- Email: [email protected]
Estimating the effect of organizational structure on knowledge transfer: A neural network approach
Expert Systems with Applications 30 (2006) 796–800 www.elsevier.com/locate/eswa
Estimating the effect of organizational structure on knowledge transfer: A neural network approach Fangcheng Tanga,*, Youmin Xib, Jun Mab a
School of Economics and Management, Tsinghua University, Beijing 100084, People’ Republic of China b School of Management, Xia´n Jiaotong University, Xia´n 710049, People’s Republic of China
Abstract Artificial neural network has been put into abundant applications in social science research recently. In this study, we investigate the topological structures of organization network, which can possibly account for the different performances of intra-organizational knowledge transfer. We construct two types of networks including hierarchy and scale-free networks, and single-layer perceptron model (SLPM) was used to simulate the knowledge transfer from a remarkable member to the others. The statistical results indicate that although the performance of knowledge transfer is related to the aspiration of the remarkable member to transfer knowledge, but the scale-free structure is more effective in knowledge transfer than that in hierarchy structure. q 2005 Elsevier Ltd. All rights reserved. Keywords: Knowledge transfer; Organization; Neural network; Scale-free
1. Introduction The volume of knowledge management research in organizations has increased in recent years, as it has in many disciplines, including organizational behavior and theory, information systems, psychology, sociology, economics and strategy. Research on knowledge management focuses on various aspects of the three fundamental questions: how organizations create, retain, and transfer knowledge (Argote, McEvily, & Reagans, 2003). Indeed, the growing empirical evidence indicates that organizations that are able to transfer knowledge effectively are more productive and more likely to survive than those are less adept at knowledge transfer (Baum & Ingram, 1998). Several academic literatures have developed that many factors affect knowledge transfer in organizations such as similarity between tasks (Darr & Kurtzberg, 2000), characteristics of the source of knowledge, the recipient, the context, and the knowledge itself (Szulanski, 2000), characteristics of individual member (Baldwin & Ford, 1988), characteristics of the social network (McEvily & Zaheer, 1999), the nature * Tel.: C86 10 62775888; fax: C86 10 62775888. E-mail address: [email protected]
0957-4174/$ - see front matter q 2005 Elsevier Ltd. All rights reserved. doi:10.1016/j.eswa.2005.07.039
of the social ties (Hansen, 1999), network structure (Reagans & McEvily, 2003). Generally speaking, knowledge transfer is considered in two aspects, one is associative learning and absorptive capacity (Cohen & Levinthal, 1990; Simon´, 1991), the other is tie strength (Hansen, 1999; Uzzi, 1997), social cohesion and network range (Reagans & McEvily, 2003). However, little concerning the effect of characteristics of network structure on knowledge transfer is taken into account. In particular, when the structure of organization network become more complex increasingly, how to estimate the performance of knowledge transfer in this type of organizational network structure have been little considered. Recently, many large networks were found that the vertex connectivities follow a scale-free power-law distribution (Baraba´si & Albert, 1999), especially in informal networks such as friendship networks, citation patterns networks and WWW networks. Here, we try to find comparatively, if this new form of organization networks will influence the convection of knowledge. In this study, we examined the topology of organization network and its effect on the ability of individuals to access to knowledge from the remarkable member in the same unit or organization. We believed that the ability of individuals to possess new knowledge would be different by changing the topology of networks. Specifically, our study focused on the different performance of knowledge transfer between two organizational network topologies: hierarchical and
F. Tang et al. / Expert Systems with Applications 30 (2006) 796–800
scale-free networks. We addressed this idea by utilizing single-layer perceptron model (SLPM) of neural network (Haykin, 1994; Hertz, Krogh, & Palmer, 1991; Lippman, 1987) to simulate the knowledge transfer process in an organization, that consists of 100 members which are linked by hierarchical or scale-free preferential attachment mechanism.
2. Organizational network models Xi and Tang (2004) think that an organization can be considered as graph, each vertex represents a member in the organization and edges indicate the communication paths between them. Thus, without loss of generality, an organization network can be measured by graph theoretic approach in this form: GZ(V, E), where V denotes the set of members and E is a token of the relations set in the networks (Xi & Tang, 2004). Two organization models both of which have 100 members are constructed. One is a simple hierarchy model, the other is a scale-free network with preferentialattachment (see Fig. 1). Fig. 1 shows the hierarchical and scale-free network model, and (a) shows hierarchical model for vertical organizations, there is a vertex at the top of the organization and each vertex in the organization have five underling vertexes. In this model, every vertex is only influenced by the vertex that superior to it directly. (b) shows scale-free model introduced by BA. At tZ0 the system consists m0Z3 isolated vertices. At every time step, a new vertex is added, which is connected to mZ2 vertices that already present in the model and absorb the knowledge from them, preferentially to the vertices with high connectivity, determined by the preferential attachment rule. These two types of network model were examined in our paper. A hierarchy model includes four layer hierarchies with one vertex at the top level, and each node has five subordinates, therefore the model has 156 vertexes. We simplify this model to a vertical organization model that can only provide vertical information flow. A scale-free model is defined in two steps (Baraba´si et al., 1999): (1) Growth. Starting with a small number (m0Z3) of vertices, a new vertex with mZ2 (%m0) edges (link the new vertex to the vertices that already exist in the
797
system) was added to the network at every time step. (In order to construct an equal size model to the hierarchy’s, we limited the scope growth up to 156 vertices). (2) Preferential attachment. The probability P that a new vertex will be connected to vertex i depends on the connectivity ki of that vertex, so that .X Y ki Z ki kj (1) j
After 153 timesteps, we get a scale-free model in which the probability that a vertex has k edges following a power law exponent gmodelZ2.9G0.1. We are going to explore the knowledge transfer process by simulating how a remarkable person spread the particular knowledge to the other colleagues. 2.1. Vertex status variable scheme We assume that every member in the organization have three ingredients VZ[S, D, A]. The first is a signal variable S, which demonstrates whether a member obtained the particular knowledge, and value 1 means he successfully accesses to the knowledge and 0 means the opposite. The second, variable D, which is independent randomized uniformly distributed on [0, 1], represents the capacity of the members to learn and absorb particular knowledge. The larger value means less capacity, it may be determined by the inherence, background or the experience of those members. The last variable is A, it shows the aspiration level of the member to transfer knowledge to others. A is also a random variable uniformly distributed on [0, 1] and the member which has strong will to spread the knowledge is able to get high numerical value. 2.2. Interacting mechanism and mathematical definition of SLPM In this study, the SLPM is used to model the knowledge transfer stage, that is the mapping of the capacity of absorb knowledge to absorption probabilities. The structure of the SLPM used in our model is shown schematically in Fig. 2. In Fig. 2, the top row of small circles represents the input layer including the bias. The middle row of large circles
Fig. 1. Hierarchical and scale-free model of organization network.
798
F. Tang et al. / Expert Systems with Applications 30 (2006) 796–800
Note that this normalization step is not traditionally part as probabilities of the SLPM. The pij can now beP Ninterpreted r because pij2[0, 1] and for ci, pij Z 1. The probabilities jZ1 that the model responds pij are interpreted as the probability with certain knowledge Ci acquired by agents when it is assigned with absorptive capacity Di. Generally, before being used as model input, the set of values for each level is standardized over all absorptive capacity using F Kmk F~ ik Z ik sk
Fig. 2. Schematic representation of the SLPM.
represents the output layer. The symbols S and s represent summation and sigmoid transformation, respectively. The SLPM consists of an input layer and an output layer and no hidden layers. The level of absorptive capacity is clamped to the input nodes, represented as the top row of circles in Fig. 2. The input nodes pass the features unchanged. One input node is assigned to each level of absorptive capacity. The SLPM in Fig. 2 has 4 input nodes. The top right circle with the number ‘1’ is the bias node. Instead of transferring a feature value, this node simply outputs a fixed value 1. All input nodes, including the bias node, are connected to all output nodes. A weight is associated with each connection. The feature value passing through the connection is multiplied by the respective weight before reaching the output node. The weights are the parameters of the model. The number of weights Nu, including biases, in the SLPM equals Nu Z ðNF C 1ÞNr
(2)
In the output nodes two processing steps take place. First, the weighted feature values of absorptive capacity Di are summed, yielding a quantity dij2R: dij Z bj C
NF X
ukj Fik
(3)
kZ1
where bj2R is the bias, which is the weight between the bias node and output node j, and ukj2R is the weight between input node k and output node j. Next, dij is passed through a sigmoid activation function yielding a quantity sij defined by 1 sij Z 1 C expðKdij Þ
(4)
Note that Sij2(0, 1), and limdij/KN sij ðdij ÞZ 0 and limdij/N sij ðdij ÞZ 1. Finally, using Luce’s (unbiased) choice rule (Luce, 1963; chapter 3), the outputs sij of the output nodes are normalized, yielding the quantity pij: sij pij Z N Pr sil lZ1
(5)
(6)
Where Fik and F~ ik are the original and standardized values of level k of absorptive capacity Di, and vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi u Nr Nr u 1 X X mk Z 1=Nr Fik ; sk Z t ðm KFik Þ2 Nr K1 iZ1 k iZ1
3. Interpretation of simulation process First, we construct a common four layers hierarchical model with 156 vertexes, and in order to compare with this model in the same scale, we create a scale-free model also including 156 vertexes following Eq. (1). Each member has an ID number to specify the sequence. The ingredients [S, D, A] were also engendered in the way that we present in Section 2.1. Then we select a vertex to be a remarkable member. In hierarchical model, we appoint the vertex in top level as remarkable member and endow it with SZ1. In scale-free model, we pick the vertex whose degree is the greatest in a whole, and set its variable S to 1. And we start to simulate the knowledge transfer, at every time step, we select a member by ordinal rank of ID and compute the S of it by using Eqs. (1)–(6). Finally, after all the S of members was calculated, we can get the results of knowledge transfer in each model.
4. Results and discussion The results of the computational experiments are summarized in Table 1 and Fig. 3. The values in Table 1 represent the performance of the knowledge transfer. We present the results in high aspiration and normal aspiration Table 1 Performance of the knowledge transfer
High aspiration Normal aspiration
Hierarchy model
Scale-free model
Mean
Std. dev.
Mean
Std. dev.
49.08 26.21
14.51 19.71
115.02 74.17
21.99 43.01
Note: Average over 400 simulations.
F. Tang et al. / Expert Systems with Applications 30 (2006) 796–800
799
Fig. 3. Histogram of the frequency count with superimposed normal density.
of the remarkable member, because we have found that if the remarkable member has a very high aspiration AZ1, then the performance of knowledge transfer will be very different from the normal aspiration A randomized in 0 to 1. The values in Table 1 indicate that the scale-free organization has higher average number of members acquiring knowledge than the hierarchy model whether the remarkable member has high aspiration or normal aspiration. And the results also indicate that if the remarkable member has extremely high aspiration to make all the colleagues which were directly linked to it access to the knowledge, the standard deviation of the results will be lower than that which has normal aspiration, even though the mean of acquired members is much higher. Fig. 3(c) and (d) shows that when the remarkable member has normal aspiration, the column 1 has the greatest amount of member which means in many simulations, especially more than half of the total simulations in hierarchical model, the knowledge will not be carried out more than 4 or 5 members. But in Fig. 3(a) and (b), as the aspiration of the remarkable member was set at 1, the average number was increased and it seems that the frequency of hierarchical structure follows the assumption of normal distribution.
5. Conclusion The findings generated by this study demonstrate that the characteristics of organization structure affect performance
of the intra-organizational knowledge transfer. As computational experiments accumulate, the scale-free structure contributes to better knowledge transfer, and the aspiration of remarkable member is also important to the results of knowledge transfer.
Acknowledgements We would like to thank Wenqiang Dai for offering the helps on computer program. Also, we would like to thank, in Part National Science Excellent Innovation Research Group Fund of China under contract No. 7012001, National Science Fund of China through the project No. 70233001, and Postdoctoral Foundation of China under contract No. 2005038072 for their support.
References Argote, L., McEvily, B., & Reagans, R. (2003). Introduction to the special issue on managing knowledge in organizations: Creating, Retaining, and Transferring knowledge. Management Science, 49(4). Baldwin, T. T., & Ford, J. K. (1988). Transfer of training: A review and direction for future research. Personnel Psychology, 41, 63–105. Baraba´si, A-L. , & Albert, R. (1999). Emergence of scaling in random networks. Science, 286, 509–512. Baraba´si, A-L. , Albert, R., & Jeong, H. (1999). Mean-field theory for scalefree random networks. Physica A, 272, 173–187. Baum, J. A. C., & Ingram, P. (1998). Survival-enhancing learning in the Manhattan hotel industry, 1898–1980. Management Science, 44, 996–1016.
800
F. Tang et al. / Expert Systems with Applications 30 (2006) 796–800
Cohen, W., & Levinthal, D. (1990). Absorptive capacity: A new perspective on learning and innovation. Administrative Science Quarterly, 35, 128–152. Darr, E. D., & Kurtzberg, T. R. (2000). An investigation of partner similarity dimensions on knowledge transfer. Organizational Behavior and Human Decision Processes, 82, 28–44. Hansen, M. (1999). The search-transfer problem: The role of weak ties in sharing knowledge across organization subunits. Administrative Science Quarterly, 44, 82–111. Haykin, S. (1994). Neural networks—A comprehensive foundation. New York: Macmillan College Publishing Company. Hertz, J., Krogh, A., & Palmer, R. G. (1991). Introduction to the theory of neural computation. Redwood City: Addison-Wesley. Lippman, R. P. (1987). An introduction to computing with neural nets. IEEE ASSP Magazine, 4, 4–22. Luce, R. D. (1963). Detection and recognition. In R. D. Luce, R. R. Bush, & S. E. Galanter, Handbook of mathematical psychology (vol. 1). New York: Wiley.
McEvily, B., & Zaheer, A. (1999). Bridging ties: A source of firm heterogeneity in competitive capabilities. Strategic Management Journal, 20, 1133–1156. Reagans, R., & McEvily, B. (2003). Network structure and knowledge transfer: The effects of cohesion and range. Administrative Science Quarterly, 48, 240–267. Simon, H. (1991). Bounded rationality and organizational learning. Organization Science, 2, 125–134. Szulanski, G. (2000). The process of knowledge transfer: A diachronic analysis of stickiness. Organizational Behavior and Human Decision Processes, 82, 9–27. Xi, Youmin, & Tang, Fangcheng (2004). Multiplex multi-core pattern of network organizations: An exploratory study. Computational & Mathematical Organization Theory, 10, 179–195. Uzzi, B. (1997). Social structure and competition in interfirm networks: The paradox of embeddedness. Administrative Science Quarterly, 42, 35–67.
Estimating the effect of organizational structure on knowledge transfer: A neural network approach Fangcheng Tanga,*, Youmin Xib, Jun Mab a
School of Economics and Management, Tsinghua University, Beijing 100084, People’ Republic of China b School of Management, Xia´n Jiaotong University, Xia´n 710049, People’s Republic of China
Abstract Artificial neural network has been put into abundant applications in social science research recently. In this study, we investigate the topological structures of organization network, which can possibly account for the different performances of intra-organizational knowledge transfer. We construct two types of networks including hierarchy and scale-free networks, and single-layer perceptron model (SLPM) was used to simulate the knowledge transfer from a remarkable member to the others. The statistical results indicate that although the performance of knowledge transfer is related to the aspiration of the remarkable member to transfer knowledge, but the scale-free structure is more effective in knowledge transfer than that in hierarchy structure. q 2005 Elsevier Ltd. All rights reserved. Keywords: Knowledge transfer; Organization; Neural network; Scale-free
1. Introduction The volume of knowledge management research in organizations has increased in recent years, as it has in many disciplines, including organizational behavior and theory, information systems, psychology, sociology, economics and strategy. Research on knowledge management focuses on various aspects of the three fundamental questions: how organizations create, retain, and transfer knowledge (Argote, McEvily, & Reagans, 2003). Indeed, the growing empirical evidence indicates that organizations that are able to transfer knowledge effectively are more productive and more likely to survive than those are less adept at knowledge transfer (Baum & Ingram, 1998). Several academic literatures have developed that many factors affect knowledge transfer in organizations such as similarity between tasks (Darr & Kurtzberg, 2000), characteristics of the source of knowledge, the recipient, the context, and the knowledge itself (Szulanski, 2000), characteristics of individual member (Baldwin & Ford, 1988), characteristics of the social network (McEvily & Zaheer, 1999), the nature * Tel.: C86 10 62775888; fax: C86 10 62775888. E-mail address: [email protected]
0957-4174/$ - see front matter q 2005 Elsevier Ltd. All rights reserved. doi:10.1016/j.eswa.2005.07.039
of the social ties (Hansen, 1999), network structure (Reagans & McEvily, 2003). Generally speaking, knowledge transfer is considered in two aspects, one is associative learning and absorptive capacity (Cohen & Levinthal, 1990; Simon´, 1991), the other is tie strength (Hansen, 1999; Uzzi, 1997), social cohesion and network range (Reagans & McEvily, 2003). However, little concerning the effect of characteristics of network structure on knowledge transfer is taken into account. In particular, when the structure of organization network become more complex increasingly, how to estimate the performance of knowledge transfer in this type of organizational network structure have been little considered. Recently, many large networks were found that the vertex connectivities follow a scale-free power-law distribution (Baraba´si & Albert, 1999), especially in informal networks such as friendship networks, citation patterns networks and WWW networks. Here, we try to find comparatively, if this new form of organization networks will influence the convection of knowledge. In this study, we examined the topology of organization network and its effect on the ability of individuals to access to knowledge from the remarkable member in the same unit or organization. We believed that the ability of individuals to possess new knowledge would be different by changing the topology of networks. Specifically, our study focused on the different performance of knowledge transfer between two organizational network topologies: hierarchical and
F. Tang et al. / Expert Systems with Applications 30 (2006) 796–800
scale-free networks. We addressed this idea by utilizing single-layer perceptron model (SLPM) of neural network (Haykin, 1994; Hertz, Krogh, & Palmer, 1991; Lippman, 1987) to simulate the knowledge transfer process in an organization, that consists of 100 members which are linked by hierarchical or scale-free preferential attachment mechanism.
2. Organizational network models Xi and Tang (2004) think that an organization can be considered as graph, each vertex represents a member in the organization and edges indicate the communication paths between them. Thus, without loss of generality, an organization network can be measured by graph theoretic approach in this form: GZ(V, E), where V denotes the set of members and E is a token of the relations set in the networks (Xi & Tang, 2004). Two organization models both of which have 100 members are constructed. One is a simple hierarchy model, the other is a scale-free network with preferentialattachment (see Fig. 1). Fig. 1 shows the hierarchical and scale-free network model, and (a) shows hierarchical model for vertical organizations, there is a vertex at the top of the organization and each vertex in the organization have five underling vertexes. In this model, every vertex is only influenced by the vertex that superior to it directly. (b) shows scale-free model introduced by BA. At tZ0 the system consists m0Z3 isolated vertices. At every time step, a new vertex is added, which is connected to mZ2 vertices that already present in the model and absorb the knowledge from them, preferentially to the vertices with high connectivity, determined by the preferential attachment rule. These two types of network model were examined in our paper. A hierarchy model includes four layer hierarchies with one vertex at the top level, and each node has five subordinates, therefore the model has 156 vertexes. We simplify this model to a vertical organization model that can only provide vertical information flow. A scale-free model is defined in two steps (Baraba´si et al., 1999): (1) Growth. Starting with a small number (m0Z3) of vertices, a new vertex with mZ2 (%m0) edges (link the new vertex to the vertices that already exist in the
797
system) was added to the network at every time step. (In order to construct an equal size model to the hierarchy’s, we limited the scope growth up to 156 vertices). (2) Preferential attachment. The probability P that a new vertex will be connected to vertex i depends on the connectivity ki of that vertex, so that .X Y ki Z ki kj (1) j
After 153 timesteps, we get a scale-free model in which the probability that a vertex has k edges following a power law exponent gmodelZ2.9G0.1. We are going to explore the knowledge transfer process by simulating how a remarkable person spread the particular knowledge to the other colleagues. 2.1. Vertex status variable scheme We assume that every member in the organization have three ingredients VZ[S, D, A]. The first is a signal variable S, which demonstrates whether a member obtained the particular knowledge, and value 1 means he successfully accesses to the knowledge and 0 means the opposite. The second, variable D, which is independent randomized uniformly distributed on [0, 1], represents the capacity of the members to learn and absorb particular knowledge. The larger value means less capacity, it may be determined by the inherence, background or the experience of those members. The last variable is A, it shows the aspiration level of the member to transfer knowledge to others. A is also a random variable uniformly distributed on [0, 1] and the member which has strong will to spread the knowledge is able to get high numerical value. 2.2. Interacting mechanism and mathematical definition of SLPM In this study, the SLPM is used to model the knowledge transfer stage, that is the mapping of the capacity of absorb knowledge to absorption probabilities. The structure of the SLPM used in our model is shown schematically in Fig. 2. In Fig. 2, the top row of small circles represents the input layer including the bias. The middle row of large circles
Fig. 1. Hierarchical and scale-free model of organization network.
798
F. Tang et al. / Expert Systems with Applications 30 (2006) 796–800
Note that this normalization step is not traditionally part as probabilities of the SLPM. The pij can now beP Ninterpreted r because pij2[0, 1] and for ci, pij Z 1. The probabilities jZ1 that the model responds pij are interpreted as the probability with certain knowledge Ci acquired by agents when it is assigned with absorptive capacity Di. Generally, before being used as model input, the set of values for each level is standardized over all absorptive capacity using F Kmk F~ ik Z ik sk
Fig. 2. Schematic representation of the SLPM.
represents the output layer. The symbols S and s represent summation and sigmoid transformation, respectively. The SLPM consists of an input layer and an output layer and no hidden layers. The level of absorptive capacity is clamped to the input nodes, represented as the top row of circles in Fig. 2. The input nodes pass the features unchanged. One input node is assigned to each level of absorptive capacity. The SLPM in Fig. 2 has 4 input nodes. The top right circle with the number ‘1’ is the bias node. Instead of transferring a feature value, this node simply outputs a fixed value 1. All input nodes, including the bias node, are connected to all output nodes. A weight is associated with each connection. The feature value passing through the connection is multiplied by the respective weight before reaching the output node. The weights are the parameters of the model. The number of weights Nu, including biases, in the SLPM equals Nu Z ðNF C 1ÞNr
(2)
In the output nodes two processing steps take place. First, the weighted feature values of absorptive capacity Di are summed, yielding a quantity dij2R: dij Z bj C
NF X
ukj Fik
(3)
kZ1
where bj2R is the bias, which is the weight between the bias node and output node j, and ukj2R is the weight between input node k and output node j. Next, dij is passed through a sigmoid activation function yielding a quantity sij defined by 1 sij Z 1 C expðKdij Þ
(4)
Note that Sij2(0, 1), and limdij/KN sij ðdij ÞZ 0 and limdij/N sij ðdij ÞZ 1. Finally, using Luce’s (unbiased) choice rule (Luce, 1963; chapter 3), the outputs sij of the output nodes are normalized, yielding the quantity pij: sij pij Z N Pr sil lZ1
(5)
(6)
Where Fik and F~ ik are the original and standardized values of level k of absorptive capacity Di, and vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi u Nr Nr u 1 X X mk Z 1=Nr Fik ; sk Z t ðm KFik Þ2 Nr K1 iZ1 k iZ1
3. Interpretation of simulation process First, we construct a common four layers hierarchical model with 156 vertexes, and in order to compare with this model in the same scale, we create a scale-free model also including 156 vertexes following Eq. (1). Each member has an ID number to specify the sequence. The ingredients [S, D, A] were also engendered in the way that we present in Section 2.1. Then we select a vertex to be a remarkable member. In hierarchical model, we appoint the vertex in top level as remarkable member and endow it with SZ1. In scale-free model, we pick the vertex whose degree is the greatest in a whole, and set its variable S to 1. And we start to simulate the knowledge transfer, at every time step, we select a member by ordinal rank of ID and compute the S of it by using Eqs. (1)–(6). Finally, after all the S of members was calculated, we can get the results of knowledge transfer in each model.
4. Results and discussion The results of the computational experiments are summarized in Table 1 and Fig. 3. The values in Table 1 represent the performance of the knowledge transfer. We present the results in high aspiration and normal aspiration Table 1 Performance of the knowledge transfer
High aspiration Normal aspiration
Hierarchy model
Scale-free model
Mean
Std. dev.
Mean
Std. dev.
49.08 26.21
14.51 19.71
115.02 74.17
21.99 43.01
Note: Average over 400 simulations.
F. Tang et al. / Expert Systems with Applications 30 (2006) 796–800
799
Fig. 3. Histogram of the frequency count with superimposed normal density.
of the remarkable member, because we have found that if the remarkable member has a very high aspiration AZ1, then the performance of knowledge transfer will be very different from the normal aspiration A randomized in 0 to 1. The values in Table 1 indicate that the scale-free organization has higher average number of members acquiring knowledge than the hierarchy model whether the remarkable member has high aspiration or normal aspiration. And the results also indicate that if the remarkable member has extremely high aspiration to make all the colleagues which were directly linked to it access to the knowledge, the standard deviation of the results will be lower than that which has normal aspiration, even though the mean of acquired members is much higher. Fig. 3(c) and (d) shows that when the remarkable member has normal aspiration, the column 1 has the greatest amount of member which means in many simulations, especially more than half of the total simulations in hierarchical model, the knowledge will not be carried out more than 4 or 5 members. But in Fig. 3(a) and (b), as the aspiration of the remarkable member was set at 1, the average number was increased and it seems that the frequency of hierarchical structure follows the assumption of normal distribution.
5. Conclusion The findings generated by this study demonstrate that the characteristics of organization structure affect performance
of the intra-organizational knowledge transfer. As computational experiments accumulate, the scale-free structure contributes to better knowledge transfer, and the aspiration of remarkable member is also important to the results of knowledge transfer.
Acknowledgements We would like to thank Wenqiang Dai for offering the helps on computer program. Also, we would like to thank, in Part National Science Excellent Innovation Research Group Fund of China under contract No. 7012001, National Science Fund of China through the project No. 70233001, and Postdoctoral Foundation of China under contract No. 2005038072 for their support.
References Argote, L., McEvily, B., & Reagans, R. (2003). Introduction to the special issue on managing knowledge in organizations: Creating, Retaining, and Transferring knowledge. Management Science, 49(4). Baldwin, T. T., & Ford, J. K. (1988). Transfer of training: A review and direction for future research. Personnel Psychology, 41, 63–105. Baraba´si, A-L. , & Albert, R. (1999). Emergence of scaling in random networks. Science, 286, 509–512. Baraba´si, A-L. , Albert, R., & Jeong, H. (1999). Mean-field theory for scalefree random networks. Physica A, 272, 173–187. Baum, J. A. C., & Ingram, P. (1998). Survival-enhancing learning in the Manhattan hotel industry, 1898–1980. Management Science, 44, 996–1016.
800
F. Tang et al. / Expert Systems with Applications 30 (2006) 796–800
Cohen, W., & Levinthal, D. (1990). Absorptive capacity: A new perspective on learning and innovation. Administrative Science Quarterly, 35, 128–152. Darr, E. D., & Kurtzberg, T. R. (2000). An investigation of partner similarity dimensions on knowledge transfer. Organizational Behavior and Human Decision Processes, 82, 28–44. Hansen, M. (1999). The search-transfer problem: The role of weak ties in sharing knowledge across organization subunits. Administrative Science Quarterly, 44, 82–111. Haykin, S. (1994). Neural networks—A comprehensive foundation. New York: Macmillan College Publishing Company. Hertz, J., Krogh, A., & Palmer, R. G. (1991). Introduction to the theory of neural computation. Redwood City: Addison-Wesley. Lippman, R. P. (1987). An introduction to computing with neural nets. IEEE ASSP Magazine, 4, 4–22. Luce, R. D. (1963). Detection and recognition. In R. D. Luce, R. R. Bush, & S. E. Galanter, Handbook of mathematical psychology (vol. 1). New York: Wiley.
McEvily, B., & Zaheer, A. (1999). Bridging ties: A source of firm heterogeneity in competitive capabilities. Strategic Management Journal, 20, 1133–1156. Reagans, R., & McEvily, B. (2003). Network structure and knowledge transfer: The effects of cohesion and range. Administrative Science Quarterly, 48, 240–267. Simon, H. (1991). Bounded rationality and organizational learning. Organization Science, 2, 125–134. Szulanski, G. (2000). The process of knowledge transfer: A diachronic analysis of stickiness. Organizational Behavior and Human Decision Processes, 82, 9–27. Xi, Youmin, & Tang, Fangcheng (2004). Multiplex multi-core pattern of network organizations: An exploratory study. Computational & Mathematical Organization Theory, 10, 179–195. Uzzi, B. (1997). Social structure and competition in interfirm networks: The paradox of embeddedness. Administrative Science Quarterly, 42, 35–67.