Effect evaluation method and device based on multi-agent virtual community simulation

Abstract

The invention relates to an effect evaluation method and device based on multi-agent virtual community simulation. The method comprises the following steps: constructing an NAA model; the NAA model comprises agents representing virtual community members, products representing community tasks and schemes and specifications representing a virtual community incentive mechanism; constructing a virtual community model and an incentive mechanism model according to the NAA model; and according to the virtual community model and the incentive mechanism model, constructing a community calculation model, and according to the simulated global emergence of the community calculation model and the real emergence of a real community, evaluating the trend similarity between the simulated global emergence and the real emergence of the real community by adopting a Pearson's correlation coefficient, and evaluating and simulating a numerical value difference between global emergence and real emergence of a real community by adopting an average value approximation degree, and obtaining global emergence similarity according to the trend similarity and the numerical value difference so as to perform effect evaluation on virtual community simulation. By adopting the method, virtual community simulation effect evaluation can be realized.

Description

Technical Field

[0001] The present application relates to the technical field of virtual community simulation, and in particular to a method and device for effect evaluation based on multi-agent virtual community simulation. Background Art

[0002] In today's digital age, the rapid development of internet technology has led to the emergence of a large number of virtual communities, such as knowledge Q&A communities, online education communities, and e-commerce platforms. These communities have become crucial in people's lives, work, and learning. These virtual communities provide convenient spaces for communication and collaboration among people with similar interests and needs, greatly promoting information sharing and knowledge dissemination.

[0003] However, current virtual community simulation technology faces numerous significant challenges. Regarding abstract model construction, most existing research fails to fully and accurately encompass the key elements of virtual communities. The portrayal of community members remains superficial, considering only a limited number of basic attributes and completely ignoring the complex and ever-changing behavioral patterns of members within the community and the psychological drivers behind them. Specifically, there is a lack of in-depth analysis of the dynamics of member behavior under different incentive scenarios and the deeper interactions between members. The representation of community tasks and contributions is based on simple definitions, omitting key factors such as the diversity of tasks, the quality of contributions, and their actual value. Consequently, it fails to truly reflect the flow of tasks and the accumulation and evolution of contributions in real communities. Incentive mechanism modeling, in particular, is overly crude and general, lacking a thorough understanding of how incentive mechanisms influence member behavior and decisions, ultimately leading to complex changes in the overall community's operations. This makes it difficult to capture the core role and critical importance of incentive mechanisms in virtual community ecosystems.

[0004] Problems are equally prominent at the computational model and simulation algorithm level. Many computational models suffer from serious mapping inaccuracies when translating abstract concepts into executable simulation programs. This leads to significant deviations between simulation results and actual community operations, failing to truly reflect the true state of virtual communities. Simulation algorithm design often fails to adequately consider the dynamic and random nature of communities, resulting in unsatisfactory performance when handling complex situations such as concurrent behavior of community members, random assignment of tasks, and real-time feedback from incentive mechanisms. Summary of the Invention

[0005] Based on this, it is necessary to provide an effect evaluation method and device based on multi-agent virtual community simulation to address the above technical problems.

[0006] A method for effect evaluation based on multi-agent virtual community simulation, the method comprising:

[0007] Constructing an NAA model; the NAA model includes agents representing virtual community members, artifacts representing community tasks and plans, and specifications representing virtual community incentive mechanisms;

[0008] Based on the NAA model, a virtual community model and an incentive mechanism model are constructed. The virtual community model includes a community member model, a contribution model, and agent behavior rules. The community member model is composed of the agents, which include agent attributes and agent behaviors for participating in community activities. The contribution model is composed of the artifacts and is used to enable interaction and collaboration among community members. The agent behavior rules are used to formulate rules for participating in community contributions, and the rules are related to the level of influence of the incentive mechanism. The incentive mechanism model includes context, trigger conditions, and reward and punishment actions. The reward and punishment actions are generated by the context and trigger conditions to represent the norms.

[0009] A community calculation model is constructed based on the virtual community model and the incentive mechanism model. Based on the simulated global emergence of the community calculation model and the real emergence of the real community, the Pearson correlation coefficient is used to evaluate the trend similarity between the simulated global emergence and the real emergence of the real community. The average value approximation is used to evaluate the numerical difference between the simulated global emergence and the real emergence of the real community. The global emergence similarity is obtained based on the trend similarity and the numerical difference, and the effect of the virtual community simulation is evaluated based on this.

[0010] In one embodiment, the community member model includes the agent's incentive value, professional ability, contribution activity, review activity, task proposal rate, and like rate. The contribution model includes artifact type, creating agent, creation days, number of solutions, likes, dislikes, and quality. The agent behavior rules include task proposal rules, solution contribution rules, likes, dislikes, and update rules.

[0011] In one embodiment, the task-proposing rule is:

[0012]

[0013] Among them, AG represents the community member model, ag represents the members in the community member model, q represents the proposed task, Q represents the proposed task set, PQ(ag,pq) represents the level of influence of the specification on the current task-proposing behavior of the agent ag is pq, QR(ag,qr) represents the task-proposing rate of the agent ag is qr, and CR(ag,p) represents the creation of the product p by the agent ag.

[0014] In one embodiment, the contribution solution rule is:

[0015]

[0016] Among them, a represents the scheme of the contribution model, A represents the scheme set of the contribution model, PA(ag,pa) represents the level of influence of the specification on the current contribution solution behavior of agent ag is pa, QR(ag,qr) represents the task proposal rate of agent ag is qr, and CR(ag,p) represents the creation of artifact p by agent ag.

[0017] In one embodiment, the like rule is:

[0018]

[0019] Among them, p represents the product, ART represents the contribution model, PU(ag,pu) represents the influence level of the norm on the current like behavior of agent ag is pu, UR(ag,ur) represents the like rate of agent ag is ur, and UV(ag,p) represents the like of agent ag to product p.

[0020] In one embodiment, the click rule is:

[0021]

[0022] Where p represents the product, ART represents the contribution model, PD(ag,pd) represents the level of influence of the norm on the current dislike behavior of agent ag, which is pd, UR(ag,ur) represents the like rate of agent ag, which is ur, and DV(ag,p) represents whether agent ag dislikes or cancels the like of product p.

[0023] In one embodiment, the update rule is:

[0024]

[0025] Among them, REW(ag,rt,rew) indicates that the rt type incentive value of agent ag is rew, EXP(ag,exp) indicates that the professional ability of agent ag is exp, PAC(ag,pac) indicates that the contribution activity of agent ag is pac, VAC(ag,vac) indicates that the average daily evaluation number of agent ag is vac, QR(ag,qr) indicates that the task proposal rate of agent ag is qr, UR(ag,ur) indicates that the like rate of agent ag is ur, f i is a function that represents the relationship between the agent’s incentive value and its other attributes, f i (rew)=a i ×rew+b i ,i∈{1,2,3,4,5}.

[0026] In one embodiment, the community computing model includes a virtual community simulation algorithm, a community data sampling algorithm, and an agent behavior decision algorithm; wherein, the virtual community simulation algorithm outputs a community member model and a set of artifacts by inputting real community member information, a description of the relationship between incentive values ​​and agent attributes, an incentive mechanism parameter set, an incentive value sparse segmentation point, a real community sampling ratio, a new agent joining rate, and a maximum number of simulations; the community data sampling algorithm outputs a community member model and specifications by inputting real community user information, an incentive mechanism parameter set, an incentive value sparse segmentation point, and a sampling ratio value; and the agent behavior decision algorithm outputs an affected set of artifacts by inputting an agent, an artifact set, and a specification.

[0027] In one embodiment, the method further includes: using the Pearson correlation coefficient to evaluate the trend similarity between the simulated global emergence and the real emergence of the real community:

[0028]

[0029] Among them, S and M are the simulated global emergence of the real community and the computational model, and the real emergence of the real community, respectively. i and M i are the i-th component of the simulated global emergence and the real emergence of the real community, respectively;

[0030] The numerical difference between the simulated global emergence and the real emergence of the real community is evaluated using the average approximation:

[0031]

[0032] The global emergence similarity obtained according to the trend similarity and the numerical difference is:

[0033] sim(S,M)=λ×pcc(S,M)+(1-λ)×ava(S,M).

[0034] An effect evaluation device based on multi-agent virtual community simulation, the device comprising:

[0035] NAA model construction module, used to construct the NAA model; the NAA model includes intelligent agents representing virtual community members, artifacts representing community tasks and plans, and specifications representing virtual community incentive mechanisms;

[0036] A virtual community model and incentive mechanism model construction module is used to construct a virtual community model and an incentive mechanism model based on the NAA model. The virtual community model includes: a community member model, a contribution model, and agent behavior rules. The community member model is composed of the agents, which include agent attributes and agent behaviors for participating in community activities. The contribution model is composed of the artifacts and is used to enable interaction and collaboration among community members. The agent behavior rules are used to formulate rules for participating in community contributions, and the rules are related to the impact level of the incentive mechanism. The incentive mechanism model includes: context, trigger conditions, and reward and punishment actions. The reward and punishment actions are generated by the context and trigger conditions to represent the norms.

[0037] The effect evaluation module is used to construct a community calculation model based on the virtual community model and the incentive mechanism model, and use the Pearson correlation coefficient to evaluate the trend similarity between the simulated global emergence of the community calculation model and the real emergence of the real community, and use the average value approximation to evaluate the numerical difference between the simulated global emergence and the real emergence of the real community. The global emergence similarity is obtained based on the trend similarity and the numerical difference, so as to evaluate the effect of the virtual community simulation.

[0038] The aforementioned multi-agent virtual community simulation effectiveness evaluation method and device, based on model construction, proposes a NAA model that abstracts community members into agents, community tasks and contributions into artifacts, and incentive mechanisms into norms, establishing a close connection between these three elements. This provides a clear framework for a deeper understanding of the workings of virtual communities. This enables a more accurate understanding of the drivers of community member behavior, the flow of tasks and contributions, and the operation of incentive mechanisms, addressing the shortcomings of previous models that have insufficiently integrated these key elements. During the community computational model development phase, a comprehensive simulation algorithm system, including a virtual community simulation algorithm, a community data sampling algorithm, and an agent behavior decision-making algorithm, was implemented. These technical approaches effectively transformed the abstract model into a runnable simulation environment, significantly improving the accuracy and efficiency of the simulation. During the simulation effectiveness evaluation phase, comprehensive metrics such as the Pearson correlation coefficient, mean approximation, and global emergent similarity were used to comprehensively and accurately measure the degree of fit between the simulation results and the real community. This provides a reliable quantitative basis for continuous model improvement and overcomes the limitations of previous evaluation methods, which were limited by their singleness and incompleteness. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 1 is a flow chart of an effect evaluation method based on multi-agent virtual community simulation in one embodiment;

[0040] Figure 2 Schematic diagram of an NAA model in one embodiment;

[0041] Figure 3 is a schematic diagram of an incentive mechanism model in one embodiment;

[0042] Figure 4 is a pseudo code diagram of a virtual community simulation algorithm in one embodiment;

[0043] Figure 5 A pseudo code diagram of a community data sampling algorithm in one embodiment;

[0044] Figure 6 A pseudo code diagram of a behavior decision algorithm of an intelligent agent in one embodiment;

[0045] Figure 7 This is a structural block diagram of an effect evaluation device based on multi-agent virtual community simulation in one embodiment. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0047] In one embodiment, Figure 1 As shown, a method for effect evaluation based on multi-agent virtual community simulation is provided, comprising the following steps:

[0048] Step 102: construct a NAA model.

[0049] The NAA model includes agents representing virtual community members, artifacts representing community tasks and plans, and specifications representing the virtual community's incentive mechanism.

[0050] like Figure 2 As shown, the agent in AG makes behavioral decisions based on its own state and available rewards and punishments. x ,Through actions, we create artifacts in ART or modify its state.,NORM,rewards or punishes the agent of AG according to the state change of ART,,changes its properties, and thus affects its behavior.

[0051] Step 104: construct a virtual community model and an incentive mechanism model based on the NAA model.

[0052] The virtual community model includes: a community member model, a contribution model, and agent behavior rules; the community member model is composed of the agents, which include agent attributes and agent behaviors that participate in community activities; the contribution model is composed of artifacts and is used to achieve interaction and collaboration among community members; the agent behavior rules are used to formulate rules for participating in community contributions, and the rules are related to the level of influence of the incentive mechanism; the incentive mechanism model includes: context, trigger conditions, and reward and punishment actions, and the norms are expressed by generating reward and punishment actions through context and trigger conditions.

[0053] Step 106: Based on the virtual community model and the incentive mechanism model, a community calculation model is constructed. Based on the simulated global emergence of the community calculation model and the real emergence of the real community, the Pearson correlation coefficient is used to evaluate the trend similarity between the simulated global emergence and the real emergence of the real community. The average value approximation is used to evaluate the numerical difference between the simulated global emergence and the real emergence of the real community. The global emergence similarity is obtained based on the trend similarity and the numerical difference, and the effect of the virtual community simulation is evaluated.

[0054] In the aforementioned multi-agent virtual community simulation effectiveness evaluation method, the proposed NAA model abstracts community members as agents, community tasks and contributions as artifacts, and incentive mechanisms as norms. This model establishes a close connection between these three elements, providing a clear framework for a deeper understanding of how virtual communities work. This allows for a more accurate understanding of the drivers of community member behavior, the flow of tasks and contributions, and the workings of incentive mechanisms, addressing the shortcomings of previous models that often lack integration of these key elements. During the community computational model development phase, a comprehensive simulation algorithm system, including a virtual community simulation algorithm, a community data sampling algorithm, and an agent behavior decision-making algorithm, was developed. These technical approaches effectively transformed the abstract model into a runnable simulation environment, significantly improving the accuracy and efficiency of the simulation. During the simulation effectiveness evaluation phase, comprehensive metrics such as the Pearson correlation coefficient, mean approximation, and global emergent similarity were used to comprehensively and accurately measure the degree of fit between simulation results and the real community. This provides a reliable quantitative basis for continuous model improvement and overcomes the limitations of previous evaluation methods, which were limited by their single nature and lacked comprehensiveness.

[0055] In one embodiment, the community member model includes the agent's incentive value, professional ability, contribution activity, evaluation activity, task proposal rate, and like rate.

[0056] Specifically, based on the NAA model, AG can represent a collection of agents that are members of a virtual community. Each agent (ag∈AG) has its own attributes and behaviors. This application focuses on the four types of behaviors that are most relevant to the core assets of the virtual community (tasks and solutions): proposing tasks, contributing solutions, evaluating solutions, and updating solutions. At the same time, it focuses on the internal states of members that affect these behaviors, including incentive values, activeness of contributions and evaluations, contribution preferences, and evaluation preferences. Therefore, ag is defined as a tuple ag=<rew,exp,pac,vac,qr,ur> .

[0057] 1) rew represents the incentive value of agent ag. It can be reputation, badges, or status level, etc.

[0058] 2) exp represents the professional ability of the agent ag. Its level is measured using a value between 0 and 1, that is, exp∈[0,1].

[0059] 3) PAC represents the contribution activity of agent ag. It is measured by its average daily contribution. PAC ≥ 0.

[0060] 4) vac represents the evaluation activity of agent ag. It is measured by its average daily evaluation number. vac ≥ 0.

[0061] 5) qr represents the agent's task proposal rate, reflecting its preference for task proposals. It is measured as the ratio of the agent's proposed tasks to its total contributions (including proposed tasks and solutions). qr∈[0,1]. Accordingly, the agent's solution proposal rate (task solution rate) is 1-qr. Furthermore, qr and 1-qr represent the overall strength of the incentive mechanism's influence on the agent's task proposal and task solution behaviors, respectively. In other words, in most cases, the incentive mechanism influences the agent's level of influence on these two types of behaviors.

[0062] 6) ur represents the agent's like (positive evaluation) rate, reflecting its preference for likes and dislikes. It is measured as the ratio of the agent's like votes to its total votes (including both likes and dislikes). ur∈[0,1]. Correspondingly, the agent's dislike (negative evaluation) rate is 1-ur. Furthermore, ur and 1-ur represent the overall strength of the incentive mechanism's influence on the agent's like and dislike behaviors, respectively. In other words, in most cases, the incentive mechanism influences the agent's level of influence on these two types of behaviors.

[0063] In one embodiment, the contribution model includes: product type, creating agent, creation days, number of solutions, number of likes, number of dislikes, and quality.

[0064] Specifically, community member contributions in virtual communities are community content in the form of text, images, audio, and video. This application categorizes this content into two categories: the first is the tasks to be solved, such as daily life problems and software requirements; the second is the solutions to these tasks. Contributions in virtual communities can only be created and modified by community members and serve as a medium for interaction and collaboration among community members. Therefore, this invention uses artifacts to represent community contributions.

[0065] According to the contribution category, the artifact set ART is the union of tasks Q and solutions A, that is, ART = Q∪A. For any artifact (task or solution) art∈ART, define art=<c,ag,t,na,nu,nd,qu> .

[0066] 1) c represents the type of artifact p. c∈{0,1}, where 0 and 1 represent tasks and solutions, respectively.

[0067] 2) ag represents the agent that creates artifact p.

[0068] 3) t represents the number of days since the creation of the product p, i.e., the product age (Age). On the day when the product p is created, t=0.

[0069] 4) na represents the number of solutions for product p. If product p is a solution, na = -1.

[0070] 5) nu represents the number of likes for product p.

[0071] 6) nd represents the number of times product p has been stepped on.

[0072] 7) qu represents the quality of product p.

[0073] In one embodiment, the agent behavior rules include: task proposal rules, solution contribution rules, like rules, dislike rules, and update rules.

[0074] Specifically, the task rules are proposed as follows:

[0075]

[0076] Among them, AG represents the community member model, ag represents the members in the community member model, q represents the proposed task, Q represents the proposed task set, PQ(ag,pq) represents the level of influence of the specification on the current task-proposing behavior of the agent ag is pq, QR(ag,qr) represents the task-proposing rate of the agent ag is qr, and CR(ag,p) represents the creation of the product p by the agent ag.

[0077] In this embodiment, community members participate in the community by proposing tasks, seeking solutions to the tasks they are interested in. To describe the task proposals submitted by members, this application introduces pq to represent the level of influence of the incentive mechanism on the current task proposal behavior of agent ag. The time at which a member expects to receive community help is random. Therefore, this application uses a random number between 0 and 1 to represent the level of influence of the incentive mechanism on a member's task proposal behavior at a certain moment, that is, pq∈[0,1]. When agent ag's pq is greater than qr, that is, the level of influence of the incentive mechanism on its current task proposal behavior exceeds its overall level, agent ag proposes a question.

[0078] In one embodiment, the contribution solution rule is:

[0079]

[0080] Among them, a represents the scheme of the contribution model, A represents the scheme set of the contribution model, PA(ag,pa) represents the level of influence of the specification on the current contribution solution behavior of agent ag is pa, QR(ag,qr) represents the task proposal rate of agent ag is qr, and CR(ag,p) represents the creation of artifact p by agent ag.

[0081] Specifically, this application uses pa to represent the level of influence of the incentive mechanism on the current solution contribution behavior of agent ag. When it is greater than its overall level (answer rate) 1-qr, agent ag contributes a solution. The level of influence of the incentive mechanism on a member's solution contribution behavior, pa, is related to the potential benefits of contributing solutions. The greater the benefits a member receives, the higher the level of influence of the incentive mechanism on their solution contribution behavior. The potential benefits of contributing solutions depend on both the rewards received and the costs incurred. In terms of benefits, the potential rewards for a member's solution contribution are first closely related to the quality of the solution. The higher the solution quality qu, the more likely it is to receive likes, thereby earning more reputation points or badge rewards. The generation of high-quality solutions is closely related to the member's expertise level exp and the individual effort e. Therefore, this application represents solution quality as qu = φ(exp, e). Secondly, the reward for contributing solutions is related to the number of existing solutions n for the task. The more solutions a task has, the less attention subsequent solutions receive, and the more difficult it is to earn rewards. Thirdly, the reward for a member's solution is related to the time t it takes to solve the task. Time t refers to the time from when the task is proposed to when the member contributes a solution to it. The larger t is, the lower the task is ranked on the content page, and the harder it is for the solution to the task to gain attention and rewards. Therefore, the member’s solution benefit is expressed as g A(n, φ(exp, e), t). The cost of a solution depends primarily on the number of solutions n for the task and the individual effort e expended by the member, denoted in this application as c(n, e). This is because contributing a solution to a task with existing solutions requires browsing previous solutions, resulting in a higher cost than for tasks with no solutions or few solutions. Furthermore, members prioritize solution quality, and the greater the individual effort e expended, the greater the cost. In summary, the degree to which a member's solution contribution behavior is influenced by the incentive mechanism can be expressed as:

[0082] p a =η(-c(n,e)+g A (n,φ(exp,e),t)).

[0083] η is a function that maps member benefits to the level affected by the incentive mechanism.

[0084] In one embodiment, the like rule is:

[0085]

[0086] Among them, p represents the product, ART represents the contribution model, PU(ag,pu) represents the influence level of the norm on the current like behavior of agent ag is pu, UR(ag,ur) represents the like rate of agent ag is ur, and UV(ag,p) represents the like of agent ag to product p.

[0087] Specifically, this application uses pu to represent the level at which the current like behavior of agent ag is affected by the incentive mechanism. When it is greater than the level (like rate) ur at which its overall like behavior is affected by the incentive mechanism, agent ag will like a product. Among them, the level pu at which a member's like behavior is affected by the incentive mechanism is related to the possible benefits of liking. The greater the benefits obtained by the member, the higher the level at which the like behavior is affected by the incentive mechanism. The possible benefits of liking depend on the reward R obtained. U (such as badges) and expenses paid C U Therefore, this application uses the following formula to express the degree to which the member's like behavior is affected by the incentive mechanism. It is a function that maps member benefits to the level at which the like behavior is affected by the incentive mechanism.

[0088]

[0089] In one embodiment, the click-and-thumb rules are:

[0090]

[0091] Where p represents the product, ART represents the contribution model, PD(ag,pd) represents the level of influence of the norm on the current dislike behavior of agent ag, which is pd, UR(ag,ur) represents the like rate of agent ag, which is ur, and DV(ag,p) represents whether agent ag dislikes or cancels the like of product p.

[0092] Specifically, this application uses pd to represent the level of influence of the incentive mechanism on the current downvoting behavior of the agent ag. When it is greater than the level of influence of the incentive mechanism on its overall downvoting behavior (downvoting rate) 1-ur, the agent ag will downvote a contribution. Similar to the level of influence of the incentive mechanism on the like behavior, the level of influence of the incentive mechanism on the member downvoting behavior pd is related to the possible benefits of downvoting. The possible benefits of downvoting depend on the reward R D and the expenses paid C D Therefore, the following formula is used to express the degree to which the member's click-and-dump behavior is affected by the incentive mechanism. It is a function that maps member benefits to the level at which the click-and-tick behavior is affected by the incentive mechanism.

[0093]

[0094] In one embodiment, the update rule is:

[0095]

[0096] Among them, REW(ag,rt,rew) indicates that the rt type incentive value of agent ag is rew, EXP(ag,exp) indicates that the professional ability of agent ag is exp, PAC(ag,pac) indicates that the contribution activity of agent ag is pac, VAC(ag,vac) indicates that the average daily evaluation number of agent ag is vac, QR(ag,qr) indicates that the task proposal rate of agent ag is qr, UR(ag,ur) indicates that the like rate of agent ag is ur, f i is a function that represents the relationship between the agent’s incentive value and its other attributes, f i (rew)=a i ×rew+b i ,i∈{1,2,3,4,5}.

[0097] Specifically, the degree to which the contribution behavior of community members is affected by the incentive mechanism changes with the change of status (corresponding to its incentive value), thereby affecting the change of its behavior pattern. This application uses the agent's update rule to simulate the change of member behavior pattern. In addition, f iThese are represented as simple functions that depict the impact of incentive rewards on member attributes. For example, pac = f1(rew) represents the relationship between a member's incentive value, rew, and their contribution activity, pac. As members continue to receive incentive rewards, their attributes change, influencing the evolution of their behavior patterns.

[0098] In one embodiment, Figure 3 As shown in Figure 1, the incentive mechanism model includes: context, trigger conditions, and reward and punishment actions. The context and trigger conditions are used to generate reward and punishment actions to express norms. Specifically:

[0099] 1) Context. The implementation context of the incentive mechanism refers to the community environment requirements for the incentive mechanism, that is, the community contributions for which the incentive mechanism is implemented. This application categorizes this into two types: single contributions and composite contributions. The former refers to a single community contribution or a single community contribution. For example, a task, solution, or vote in the community; the latter refers to the combination of tasks, solutions, and votes in the community. For example, the number of solutions for a task in the community, the total number of votes cast by a member, etc.

[0100] 2) Trigger conditions. The trigger conditions for implementing an incentive mechanism refer to the circumstances under which the incentive mechanism can be implemented. Virtual community incentive mechanisms are primarily triggered through indirect interactions between community members based on tasks or contributions. For example, when a member votes on a community contribution, this triggers a reward or penalty for that member's reputation.

[0101] 3) Reward and punishment actions. In order to induce members to perform specific behaviors, the virtual community implements one or more reward and punishment actions on them. According to the form of incentive factors, this application divides reward and punishment actions into three types: ① Preset rewards and punishments - a type of reward and punishment clearly stipulated by the virtual community. For example, members receive a number of reputation rewards; after reaching a certain reputation threshold, they obtain corresponding status, etc. ② Ranking changes - rewards and punishments after the preset rewards and punishments are completed. Its changes cannot be predicted and need to be determined after comparison with peers in the community. For example, the reputation ranking and influence ranking of community members. ③ Psychological needs - refers to whether the internal needs of members are met, mainly reflected in identification incentives, integration incentives and internal incentives. For example, whether members recognize the value of community contributions, whether they feel reciprocity, whether they feel that the community can meet their social needs, etc.

[0102] Based on the above-mentioned incentive mechanism components, this application defines each incentive rule norm in the incentive mechanism set NORM as:

[0103] norm:ct×ts→rp.

[0104] Where norm∈NORM; ct represents the context of the implementation of the incentive mechanism; ts represents the triggering condition for the implementation of the incentive mechanism; and rp represents the implemented reward or punishment action.

[0105] In one embodiment, the community computing model includes a virtual community simulation algorithm, a community data sampling algorithm, and an agent behavior decision algorithm; wherein the virtual community simulation algorithm outputs a community member model and a set of artifacts by inputting real community member information, a description of the relationship between incentive values ​​and agent attributes, an incentive mechanism parameter set, an incentive value sparse segmentation point, a real community sampling ratio, a new agent joining rate, and a maximum number of simulations; the community data sampling algorithm outputs a community member model and specifications by inputting real community user information, an incentive mechanism parameter set, an incentive value sparse segmentation point, and a sampling ratio value; and the agent behavior decision algorithm outputs an affected set of artifacts by inputting an agent, an artifact set, and a specification.

[0106] Specifically, the virtual community simulation algorithm is as follows Figure 4 As shown, the input of the algorithm is:

[0107] 1)DS: real community member information.

[0108] 2) QA: Description of the relationship between the incentive value and the agent attributes, that is, the parameters of Equation 9.

[0109] 3) PT: incentive mechanism parameter set, determined by the incentive mechanism being simulated.

[0110] 4) sparsePoint: The incentive value sparse segmentation point. When the incentive value of a member is greater than a certain value (the sparse point), the number of members decreases significantly and the incentive value distribution span increases.

[0111] 5) ratio: The sampling ratio of real communities. Due to the large scale of real community data, simulation experiments directly using real community data require a long time to produce results. To improve simulation efficiency, this application designed an algorithm to sample smaller amounts of real community data. Here, ratio ≥ 1. If it is equal to 1, it means that all real community data is used; otherwise, the real community data is sampled at a ratio of 1 / ratio.

[0112] 6) joinRate: The rate at which new agents join, determined by the growth rate of new members in the simulated community.

[0113] 7) maxTicks: The maximum number of simulations, determined by the number of days the simulated community has been running.

[0114] The algorithm outputs the agent set AG and artifact set ART of the computational model. The algorithm steps are described as follows.

[0115] 1) Initialize the agent and specification. Line 1 generates simulated experimental data, samples users from the real community, and initializes the agent and specification. The implementation process is shown in Algorithm 2.

[0116] 2) Initialize the environment. Lines 2-3 reset the simulation count tick and clear the agent interaction environment ART.

[0117] 3) New Agents: Line 5 shows the agents that are continuously added to the simulator.

[0118] 4) The agent generates behavior. Lines 6-8 represent the agent's continuous contribution process. The specific implementation of the agent's contribution process is described in detail in Algorithm 3. The number of simulations is controlled by lines 4 and 9.

[0119] In one embodiment, the community data sampling algorithm is as follows Figure 5 As shown, the input of the algorithm is:

[0120] 1) DS: real community user information;

[0121] 2) PT: incentive mechanism parameter set;

[0122] 3) sparsePoint: sparse segmentation point of the stimulus value;

[0123] 4)ratio: sampling ratio value.

[0124] The output is the sampled agent set AG and the representation of the incentive mechanism NORM. The algorithm steps include

[0125] 1) Initialize variables. Lines 1-2 initialize the agent set AG and the loop variable incentive value rew;

[0126] 2) Extract non-sparse users. Lines 4-6 directly sample users whose incentive value is less than or equal to sparsePoint in proportion;

[0127] 3) Extracting sparse users. Lines 8-10 show that for users whose incentive values ​​are greater than sparsePoint, the algorithm first analyzes the minimum and maximum incentive values ​​and their number (minRew, maxRew, sum) of users with incentive values ​​of ratio. Then, it randomly samples users within the range [minRew, maxRew] according to the ratio. Since the distribution of the sampled data is consistent with the real community data, it can reflect the overall status of users in the real community.

[0128] 4) Add the agent. Line 12 adds the sampled agent to the collection;

[0129] 5) Instantiate the incentive mechanism. Line 14 indicates that the incentive mechanism model is instantiated by inputting the incentive mechanism rule parameter set PT;

[0130] 6) Return the result. Line 15 returns the generated agent set and incentive mechanism.

[0131] In one embodiment, the agent's behavior decision algorithm is as follows: Figure 6 As shown, the input of the algorithm is:

[0132] 1) ag: contributing agent;

[0133] 2) ART: product collection;

[0134] 3) NORM: A normative model for simulating community incentive mechanisms.

[0135] The output is the affected artifact set ART after the agent contributes. The algorithm steps include:

[0136] 1) Simulate the agent's task-proposing process. Lines 1-4 describe agent ag's task-proposing process. When the level of incentive influence pq on agent ag's current task-proposing behavior exceeds the overall level of incentive influence qr on its overall task-proposing behavior, a question is raised. Task q is then added to the ART.

[0137] 2) Simulate the process of agent contribution solutions. Lines 5-11 describe the process of agent ag contributing solutions. Line 7 describes the agent contributing solutions to tasks with na solutions and tasks created t days ago, subject to the level of incentive influence pa. If the agent's pa is greater than the level of incentive influence 1-qr for its overall contribution solution, it contributes a solution.

[0138] 3) Simulate the agent voting process. Lines 13-22 describe the voting process for agent ag. Line 13 shows the influence level (pu) of the incentive mechanism on agent ag's current like behavior, which is related to the like benefits and costs. When pu is greater than the overall influence level (ur) of the incentive mechanism on its like behavior, it likes the selected product. The dislike process is similar.

[0139] 4) Simulate incentive mechanism rewards and punishments. Line 23 shows that the simulator rewards voters and artifact owners according to the incentive mechanism rules.

[0140] 5) Return the result. Line 24 returns the affected artifact set ART.

[0141] In one embodiment, the Pearson correlation coefficient is used to evaluate the trend similarity between the simulated global emergence and the real emergence of the real community:

[0142]

[0143] Among them, S and M are the simulated global emergence of the real community and the computational model, and the real emergence of the real community, respectively. i and M i are the i-th component of the simulated global emergence and the real emergence of the real community, respectively; |pcc|>0.5 indicates a strong correlation between the two; |pcc|∈[0.3, 0.5] indicates a moderate correlation between the two; |pcc|<0.3 indicates a weak correlation or no correlation between the two.

[0144] The numerical difference between the simulated global emergence and the real emergence of the real community is evaluated using the average approximation:

[0145]

[0146] ava>0.5 means that the two means are highly close; ava∈[0.3,0.5] means that the two means are moderately close; ava<0.3 means that the two means are weakly close or not close.

[0147] According to the trend similarity and the numerical gap, the global emergence similarity is obtained as follows:

[0148] sim(S,M)=λ×pcc(S,M)+(1-λ)×ava(S,M).

[0149] This application proposes the concept of global emergent similarity (sim). This indicator is a balance between the evaluation results of trend similarity and mean similarity. λ is used as a balance coefficient to comprehensively consider the influence of trend similarity and mean similarity. This application emphasizes the importance of the two types of indicators in a balanced manner and sets λ = 0.5. The larger the value of the indicator sim, the higher the accuracy of the model simulation results. Based on the analysis of the above two types of indicators, this chapter believes that sim>0.5 indicates high similarity between the two; sim∈[0.3, 0.5] indicates that there is a moderate degree of similarity between the two; sim<0.3 indicates weak similarity or dissimilarity between the two.

[0150] It should be understood that although Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. In addition, Figure 1 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0151] In one embodiment, Figure 7 As shown, a device for effect evaluation based on multi-agent virtual community simulation is provided, comprising: an NAA model construction module 702, a virtual community model and incentive mechanism model construction module 704, and an effect evaluation module 706, wherein:

[0152] NAA model construction module 702, for constructing an NAA model; the NAA model includes agents representing virtual community members, artifacts representing community tasks and plans, and specifications representing virtual community incentive mechanisms;

[0153] A virtual community model and incentive mechanism model construction module 704 is configured to construct a virtual community model and an incentive mechanism model based on the NAA model. The virtual community model includes a community member model, a contribution model, and agent behavior rules. The community member model is composed of the agents, which include agent attributes and agent behaviors for participating in community activities. The contribution model is composed of the artifacts and is used to enable interaction and collaboration among community members. The agent behavior rules are used to formulate rules for participating in community contributions, and the rules are related to the level of influence of the incentive mechanism. The incentive mechanism model includes context, trigger conditions, and reward and punishment actions, with reward and punishment actions generated through context and trigger conditions to represent the norms.

[0154] The virtual effect evaluation module 706 is used to construct a community calculation model based on the virtual community model and the incentive mechanism model, and use the Pearson correlation coefficient to evaluate the trend similarity between the simulated global emergence of the community calculation model and the real emergence of the real community. The average value approximation is used to evaluate the numerical difference between the simulated global emergence and the real emergence of the real community. The global emergence similarity is obtained based on the trend similarity and the numerical difference, so as to evaluate the effect of the virtual community simulation.

[0155] In one embodiment, the community member model includes the agent's incentive value, professional ability, contribution activity, review activity, task proposal rate, and like rate. The contribution model includes artifact type, creating agent, creation days, number of solutions, likes, dislikes, and quality. The agent behavior rules include task proposal rules, solution contribution rules, likes, dislikes, and update rules.

[0156] In one embodiment, the task-proposing rule is:

[0157]

[0158] Among them, AG represents the community member model, ag represents the members in the community member model, q represents the proposed task, Q represents the proposed task set, PQ(ag,pq) represents the level of influence of the specification on the current task-proposing behavior of the agent ag is pq, QR(ag,qr) represents the task-proposing rate of the agent ag is qr, and CR(ag,p) represents the creation of the product p by the agent ag.

[0159] In one embodiment, the contribution solution rule is:

[0160]

[0161] Among them, a represents the scheme of the contribution model, A represents the scheme set of the contribution model, PA(ag,pa) represents the level of influence of the specification on the current contribution solution behavior of agent ag is pa, QR(ag,qr) represents the task proposal rate of agent ag is qr, and CR(ag,p) represents the creation of artifact p by agent ag.

[0162] In one embodiment, the like rule is:

[0163]

[0164] Among them, p represents the product, ART represents the contribution model, PU(ag,pu) represents the influence level of the norm on the current like behavior of agent ag is pu, UR(ag,ur) represents the like rate of agent ag is ur, and UV(ag,p) represents the like of agent ag to product p.

[0165] In one embodiment, the click rule is:

[0166]

[0167] Where p represents the product, ART represents the contribution model, PD(ag,pd) represents the level of influence of the norm on the current dislike behavior of agent ag, which is pd, UR(ag,ur) represents the like rate of agent ag, which is ur, and DV(ag,p) represents whether agent ag dislikes or cancels the like of product p.

[0168] In one embodiment, the update rule is:

[0169]

[0170] Among them, REW(ag,rt,rew) indicates that the rt type incentive value of agent ag is rew, EXP(ag,exp) indicates that the professional ability of agent ag is exp, PAC(ag,pac) indicates that the contribution activity of agent ag is pac, VAC(ag,vac) indicates that the average daily evaluation number of agent ag is vac, QR(ag,qr) indicates that the task proposal rate of agent ag is qr, UR(ag,ur) indicates that the like rate of agent ag is ur, f i is a function that represents the relationship between the agent’s incentive value and its other attributes, f i (rew)=ai×rew+b i , i∈{1, 2, 3, 4, 5}.

[0171] In one embodiment, the community computing model includes a virtual community simulation algorithm, a community data sampling algorithm, and an agent behavior decision algorithm; wherein, the virtual community simulation algorithm outputs a community member model and a set of artifacts by inputting real community member information, a description of the relationship between incentive values ​​and agent attributes, an incentive mechanism parameter set, an incentive value sparse segmentation point, a real community sampling ratio, a new agent joining rate, and a maximum number of simulations; the community data sampling algorithm outputs a community member model and specifications by inputting real community user information, an incentive mechanism parameter set, an incentive value sparse segmentation point, and a sampling ratio value; and the agent behavior decision algorithm outputs an affected set of artifacts by inputting an agent, an artifact set, and a specification.

[0172] In one embodiment, the effect evaluation module 706 is further configured to use the Pearson correlation coefficient to evaluate the trend similarity between the simulated global emergence and the real emergence of the real community:

[0173]

[0174] Among them, S and M are the simulated global emergence of the real community and the computational model, and the real emergence of the real community, respectively. i and M i are the i-th component of the simulated global emergence and the real emergence of the real community, respectively;

[0175] The numerical difference between the simulated global emergence and the real emergence of the real community is evaluated using the average approximation:

[0176]

[0177] The global emergence similarity obtained according to the trend similarity and the numerical difference is:

[0178] sim(S,M)=λ×pcc(S,M)+(1-λ)×ava(S,M).

[0179] Regarding the specific limitations of the effect evaluation device based on multi-agent virtual community simulation, please refer to the limitations of the effect evaluation method based on multi-agent virtual community simulation above, and will not be repeated here. The various modules in the above-mentioned effect evaluation device based on multi-agent virtual community simulation can be implemented in whole or in part through software, hardware, or a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory of the computer device in software form, so that the processor can call and execute the corresponding operations of the above modules.

[0180] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0181] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0182] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A method for effect evaluation based on multi-agent virtual community simulation, characterized in that: The method comprises: Constructing an NAA model; the NAA model includes agents representing virtual community members, artifacts representing community tasks and plans, and specifications representing virtual community incentive mechanisms; Based on the NAA model, a virtual community model and an incentive mechanism model are constructed. The virtual community model includes a community member model, a contribution model, and agent behavior rules. The community member model is composed of the agents, which include agent attributes and agent behaviors for participating in community activities. The contribution model is composed of the artifacts and is used to enable interaction and collaboration among community members. The agent behavior rules are used to formulate rules for participating in community contributions, and the rules are related to the level of influence of the incentive mechanism. The incentive mechanism model includes context, trigger conditions, and reward and punishment actions. The reward and punishment actions are generated by the context and trigger conditions to represent the norms. A community calculation model is constructed based on the virtual community model and the incentive mechanism model. Based on the simulated global emergence of the community calculation model and the real emergence of the real community, the Pearson correlation coefficient is used to evaluate the trend similarity between the simulated global emergence and the real emergence of the real community. The average value approximation is used to evaluate the numerical difference between the simulated global emergence and the real emergence of the real community. The global emergence similarity is obtained based on the trend similarity and the numerical difference, and the effect of the virtual community simulation is evaluated based on this.

2. The method according to claim 1, characterized in that The community member model includes the agent's incentive value, professional ability, contribution activity, evaluation activity, task proposal rate, and like rate; the contribution model includes: product type, creating agent, creation days, number of solutions, number of likes, number of dislikes, and quality; the agent behavior rules include: task proposal rules, solution contribution rules, like rules, dislike rules, and update rules.

3. The method according to claim 2, characterized in that The proposed task rules are: Among them, AG represents the community member model, ag represents the members in the community member model, q represents the proposed task, Q represents the proposed task set, PQ(ag,pq) represents the level of influence of the specification on the current task-proposing behavior of the agent ag is pq, QR(ag,qr) represents the task-proposing rate of the agent ag is qr, and CR(ag,p) represents the creation of the product p by the agent ag.

4. The method according to claim 2, characterized in that The contribution solution rules are: Among them, a represents the scheme of the contribution model, A represents the scheme set of the contribution model, PA(ag,pa) represents the level of influence of the specification on the current contribution solution behavior of agent ag is pa, QR(ag,qr) represents the task proposal rate of agent ag is qr, and CR(ag,p) represents the creation of artifact p by agent ag.

5. The method according to claim 2, characterized in that The likes rules are: Among them, p represents the product, ART represents the contribution model, PU(ag,pu) represents the influence level of the norm on the current like behavior of agent ag is pu, UR(ag,ur) represents the like rate of agent ag is ur, and UV(ag,p) represents the like of agent ag to product p.

6. The method according to claim 2, characterized in that The click-and-tick rules are: Where p represents the product, ART represents the contribution model, PD(ag,pd) represents the level of influence of the norm on the current dislike behavior of agent ag, which is pd, UR(ag,ur) represents the like rate of agent ag, which is ur, and DV(ag,p) represents whether agent ag dislikes or cancels the like of product p.

7. The method according to claim 2, characterized in that The update rules are: Among them, REW(ag,rt,rew) indicates that the rt type incentive value of agent ag is rew, EXP(ag,exp) indicates that the professional ability of agent ag is exp, PAC(ag,pac) indicates that the contribution activity of agent ag is pac, VAC(ag,vac) indicates that the average daily evaluation number of agent ag is vac, QR(ag,qr) indicates that the task proposal rate of agent ag is qr, UR(ag,ur) indicates that the like rate of agent ag is ur, f i is a function that represents the relationship between the agent’s incentive value and its other attributes, f i (rew)=a i ×rew+b i ,i∈{1,2,3,4,5}.

8. The method according to any one of claims 1 to 7, characterized in that The community computing model includes a virtual community simulation algorithm, a community data sampling algorithm, and an agent behavior decision algorithm; wherein, the virtual community simulation algorithm outputs a community member model and a set of artifacts by inputting real community member information, a description of the relationship between incentive values ​​and agent attributes, an incentive mechanism parameter set, an incentive value sparse segmentation point, a real community sampling ratio, a new agent joining rate, and a maximum number of simulations; the community data sampling algorithm outputs a community member model and specifications by inputting real community user information, an incentive mechanism parameter set, an incentive value sparse segmentation point, and a sampling ratio value; and the agent behavior decision algorithm outputs an affected set of artifacts by inputting an agent, an artifact set, and a specification.

9. The method according to any one of claims 1 to 7, characterized in that The Pearson correlation coefficient is used to evaluate the trend similarity between the simulated global emergence and the real emergence of the real community. The average approximation is used to evaluate the numerical difference between the simulated global emergence and the real emergence of the real community. The global emergence similarity is obtained based on the trend similarity and the numerical difference, including: The Pearson correlation coefficient is used to evaluate the trend similarity between the simulated global emergence and the real emergence of the real community: Among them, S and M are the simulated global emergence of the real community and the computational model, and the real emergence of the real community, respectively. i and M i are the i-th component of the simulated global emergence and the real emergence of the real community, respectively; The numerical difference between the simulated global emergence and the real emergence of the real community is evaluated using the average approximation: The global emergence similarity obtained according to the trend similarity and the numerical difference is: sim(S,M)=λ×pcc(S,M)+(1-λ)×ava(S,M).

10. An effect evaluation device based on multi-agent virtual community simulation, characterized in that: The device comprises: NAA model construction module, used to construct the NAA model; the NAA model includes intelligent agents representing virtual community members, artifacts representing community tasks and plans, and specifications representing virtual community incentive mechanisms; A virtual community model and incentive mechanism model construction module is used to construct a virtual community model and an incentive mechanism model based on the NAA model. The virtual community model includes: a community member model, a contribution model, and agent behavior rules. The community member model is composed of the agents, which include agent attributes and agent behaviors for participating in community activities. The contribution model is composed of the artifacts and is used to enable interaction and collaboration among community members. The agent behavior rules are used to formulate rules for participating in community contributions, and the rules are related to the impact level of the incentive mechanism. The incentive mechanism model includes: context, trigger conditions, and reward and punishment actions. The reward and punishment actions are generated by the context and trigger conditions to represent the norms. The effect evaluation module is used to construct a community calculation model based on the virtual community model and the incentive mechanism model, and use the Pearson correlation coefficient to evaluate the trend similarity between the simulated global emergence of the community calculation model and the real emergence of the real community, and use the average value approximation to evaluate the numerical difference between the simulated global emergence and the real emergence of the real community. The global emergence similarity is obtained based on the trend similarity and the numerical difference, so as to evaluate the effect of the virtual community simulation.

Images