Adaptive information sharing approach for crowd networks based on two stage optimization


Purpose
This paper aims to optimize and evaluating the performance of the crowd networks through analyzing their information sharing patterns. That is, in a crowd network, the qualities of accomplishing tasks are highly dependent on the effective information sharing among intelligent subjects within the networks. Hence, proposing an adaptive information-sharing approach can help improve the performance of crowd networks on accomplishing tasks that are assigned to them.


Design/methodology/approach
This paper first introduces the factors that affect effectiveness of information-sharing pattern: the network topology, the resources owned by intelligent subjects and the degree of information demand. By analyzing the correlation between these factors and the performance of crowd networks, an Adaptive Information Sharing Approach for Crowd Networks (AISCN approach) is proposed. By referring to information needed for accomplishing the historical tasks that are assigned to a crowd network, the AISCN approach can explore the optimized information-sharing pattern based on the predefined composite objective function. The authors implement their approach on two crowd networks including bee colony and supply chain, to prove the effectiveness of the approach.


Findings
The shared information among intelligent subjects affects the efficiency of task completion in the crowd network. The factors that can be used to describe the effectiveness of information-sharing patterns include the network topology, the resources owned by intelligent subjects and the degree of information demand. The AISCN approach used heuristic algorithm to solve a composite objective function which takes all these factors into consideration, so that the optimized information-sharing pattern can be obtained.


Originality/value
This paper introduces a set of factors that can be used to describe the correlation between information-sharing pattern and performance of crowd network. By quantifying such correlation based on these factors, this paper proposes an adaptive information-sharing approach which can explore the optimized information-sharing pattern for a variety of crowd networks. As the approach is a data-driven approach that explores the information-sharing pattern based on the network’s performance on historical tasks and network’s characteristics, it is adaptive to the dynamic change (change of incoming tasks, change of network characteristics) of the target crowd network. To ensure the commonality of the information-sharing approach, the proposed approach is not designed for a specific optimization algorithm. In this way, during the implementation of the proposed approach, heuristic algorithms can be chosen according to the complexity of the target crowd network.



Introduction
The phenomenon of crowd intelligence can be widely found in nature and human society. In the nature, the crowd intelligence leads to the ant colony effect, the formation of birds flying, etc. In the human society, the phenomenon of crowd intelligence can be leveraged to enterprises' operation and management process, industrial chain, etc. (Jiang and Jiang, 2004;Li and Qiao, 2015). Unlike the networks (such as computer networks, IoT networks) that are designed for accomplishing specific tasks, as a network that consists of human subjects, asset subjects and digital agent subjects, crowd network is required to accomplish different tasks (Prpic and Shukla, 2017). To accomplish these tasks, intelligent subjects within the crowd networks need to cooperate with each other through making interactions carried by information and behaviors. The completion of different tasks requires the possession of different information (Fehrenbacher and Djamasbi, 2017). In a crowd network with high network complexity, redundant information and limited resources, information should be exchanged selectively to achieve "low cost and high efficiency" under limited resources. When cooperating to complete tasks, intelligent subjects in the crowd network can choose to share specific information among their neighbors or spread to network to help each other better accomplish their tasks (He et al., 2006;Rose et al., 2003). Therefore, it is particularly important to determine the object and scope of information sharing. Because of the complexity of the topological structure of the crowd network, it is more difficult to consider the information-sharing scope of an intelligent subject in the whole network (under global conditions) than that in the neighborhood (under local conditions). Therefore, this paper proposes an Adaptive Information Sharing Approach for Crowd Networks (AISCN approach) to explore the optimized information-sharing pattern for crowd networks The approach adopts the idea of first local optimization and then overall optimization, and quantifies the effectiveness value of shared information IV(t) for a task t in the crowd network. As different intelligent subjects have different needs for information when doing different tasks, the demands of different information are also different. Moreover, a crowd network has limited resources instead of redundant resources. Therefore, we defined a set of factors that can be used to describe the effectiveness of information-sharing patterns, which include the network topology, the resources owned by intelligent subjects and the degree of information demand. The AISCN approach used heuristic algorithm to solve a composite objective function which takes all these factors into consideration, so that the optimized information-sharing pattern can be obtained. We implement our approach on two crowd networks including bee colony and supply chain, to prove the effectiveness of our approach.

Adaptive informationsharing approach
This paper is organized as follows. The background and related works are introduced in Section 2. The shared information content is introduced in Section 3. Section 4 analyses factors of shared information. Section 5 introduces AISCN approach based on task. Section 6 describes using AISCN approach to analyze the applications of shared information in bee colony and supply chain. In Section 7, the paper is summarized.
2. Background and related work 2.1 Background Intelligence refers to the responsiveness of an intelligent subject to environmental or external stimuli, including the corresponding speed, quality and learning ability. It is a measure of an intelligent subject's ability to achieve goals in various environments (Legg and Hutter, 2007;Legg and Veness, 2011). Intelligent subjects are entities with certain intelligence, including any human agent (individual, enterprise, institution) and machine agent (intelligent equipment, intelligent robot or goods). Crowd network is a simple regular network, random network or any complex network where complexity is between regular network and random network. A crowd network is usually formed by many intelligent subjects (network nodes) and the relationship among these subjects includes administrators, subordinators, cooperators, competitors, interest-related parties, irrelevant parties, etc. Figure 1 illustrates the general workflow of a crowd network in which intelligent subjects interconnect, influence and interact with each other. The crowd network needs to complete different tasks assigned by the external environment. Tasks can be divided into a series of subtasks, which require intelligent subjects with specific professional abilities to complete. Action1 refers to the interaction between intelligent subjects (the interaction of information, behavior, etc.). Action2 refers to the interaction between different types of professional intelligent subjects, to interact with the completion results of subtasks and finally output the completion results of tasks. Accomplishment is the completion results of tasks by the crowd network. Rewards is the evaluation of the external environment for the completion results of tasks. The performance of a crowd network refers to its expected finishing quality (also called expected rewards) of a series of tasks that are assigned to the crowd network by the external environment. It can be seen that the interaction between intelligent subjects (such as information interaction) has an impact on the accomplishment rewards of tasks, and hence it affects the performance of the crowd network. Therefore, the objective of this paper is to improve the performance of a crowd network through optimizing the information-sharing pattern of the network.

Related work
In the crowd network, a large number of intelligent subjects cooperate with each other to complete specific complex task (Liang et al., 2018). In the crowd network, an intelligent  A sample of the  crowd network   IJCS  3,3 subject may be an individual, organization, enterprise or a robot or other intelligent tool. The types of intelligent subjects in the crowd network are heterogeneous, and they operate in a complex environment which includes the information space, physical space and intelligence space (Liang et al., 2018). One form of complexity in intelligent environments arises from their heterogeneous nature. The growing variety of environments and countless stereotypes of users operating intelligent environments will, theoretically, increase the complexity and resources needed to use them (Ananta et al., 2013). At the same time, the structure of the hybrid space will be affected by the changes of the intelligent subject's class and task. The problem of big data in the network also has a certain impact on target analysis (Wang et al., 2018). Therefore, it is not feasible to comprehensively evaluate the performance of intelligent subjects. Nowadays, people and computers can interact with each other in different ways, which is a process of information exchange between people and computers to complete certain tasks. Information interaction between people is also carried out in a certain way (Russell et al., 2017). Forbrig et al. (2017) introduced the classification of human-computer interaction. Meanwhile, task response theory estimation and algorithm information theory can be used to test machine intelligence (Araiza-Illan et al., 2016;He et al., 2006). Sharing information should also be applied in various fields to promote the development of corresponding fields.
In medicine, information sharing integrates genome-based informatics, modernizes global disease surveillance and ensures more effective detection, prevention and control of endemic diseases (Aarestrup et al., 2012). In modern commerce, supply chain also has competition, which is intrinsically complex. Information sharing and collaboration have shown an important role in the performance of supply chain (Wu et al., 2014). Schloetzer (2012) also analyzed information sharing in supply chains.  proposed an information-sharing mechanism (ISM) to improve the performance of particle swarm optimization (PSO). The ISM allows each particle to share its best search information, so that all the other particles can take advantage of the shared information by communicating with it. In this way, the particles could enhance the mutual interaction with the others sufficiently and heighten their search ability greatly by using the search information of the whole swarm. Sharing of knowledge, information and practices across cultural and national boundaries has become a means to address critical global problems. Sharon et al. (2012) proposed a conceptual dynamic model to explain the further application of information sharing.
The existing works regarding optimizing the information sharing patterns have the following limitations: all the existing approaches are proposed for specific applications such as supply chain and information networks, and no comprehensive approach has been proposed; because most of the existing approaches did not take both the information demand of different kinds of tasks and dynamic characteristic of a network into consideration, these approaches cannot operate in an adaptive manner; the correlation between information sharing and network performance is not described in detail by most of the existing works. To explicitly describe the correlation between information sharing and network performance, this paper introduces a set of factors that can be used to describe such correlation. These factors are comprehensive enough to be mapped into most of the crowd networks. By quantifying such correlation based on these factors, this paper proposes an adaptive information-sharing approach which can explore the optimized informationsharing pattern for a variety of crowd networks. As the approach is a data-driven approach that explores the information-sharing pattern based on the network's performance on historical tasks and network's characteristics, the approach is adaptive to the dynamic Adaptive informationsharing approach change (change of incoming tasks, change of network characteristics) of the target crowd network.

Shared information content
Every individual in the crowd network has its own ontology information. The more information it has, the more predictability the behavior of the intelligent subject is (He et al., 2006;Hu and Yang, 2009). This is because the intelligent subject who obtains more information and has higher intelligence level are more likely to trigger the optimized behavior scenario for a certain task. Therefore, to improve the performance of the crowd network, the information obtained by intelligent subjects within the crowd network should be maximized.
There are three basic elements in the crowd network: environment, individual and individual behavior rules (Liu and Wang, 2000;Wu et al., 2003). All intelligent subjects perform tasks according to a cluster of rules in a specific environment, and complex collective behavior is carried out by the interactions among subjects. Through interacting with their neighbors or subjects that can be reached, the intelligent subjects can acquire information/knowledge, such as detail of the sub-tasks and collaborative rules, so that the sub-tasks that are assigned to the subject can be accomplished.
To describe the ISM, we need to describe the information-sharing behavior pattern between intelligent subjects. Specifically, the information-sharing behavior between intelligent subjects can be described by the amount of information shared between intelligent subjects by a behavior. The amount of information can be calculated through giving weight to all kinds of information shared and determining the scope (class) of information. The calculation equation (1) is as follows: where R j is the scope of information exchanged and shared by class j of the first i intelligent subjects, R j = N (number of global sharing behavior) or K (number of local sharing behavior), W j is the weight of exchangeable and sharable information in class j of the first i intelligent subjects, F j = 1 (class j information exchanged and shared by the first i intelligent subjects) or 0 (class j information not exchanged and shared by the first i intelligent subjects). According to equation (1), it is necessary to determine R j and W j of the information j of intelligent subject i in the network when calculating I s , which is relatively difficult to determine effectively. Through exploring the factors that may affect the effectiveness of information-sharing patterns, we can explicitly describe the detail of R j and W j . In the following sections, we introduce these factors and propose an approach to obtain the optimal information-sharing pattern.

The analysis of factors of information sharing
In the crowd network, each individual can obtain local information from neighbors and obtain global information from intelligent subjects other than neighbors. Through learning from neighbors (also known as local information sharing), intelligent subjects can collect all the information owned by their neighbors (Bala and Goyal, 2010). On the other hand, the learning from intelligent subject other than neighbors (also known as global information sharing) can make intelligent subject escape from the local optimum to explore more IJCS 3,3 information for completing the tasks. Hence, both local information sharing and global information sharing can promote the improvement of intelligence level of an intelligent subject. In this way, the whole crowd network will be evolved to gain higher finishing qualities of tasks and the high efficiency of task accomplishments. Figure 2 illustrates a typical ISM in the crowd network. We use Ind to represent the individual, and individuals that can directly interact with the ith individual are represented by 1-nei. Individuals that need to interact through 1-nei are represented by 2-nei, and so on. Individuals can directly obtain 1-nei information, but it needs to be obtained indirectly by a-nei (1 # a < b) to obtain b-nei information (1 < b).

The topological structure of the crowd network
In the crowd network, it is difficult to directly determine the optimal sharing scope for information owned by an intelligent subject. According to the content and scope of information and the topology of the network, the intelligent subject chooses whether to share information in the whole network (through the n-nei interaction mode in Figure 2) or only in the neighborhood. When determining this condition, it is necessary to consider whether the network structure can achieve global sharing and whether the resources owned by intelligent subjects can achieve global sharing. As shown in Figure 3(b), a network is divided into several sub-networks, in which case it is difficult to achieve global sharing. For a more regular network, as shown in Figure 3(a), global or local sharing is selected while considering resources. According to the x-nei (1 # x < n) interaction mode in Figure 2, the cost, benefit and resources to obtain the optimal implementation performance should be considered. Here we simplify the scope of information sharing. When a certain information is shared with the neighbor, the information owned by the neighbor changes. When sharing information with its neighbors, it is necessary to consider both information the received shared and the original, so as to realize the dynamic change of the information owned by the intelligent subject.

The degree of information demand
Intelligent subjects have their own information, and different intelligent subjects may contain different kinds of information. Suppose there are intelligent subjects a and b, where a contains two kinds of information, x 1 and x 2 . Intelligent subject a can share these two kinds of information to subject b, where x 1 is useful for subject b to complete task t 1 and x 2 is Adaptive informationsharing approach useless. And x 2 is useful when subject b completes task t 2 , but x 1 is useless. Both x 1 and x 2 are useful when subject b completes task t 3 , and x 2 is more useful. Therefore, in the process of information sharing, subject b receives the information from subject a equally without considering any conditions. It will be useless to share the information x 2 when task t 1 is completed, which results in unnecessary operations and waste of resources. Similarly, when subject b completes task t 2 , the sharing of x 1 is also unnecessary. When task t 3 is completed, subject b is more likely to give priority to the information that plays a greater role in receiving shared information. For different tasks, different intelligent subjects have different needs for information. For a task, when an intelligent subject shares information, it should consider the degree of information demand of other intelligent subjects. The higher the degree of information demand ID, the greater the reward R for sharing the information. It is assumed that the two factors satisfy the polynomial relationship R = a * ID 3 þ b (a and b are polynomial coefficients), which can be expressed by R = f (ID).

The resources owned by intelligent subjects
The resources owned by intelligent subjects are limited, and when sharing information, certain resources are needed. Therefore, when intelligent subjects share information, they need to do it under certain conditions. Once they exceed this limit, they cannot share information. During the information sharing, it is necessary to analyze the resources needed by information and the total resources owned by intelligent subjects. The resources needed for information are related to many factors, such as the form of information sharing, the type, storage and dissemination of information.
5. Adaptive information sharing approach for crowd networks based on task 5.1 Adaptive information sharing approach for crowd networks By analyzing the factors of information sharing in Section 4, we can know that the topology structure of the crowd network, the degree of information demand and the resources owned by intelligent subjects will have an impact on information sharing. Given a complicated network topology, it is much more complicate to determine the sharing scope of each intelligent subject's information by taking both the limited 290 resources and information demand into consideration. To make such sophisticated decision, the information-sharing process should be operated in a recurrent manner. For instance, given the information demand of information i for the current task and the resources owned by the target network, intelligent subject a makes decision whether to share information i with neighbor b and then updates the state of crowd network if information i is shared. In such a new state, subject b makes decision whether to share i with neighbors of b. After a set of iteration, the sharing pattern of information i to the whole network is finally determined.
The AISCN approach is proposed by combining the local optimization and global optimization together as a two-stage process. Here, we use Cost-effectiveness ratio (EC) and Reward as optimization criteria. EC evaluates the relative magnitude of improved effectiveness and increased costs, calculated by equation (3). Reward is the total reward for sharing information among the neighbors, Reward i = Sum(R i *X i ) is the total reward of shared information by intelligent subject i and its neighbors. In the first stage, which is for performing local optimization, we select an appropriate optimization algorithm, to find the optimal information-sharing pattern among neighbors (information owned by each intelligent subject is changing dynamically, and information sharing is also changing dynamically). When obtaining optimal information-sharing pattern, Reward i is mainly considered. If Reward i > 0, then EC i is calculated. The maximum value of Reward i is obtained when EC i > 1, and then the optimal information-sharing pattern of intelligent subject i is obtained. In the second stage, which is for performing global optimization, we use optimization algorithm to adjust the information-sharing pattern of each intelligent subject, by considering both Reward and EC. For each intelligent subject i, to get Reward i and EC i in equilibrium, where Reward i and EC i are relatively large. After the second stage, the optimized information-sharing pattern corresponding to the current task will be obtained.
The input of the optimization algorithm is TS i , M i , ID i , S i , Nei(i), X i . TS i refers to the total amount of resources of intelligent subject i; M i refers to the total information of intelligent subject i; ID i refers to the degree of demand neighbors of i for information owned by intelligent subject i, specifically, the demand of subject i's neighbor k for information j is ID ijk ; S i refers to the resources needed by intelligent subject i to share information with its neighbors, specifically, resources needed when sharing information j with neighbor k is S ijk ; Nei(i) is an array which refers to the neighbors of subject i; X i represents the information-sharing pattern of intelligent subject i, X ijk is 0 or 1 (1 indicates the information j of intelligent subject i shared with neighbor k, 0 indicates the information is not shared). Here, TS i is a number, and M i , ID i , S i , Nei(i), X i are all matrices with sizes of 1 Â M, M Â P, M Â P, 1 Â P, M Â P, respectively, where M represents the number of information owned by intelligent subject i, and P represents the number of neighbors of intelligent subject i. Therefore, the objective function is: where R i is a matrix with size M Â P, which refers to the reward for intelligent subject i after sharing information with neighbors; the reward of sharing information j with neighbor k is R ijk (R ijk = a*ID 3 ijk þ b ). M i , R i , S i and X i all change in real time according to the change of information they own after sharing. Reward i is the total reward for sharing information among the neighbors, and it is meaningful to share information Adaptive informationsharing approach when Reward i is greater than 0 [Reward i > 0, u (Reward i ) = 1, and Reward i # 0, u (Reward i ) = 0.]. If EC i > 1 (calculated by equation (3)), X i is the required informationsharing pattern when Reward i is maximum: where C s and E s , respectively, represent the cost and effectiveness under the condition of sharing information, while C and E, respectively, represent the cost and effectiveness under the condition of non-sharing information (the cost and effectiveness need to be obtained according to the actual situation). EC i evaluates the relative magnitude of improved effectiveness and increased costs. Only when EC i > 1 (effectiveness is relatively higher than cost) can we choose to share information. On the contrary, we do not share information, and the sharing of intelligent subjects in this pattern is updated. The value of EC i can be determined based on the following rules: EC i > 1, u (EC i À 1) = 1, and EC i # 1, u (EC i À 1) = 0. Solving the EC i and Reward i of intelligent subject i in the optimal informationsharing pattern (EC i > 1 and Reward i > 0, to avoid these two values being too large and causing calculation trouble, ln function can be used to standardize), and then using the optimization algorithm to optimize the information-sharing pattern of the whole network. Considering that the effectiveness improvement of each intelligent subject is large relative to the cost increase and the reward of information sharing is relatively large. Finally, under a certain task, the objective function is where N is the number of intelligent subjects in the network; IV(t) is an evaluation of the efficiency of information-sharing pattern for the whole network under task t, and the goal is to find the maximum value of IV(t); N represents the number of intelligent subjects in the crowd network; u (x) is the heavyside function, where x > 0, u (x)=1 and x # 0, u (x) = 0. AISCN based on task t is mainly divided into two stages: local optimization and global optimization.
Stage 1: local optimization. For the input of TS i , M i , ID i , S i , Nei(i) and X i , according to the characteristics of case to select an appropriate optimization algorithm, to obtain the optimal information-sharing pattern X i for intelligent subject i. The goal is as follows: Max(Reward i ) = Max(Sum(R i *X i ). If Reward i > 0, Sum(S i * X i ) # TS i and EC i > 1, X i are desired. Stage 2: global optimization. Reward i and EC i are obtained by Stage 1, and equation (4) is used to calculate IV(t). Using optimization algorithm to adjust the information-sharing pattern of some intelligent subjects, the goal is as follows: Max (IV(t)). Finally, the optimal information-sharing pattern of the whole crowd network is obtained. IJCS 3,3

AISCN implementation process
The implementation process of AISCN approach is proposed as follows: Algorithm: AISCN implementation process Input: Task t, Network topology TL, Degree of information demand ID, Total resources of each intelligent subject TS, Resources needed for each type of information S, Information of each intelligent subject M. Output: Sharing pattern X, Sharing information value IV(t). Initialization X, X ijk = 0; while( IV(t) is not max ) do for i = 1,2,3,. . ., N do get Nei(i) from TL ; for j = 1,2,3,. . ., M do for k = 1,2,3,. . ., P do then Adjust X i end Adjust M, need, S according to X i end Take Reward 1 ,. . ., Reward N , EC 1 ,. . ., EC N as input of algorithm, to get max IV(t) Adjusting the information sharing pattern of intelligent subjects end 6. Case analysis 6.1 Information sharing in bee colony In nature, the crowd network is equivalent to an ecological network, which contains all kinds of information, including physical information, chemical information and behavioral information. When an individual receives this information, they will make corresponding choices according to their needs or their environment, and first respond to the most useful or needed information. Bee colony is typical ISM in nature; when a worker bee finds a source of honey within 100 meters from the nest, it tells its peers through a waltz. When it finds a Adaptive informationsharing approach source of honey 100 meters away from the nest, it tells its peers through a wagging dance. They dance to convey information and share it throughout the colony, avoiding the repetition of useless behavior by other bees and reaping more food by more bees.
Stage 2: Reward 1 ,. . ., Reward 6 and EC 1 ,. . ., EC 6 are substituted into equation (4), IV(t1) = 11.0904. Here the optimization algorithm is used to optimize and adjust the intelligent subject's information-sharing pattern, until the optimal information-sharing pattern of the whole network is obtained.
The output of AISCN approach is as follows: under the optimal information-sharing pattern, IV(t1) = 11.0904, X 313 , X 412 = 1 and the rest of X is 0. The optimal informationsharing pattern is as follows: honey bee 1 shares the information, that is, "specific information of honey source within 100 meters from the nest" to honey bee 2, and honey bee 2 shares the information, that is, "specific information of honey source beyond 100 meters from the nest" to honey bee 1. And queen bee, male peak bee, scout bee1 and scout bee 2 do not share information. Finally, honey bee 1 and honey bee 2 all have these information: "specific information of honey source within 100 meters from the nest" and "specific information of honey source beyond 100 meters from the nest." According to the information of honey source been detected, the honey bee resources should be allocated reasonably. Specifically, honey bees should be allocated to collect honey source with higher quality honey.
To prove the adaptivity of our approach on different tasks and on different network topologies, we conducted another four implementations for the AISCN approach. To test the adaptivity of our approach on different network topologies, we implement our approach on a bee colony network with another network topology [see Figure 4(b)]. We represent queen bee, male peak bee, scout bee 1, scout bee 2 and scout bee 3 within the figure with discrete labels (1, 2, 3, 4, 5). To test the adaptivity of our approach on different tasks, we implement our approach on the case when bee colony network receives other kinds of tasks including t 2 = "Queen bee fertilization" and t 3 = "Get a new nest." Through implementing the AISCN approach, the optimal information-sharing patterns corresponding to two network topologies and two tasks are shown in Table I.

Information sharing in the supply chain
For supply chain management, through the effective information sharing among members of the supply chain, the efficiency of the supply chain can be improved. In practical operation, coordination and collaboration among supply chain members are mainly reflected in information sharing (Lu et al., 2004). Obviously, because of the existence of organizational barriers and the needs of self-interests, effective information sharing is not easy to achieve. The quantification of the value of shared information will greatly promote the coordination of supply chain members in such case. The profit that can be made by information sharing includes promoting the investment of supply chain members in information infrastructure, improving the informatization level of supply chain members and promoting the integration among supply chain members. We used two simple supply chains to implement our approach, which are shown in Figures 5 and 6.
The output of AISCN approach is as follows: under the optimal information-sharing pattern, IV(t2) = 9.6339, X 111 , X 121 , X 311 , X 321 = 1, and the rest of X is 0. The optimal information-sharing pattern is as follows: customers share the information "the consumption of the product" and "the highest acceptable price of the product" to retailers, and distributors share the information "the storage of the product" and "price of the product" to retailers. And retailers, manufacturers, transporter 1, transporter 2 and transporter 3 do not share information. Finally, retailers have these information: "the purchase of the product," "pricing of the product," "the consumption of the product," "the highest acceptable price of the product for customers," "the storage of the product" and "price of the product." According to t, the consumer's consumption and the highest acceptable price, as well as the reserve and pricing of the product by the distributor, the retailer determines the quantity of the product purchased and the selling price of the product. Thus, the net income of retail sales of this product is the highest, that is, the effectiveness of task t is the largest.
In order to prove the adaptivity of our approach on different tasks and on different network topologies, we conducted another four implementations for the AISCN approach. To test the adaptivity of our approach on different network topologies, we implement our approach on a supply chain network with another network topology [see Figure 5(b)]. We represent customers, retailers, distributors, manufacturers and transporters within the figure as discrete labels (1, 2, 3, 4). To test the adaptivity of our approach on different tasks, we implement our approach on the case when supply chain networks receive other kinds of tasks, including t 2 = "Storage and price of a product for distributors" and t 3 = "Production of a product for manufacturers." Through implementing the AISCN approach, the optimal information-sharing patterns corresponding to two network topologies and three tasks are shown in Table II.

Conclusion
To optimize and evaluate the performance of the crowd networks through analyzing their information-sharing patterns, this paper introduced the factors that affect effectiveness of information-sharing pattern: the network topology, the resources owned by intelligent subjects and the degree of information demand. To analyze the correlation between these factors and the performance of crowd networks, an Adaptive Information Sharing Approach for Crowd Networks (AISCN approach) is proposed. By referring to information needed for accomplishing the historical tasks that are assigned to a crowd network, the AISCN approach can explore the optimized information-sharing pattern based on the predefined composite objective function. As the approach is a data-driven approach that explores the information-sharing pattern based on the network's performance on historical tasks and network's characteristics, the approach is adaptive to the dynamic change (change of incoming tasks, change of network characteristics) of the target crowd network. To ensure the commonality of the information-sharing approach, the proposed approach is not designed for a specific optimization algorithm. In this way, during the implementation of our approach, heuristic algorithms can be chosen according to the complexity of the target crowd network. We implement our approach on two crowd networks, including bee colony IJCS 3,3 and supply chain, to prove the effectiveness of our approach. Based on network topologies and incoming tasks of different bee colony and supply chain networks, the AISCN approach is implemented to adjust adaptively to obtain the optimal information-sharing pattern for each task. In the future, we will further study different complex network structures and propose appropriate optimization algorithms to make the AISCN approach less complex and more efficient, and make a deeper analysis of the impact of shared information on the crowd network.