Enterprise consulting service management system based on knowledge graph

Abstract

The application discloses an enterprise consulting service management system based on a knowledge graph, and belongs to the technical field of artificial intelligence. The system comprises a problem decomposition module, which is used for mapping and decomposing a complex consulting request into a subgraph in a pre-constructed dynamic knowledge graph in response to the received complex consulting request; an expert wisdom team synthesis module, which is used for calculating a team synthesis index based on the subgraph and a pre-constructed distributed expert wisdom network, and determining an optimal expert team according to the team synthesis index; a solution path identification module, which is used for identifying a solution path connecting different knowledge nodes on the subgraph; and a path evaluation and screening module, which is used for calculating a novelty score of each solution path, and comparing the novelty score with a preset novelty threshold value to screen and output a solution path higher than the novelty threshold value. The application ensures the timeliness and frontality of all subsequent analyses.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, specifically to a knowledge graph-based enterprise consulting service management system. Background Technology

[0002] In the modern business environment, companies face an increasing number of emerging, cross-disciplinary, and unprecedented complex challenges.

[0003] This situation and its shortcomings mainly stem from the following technical limitations:

[0004] Static and outdated knowledge base: Traditional systems generally rely on static knowledge bases, which are usually updated manually or imported in batches. These knowledge bases cannot reflect the evolution of knowledge in the external world in real time. This results in an outdated knowledge base for the system, making it difficult to cope with sudden innovation needs. The timeliness and cutting-edge nature of all subsequent analyses cannot be guaranteed.

[0005] The team building approach is crude. When forming an expert team, existing technologies often rely solely on the simple ranking and selection of individual experts based on their relevance. This approach lacks quantitative consideration of the synergistic effect of the team's overall knowledge structure, which can easily lead to groupthink and knowledge blind spots, making it difficult to maximize the team's collective wisdom.

[0006] Traditional systems often offer only the shortest or most direct path based on existing knowledge when searching for solutions. They lack the ability to discover and evaluate non-obvious solutions with potential breakthrough value, and the output is often a reorganization of outdated knowledge.

[0007] As a result, traditional systems are unable to provide effective assistance when businesses seek support for innovative decision-making.

[0008] The information disclosed in the background section above is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0009] The purpose of this invention is to provide a knowledge graph-based enterprise consulting service management system to solve the problems mentioned in the background art.

[0010] The technical solution of the present invention includes:

[0011] The problem decomposition module is used to respond to received complex consultation requests by mapping and decomposing the complex consultation requests into subgraphs in a pre-built dynamic knowledge graph.

[0012] The expert team synthesis module is used to calculate the team synthesis index by combining the total relevance of each expert in the network and the knowledge overlap between any two experts, and to determine the optimal expert team based on the team synthesis index.

[0013] The solution path identification module is used to identify solution paths that connect different knowledge nodes on the subgraph;

[0014] The path evaluation and screening module is used to determine the novelty score of each solution path as the arithmetic mean of the logarithmic inverse weights of all edges along the path, and compare the novelty score with a preset novelty threshold to screen and output solution paths that are higher than the novelty threshold.

[0015] Preferably, the expert team synthesis module determines the optimal expert team, including:

[0016] Determine the total relevance of each expert in the distributed expert intelligence network;

[0017] Determine the degree of knowledge overlap between any two experts in the network;

[0018] The team composite index is calculated by combining the total relevance and knowledge overlap of all experts.

[0019] The expert combination that maximizes the team's composite index is identified as the optimal expert team.

[0020] Preferably, the determination of total relevance includes:

[0021] The pre-calculated expert relevance scores for each expert to each knowledge node in the subgraph are used to determine the total relevance of an expert as the arithmetic sum of the expert relevance scores for that expert to all knowledge nodes in the subgraph.

[0022] Preferably, the calculation of the expert relevance score includes:

[0023] Obtain semantic similarity scores between expert publications and knowledge nodes;

[0024] Obtain semantic similarity scores between expert projects and knowledge nodes;

[0025] Based on preset weighting coefficients, the semantic similarity scores of publications and projects are weighted and summed to generate an expert relevance score.

[0026] Preferably, the determination of knowledge overlap includes:

[0027] Generate an expert feature vector for each expert, and calculate the cosine similarity between the expert feature vectors of any two experts as the knowledge overlap between the two experts.

[0028] Preferably, the generation of expert feature vectors includes:

[0029] Retrieve the feature vectors of each knowledge node associated with the expert, as well as the expert relevance score of that expert to each knowledge node;

[0030] The feature vector of an expert is generated by weighting the feature vectors of related knowledge nodes. The weight of the weighted average is the expert relevance score.

[0031] Preferably, the calculation of the novelty score includes:

[0032] Obtain the edge weight of each edge that constitutes the solution path, and determine the novelty score as the arithmetic mean of the logarithmic inverse weight values ​​of all edges along the path.

[0033] Preferably, the calculation of edge weights includes:

[0034] Obtain the semantic similarity between two knowledge nodes;

[0035] Get the most recent observation timestamp and the current system timestamp of the related relationship;

[0036] By combining semantic similarity, a preset time decay coefficient, and the difference between two timestamps, the edge weights are calculated using a time decay model.

[0037] Preferred options also include:

[0038] The data acquisition and topic identification module is used to acquire data streams from preset external heterogeneous data sources and identify emerging topics through a natural language processing engine to serve as the basic input for building a dynamic knowledge graph.

[0039] This invention provides an improved enterprise consulting service management system based on knowledge graphs, which has the following improvements and advantages compared with the prior art:

[0040] 1. By setting up data acquisition and topic identification modules, the system can continuously absorb new knowledge from external heterogeneous data sources and identify emerging topics through a natural language processing engine. This design makes its core dynamic knowledge graph not a static database, but one that reflects the evolution of knowledge in the real world. This is fundamentally different from the static knowledge bases in the prior art that rely on manual updates or batch imports, ensuring the timeliness and cutting-edge nature of all subsequent analyses.

[0041] 2. This solution has a deep understanding of the physical world in measuring knowledge connections. The strength of the connection between any knowledge nodes will naturally decay over time unless it is strengthened by new observation data. This can automatically reduce the weight of outdated connections and highlight the importance of emerging connections, thereby accurately capturing the latest dynamics of technology and business models.

[0042] 3. This solution ensures that the selection of the optimal team pursues high professional capabilities while actively avoiding groupthink and cognitive blind spots that may result from overly similar knowledge backgrounds; existing technologies often rely solely on the simple ranking and selection of individual experts when forming teams, lacking quantitative consideration of the overall synergistic effect of the team structure.

[0043] 4. This solution provides quantifiable evaluation criteria for innovation, enabling the system to automatically identify and prioritize paths that are likely to lead to breakthroughs from among many possible solution paths; this contrasts with existing technologies that can only provide the shortest or most direct path based on existing knowledge. Attached Figure Description

[0044] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0045] Figure 1 This is a flowchart of the system of the present invention. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0047] Example 1:

[0048] Please see Figure 1 This invention provides a knowledge graph-based enterprise consulting service management system, comprising:

[0049] The problem decomposition module is used to respond to received complex consultation requests by mapping and decomposing the complex consultation requests into subgraphs in a pre-built dynamic knowledge graph.

[0050] The expert team synthesis module is used to calculate the team synthesis index by combining the total relevance of each expert in the network and the knowledge overlap between any two experts, and to determine the optimal expert team based on the team synthesis index.

[0051] The solution path identification module is used to identify solution paths that connect different knowledge nodes on the subgraph;

[0052] The path evaluation and screening module is used to determine the novelty score of each solution path as the arithmetic mean of the logarithmic inverse weights of all edges along the path, and compare the novelty score with a preset novelty threshold to screen and output solution paths that are higher than the novelty threshold.

[0053] The knowledge graph-based enterprise consulting service management system described in this embodiment is designed as an integrated technology closed loop to respond to and solve the highly complex consulting needs faced by enterprises; the system includes several core modules that work together.

[0054] The problem decomposition module aims to transform a macroscopic and complex consultation request into a structured problem that can be analyzed and processed within a knowledge network. In this embodiment, this module is configured to respond to complex consultation requests received by the system. Complex consultation requests here This refers to strategic questions input by users that typically cannot be directly answered in existing knowledge bases and involve cross-domain or emerging fields; this module invokes pre-built dynamic knowledge graphs. and the request Mapped into the atlas;

[0055] In this embodiment, the dynamic knowledge graph The pre-construction, or cold start process, includes: loading a publicly available or purchased industry-based ontology library as the initial schema layer of the graph, which defines the core entity types and relation types; collecting a large number of industry documents, patents, reports, etc. as seed corpora; using the aforementioned natural language processing engine to perform entity recognition and relation extraction; and filling the extracted structured knowledge, entities, relations, and attributes into the ontology library to form the initial version of the knowledge graph.

[0056] The mapping process uses natural language processing technology to identify the core concepts in the request and locate them in the graph. The corresponding knowledge nodes This module will handle complex consultation requests. It is decomposed into subgraphs consisting of multiple interconnected knowledge nodes and their connections. ; Subgraph here Refers to dynamic knowledge graph The partial view structurally and completely expresses the original complex request. The multi-dimensional knowledge requirements it contains;

[0057] To enable technical personnel in the relevant field to implement this module without barriers, complex consultation requests will be handled in this module. Map and decompose into subgraphs The steps are as follows:

[0058] Core concept identification and entity linking employ a pre-trained language model, such as BERT's named entity recognition model, which has been fine-tuned on a professional literature corpus of the target industry; this model identifies key concepts from request texts. Core technical terms and concepts are extracted from the data, and entity linking technology is used to connect the extracted entities with a dynamic knowledge graph. The only knowledge node in By performing matching and disambiguation, a set of core nodes can be located. ;

[0059] Subgraph expansion construction, based on the core node set identified in the first step. Starting from the left, the k-hop neighbor expansion algorithm is used to construct the subgraph. The system will contain All nodes in the graph, and starting from these nodes, in the graph A subgraph can be formed by traversing no more than k edges, where k is a preset integer, for example, k=2, and identifying all neighboring nodes and their edges; It not only includes the core concepts directly mentioned by the user, but also the contextual knowledge closely related to these concepts, thus fully expressing the multi-dimensional requirements of the original request;

[0060] The expert team synthesis module aims to dynamically assemble an expert team with the optimal knowledge structure and the highest probability of generating innovative solutions based on the decomposed problem. In this embodiment, the input of this module is the subgraph generated in the previous steps. Access a pre-built distributed expert intelligence network, which contains a large amount of digital profile information of experts, derived from the analysis of publicly available information of experts, such as personal homepages, academic papers, patent applications, and projects participated in.

[0061] In this embodiment, the pre-construction process of the distributed expert intelligence network includes: using technologies such as web crawlers, selectively collecting personal information of experts from publicly available channels such as academic databases, Google Scholar, professional social networking sites, and institutional websites. This information includes, but is not limited to, lists of their published works, brief descriptions of projects they have participated in, and keywords related to their research fields. The system extracts information from the collected unstructured text and organizes it into structured expert profile data. For each expert, the system calculates their correlation with the knowledge graph. Expert relevance score for each knowledge node And based on the score, generate the expert's initial expert feature vector. They are jointly stored in the expert wisdom network database for subsequent use by the team synthesis module;

[0062] This module uses a specific algorithm to calculate a team composite index, which measures the overall effectiveness of candidate teams. Team Synergy Index (here) It is a quantitative scoring system designed to balance the complementarity of team members' professional skills and knowledge structures; this module is based on the team's composite index. To determine the optimal expert team The team Considered to be for the current subgraph The issues it represents;

[0063] The solution path identification module aims to explore and discover potential, non-obvious problem-solving approaches within a defined problem framework; this is illustrated by the subgraph output by the problem decomposition module. As the object of operation; this module uses a graph search algorithm to search within the subgraph The above identifies the ability to connect different knowledge nodes. Solution paths with potential innovative value ;

[0064] To ensure the discovery of diverse and potentially innovative paths, rather than simply the shortest or optimal paths in the traditional sense, this module may employ at least one of the following graph search strategies:

[0065] k-Shortest Path Algorithm: Unlike Dijkstra's algorithm, which returns only one optimal path, this module can use algorithms such as Yens' or Eppstein's to identify k shortest, or lowest-cost, candidate paths from the starting point to the ending point; the cost here can be related to the edge weights. Inversely proportional; by generating a set containing multiple suboptimal paths, the system can provide richer candidate materials that may contain non-obvious connections for the subsequent novelty assessment module; for example, k=10 can be set to generate 10 alternative paths;

[0066] Biased random walk algorithm: To actively explore less popular but potentially breakthrough connections in a graph, a random walk algorithm can be implemented; the walker starts from a core knowledge node and selects the transition probability of the next node at each step. Not proportional to edge weight It is inversely proportional to the edge weight, for example, defined as:

[0067]

[0068] in, From the current node Move to the next node The probability of; The current knowledge node; The next neighbor node to which the transfer will take place; : Connecting nodes and Edge weights; :node The set of all neighboring nodes; : One of the neighboring nodes; : Connecting nodes and its neighboring nodes Edge weights;

[0069] in It is a node The set of neighboring nodes; this design makes the algorithm more inclined to traverse edges with lower weights, i.e., those with rarer and more novel associations, thus naturally embedding an exploratory and novel orientation into the generated paths; through multiple independent random walks, a series of structurally diverse solution paths can be collected.

[0070] By employing the specific algorithmic strategies described above, this module can systematically and purposefully generate a series of paths with potential innovative value, providing high-quality input for subsequent screening and evaluation steps.

[0071] Solution path here Refers to subgraph An ordered set of edges in a given context represents, semantically, a logical reasoning chain or technical implementation route from one or more knowledge starting points to one or more knowledge ending points.

[0072] The path evaluation and screening module aims to quantitatively evaluate the numerous potential solution paths discovered in order to select the most innovative solutions. In this embodiment, this module evaluates each solution path identified in the previous steps. Processing is performed; each path is calculated using a specific mathematical model. Novelty score Novelty score here It is a quantitative indicator used to evaluate the amount of information or surprise contained in a solution path. The higher the score, the more unusual the path and the greater its innovative potential. This module calculates the novelty score. With a preset novelty threshold Comparison; this novelty threshold It is the minimum score set by the system administrator to filter out routine or low-value solutions. The setting can be based on statistical analysis of the novelty scores of a large number of historical solutions. For example, the 80th percentile of the score distribution can be used as the threshold to ensure that only the top 20% of innovative paths are given priority.

[0073] In the case of a system cold start where there is no large number of historical solutions available for statistical analysis, a batch of solutions, such as 1000, can be generated randomly on a pre-built knowledge graph or through breadth-first search, and their novelty scores can be calculated. Based on the score distribution of this initial sample set, the 80th percentile is determined as the initial novelty threshold. ;

[0074] This module filters and outputs all scores higher than the novelty threshold. The following solution path is provided for user reference;

[0075] Through the sequential activation and data flow of the above-mentioned modules such as problem decomposition, expert synthesis, path identification and evaluation screening, this embodiment constructs a complete technical closed loop; the overall technical effect is that it integrates real-time data perception, dynamic knowledge graph construction, distributed expert network matching and innovative path discovery into one.

[0076] Existing consulting service management systems are good at handling known unknown problems, that is, finding answers within existing knowledge frameworks. However, they are inadequate for exploring and solving unknown unknown problems, that is, dealing with emerging, cross-disciplinary, and unprecedented complex challenges.

[0077] Example 2

[0078] The expert team synthesis module determines the optimal expert team, including:

[0079] Determine the total relevance of each expert in the distributed expert intelligence network;

[0080] Determine the degree of knowledge overlap between any two experts in the network;

[0081] The team composite index is calculated by combining the total relevance and knowledge overlap of all experts.

[0082] The expert combination that maximizes the team's composite index is identified as the optimal expert team.

[0083] The determination of total relevance includes:

[0084] The pre-calculated expert relevance scores for each expert to each knowledge node in the subgraph are used to determine the total relevance of an expert as the arithmetic sum of the expert relevance scores for that expert to all knowledge nodes in the subgraph.

[0085] The calculation of expert relevance score includes:

[0086] Obtain semantic similarity scores between expert publications and knowledge nodes;

[0087] Obtain semantic similarity scores between expert projects and knowledge nodes;

[0088] Based on preset weighting coefficients, the semantic similarity scores of publications and projects are weighted and summed to generate an expert relevance score.

[0089] Determining the degree of knowledge overlap includes:

[0090] Generate an expert feature vector for each expert, and calculate the cosine similarity between the expert feature vectors of any two experts as the knowledge overlap between the two experts.

[0091] The generation of expert feature vectors includes:

[0092] Retrieve the feature vectors of each knowledge node associated with the expert, as well as the expert relevance score of that expert to each knowledge node;

[0093] By taking a weighted average of the feature vectors of related knowledge nodes, the feature vector of the expert is generated, and the weight of the weighted average is the expert relevance score.

[0094] Based on the system in Example 1, the expert team synthesis module for determining the optimal implementation of the expert team has been further optimized, and the internal logic has been refined into multiple related calculation steps to ensure the accuracy and effectiveness of the team synthesis index.

[0095] To achieve accurate calculation of the aforementioned team synthesis index, this module identifies each expert in the distributed expert intelligence network. For the current problem subgraph Total relevance Overall relevance It's a quantitative indicator used to assess the overall match between a single expert and the current consultation question; this module identifies any two experts in the network. and Knowledge overlap between Knowledge overlap This is a quantitative indicator used to measure the similarity of knowledge backgrounds between two experts, thus assessing cognitive diversity within the team. Based on the above results, this module combines the total relevance of all candidate experts. Knowledge overlap between and any expert pair The team composite index is calculated using a specific mathematical model. The calculation of this index is initially based on the financial portfolio theory, aiming to reduce risk while pursuing high returns through diversified allocation. In this embodiment, the calculation method of this index is defined as follows:

[0096]

[0097] in, Team Synergy Index Overall expert relevance Knowledge overlap Number of experts in the team; Refers to an expert in the team; Refers to the best expert team; : Refers to any two different experts in the team;

[0098] This formula balances the depth and breadth of expertise by multiplying the sum of the individual abilities and total relevance of team members by the team's overall cognitive diversity (1 minus the average knowledge overlap). The system uses an optimization algorithm, such as a genetic algorithm or simulated annealing, to search among all possible combinations of experts to determine the combination that maximizes the team's composite index. The expert combination that achieves the maximum value is selected as the optimal expert team. Output;

[0099] To achieve the above calculations, the overall relevance... The determination process is as follows: The system calls the pre-calculated values ​​for each expert. Pair diagram Each knowledge node Expert relevance score Expert relevance score This is a quantitative scoring system; its definition and calculation will be detailed later. (An expert) Total relevance It was identified as the expert's view on the subgraph. Expert relevance scores for all knowledge nodes within the scope The arithmetic sum; the calculation method is:

[0100]

[0101] in, Overall expert relevance Experts' relevance scores for nodes. Problem subgraph; Problem Subgraph Knowledge nodes within;

[0102] This step accumulates the experts' expertise in each specific knowledge point involved in the problem, thereby comprehensively reflecting their matching level with the entire complex problem.

[0103] Furthermore, expert relevance score The calculation method is as follows: This calculation is based on the principle of multidimensional evidence-weighted evaluation, aiming to comprehensively evaluate the expert's capabilities in both theory and practice; the system first obtains expert... Publications such as papers, patents and knowledge nodes semantic similarity score between and experts Past projects, such as engineering cases and consulting projects, and this knowledge node semantic similarity score The semantic similarity score here is calculated using a pre-trained language model, such as BERT, fine-tuned based on an industry corpus. The output is normalized to the [0, 1] interval to ensure the comparability of similarity scores from different sources. This is based on a pair of pre-defined weight coefficients. and The semantic similarity score of the publications semantic similarity score with the project Perform a weighted summation to generate the final expert relevance score. ;

[0104]

[0105] in, Expert relevance score Weighting coefficients Publication similarity Project similarity;

[0106] Weighting coefficient and These are two dimensionless, adjustable parameters used to balance the importance of theoretical contributions and practical experience in the evaluation system; their sum is usually set to 1. Initial values ​​can be preset based on the nature of the consulting task, for example, for tasks biased towards theoretical research. It can be set to 0.7. It is 0.3; and vice versa.

[0107] In the initial stage, when historical data is lacking for regression analysis to optimize these two parameters, one can... and The initial values ​​are all set to 0.5, that is... This means that theoretical contributions and practical experience are given equal importance when assessing the relevance of experts;

[0108] These two parameters can be iteratively optimized through regression analysis of the expert team composition in historical success cases to maximize the predictive accuracy of the model;

[0109] In parallel, knowledge overlap The determination process is as follows: To perform this calculation, the system needs to generate an expert feature vector for each expert. Expert feature vector It is a mathematical representation of the comprehensive knowledge profile of an expert in a high-dimensional vector space; any two experts and Knowledge overlap between That is, the expert feature vectors identified as those of the two experts. and Cosine similarity between two vectors; cosine similarity can effectively measure the difference in direction between two vectors, and its value range is [-1, 1]. The closer the value is to 1, the more similar the knowledge structures of the two experts are.

[0110] expert feature vectors The generation method is: system call and expert Related knowledge nodes The feature vectors, and the expert relevance score of the expert for these knowledge nodes. The feature vectors of the knowledge nodes here are vectorized representations of knowledge concepts generated by the dynamic knowledge graph construction module. They originate from embedding vectors obtained after training a model on massive amounts of industry literature. The feature vector of an expert is generated by weighted averaging the feature vectors of all their associated knowledge nodes. In this weighted average calculation, the weights used are the expert relevance scores for each knowledge node. This process ensures that the expert's profile vector accurately reflects the distribution and strength of their professional knowledge. At knowledge nodes with higher relevance, the profile vector is more significantly affected by that node.

[0111] Through the synergistic implementation of the aforementioned technical features, this invention brings significant technical benefits. It goes beyond simply selecting the most relevant experts; instead, it establishes a scientific team-building mechanism by introducing the calculation of knowledge overlap and the optimization of the team synthesis index. This mechanism can maximize the cognitive heterogeneity among team members while ensuring that the overall professional capabilities of the team are high enough, thereby effectively avoiding groupthink and knowledge blind spots, increasing the probability of the expert team generating breakthrough and innovative solutions, and achieving a synergistic effect.

[0112] This solution establishes a dynamically adaptable knowledge base. By setting up data acquisition and topic identification modules, the system can continuously absorb new knowledge from external heterogeneous data sources and identify emerging topics through a natural language processing engine. This design ensures that its core dynamic knowledge graph is not a static database, but rather reflects the evolution of knowledge in the real world. This is fundamentally different from the static knowledge bases in existing technologies that rely on manual updates or batch imports, ensuring the timeliness and cutting-edge nature of all subsequent analyses.

[0113] This solution establishes a scientific and optimizable team building mechanism; the core of the expert team synthesis module lies in the team synthesis index. ,in, Team Synergy Index Overall expert relevance Knowledge overlap Calculation and maximization of the number of team experts; Refers to an expert in the team; Refers to the best expert team; : Refers to any two different experts in the team; the first term on the right side of the formula This represents the sum of the individual professional abilities of the team members and is a guarantee of the team's professional depth; the second item This represents the cognitive diversity of the team, among which As the cosine similarity of the knowledge structures of any two experts, the higher the average value, the more homogeneous the team and the lower the cognitive diversity. By multiplying the two, this index ensures that the selection of the optimal team is to pursue high professional capabilities while actively avoiding groupthink and cognitive blind spots that may be caused by overly similar knowledge backgrounds. Existing technologies often rely solely on the simple ranking and selection of the relevance of individual experts when forming teams, lacking quantitative consideration of the synergistic effect of the overall team structure.

[0114] This solution provides quantifiable evaluation criteria for innovation; the novelty score calculation formula used in the path evaluation and screening module. ,in, Novelty score Path length Edge weight; : An edge that forms the solution path P, connecting knowledge nodes i and j; The solution path is an ordered set of edges that connect different knowledge nodes in a subgraph. It is the natural logarithm; the essence of this formula is to calculate the average information entropy or surprise of a solution path; a path consisting of high-weight, high-probability, regular edges, its The value is close to 1. Its novelty score approaches 0. The corresponding value is also very low; conversely, a path that traverses multiple low-weight, low-probability, and unconventional edges has a lower value. The value will increase significantly; this mechanism enables the system to automatically identify and prioritize the most extraordinary and potentially breakthrough paths from among many possible solution paths; this is in stark contrast to existing technologies that can only provide solutions based on the shortest or most direct path of existing knowledge.

[0115] Example 3

[0116] The calculation of the novelty score includes:

[0117] Obtain the edge weight of each edge that constitutes the solution path, and determine the novelty score as the arithmetic mean of the logarithmic inverse weight values ​​of all edges along the path.

[0118] The calculation of edge weights includes:

[0119] Obtain the semantic similarity between two knowledge nodes;

[0120] Get the most recent observation timestamp and the current system timestamp of the related relationship;

[0121] By combining semantic similarity, a preset time decay coefficient, and the difference between two timestamps, the edge weights are calculated using a time decay model.

[0122] Based on the system in Example 1, the calculation method of novelty score in the path evaluation and screening module is further specified to ensure the objectivity and accuracy of the evaluation results;

[0123] Novelty score The core idea behind this calculation comes from the concept of surprise in information theory, which states that the lower the probability of an event, the greater the amount of information it contains when it occurs. In this embodiment, the system obtains a solution path that constitutes a solution to be evaluated. Each edge edge weight The edge weight here It is a quantitative indicator, the definition of which will be detailed later, reflecting two knowledge nodes in the knowledge graph. and The universality or predictability of the relationship between them, with the value range normalized to the [0, 1] interval, can be regarded as a probability value; path Novelty score It was determined to be along that path The arithmetic mean of the logarithmic inverse weights of all edges; calculated as follows:

[0124]

[0125] in, Novelty score Path length Edge weight; : An edge that forms the solution path P, connecting knowledge nodes i and j; The solution path is an ordered set of edges that connect different knowledge nodes in a subgraph. It is the natural logarithm;

[0126] This formula clearly shows the edge weights traversed by a solution path. The lower the value, the rarer and more unexpected the association; the lower the log-inverse weight value. The larger the value, the higher the average novelty score of the path. The higher the value, the better; this calculation method provides a quantifiable, information theory-based objective measure of innovation.

[0127] Furthermore, the aforementioned edge weights The calculation is designed as a time-sensitive model to reflect the dynamic changes in the strength of knowledge associations; this model integrates two dimensions: semantic association and time decay; the system obtains the constituent edges Two knowledge nodes and semantic similarity between Semantic similarity here This is calculated using the aforementioned pre-trained language model, reflecting the inherent semantic correlation strength between two knowledge concepts; simultaneously, the system obtains the timestamp of the most recent observation of this correlation in the data source. and the current system timestamp Combining semantic similarity Preset time decay coefficient And the difference between the two timestamps The system uses a time decay model to calculate the final edge weights. ;

[0128]

[0129] in, Edge weight, Semantic similarity Time decay coefficient Current timestamp Observation timestamp; The base of the natural logarithm;

[0130] Time decay coefficient It is an tunable hyperparameter that controls the rate at which the strength of knowledge associations is forgotten over time. Its dimension is the reciprocal of time, ensuring the exponential term... The values ​​are dimensionless, ensuring consistent dimensions throughout the formula; the initial values ​​are preset based on the knowledge update rate of a specific industry, for example, a higher value can be set for the IT industry with rapid technological iteration. For traditional industries where knowledge is relatively stable, a lower value is set. value;

[0131] The value of the time decay coefficient λ can be based on the half-life of knowledge. This is defined as the time required for the weight of an association to decay to half of its initial semantic similarity over time. The relationship can be expressed by the formula... It is derived that ;in, Time decay coefficient; The natural logarithm of 2; The half-life of knowledge refers to the time it takes for the weight of an association to decay to half of its initial value over time; for example, in the IT industry where the technology iteration cycle is approximately two years, this can be set as follows: The year is used to calculate the initial λ value; for traditional industries with relatively stable knowledge, a value can be set. Year; when performing weight calculations, the time difference must be ensured. Units and half-life The units must be consistent; for example, if If set in years, then The value must also be converted to a value in years;

[0132] This parameter can be optimized by backtesting on historical data with the goal of improving the accuracy of link prediction in the knowledge graph.

[0133] Through the implementation of the above-mentioned technical features, the present invention has achieved significant technical benefits; it is not merely a simple path finding, but provides a rigorous quantitative evaluation system for the quality of the path—especially its innovativeness; by introducing time-sensitive edge weight calculation and a novelty scoring model based on information theory, the system can effectively distinguish between conventional paths that merely connect related knowledge nodes and innovative paths that cross seemingly distant but potentially deeply connected knowledge nodes and have timeliness.

[0134] Formula for calculating edge weights in dynamic knowledge graphs ,in, Edge weight, Semantic similarity Time decay coefficient Current timestamp The observation timestamp is given a clear physical meaning; the formula determines the strength of the inherent semantic connection between two knowledge nodes. With a time decay term Multiplication; the introduction of the decay term is based on the logic that the strength of the connection between any knowledge nodes will naturally decay over time unless it is strengthened by new observation data; this can automatically reduce the weight of outdated connections and highlight the importance of emerging connections, thereby accurately capturing the latest dynamics of technology and business models.

[0135] Example 4

[0136] Also includes:

[0137] The data acquisition and topic identification module is used to acquire data streams from preset external heterogeneous data sources and identify emerging topics through a natural language processing engine to serve as the basic input for building a dynamic knowledge graph.

[0138] Based on the system in Example 1, this example further includes a data acquisition and topic identification module as the starting point of the system; the purpose of setting up the module is to provide continuous and real-time data input for the construction of the entire dynamic knowledge graph and expert network, so as to ensure that the knowledge base of the system is synchronized with the external world;

[0139] In this embodiment, the module is configured to continuously acquire data streams from multiple preset external heterogeneous data sources. These external heterogeneous data sources refer to a series of filtered online information sources highly relevant to the application field of this invention, such as industry news portals, academic paper databases (e.g., arXiv), technical blogs, open-source project code repositories (e.g., GitHub), and patent databases. The selection of these data sources aims to cover multiple dimensions, including theoretical frontiers, technological applications, and market dynamics. After acquiring the data streams, the module performs deep processing on unstructured and semi-structured text data using a built-in natural language processing (NLP) engine built on standard tools in the field, such as Scikit-learn or spaCy. The NLP engine performs tasks including entity recognition, relation extraction, and topic modeling to dynamically discover emerging or rapidly gaining popularity in new technologies or business topics. These identified emerging topics will be used to construct or update a dynamic knowledge graph. China-Singapore Knowledge Node and New Border The basic input drives the evolution of the entire knowledge system;

[0140] The addition of data acquisition and topic identification modules brings key technical benefits; they constitute the data input and preprocessing layer of the entire technical solution, making the system no longer a closed, static system that can only process internal existing knowledge; by connecting to the data deluge of the external world in real time and automatically identifying emerging topics, this invention solves the problem of outdated knowledge and inability to cope with sudden innovations in traditional consulting systems; it ensures that all subsequent analyses, including problem decomposition, expert matching, and path discovery, are based on a latest and most cutting-edge knowledge graph, thereby guaranteeing that the output consulting results have high timeliness and practical guiding significance;

[0141] Dynamic discovery of emerging directions: The data acquisition and topic identification module may discover an emerging trend from the latest academic papers and technical blogs: the application of digital twin technology in early warning of battery thermal runaway;

[0142] Constructing a dynamic problem space: The problem decomposition module will construct a subgraph containing nodes such as BMS, SOC estimation, digital twin, and thermal runaway early warning. Among them, the edge weights between digital twins and thermal runaway early warning systems The lower price may be due to it being a relatively new concept.

[0143] Building a collaborative and efficient team: When synthesizing expert teams, the system considers not only top experts in BMS and AI, but also the team synthesis index. The optimization process may involve an expert with extensive experience in digital twin modeling in the field of aero-engines; although the expert's direct relevance score in the battery field... Not high, but its knowledge overlaps with that of existing experts. Extremely low, which can enhance the cognitive diversity of the team, thereby improving the overall... The value reaches its maximum;

[0144] Outputting innovative solutions: The solution path identification module identifies a path that passes through digital twin nodes, while the path evaluation and screening module assigns a high novelty score to the path because it traverses low-weight edges. The system output will not only be general suggestions for AI-optimized BMS, but also a specific and highly innovative solution path for building a digital twin model of the battery system and using the model for large-scale simulation to train an AI early warning system that can predict thermal runaway in advance.

[0145] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A knowledge graph-based enterprise consulting service management system, characterized in that, include: The problem decomposition module is used to respond to received complex consultation requests by mapping and decomposing the complex consultation requests into subgraphs in a pre-built dynamic knowledge graph. The expert team synthesis module is used to calculate the team synthesis index by combining the total relevance of each expert in the network and the knowledge overlap between any two experts, and to determine the optimal expert team based on the team synthesis index. The solution path identification module is used to identify solution paths that connect different knowledge nodes on the subgraph; The path evaluation and screening module is used to determine the novelty score of each solution path as the arithmetic mean of the logarithmic inverse weights of all edges along the path, and compare the novelty score with a preset novelty threshold to screen and output solution paths that are higher than the novelty threshold. The expert team synthesis module determines the optimal expert team, including: Determine the total relevance of each expert in the distributed expert intelligence network; Determine the degree of knowledge overlap between any two experts in the network; The team composite index is calculated by combining the total relevance and knowledge overlap of all experts. The team synthesis index is calculated as follows: in, Team Synergy Index Overall expert relevance Knowledge overlap Number of experts in the team; Refers to an expert in the team; Refers to the best expert team; : Refers to any two different experts in the team; The system uses either a genetic algorithm or simulated annealing algorithm to search among all possible combinations of experts to determine the combination that can improve the team's synergistic index. The expert combination that achieves the maximum value is selected as the optimal expert team. Output; Determining the degree of knowledge overlap includes: Generate an expert feature vector for each expert, and calculate the cosine similarity between the expert feature vectors of any two experts as the knowledge overlap between the two experts.

2. The knowledge graph-based enterprise consulting service management system according to claim 1, characterized in that, The determination of total relevance includes: The pre-calculated expert relevance scores for each expert to each knowledge node in the subgraph are used to determine the total relevance of an expert as the arithmetic sum of the expert relevance scores for that expert to all knowledge nodes in the subgraph.

3. The knowledge graph-based enterprise consulting service management system according to claim 2, characterized in that, The calculation of expert relevance score includes: Obtain semantic similarity scores between expert publications and knowledge nodes; Obtain semantic similarity scores between expert projects and knowledge nodes; Based on preset weighting coefficients, the semantic similarity scores of publications and projects are weighted and summed to generate an expert relevance score.

4. The knowledge graph-based enterprise consulting service management system according to claim 3, characterized in that, The generation of expert feature vectors includes: Retrieve the feature vectors of each knowledge node associated with the expert, as well as the expert relevance score of that expert to each knowledge node; The feature vector of an expert is generated by weighting the feature vectors of related knowledge nodes. The weight of the weighted average is the expert relevance score.

5. The knowledge graph-based enterprise consulting service management system according to claim 1, characterized in that, The calculation of the novelty score includes: Obtain the edge weight of each edge that constitutes the solution path, and determine the novelty score as the arithmetic mean of the logarithmic inverse weight values ​​of all edges along the path.

6. The knowledge graph-based enterprise consulting service management system according to claim 5, characterized in that, The calculation of edge weights includes: Obtain the semantic similarity between two knowledge nodes; Get the most recent observation timestamp and the current system timestamp of the related relationship; By combining semantic similarity, a preset time decay coefficient, and the difference between two timestamps, the edge weights are calculated using a time decay model.

7. The knowledge graph-based enterprise consulting service management system according to claim 1, characterized in that, Also includes: The data acquisition and topic identification module is used to acquire data streams from preset external heterogeneous data sources and identify emerging topics through a natural language processing engine to serve as the basic input for building a dynamic knowledge graph.

Images