definition 1: Time weight function

[0008] The half-sine time weight function is used as the fitting curve to reflect the influence weight of historical records within a certain period T on the feature. The expression of the time weight function is:

[0009]

[0010] Among them, T is the time period included in the statistics. Historical data longer than the time period is not considered, and data shorter than the time period T is included in the relevant calculation of the tester's characteristics. t is the time from the start of the period.

[0011] Definition 2 Ability Value

[0012] The ability value represents the tester's evaluation of the ability to perform various types of tasks. For a tester, the ability value is related to the scores of the test reports he submitted in the past. The higher the score of a single report, the higher the total ability value, and the more recent the score, the greater the influence weight on the tester's ability value. j , whose capability value is:

[0013]

[0014] Where N tot Indicates the tester w j The total number of crowdsourcing test tasks participated in, represents the score of the tester's k-th test report, t k Indicates the time from the start of the cycle, weight(t k ) represents the influence weight of the score of the kth report on the existing ability value.

[0015] Definition 3: Credibility

[0016] It indicates the possibility of the tester to perform the assigned task. Each time the tester completes the assigned task, he will receive a reputation reward value. All his historical reputations determine his current reputation, and the more recent the reputation reward, the greater the influence weight on the tester's reputation. j , its credibility is:

[0017] β = 0 or 1

[0018] Among them, N tot Indicates the tester w j The total number of crowdsourcing test tasks participated in, β represents the reputation reward for the last completed task. If the most recent task is qualified, β is 1, otherwise 0, weight(t k ) represents the weight of the influence of the reputation reward of the kth task on the existing reputation.

[0019] Definition 4 Experience Value

[0020] The experience value indicates the proficiency of the tester in writing test reports. The more crowdsourced test reports a tester has passed in the past, the higher his experience value is, and the more recent the pass is, the greater the weight of the tester's experience value is. j , its empirical value is:

[0021]

[0022] Among them, N tot Indicates the tester w j The total number of crowdsourcing test tasks participated in, weight(t k ) represents the weight of the impact of the experience of the k-th task on the existing experience value.

[0023] Definition 3.5 Activity and Activity Factor

[0024] Represents the tester's recent activity level evaluation value. The tester's activity level is related to the number of active days d in the past 30 days and the manually set activity factor σ. j , whose activity is:

[0025]

[0026] Where d represents the tester wj The number of active days in the past month, σ is the activity factor that can be adjusted by the platform, and d, σ∈[0,30]. During the growth of active days, when d<σ, S act The growth rate is slow at first and then fast, gradually approaching 0.5; when d = σ, the activity S act =0.5; when d>σ, S act The growth rate first becomes faster and then slows down, gradually approaching 1. Using d = σ as the boundary of the tester's activity, the crowd testing platform can adjust the expected value of the tester's activity by controlling the value of σ, thereby achieving a certain incentive effect.

[0027] In the actual task allocation process, there will be some newly registered testers in each allocation. They have no records of relevant historical tasks and reports, so the characteristic values ​​of newly registered testers cannot be calculated according to the above formula. Therefore, for newly registered task testers without historical data, this article sets the platform to require these new users to fill in a personal information report. In addition to their basic information, it also includes their own evaluation of their own abilities, including their understanding of various test types and test-related experience (expressed by triangular fuzzy numbers, such as the possible values ​​of the understanding of crowd testing are "not good at, slightly good at, moderately good at, relatively good at, very good at"). The corresponding triangular fuzzy numbers are (0, 0, 0.1), (0.1, 0.3, 0.5), (0.3, 0.5, 0.7), (0.5, 0.7, 0.9), (0.7, 0.9, 1.0), etc., which can be used to calculate the adaptation value of the worker's ability and experience; considering that new users often have a higher probability of participating in the first task due to curiosity and novelty, the initial value of their reputation value is set to 1, and they are encouraged to successfully submit reports; at the same time, the initial value of the new tester's activity is set to 0.5, because in the activity calculation of this article, if the newly registered user continues to be active, his activity will rise rapidly, otherwise it will fall rapidly, in order to distinguish the loyalty of the newly registered user to the crowd testing platform.

[0028] 1.2 Feature weight calculation

[0029] The importance of different features of the tester will vary depending on the task, so when evaluating the suitability of the tester, it is necessary to derive the importance of each feature for each task, that is, the feature weight. The weight of the tester's features is generally set by the task publisher, but in fact, the task publisher may be vague about the progress of task completion and the impact of worker features on the entire crowdsourcing test activity, and it is difficult to clearly give an accurate value. Based on this, the present invention sets the task publisher to describe the relative importance of each feature when publishing the task, and uses the fuzzy analytic hierarchy process (FAHP) to perform weight evaluation on the various features of the crowdsourcing workers (tester), thereby quantifying the fuzzy information and helping the task publisher find the desired tester type.

[0030] 1.1 Matching calculation strategy

[0031] Definition 6: Matching Degree

[0032] Matching, that is, the tester-task matching, indicates the suitability of the tester and the task, and is also an estimate of the expected quality of the task. The higher the matching, the higher the expected quality of the tester's completion of the task. Define the i-th test task u i For the jth tester w j The matching degree is:

[0033]

[0034] in, Represents the i-th test task u i Evaluation weights for tester ability, activity, credibility, and experience. The jth tester w j ability, activity, credibility and experience.

[0035] Based on the calculations in the above chapters, this paper proposes an algorithm for calculating the matching degree between testers and tasks. By combining the historical data of the tester cycle and the test task requirements, the matching degree between each tester and each task is calculated to obtain a matching degree matrix.

[0036] In this algorithm, we first need to determine the calculation cycle T according to the current status of the platform, determine the statistical cycle of historical data, and obtain the time weight function, which is used to calculate the calculation weights of different time data within the cycle. Then collect the relevant historical data of the testers within the cycle T, including the number of crowd-testing tasks each tester participated in, whether they participated after each task was assigned, the score of each task report, the number of qualified reports, the number of active days within 30 days, etc. Then, according to the calculation formula of each feature, calculate the relevant features of each tester, set the value of the active factor according to the platform requirements, and calculate the tester's activity. At the same time, when the task publisher publishes a task, the platform will require the task publisher to evaluate the importance of the four features of the tester according to the type and requirements of this task, and then calculate the weight distribution of the tester's features in each task according to the FAHP method. Finally, calculate the matching degree between each tester and each task to obtain the matching degree matrix.

[0037] 2. Kuhn-Munkres-based crowdsourcing test task allocation mechanism

[0038] 2.1 Crowd-testing task allocation model

[0039] Definition 7 Test Task

[0040] Refers to the software testing crowdsourcing task provided by the task requester, which contains the test object (target system or APP) and test requirements described in natural language. Task u includes the number of people expected to be tested, and also has the evaluation weights of different tester characteristics, which can be expressed as u= money, u time ,u cnt ,ω cap ,ω cre ,ω exp ,ω act >. Among them, u money represents the expected total budget of the task, u time represents the expected completion time of the task, u cnt represents the expected number of testers for this task, ω cap ,ω act ,ω cre ,ω exp They represent the evaluation weights of the task on ability, activity, credibility and experience. The value range of each weight is [0, 1], and the sum of the four is 1, that is, ω cap +ω act +ω cre +ω exp =1.

[0041] Definition 8 Crowdsourced Tester

[0042] ​Testers refer to registered users on the crowd testing platform who are willing to participate in crowdsourcing testing. Through some data calculations, we can characterize the worker portrait w from four dimensions: ability, activity, credibility, and experience value, expressed as w= cap , S cre , S exp , S act > Where S cap represents the tester's ability evaluation value, S act Represents the activity evaluation value of the tester, S cre represents the reputation evaluation value of the tester, S exp Represents the tester's experience evaluation value.

[0043] Because the tester-task matching value is calculated based on the feature values ​​of each tester and the weight of each task to each feature value, the matching value reflects the degree of adaptation between the tester and the task. In other words, the higher the matching value, the higher the possibility of the tester completing the task and the higher the completion quality. Therefore, when the overall matching degree S is maximized as the goal, the crowd testing task allocation problem can be transformed into a global optimization problem. That is, given the crowd testing task set U, the candidate tester set w, and the task-tester matching degree calculation method, from any possible allocation scheme, find an allocation scheme that maximizes the overall matching degree. Based on the above definition and related calculations, the test task allocation problem is reduced to an ILP problem. The decision variable X ij Defined as whether to put the i-th test task u i Assigned to the jth tester w j When X ij =1, it means that the i-th test task u i Assigned to the jth tester w j , when X ij =0 means no allocation. At this time, the target of the crowd testing task allocation model is:

[0044]

[0045] 2.2 Crowd-testing task allocation strategy based on Kuhn-Munkres

[0046] Taking the task-tester matching matrix obtained by the matching algorithm and the expected number of testers of the task as input, we first need to decompose the task and construct the final weighted bipartite graph described in the previous section. Then, based on the idea of ​​depth-first search and augmenting path, we continuously expand the matching to find the largest matching. The workflow of the Kuhn-Munkres method is as follows:

[0047] ​(1) Initialization of the top label value of each vertex. Select a vertex set with a smaller number of vertices as the V1 part of the weighted bipartite graph G, and another vertex set as the V2 part of G. Set the top label value of each vertex in V1 to the maximum weight of all edges associated with the vertex, and set the top label value of each vertex in V2 to 0.

[0048] (2) Find the edge e that meets the conditions ij Construct an equal subgraph and use the Hungarian algorithm to find a complete matching of the equal subgraph. For each vertex in V1, use depth-first search or breadth-first search to find an augmenting path in the equal subgraph.

[0049] (3) If a complete match of the equal subgraph is found, the algorithm ends; otherwise, the vertex values ​​of the relevant vertices are adjusted. The method for modifying the vertex value is as follows: for all vertices in V1 involved in the process of finding the augmenting path, find the edge with the smallest difference between the vertex values ​​at both ends and the weight of the edge in the bipartite graph, set the difference value to d, reduce the vertex values ​​of all vertices in V1 by d, and increase the vertex values ​​of all vertices in V2 involved by d. Each time the vertex value is adjusted, the edges that originally belonged to the equal subgraph will still belong to the equal subgraph after the adjustment, thereby expanding the equal subgraph and giving the equal subgraph a chance to find a complete match.

[0050] (4) Repeat steps (2) and (3) until a complete match is found. Continue to search for augmenting paths. When each point can find an augmenting path, each point is in the match, that is, a complete match of the bipartite graph is found. This complete match is the maximum weighted match of the bipartite graph.

[0051] Solve the final weighted bipartite graph ( Figure 3 ), that is, the tester allocation result corresponding to each subtask, is an expanded initial 0-1 allocation matrix. Then it is necessary to traverse the tasks, merge the matrices according to the expected number of people for each task, and output the final 0-1 allocation matrix. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 The architecture diagram of the matching calculation strategy based on the tester model

[0053] Figure 2 The workflow diagram of fuzzy hierarchy process

[0054] Figure 3 is a weighted bipartite graph DETAILED DESCRIPTION

[0055] In order to better understand the technical content of the present invention, specific embodiments are given and described as follows in conjunction with the accompanying drawings.

[0056] Figure 1The architecture diagram of the matching degree calculation strategy based on the tester model includes the following steps:

[0057] Step 1: Set the time period T, calculate the time weight function wight(t), and set the activity factor σ;

[0058] Step 2: For each candidate tester w j ∈W;

[0059] Step 3: Determine the tester w j Whether it is the first allocation;

[0060] Step 4: If yes, calculate the tester w j Ability value, reputation value, experience value and activity level;

[0061] If not, set the initial characteristic value for it;

[0062] Step 5: For each task t to be assigned i ∈T;

[0063] Step 6: Calculate the test subject feature weight (ω cap ,ω act ,ω cre ,ω exp );

[0064] Step 7: For each task t to be assigned i ∈T, each candidate tester w j ∈W;

[0065] Step 8: Calculate the matching degree

[0066] Step 9: Return to the task-tester matching matrix.

[0067] Figure 2 This is the workflow diagram of the fuzzy hierarchy process. When using FAHP to calculate the tester's feature weights, the specific steps include:

[0068] Step 1: Build a hierarchy of decision problems: Identify the primary objectives and break them down into a hierarchy of criteria and alternatives.

[0069] Step 2: Identify the criteria and alternatives to be evaluated: Identify the specific criteria and alternatives that will be used in the decision-making process.

[0070] Step 3: Create a pairwise comparison matrix for criteria and alternatives: Construct a matrix that compares the importance of each criterion and alternative relative to the others.

[0071] Step 4: Use linguistic terms to express relative importance in the matrix: Use linguistic terms (such as "equal", "slightly more important", "very important", etc.) to express the relative importance between the elements in the matrix.

[0072] Step 5: Convert the linguistic terms into fuzzy numerical values ​​using appropriate conversion methods: Convert the linguistic terms into fuzzy numerical values ​​to express the degree of importance more quantitatively.

[0073] Step 6: Calculate the weighted normalized fuzzy matrix of the criteria and alternatives: Calculate the weights of the criteria and alternatives based on the fuzzy matrix.

[0074] Step 7: Calculate the overall priority of the criteria and alternatives: Combine the weights to determine the overall priority of the criteria and alternatives.

[0075] Step 8: Check the consistency of the comparison: Verify the consistency of the comparison through mathematical checks to ensure the reliability of the results.

[0076] Step 9: Conduct sensitivity analysis to assess the robustness of the results: Analyze how changes in input parameters affect the results and assess the stability and reliability of the decision-making process.

[0077] Figure 3 The final weighted bipartite graph obtained by using the Kuhn-Munkres method has its specific working method described in detail in "2.2 Crowd-testing task allocation strategy based on Kuhn-Munkres" of the invention content, and will not be described here in detail.

[0078] The present invention involves several key technologies, which are described as follows:

[0079] 1.Angular2

[0080] Angular 2 is an open source JavaScript framework that aims to build web applications using HTML and JavaScript technology stacks. It is considered a technology approach that prioritizes mobile devices. The framework uses server-side rendering technology to achieve fast loading of views on mobile devices, and supports ECMAScript and other programming languages ​​that can be compiled into JavaScript. Through the dependency injection mechanism, Angular 2 ensures the maintainability of the application and fully adopts a component-based development paradigm. In terms of building the front end, the platform chose the Angular 2 framework, and its componentized design concept makes it easier for the front-end module to be seamlessly integrated with the front-end system of the Muche main site. In addition, the use of Angular 2 also brings about an increase in the speed of view rendering, thereby further optimizing the user experience.

[0081] 2. Echarts

[0082] ECharts is an open source visualization library based on JavaScript. It is designed to run smoothly on both PC and mobile devices and is widely compatible with mainstream browsers on the market. The underlying layer of the library relies on the lightweight and efficient vector graphics engine ZRender, which can provide users with intuitive, interactive and highly customized data visualization charts. In the scene application of the crowdsourcing platform, both the guided visualization diagram and the tree structure diagram are built with ECharts. These visualization charts are not only concise and clear in design, but also intuitive, which greatly improves the convenience and efficiency of users viewing data. By leveraging the powerful functions of ECharts, the crowdsourcing platform is able to show users a clearer and more intuitive data view, thereby further optimizing the user experience.

[0083] 3. SpringBoot

[0084] SpringBoot is a lightweight, non-invasive and highly integrated application development framework, which is designed to simplify the development process of enterprise-level applications. The framework has a one-stop feature and supports modular development, which provides great convenience for developers. In the back-end construction of the crowdsourcing platform, SpringBoot is used as the core framework. With its powerful functions, the platform can efficiently respond to front-end requests and easily integrate key modules such as log management, Redis cache, and MongoDB database. This design not only improves the overall performance of the platform, but also enhances the scalability and maintainability of the system.

[0085] 4. MongoDB

[0086] MongoDB is a product between relational databases and non-relational databases. It has the most comprehensive functions among non-relational databases and is closest to the characteristics of relational databases. The data structure supported by this database is extremely flexible, using the Bson format similar to JSON, which enables it to store complex and diverse data types. In the back-end architecture of the crowdsourcing platform, MongoDB is used to store important information such as collaborative data. With its efficient access speed and excellent concurrent processing capabilities, MongoDB provides powerful data storage support for the platform. This design not only improves the data processing efficiency of the platform, but also ensures stable operation in high-concurrency scenarios.

[0087] In summary, the present invention determines the appropriate tester characteristics, provides an analysis, and provides a method for calculating the feature weights; considering that the historical task completion status at different time intervals should have different influences on the tester's current feature evaluation, a historical weight function is introduced, and a series of feature calculations for the task tester are performed based on this function. At the same time, in the tester activity calculation, an manually adjustable activity factor is set, and the tester's activity in recent period of time is evaluated by a nonlinear function, and a complete tester model is established; then, based on the tester model, a matching method is given to estimate the tester's expected completion quality for the task; further, the crowd-testing task allocation problem is transformed into an optimal matching problem of a weighted bipartite graph with matching as the weight, and the task is decomposed according to the multi-personnel matching requirements of the task, and a weighted bipartite graph is constructed and expanded. Based on this bipartite graph combined with the ideas of depth-first search and augmenting path, a crowd-testing task allocation method based on KM is obtained. This method takes the maximum total matching degree as the optimization goal and is committed to improving the overall completion quality of the crowd-testing task. The platform backend is built in the form of microservices, and the frontend coordinates and interacts with the backend of the Muche main site and the backend of the crowdsourcing platform to create a high-performance and highly available crowdsourcing testing platform.

Claims

1. A crowdsourcing test task allocation method based on Kuhn-Munkres, characterized by According to the characteristics and requirements of the crowd testing task, appropriate tester characteristics are selected, and combined with the evaluation of the importance of tester characteristics by the task publisher, a complete tester model is constructed to distinguish testers of different qualities. A matching algorithm for calculating the degree of task-tester adaptation is proposed to estimate the expected completion quality of the tester. According to the crowd-testing task allocation restrictions, a crowd-testing task allocation model without task priority is established with the goal of maximizing the overall matching degree. Then, in the solution, the task allocation problem with the maximum comprehensive matching degree is modeled as a bipartite graph optimal matching problem. The tasks are decomposed and an extended weighted bipartite graph is constructed. After solving and merging using the Kuhn-Munkres algorithm, a crowd-testing task allocation method based on Kuhn-Munkres is obtained, which ensures the degree of adaptation between tasks and testers and improves the overall task allocation quality, including the following steps: 1) Considering the changes of tester characteristics over time, a time weight function is introduced, and a calculation method for the tester's ability, credibility, and activity experience value is designed; 2) Introduce active factors and statistical cycles to calculate the matching degree between each task and each tester; 3) According to the characteristics of crowd-testing task allocation problem, a series of tester characteristics and related optimization goals, a crowd-testing task allocation model based on matching degree is established; 4) Based on the Kuhn-Munkres algorithm, a reasonable allocation between tasks and testers is achieved and the allocation matrix is ​​output.

2. According to the method for allocating crowdsourcing test tasks based on Kuhn-Munkres in claim 1, the method captures the tester's ability, credibility, experience, activity and other characteristics, and adopts the fuzzy analytic hierarchy process (FAHP) to evaluate the weights of various characteristics of crowdsourcing workers (tester) when determining the weights, thereby quantifying fuzzy information, helping the task publisher to find the desired tester type, and decomposing the problem in an orderly manner according to the levels to form a bottom-up pyramid structure; considering the virtuality and spatio-temporal asymmetry of the crowd testing platform, a crowd testing personnel feature model is established to screen out low-quality workers, thereby achieving the matching optimization between crowd testing workers and tasks and ensuring the final task completion quality.

3. The method for allocating crowdsourcing test tasks based on Kuhn-Munkres according to claim 1 is characterized in that Determination of matching degree. Matching degree, i.e., tester-task matching degree, indicates the suitability between the tester and the task, and is also an estimate of the expected completion quality of the task. The tester's ability, credibility, experience and activity are selected in the calculation of matching degree, and these indicators are all related to the expected completion quality of the task. The tester's feature weight is dynamically calculated through the task publisher's score. For each task, the task-tester matching degree is calculated by the weighted average of the four indicator values.

4. The method for allocating crowdsourcing test tasks based on Kuhn-Munkres according to claim 1, characterized in that Considering the number of people constraints, execution constraints, etc., among all the test task execution plans that meet the requirements, the execution plan with the highest task completion quality is selected. The appropriateness of task allocation and the final completion quality of the task are the main considerations. In order to assign tasks to appropriate testers, the crowd testing platform will collect the basic information and historical data of the testers before assignment, and conduct a brief evaluation based on these data, and then calculate the suitability between the tester and each task.

5. The method for allocating crowdsourcing test tasks based on Kuhn-Munkres according to claim 1, characterized in that The crowd-testing task allocation problem is transformed into a weighted bipartite graph optimal matching problem with matching degree as weight. Unlike the traditional optimal matching problem, the crowd-testing task needs to be completed by multiple testers, so it is necessary to construct a modified and expanded weighted bipartite graph, apply the Kuhn-Munkres algorithm to solve the task allocation, and finally merge the allocation results to improve the traditional algorithm.

Images