Mining and explicitation of tacit knowledge in enterprise knowledge base

Abstract

The application relates to the technical field of knowledge management and data processing in the field of enterprise information, and discloses a method for mining and making explicit implicit knowledge in an enterprise knowledge base. In a preset period, collaboration documents and employees are extracted from the knowledge base, a binary collaboration adjacency matrix is constructed according to participation relations, a standard Laplace is formed according to a degree vector, a degree matrix and the adjacency matrix, characteristic decomposition is carried out, spectral representation is formed according to the employee dimension, the mean value of the characteristic value and spectral truncation are calculated, a spectral enhancement function is constructed, spectral energy is normalized, spectral intermediate enhancement values are obtained by weighting the normalized collaboration degree, spectral enhancement and spectral components, implicit collaborators are sorted and screened after eliminating explicit roles, an explicit mapping matrix is generated according to the participation relations between the implicit collaborators and the documents, and long-term implicit collaborators are marked across periods.

Description

Technical Field

[0001] This invention relates to the field of knowledge management and data processing technology in the field of enterprise informatization, specifically to methods for mining and making explicit tacit knowledge in enterprise knowledge bases. Background Technology

[0002] As enterprises deepen their digital transformation, enterprise knowledge bases, as centralized carriers of internal information and experience, play an increasingly important role in supporting project management, employee collaboration, and innovation activities. Enterprise knowledge bases not only contain a large amount of explicit knowledge (such as process specifications, technical documents, and project reports), but also a wealth of tacit knowledge. This tacit knowledge is often hidden within the data structure in the form of collaborative relationships, project interactions, and communication networks. Tacit knowledge includes, but is not limited to, informal collaboration networks among employees, knowledge transfer at key nodes in projects, and implicit resource allocation in cross-departmental collaborations. This content is difficult to directly identify and effectively utilize under traditional document management and information retrieval systems.

[0003] Existing technologies primarily employ methods based on text mining, rule statistics, and simple graph analysis to discover tacit knowledge in enterprise knowledge bases. For example, they use keyword matching, entity recognition, and social network analysis to extract explicit relationships between documents or individuals; or they utilize basic indicators such as graph centrality and community segmentation to quantitatively analyze knowledge networks. These methods often only uncover surface-level collaborative relationships, failing to deeply reveal tacit knowledge that is not directly recorded but plays a crucial role in enterprise collaboration, such as hidden opinion leaders, knowledge flow in weak connections, and the bridging role of key nodes. Especially in large enterprises or complex project networks, due to the multidimensional heterogeneity and non-linear characteristics of knowledge flow, traditional centrality analysis and clustering techniques struggle to characterize potential knowledge dissemination pathways, easily overlooking structural "mediators" or collaborative relationships with significant actual influence. Furthermore, current mainstream technologies still have many limitations in practical applications. For example, some methods rely heavily on labeled data, rule settings, or pre-defined social behavior patterns, making it difficult to adapt to the complex scenarios of continuously evolving enterprise knowledge bases and dynamic changes in members. Some machine learning or deep model-based technologies face practical obstacles such as scarce training samples, weak interpretability, and untraceable parameters. At the same time, existing collaborative network mining methods mostly adopt a static perspective, ignoring the dynamic evolution and temporal characteristics of the knowledge collaboration process, resulting in problems of lag and fragmentation in the mining of tacit knowledge.

[0004] Therefore, this case aims to propose a method for mining and making explicit tacit knowledge in enterprise knowledge bases. First, based on enterprise collaborative documents and employee participation relationships, a binary collaborative adjacency matrix is ​​constructed. Then, through spectral decomposition of the degree matrix and standard Laplace matrix in graph theory, the network structure is mapped to a low-dimensional spectral space. On this basis, a "spectral enhancement function" is introduced to reweight the energy distribution of different spectral indices to highlight employees with more prominent network mediation roles. Finally, by combining normalized collaboration degree and spectral mediation enhancement value, explicit roles are eliminated, implicit collaborators are screened out, and through multi-period analysis and mapping matrix visualization, persistent and stable implicit collaborators are marked as long-term implicit collaborators. Summary of the Invention

[0005] This invention provides a method for mining and making explicit tacit knowledge in enterprise knowledge bases, which helps to solve the problems mentioned in the background art above.

[0006] This invention provides the following technical solution: a method for mining and making explicit tacit knowledge in an enterprise knowledge base, comprising: Within a preset analysis period, retrieve the set of collaborative documents and the set of employees corresponding to the preset analysis period from the enterprise knowledge base, construct an employee collaboration graph according to the document participation relationship, and form a binary collaborative adjacency matrix. The number of edges for each employee is counted on the binary cooperative adjacency matrix to form a cooperative degree vector, a degree matrix is ​​constructed, and it is combined with the binary cooperative adjacency matrix to obtain the standard Laplace matrix. Perform eigenvalue decomposition on the standard Laplacian matrix to obtain eigenvalues ​​and corresponding eigenvectors, and combine the components according to the employee dimension to form a spectral space representation matrix. Calculate the mean of the eigenvalues, set the spectral cutoff dimension, use the index of each eigenvalue as the spectral index, construct the spectral enhancement function within the selected spectral index range, and normalize the spectral energy distribution in the spectral space; Calculate the normalized cooperation degree of each employee, and weight the normalized cooperation degree with the spectral enhancement function and the employee components in the spectral space representation matrix to obtain the spectral mediation enhancement value of each employee. Employees with explicit roles are removed from the explicit role set. The spectral mediation enhancement values ​​of the remaining employees are used to generate implicit spectral values. The values ​​are then sorted and the implicit collaborator set is selected according to a preset ratio. Construct an implicit-document explicit mapping matrix based on the participation relationship between the implicit collaborator set and the collaborative document set; The set of implicit collaborators for each period is recorded in multiple analysis periods. Employees who are continuously identified are marked as long-term implicit collaborators and their labels are written into the enterprise knowledge base.

[0007] Optionally, the step of obtaining the set of collaborative documents and the set of employees corresponding to the preset analysis period from the enterprise knowledge base within the preset analysis period, constructing an employee collaboration graph according to the document participation relationship, and forming a binary collaboration adjacency matrix specifically includes: Set the analysis cycle length, collect all collaborative documents from the enterprise knowledge base according to the preset analysis cycle, and number each collaborative document in time or index order to form the collaborative document set for this analysis cycle. Extract all employees that appeared in the collaborative document set, number each employee, and form the set of employees who participated in the collaboration in this analysis period. Construct a document participation matrix based on the collaborative document number and employee number. Record the participation status for each document and each employee in the matrix. When an employee participates in a collaborative document, record the participation status in the corresponding position. When an employee does not participate in the corresponding collaborative document, record the non-participation status in the corresponding position. Based on the participation relationships recorded in the document participation matrix, a collaborative relationship is established between employees. For each pair of different employees, it is checked whether they participate in at least one collaborative document at the same time. If there are documents in common participation, the corresponding position in the binary collaborative adjacency matrix is ​​recorded as having collaboration. If there are no documents in common participation, it is recorded as not having collaboration, thus forming a binary collaborative adjacency matrix with employees as nodes. Summing all elements in the binary cooperative adjacency matrix yields the total number of cooperative edges in the current analysis period. When the total number of cooperative edges is zero, the spectral mediation enhancement value of all employees is uniformly set to zero, and the spectral analysis process ends. When the total number of cooperative edges is greater than zero, the spectral mediation enhancement value remains in a state to be calculated, and the process proceeds to the next step.

[0008] Optionally, the step of counting the number of edges for each employee on the binary cooperative adjacency matrix to form a cooperative degree vector, constructing a degree matrix, and combining it with the binary cooperative adjacency matrix to obtain a standard Laplace matrix specifically includes: In the binary collaborative adjacency matrix, the sum of the rows or columns corresponding to each employee is calculated to count the number of edges that the corresponding employee has collaborative relationships with other employees in the current analysis period. The counted number of edges is taken as the degree of collaboration of the corresponding employee. Arrange all employees' collaboration scores in order of employee ID, fill in the collaboration scores of each employee in the diagonal positions, and fill in zero values ​​in the off-diagonal positions to construct a degree matrix with collaboration scores located in the diagonal elements. Based on the degree matrix and the binary cooperative adjacency matrix, subtraction is performed on corresponding positions of the matrix to obtain the standard Laplace matrix.

[0009] Optionally, the step of performing eigenvalue decomposition on the standard Laplacian matrix to obtain eigenvalues ​​and corresponding eigenvectors, and combining the components according to the employee dimension to form a spectral space representation matrix, specifically includes: Perform eigenvalue decomposition on the standard Laplacian matrix to obtain multiple combinations of eigenvalues ​​and eigenvectors. Each combination contains an eigenvalue and an eigenvector corresponding to the eigenvalue, and the index of each eigenvalue is used as the spectral index. For each set of feature vectors, the component values ​​corresponding to each employee are read one by one in the employee dimension. The components of the corresponding feature vectors are arranged into an employee component sequence according to the employee number order. The employee components of all feature vectors are organized in the spectral index dimension to form a spectral space representation matrix.

[0010] Optionally, the calculation of the mean of eigenvalues, setting the spectral truncation dimension, using the index of each eigenvalue as a spectral index, constructing a spectral enhancement function within the selected spectral index range, and normalizing the spectral energy distribution in the spectral space specifically includes: After summing all the eigenvalues, divide the sum by the number of eigenvalues ​​to obtain the mean of the eigenvalues; The spectral truncation dimension is set based on the total number of employees. Specifically, the total number of employees is calculated as half and rounded down. The rounded value is then compared with a constant two and the larger value is taken. Finally, the comparison result is compared with the total number of employees and the smaller value is taken. The final result is used as the spectral truncation dimension. Within the range of the spectral index from the second to the spectral truncation dimension, for each spectral index corresponding to an eigenvalue, a spectral enhancement function is constructed based on the sum of the corresponding eigenvalue and the mean of the eigenvalues. When the mean of the eigenvalues ​​is greater than zero, the reciprocal of the sum of the eigenvalues ​​and the mean of the eigenvalues ​​is used as the spectral enhancement function value at the corresponding spectral index. When the mean of the eigenvalues ​​is equal to zero, a constant value of one is used as the spectral enhancement function value at the corresponding spectral index. Within the selected spectral index range, a one-to-one correspondence is established between the spectral enhancement function values ​​and the corresponding eigenvector components. A weighted operation is then performed on the spectral enhancement function values ​​and corresponding components at each spectral index to form a spectral component sequence after spectral enhancement.

[0011] Optionally, the calculation of the normalized cooperation degree of each employee, by weighting the normalized cooperation degree with the spectral enhancement function and the employee components in the spectral space representation matrix to obtain the spectral mediation enhancement value of each employee, specifically includes: Traverse the collaboration degree of each employee in the binary collaboration adjacency matrix, select the maximum collaboration degree from all collaboration degrees, and record the maximum collaboration degree as the maximum collaboration degree of the current analysis period. When the maximum collaboration degree is greater than zero, for each employee, the collaboration degree of the employee is divided by the maximum collaboration degree to obtain the normalized collaboration degree of each employee. When the maximum collaboration degree is equal to zero, the normalized collaboration degree of all employees is uniformly set to zero. When the total number of collaborative edges is zero, the spectral mediation enhancement value of all employees is uniformly set to zero. When the total number of collaborative edges is greater than zero, for each employee, in the order from the second spectral index to the spectral truncation dimension, the spectral enhancement function value at the corresponding spectral index and the square of the corresponding component in the employee's spectral space representation matrix are multiplied item by item and summed. The summation result is multiplied by the result of adding one to the normalized collaboration degree of each employee to obtain the spectral mediation enhancement value of the employee.

[0012] Optionally, the step of removing employees with explicit roles from the explicit role set, generating implicit spectral values ​​from the spectral mediation enhancement values ​​of the remaining employees, sorting them, and filtering the implicit collaborator set according to a preset ratio specifically includes: Based on organizational management information, a set of explicit roles is constructed. Employees who play a leading role in the organization are added to the set of explicit roles, and explicit role index information is recorded for each employee in the set of explicit roles. When the total number of collaborative edges is zero, the implicit spectral values ​​of all employees are uniformly set to zero, and the set of implicit collaborators is set to an empty set; When the total number of collaborative edges is greater than zero, for each employee, a latent spectral value is constructed based on whether it belongs to the set of explicit roles; for employees who do not belong to the set of explicit roles, the spectral mediation enhancement value of the corresponding employee is used as the latent spectral value; for employees who belong to the set of explicit roles, the latent spectral value of the corresponding employee is set to zero. Provided that the total number of collaborative edges is greater than zero, all employees are sorted from largest to smallest according to their latent spectral values ​​to obtain the sorted employee number sequence. For the sorted employee ID sequence, the number of target hidden collaborators is calculated based on the total number of employees. Specifically, the number of hidden collaborators is calculated as follows: first, 10% of the total number of employees is calculated and rounded down, then the rounded value is compared with the number of employees and the larger value is taken as the number of target hidden collaborators. The number of employees with a latent spectrum value greater than zero is counted to obtain the number of latent candidates. The actual selection number is obtained based on the number of latent candidates and the number of target latent collaborators. When the number of latent candidates is zero, the actual selection number is set to zero. When the number of latent candidates is greater than zero, the smaller value between the number of latent candidates and the number of target latent collaborators is taken as the actual selection number. When the actual number of selections is zero, the set of implicit collaborators is set to empty. When the actual number of selections is greater than zero, the same number of employees as the actual number of selections are selected from the first few positions of the sorted employee number sequence, and the corresponding employees are added to the set of implicit collaborators.

[0013] Optionally, the construction of the implicit-document explicit mapping matrix based on the participation relationship between the implicit collaborator set and the collaborative document set specifically includes: Using each collaborative document in the collaborative document set as a row index and each employee in the employee set as a column index, an implicit mapping matrix between the collaborative document number and the employee number is constructed. For each combination of collaborative document and employee in the mapping matrix, check whether the employee in the combination belongs to the set of implicit collaborators and whether the collaborative document in the combination is marked as participating in the document participation matrix. For combinations that satisfy both the conditions of belonging to the set of implicit collaborators and being marked as participating in the document participation matrix, record one at the corresponding position in the mapping matrix; for combinations that do not satisfy both conditions at the same time, record zero at the corresponding position.

[0014] Optionally, the step of recording the set of implicit collaborators for each analysis period, marking continuously identified employees as long-term implicit collaborators, and writing long-term implicit collaborator tags into the enterprise knowledge base specifically includes: Set the number of analysis cycles, and perform multiple collaborative analyses on the same enterprise knowledge base consecutively according to the number of analysis cycles. Record the set of implicit collaborators obtained in the analysis cycle for each analysis, and assign a cycle index to each analysis. For each employee in the employee set, a long-term implicit count is constructed and initialized to zero. As the period index increases from the first period to the set number of analysis periods, it is determined period by period whether the employee belongs to the implicit collaborator set of the corresponding period. If the employee belongs to the implicit collaborator set, the long-term implicit count is incremented by one. If the employee does not belong to the implicit collaborator set, the long-term implicit count remains unchanged. After completing all analysis cycles, the long-term implicit counts of all employees are checked. Employees whose long-term implicit counts equal the number of analysis cycles are marked as long-term implicit collaborators. The corresponding long-term implicit collaborator labels are written into the explicit knowledge table of the enterprise knowledge base and the label status is maintained.

[0015] The present invention has the following beneficial effects: This approach accurately characterizes employee collaboration relationships using collaborative documents as the medium, rather than simply relying on project affiliation, departmental structure, or subjective claims. By extracting the set of collaborative documents and the set of participating employees within a preset analysis period, a document participation matrix is ​​first constructed. Then, a binary collaborative adjacency matrix between employees is derived from this matrix. Only employee pairs who have actually co-authored at least one document are considered to have collaborative edges. In this way, the network structure directly originates from knowledge creation and modification behaviors, eliminating noise from those who are only nominally affiliated with the same project but have no actual collaboration. Compared to existing methods that only count "shared projects" or "same department," this approach reflects the substantive collaborative relationships at the knowledge level with finer granularity, and can identify technical personnel who jointly write proposals across departments and repeatedly participate in modifying others' documents. In addition, this approach incorporates a judgment on the total number of collaborative edges. When there is no collaborative behavior within a period, the total number of edges is uniformly assigned to zero, and subsequent spectral analysis is terminated, avoiding artificially amplified spurious results in extremely sparse scenarios.

[0016] This approach transforms the intuitive concept of "cooperation frequency" into a structured matrix suitable for spectral analysis. First, the cooperation degree of each employee is accumulated row-wise or column-wise on the cooperation adjacency matrix, organized into a pairwise angle matrix, and then combined with the cooperation adjacency matrix to construct a standard Laplace matrix. Unlike simply ranking collaboration frequency, this scheme characterizes the overall network structure using standard graph theory structures: cooperation degree reflects the connection strength between employees and the network, while the standard Laplace matrix, while preserving degree information, characterizes the local and global connectivity properties of the network, providing a unified operator basis for subsequent spectral decomposition. Compared to existing schemes that rely solely on single indicators such as degree centrality and betweenness centrality, this scheme lays the foundation for multi-scale, multi-frequency analysis, enabling the simultaneous expression of "cooperation quantity" and "cooperation structure position" on the same matrix. This allows for the differentiation between "high-frequency but marginal collaborators" and "collaborators with moderate cooperation but in key connection positions."

[0017] The solution performs eigenvalue decomposition on the standard Laplacian matrix, mapping each employee to a spectral space vector. Each dimension in the spectral space corresponds to a structural pattern of the network, and the employee's component in that dimension reflects its position and role within that structural pattern. Instead of simply performing degree or path analysis on the original collaborative network, this approach introduces the "frequency domain" perspective of the spectral space, allowing for intuitive measurement of higher-order similarities, community structure, and bridging positions among employees through vector geometric relationships. While traditional methods for analyzing collaborative relationships often only handle local adjacency or betweenness based on the shortest path, they struggle to simultaneously consider the global structure. The spectral space representation compresses the overall network information into a finite-dimensional vector, facilitating subsequent combinations with enhancement functions, normalized degrees, etc. Furthermore, organizing the feature vectors into a unified spectral space representation matrix according to the employee dimension provides structural convenience for subsequent weighting and enhancement along the spectral index dimension, avoiding redundant calculations and unnecessary complexity.

[0018] The proposed solution first calculates the mean based on all eigenvalues, then sets the spectral truncation dimension according to the number of employees, avoiding distortion in large-scale or extremely sparse networks caused by fixed selection of the first few eigenvalues. Subsequently, the eigenvalue indices are used as spectral indices to construct an enhancement function within a selected range, assigning different weights to different spectral components based on their relationship to the mean. Compared to directly using unweighted eigenvectors or simple truncation, the spectral enhancement function can suppress unstable high-frequency noise components, improve sensitivity to structural intermediate locations, and avoid the over-dominance of low-frequency components that would prevent the inability to distinguish fine-grained structural differences. Compared to the common practice of simple spectral clustering or taking only a few principal eigenvectors in existing technologies, this solution, through enhancement functions and energy gauge, allows spectral analysis to both preserve the overall structure and actively amplify information density in key frequency bands, helping to more accurately characterize implicit key employees located at the intersection of multiple communities and bridging across teams.

[0019] The spectral mediation enhancement value is formed by combining the local connectivity strength of nodes with the structural information they carry in the spectral space. First, the collaboration degree of each employee is normalized using the maximum collaboration degree to avoid the incomparability of absolute values ​​between networks of different sizes, while preserving relative activity. Then, the normalized collaboration degree is combined with the spectral enhancement function and the employee's component in each spectral component to construct a comprehensive index that considers both "collaboration frequency" and "structural position." Unlike traditional betweenness centrality or specific path length indices, the spectral mediation enhancement value does not rely on global shortest path enumeration, making computational complexity easier to control, and it measures the influence of nodes on information propagation in the network from the perspective of spectral energy. This approach can identify employees who play a connecting role between multiple collaborative groups without misjudging employees who are structurally peripheral but participate in too many documents as core members, thus better aligning with the practical business meaning of the concept of implicit collaborators.

[0020] After obtaining the spectral mediation enhancement value, this solution introduces an explicit role set, eliminating personnel with existing formal authority or management responsibilities, and only ranking and screening the remaining employees based on their implicit spectral values. Traditional organizational analysis or expert mining methods often naturally regard employees with high-ranking and explicit positions as "key talents," which makes it difficult to discover those who, without management titles, play a crucial role in connecting key knowledge groups. This solution explicitly describes existing personnel with dominant functions through an explicit role set, sets their spectral mediation enhancement value to zero, removes them from the ranking, and then ranks them from high to low according to their implicit spectral values. It also sets a relationship between the number of target implicit collaborators and the number of effective candidates, ensuring a certain proportion of coverage while avoiding forced selection when there are too few candidates. Compared with existing techniques that only perform overall network ranking, this solution emphasizes the "implicit" perspective, effectively preventing the selected list from being dominated by formal managers, making the identification results closer to the "overlooked but crucial" target group, and possessing strong management implications and differentiated value.

[0021] After identifying implicit collaborators, this solution constructs an explicit mapping matrix between implicit and explicit documents, establishing a direct correspondence between implicit collaborators and the specific collaborative documents they participate in. The advantage of this design lies in bridging the "last mile" between "algorithm results" and "management operations": managers can not only see that an employee is marked as an implicit collaborator, but also clearly see in which key documents or knowledge areas that employee plays a role, and which project teams highly rely on their participation. For example, when it is discovered that an implicit collaborator is concentrated in a certain type of core technical document, targeted arrangements can be made for document co-editing, knowledge transfer, and job rotation, reducing the risk of knowledge gaps caused by the loss of key personnel.

[0022] The solution repeatedly identifies implicit collaborators across multiple analysis cycles, accumulating the number of times each employee is identified as an implicit collaborator in different cycles, forming a long-term implicit count. Only when an employee is consistently identified as an implicit collaborator throughout all or most analysis cycles is they marked as a long-term implicit collaborator, and this label is added to the explicit knowledge table of the enterprise knowledge base. Compared to "short-term hotspots" obtained from a one-time analysis, this multi-cycle stability determination better eliminates randomness, such as situations where temporarily taking on emergency response tasks at a certain stage leads to an abnormally high level of collaboration. Existing technologies mostly rely on static networks or single snapshot analysis; this solution, through time series overlay, distinguishes between "stable contributions" and "short-term fluctuations," ensuring that the long-term implicit collaborator label truly represents continuous cross-team collaboration value, providing more solid data support for performance evaluation, talent inventory, and key talent retention. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] Example, refer to Figure 1 Methods for mining and making explicit tacit knowledge in enterprise knowledge bases include: Within a preset analysis period, retrieve the set of collaborative documents and the set of employees corresponding to the preset analysis period from the enterprise knowledge base, construct an employee collaboration graph according to the document participation relationship, and form a binary collaborative adjacency matrix. The number of edges for each employee is counted on the binary cooperative adjacency matrix to form a cooperative degree vector, a degree matrix is ​​constructed, and it is combined with the binary cooperative adjacency matrix to obtain the standard Laplace matrix. Perform eigenvalue decomposition on the standard Laplacian matrix to obtain eigenvalues ​​and corresponding eigenvectors, and combine the components according to the employee dimension to form a spectral space representation matrix. Calculate the mean of the eigenvalues, set the spectral cutoff dimension, use the index of each eigenvalue as the spectral index, construct the spectral enhancement function within the selected spectral index range, and normalize the spectral energy distribution in the spectral space; Calculate the normalized cooperation degree of each employee, and weight the normalized cooperation degree with the spectral enhancement function and the employee components in the spectral space representation matrix to obtain the spectral mediation enhancement value of each employee. Employees with explicit roles are removed from the explicit role set. The spectral mediation enhancement values ​​of the remaining employees are used to generate implicit spectral values. The values ​​are then sorted and the implicit collaborator set is selected according to a preset ratio. Construct an implicit-document explicit mapping matrix based on the participation relationship between the implicit collaborator set and the collaborative document set; The set of implicit collaborators for each period is recorded in multiple analysis periods. Employees who are continuously identified are marked as long-term implicit collaborators and their labels are written into the enterprise knowledge base.

[0026] By systematically acquiring collaborative documents and employee sets within a preset analysis period and generating a binary collaborative adjacency matrix based on document participation relationships, this approach eliminates misjudgments of collaborative relationships based solely on nominal team or project affiliation. It directly bases the analysis on actual collaborative behavior, addressing the problem in existing technologies that often fail to reflect the true state of knowledge collaboration due to reliance on organizational structure or coarse-grained project labels. Subsequently, using the degree matrix and standard Laplacian matrix, the number of collaborative nodes is integrated with the overall network structural characteristics, providing a unified operator for subsequent spectral decomposition. This overcomes the limitation of using only degree centrality or path betweenness to simultaneously consider both global and local structures. In the spectral space, employees are mapped to multi-dimensional structural patterns through eigenvector components. Combined with adaptive spectral truncation and spectral enhancement functions, the energy distribution is normalized, suppressing spectral noise and enhancing key frequency band information. This overcomes the information loss or noise amplification caused by simple truncation or principal component selection in traditional spectral clustering. The spectral mediation enhancement value, formed by weighting normalized collaboration degree and spectral enhancement value, comprehensively reflects the activity and structural mediation role of nodes. It can identify cross-team bridging figures while avoiding misjudging those with high document participation but peripheral structures as core members. Based on this, explicit roles with formal management or key functions are eliminated, focusing on uncovering "overlooked but crucial" implicit collaborators, filling the blind spot of traditional methods that only focus on high-level positions or project leaders. By constructing an implicit-document explicit mapping matrix and multi-period stabilization labels, the algorithm results are closely integrated with the knowledge base, truly achieving a closed loop from abstract scores to specific personnel-document mapping, and then to long-term labels. This helps human resources and knowledge management departments develop targeted succession and backup strategies, improving the accuracy and operability of tacit knowledge identification.

[0027] The step of obtaining the set of collaborative documents and the set of employees corresponding to the preset analysis period from the enterprise knowledge base within the preset analysis period, constructing an employee collaboration graph according to the document participation relationship, and forming a binary collaboration adjacency matrix specifically includes: Set the analysis cycle length, collect all collaborative documents from the enterprise knowledge base according to the preset analysis cycle, and number each collaborative document in time or index order to form the collaborative document set for this analysis cycle. Extract all employees that appeared in the collaborative document set, number each employee, and form the set of employees who participated in the collaboration in this analysis period. Construct a document participation matrix based on the collaborative document number and employee number. Record the participation status for each document and each employee in the matrix. When an employee participates in a collaborative document, record the participation status in the corresponding position. When an employee does not participate in the corresponding collaborative document, record the non-participation status in the corresponding position. Based on the participation relationships recorded in the document participation matrix, a collaborative relationship is established between employees. For each pair of different employees, it is checked whether they participate in at least one collaborative document at the same time. If there are documents in common participation, the corresponding position in the binary collaborative adjacency matrix is ​​recorded as having collaboration. If there are no documents in common participation, it is recorded as not having collaboration, thus forming a binary collaborative adjacency matrix with employees as nodes. Summing all elements in the binary cooperative adjacency matrix yields the total number of cooperative edges in the current analysis period. When the total number of cooperative edges is zero, the spectral mediation enhancement value of all employees is uniformly set to zero, and the spectral analysis process ends. When the total number of cooperative edges is greater than zero, the spectral mediation enhancement value remains in a state to be calculated, and the process proceeds to the next step.

[0028] Further specific implementation steps include: setting the analysis period to... On that day, all knowledge collaboration documents from the enterprise knowledge base within that period were extracted, numbered as follows: in, No. One document; Total number of documents; It is a set of positive integers; This refers to the set of collaborative documents included in the calculations during this analysis period. For document indexing; Build all in The composition of employees that have appeared in the content: in, This is a collection of employees who participated in collaborative document processing during this period. For the number Employees; Total number of employees; For employees in the collection The number in; Building a document participation matrix The document participates in the matrix elements as follows: in, The document participation matrix has the following dimensions: ; , indicating employees Did you participate in the document? A binary indicator; Constructing a cooperative adjacency matrix The elements of the cooperative adjacency matrix are: ;in, This is a cooperative adjacency matrix with dimension . ; , indicating employees With employees A binary indicator indicating whether there is at least one collaborative relationship involving the same document within this period; For the number Employees; Index for employees; Calculate the total number of cooperative edges ;when When this condition is met, it indicates that there are no collaborative relationships within the selected analysis period. In this case, the spectral enhancement mediation value for all employees is directly defined as follows: , in, For employees Spectral enhancement mediation value; when If necessary, continue executing the process.

[0029] The process of counting the number of edges for each employee on the binary cooperative adjacency matrix to form a cooperative degree vector, constructing a degree matrix, and combining it with the binary cooperative adjacency matrix to obtain the standard Laplace matrix specifically includes: In the binary collaborative adjacency matrix, the sum of the rows or columns corresponding to each employee is calculated to count the number of edges that the corresponding employee has collaborative relationships with other employees in the current analysis period. The counted number of edges is taken as the degree of collaboration of the corresponding employee. Arrange all employees' collaboration scores in order of employee ID, fill in the collaboration scores of each employee in the diagonal positions, and fill in zero values ​​in the off-diagonal positions to construct a degree matrix with collaboration scores located in the diagonal elements. Based on the degree matrix and the binary cooperative adjacency matrix, subtraction is performed on corresponding positions of the matrix to obtain the standard Laplace matrix.

[0030] Further specific implementation steps include: For each employee, calculate the degree of collaboration: ;in, For employees The degree of cooperation; Let be a set of non-negative integers; combine all degree values ​​into a diagonal matrix: in, It is a degree matrix with dimension 1. .

Claims

1. A method for mining and making explicit tacit knowledge in an enterprise knowledge base, characterized in that, include: Within a preset analysis period, retrieve the set of collaborative documents and the set of employees corresponding to the preset analysis period from the enterprise knowledge base, construct an employee collaboration graph according to the document participation relationship, and form a binary collaborative adjacency matrix. The number of edges for each employee is counted on the binary cooperative adjacency matrix to form a cooperative degree vector, a degree matrix is ​​constructed, and it is combined with the binary cooperative adjacency matrix to obtain the standard Laplace matrix. Perform eigenvalue decomposition on the standard Laplacian matrix to obtain eigenvalues ​​and corresponding eigenvectors, and combine the components according to the employee dimension to form a spectral space representation matrix. Calculate the mean of the eigenvalues, set the spectral cutoff dimension, use the index of each eigenvalue as the spectral index, construct the spectral enhancement function within the selected spectral index range, and normalize the spectral energy distribution in the spectral space; Calculate the normalized cooperation degree of each employee, and weight the normalized cooperation degree with the spectral enhancement function and the employee components in the spectral space representation matrix to obtain the spectral mediation enhancement value of each employee. Employees with explicit roles are removed from the explicit role set. The spectral mediation enhancement values ​​of the remaining employees are used to generate implicit spectral values. The values ​​are then sorted and the implicit collaborator set is selected according to a preset ratio. Construct an implicit-document explicit mapping matrix based on the participation relationship between the implicit collaborator set and the collaborative document set; The set of implicit collaborators for each period is recorded in multiple analysis periods. Employees who are continuously identified are marked as long-term implicit collaborators and their labels are written into the enterprise knowledge base.

2. The method for mining and making explicit tacit knowledge in an enterprise knowledge base according to claim 1, characterized in that, The step of obtaining the set of collaborative documents and the set of employees corresponding to the preset analysis period from the enterprise knowledge base within the preset analysis period, constructing an employee collaboration graph according to the document participation relationship, and forming a binary collaboration adjacency matrix specifically includes: Set the analysis cycle length, collect all collaborative documents from the enterprise knowledge base according to the preset analysis cycle, and number each collaborative document in time or index order to form the collaborative document set for this analysis cycle. Extract all employees that appeared in the collaborative document set, number each employee, and form the set of employees who participated in the collaboration in this analysis period. Construct a document participation matrix based on the collaborative document number and employee number. Record the participation status for each document and each employee in the matrix. When an employee participates in a collaborative document, record the participation status in the corresponding position. When an employee does not participate in the corresponding collaborative document, record the non-participation status in the corresponding position. Based on the participation relationships recorded in the document participation matrix, a collaborative relationship is established between employees. For each pair of different employees, it is checked whether they participate in at least one collaborative document at the same time. If there are documents in common participation, the corresponding position in the binary collaborative adjacency matrix is ​​recorded as having collaboration. If there are no documents in common participation, it is recorded as not having collaboration, thus forming a binary collaborative adjacency matrix with employees as nodes. Summing all elements in the binary cooperative adjacency matrix yields the total number of cooperative edges in the current analysis period. When the total number of cooperative edges is zero, the spectral mediation enhancement value of all employees is uniformly set to zero, and the spectral analysis process ends. When the total number of cooperative edges is greater than zero, the spectral mediation enhancement value remains in a state to be calculated, and the process proceeds to the next step.

3. The method for mining and making explicit tacit knowledge in an enterprise knowledge base according to claim 2, characterized in that, The process of counting the number of edges for each employee on the binary cooperative adjacency matrix to form a cooperative degree vector, constructing a degree matrix, and combining it with the binary cooperative adjacency matrix to obtain the standard Laplace matrix specifically includes: In the binary collaborative adjacency matrix, the sum of the rows or columns corresponding to each employee is calculated to count the number of edges that the corresponding employee has collaborative relationships with other employees in the current analysis period. The counted number of edges is taken as the degree of collaboration of the corresponding employee. Arrange all employees' collaboration scores in order of employee ID, fill in the collaboration scores of each employee in the diagonal positions, and fill in zero values ​​in the off-diagonal positions to construct a degree matrix with collaboration scores located in the diagonal elements. Based on the degree matrix and the binary cooperative adjacency matrix, subtraction is performed on corresponding positions of the matrix to obtain the standard Laplace matrix.

4. The method for mining and making explicit tacit knowledge in an enterprise knowledge base according to claim 3, characterized in that, The process of performing eigenvalue decomposition on the standard Laplacian matrix to obtain eigenvalues ​​and corresponding eigenvectors, and combining the components according to the employee dimension to form a spectral space representation matrix, specifically includes: Perform eigenvalue decomposition on the standard Laplacian matrix to obtain multiple combinations of eigenvalues ​​and eigenvectors. Each combination contains an eigenvalue and an eigenvector corresponding to the eigenvalue, and the index of each eigenvalue is used as the spectral index. For each set of feature vectors, the component values ​​corresponding to each employee are read one by one in the employee dimension. The components of the corresponding feature vectors are arranged into an employee component sequence according to the employee number order. The employee components of all feature vectors are organized in the spectral index dimension to form a spectral space representation matrix.

5. The method for mining and making explicit tacit knowledge in an enterprise knowledge base according to claim 4, characterized in that, The calculation of the mean of eigenvalues, setting the spectral cutoff dimension, using the index of each eigenvalue as a spectral index, constructing a spectral enhancement function within the selected spectral index range, and normalizing the spectral energy distribution in the spectral space specifically includes: After summing all the eigenvalues, divide the sum by the number of eigenvalues ​​to obtain the mean of the eigenvalues; The spectral truncation dimension is set based on the total number of employees. Specifically, half of the total number of employees is calculated and rounded down. The rounded value is then compared with a constant two and the larger value is taken. Finally, the comparison result is compared with the total number of employees and the smaller value is taken. The final result is used as the spectral truncation dimension. Within the range of the spectral index from the second to the spectral truncation dimension, for each spectral index corresponding to an eigenvalue, a spectral enhancement function is constructed based on the sum of the corresponding eigenvalue and the mean of the eigenvalues. When the mean of the eigenvalues ​​is greater than zero, the reciprocal of the sum of the eigenvalues ​​and the mean of the eigenvalues ​​is used as the spectral enhancement function value at the corresponding spectral index. When the mean of the eigenvalues ​​is equal to zero, a constant value of one is used as the spectral enhancement function value at the corresponding spectral index. Within the selected spectral index range, a one-to-one correspondence is established between the spectral enhancement function values ​​and the corresponding eigenvector components. A weighted operation is then performed on the spectral enhancement function values ​​and corresponding components at each spectral index to form a sequence of spectral components after spectral enhancement.

6. The method for mining and making explicit tacit knowledge in an enterprise knowledge base according to claim 5, characterized in that, The calculation of the normalized cooperation degree of each employee involves weighting the normalized cooperation degree with the spectral enhancement function and the employee components in the spectral space representation matrix to obtain the spectral mediation enhancement value of each employee. Specifically, this includes: Traverse the collaboration degree of each employee in the binary collaboration adjacency matrix, select the maximum collaboration degree from all collaboration degrees, and record the maximum collaboration degree as the maximum collaboration degree of the current analysis period. When the maximum collaboration degree is greater than zero, for each employee, the collaboration degree of the employee is divided by the maximum collaboration degree to obtain the normalized collaboration degree of each employee. When the maximum collaboration degree is equal to zero, the normalized collaboration degree of all employees is uniformly set to zero. When the total number of collaborative edges is zero, the spectral mediation enhancement value of all employees is uniformly set to zero. When the total number of collaborative edges is greater than zero, for each employee, in the order from the second spectral index to the spectral truncation dimension, the spectral enhancement function value at the corresponding spectral index and the square of the corresponding component in the employee's spectral space representation matrix are multiplied item by item and summed. The summation result is multiplied by the result of adding one to the normalized collaboration degree of each employee to obtain the spectral mediation enhancement value of the employee.

7. The method for mining and making explicit tacit knowledge in an enterprise knowledge base according to claim 6, characterized in that, The process of removing employees with explicit roles from the explicit role set, generating implicit spectral values ​​from the spectral mediation enhancement values ​​of the remaining employees, sorting them, and filtering the implicit collaborator set according to a preset ratio specifically includes: Based on organizational management information, a set of explicit roles is constructed. Employees who play a leading role in the organization are added to the set of explicit roles, and explicit role index information is recorded for each employee in the set of explicit roles. When the total number of collaborative edges is zero, the implicit spectral values ​​of all employees are uniformly set to zero, and the set of implicit collaborators is set to an empty set; When the total number of collaborative edges is greater than zero, for each employee, a latent spectral value is constructed based on whether it belongs to the set of explicit roles; for employees who do not belong to the set of explicit roles, the spectral mediation enhancement value of the corresponding employee is used as the latent spectral value; for employees who belong to the set of explicit roles, the latent spectral value of the corresponding employee is set to zero. Provided that the total number of collaborative edges is greater than zero, all employees are sorted from largest to smallest according to their latent spectral values ​​to obtain the sorted employee number sequence. For the sorted employee ID sequence, the number of target hidden collaborators is calculated based on the total number of employees. Specifically, the number of hidden collaborators is calculated as follows: first, 10% of the total number of employees is calculated and rounded down, then the rounded value is compared with the number of employees and the larger value is taken as the number of target hidden collaborators. The number of employees with a latent spectrum value greater than zero is counted to obtain the number of latent candidates. The actual selection number is obtained based on the number of latent candidates and the number of target latent collaborators. When the number of latent candidates is zero, the actual selection number is set to zero. When the number of latent candidates is greater than zero, the smaller value between the number of latent candidates and the number of target latent collaborators is taken as the actual selection number. When the actual number of selections is zero, the set of implicit collaborators is set to empty. When the actual number of selections is greater than zero, the same number of employees as the actual number of selections are selected from the first few positions of the sorted employee number sequence, and the corresponding employees are added to the set of implicit collaborators.

8. The method for mining and making explicit tacit knowledge in an enterprise knowledge base according to claim 7, characterized in that, The construction of the implicit-to-explicit document mapping matrix based on the participation relationship between the implicit collaborator set and the collaborative document set specifically includes: Using each collaborative document in the collaborative document set as a row index and each employee in the employee set as a column index, an implicit mapping matrix between the collaborative document number and the employee number is constructed. For each combination of collaborative document and employee in the mapping matrix, check whether the employee in the combination belongs to the set of implicit collaborators and whether the collaborative document in the combination is marked as participating in the document participation matrix. For combinations that satisfy both the conditions of belonging to the set of implicit collaborators and being marked as participating in the document participation matrix, record one at the corresponding position in the mapping matrix; for combinations that do not satisfy both conditions at the same time, record zero at the corresponding position.

9. The method for mining and making explicit tacit knowledge in an enterprise knowledge base according to claim 8, characterized in that, The process of recording the set of implicit collaborators for each analysis period, marking continuously identified employees as long-term implicit collaborators, and writing long-term implicit collaborator tags into the enterprise knowledge base specifically includes: Set the number of analysis cycles, and perform multiple collaborative analyses on the same enterprise knowledge base consecutively according to the number of analysis cycles. Record the set of implicit collaborators obtained in the analysis cycle for each analysis, and assign a cycle index to each analysis. For each employee in the employee set, a long-term implicit count is constructed and initialized to zero. As the period index increases from the first period to the set number of analysis periods, it is determined period by period whether the employee belongs to the implicit collaborator set of the corresponding period. If the employee belongs to the implicit collaborator set, the long-term implicit count is incremented by one. If the employee does not belong to the implicit collaborator set, the long-term implicit count remains unchanged. After completing all analysis cycles, the long-term implicit counts of all employees are checked. Employees whose long-term implicit counts equal the number of analysis cycles are marked as long-term implicit collaborators. The corresponding long-term implicit collaborator labels are written into the explicit knowledge table of the enterprise knowledge base and the label status is maintained.

Images