A knowledge graph-based scientific and technological achievement transformation risk prediction system

Abstract

The application discloses a scientific and technological achievement transformation risk prediction system based on a knowledge graph, comprising: a knowledge graph construction module, which is used for constructing a scientific and technological achievement transformation risk knowledge graph; a risk topology generation module, which is used for determining a sink and an unburned entity set; a script configuration calculation module, which is used for determining a configuration inspection quantity set and a directed internal correlation quantity set; a burning entity screening module, which is used for comparing the configuration inspection quantity and the directed internal correlation quantity by adopting a Dhar burning algorithm, screening a candidate burning entity, and updating the directed internal correlation quantity; a risk lock identification module, which is used for repeatedly performing Dhar burning inspection, and determining remaining entities as a risk lock entity set; a critical risk inversion module, which is used for limiting a critical risk inversion range, and generating a critical risk configuration; and a risk prediction output module, which is used for generating a scientific and technological achievement transformation risk prediction result. The application improves the accuracy of the risk prediction result and the risk disposal pertinence.

Description

Technical Field

[0001] This invention relates to the field of data processing and risk prediction technology, and in particular to a risk prediction system for the transformation of scientific and technological achievements based on knowledge graphs. Background Technology

[0002] With the continuous growth in the number of scientific and technological achievements, technology transfer activities involve achievement evaluation, technology transactions, resource input, market demand, industrialization implementation, and inter-entity collaboration. The data generated during the transfer process comes from a wide range of sources and has significant structural differences, and there are complex transmission relationships between risk factors. Existing technology transfer risk prediction technologies mainly use indicator evaluation, expert scoring, statistical analysis, and machine learning models to comprehensively calculate technology maturity, capital investment, market demand, team capabilities, and industrialization conditions, thereby outputting risk levels, risk probabilities, or evaluation scores.

[0003] Existing technologies typically treat risk indicators as independent input variables, focusing on calculating the impact of individual risk factors on prediction results. This makes it difficult to accurately describe the directional dependencies between risk factors and the strength of risk transmission. While knowledge graphs can preserve entity relationships, existing prediction methods often lack a mechanism to convert graph relationships into a computable directed risk topology. This makes it impossible to identify risk-locked entities constrained by internal risk dependencies based on the risk topology state, and it is also difficult to determine the dominant risk dependency path that propagates risk along associated entities to the target node. Consequently, risk prediction results lack a clear structural interpretation.

[0004] Furthermore, existing technologies can typically only provide risk levels or risk probabilities, making it difficult to determine the critical configuration adjustment range required for the transformation of scientific and technological achievements from a risk-locked state to a risk-resolvable state, and also unable to limit the scope of risk inversion around the risk-locked entity and the dominant risk-dependent path.

[0005] Therefore, how to provide a risk prediction system for the transformation of scientific and technological achievements based on knowledge graphs is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a knowledge graph-based risk prediction system for technology transfer. This invention constructs a knowledge graph of technology transfer risks, combines the Dhar burning test to identify risk-locked entities and dominant risk-dependent paths, and uses critical risk inversion to determine the critical risk boundary and minimum configuration adjustment requirements, thereby improving the accuracy, interpretability, and targeted nature of risk prediction for technology transfer.

[0007] A technology transfer risk prediction system based on knowledge graphs according to an embodiment of the present invention includes: The knowledge graph construction module is used to acquire data related to the transformation of scientific and technological achievements, and to perform entity recognition, relationship extraction and entity alignment to construct a knowledge graph of risks in the transformation of scientific and technological achievements. The risk topology generation module is used to construct a directed risk topology matrix based on the knowledge graph of technology transfer risks, generate entity risk mitigation configuration quantities, and determine the sink point and the set of unburned entities; The script configuration calculation module is used to generate script vectors from the directed risk topology matrix, determine the set of configuration inspection quantities by combining the entity risk mitigation configuration quantity, and determine the set of directed internal correlation quantities based on the set of unburned entities. The combustion entity screening module is used to compare the configuration check quantity and the directed internal correlation quantity using the Dhar combustion algorithm to screen out candidate combustion entities, determine the combustion entity and parent relationship according to the preset relational ordering parameters, remove combustion entities from the set of unburned entities and update the directed internal correlation quantity. The risk lockout identification module is used to repeatedly perform the Dhar burning test based on the updated directed internal correlation quantity until no entity meets the Dhar burning condition. Based on the parent relationship, a dominant risk dependency tree with the sink node as the root node is generated, and the remaining entities are identified as the risk lockout entity set. The critical risk inversion module is used to limit the critical risk inversion range by the risk-locked entity set and the dominant risk dependency tree, and to obtain the critical configuration adjustment vector by combining the directed risk topology matrix to generate the critical risk configuration. The risk prediction output module is used to perform a second Dhar combustion test on the critical risk configuration based on the script vector, and generate risk prediction results for the transformation of scientific and technological achievements.

[0008] Optionally, the knowledge graph construction module includes: Obtain the technology transfer-related data corresponding to the target scientific and technological achievement, perform association identification binding and data standardization processing on the technology transfer-related data, and collect the processed technology transfer-related data to generate standardized technology transfer-related data. Entity parsing and entity alignment are performed on the standardized scientific and technological achievement transformation-related data to obtain a set of related entities and a set of entity attributes. Based on standardized data related to the transformation of scientific and technological achievements, sets of related entities, and sets of entity attributes, relationship parsing and graph construction are performed to obtain a knowledge graph of risks in the transformation of scientific and technological achievements.

[0009] Optionally, the risk topology generation module includes: Read the associated entities, directed risk dependencies, and relationship confidence in the knowledge graph of technology transfer risks, perform topological quantification on the directed risk dependencies, and perform matrix construction based on the topological quantification results to obtain the directed risk topological matrix; Read the entity attributes in the knowledge graph of technology transfer risk, perform risk configuration mapping on the entity attributes corresponding to each associated entity, obtain the entity risk mitigation configuration quantity corresponding to each associated entity, and form a set of entity risk mitigation configuration quantities. Based on the directed risk topology matrix, the associated entities are screened to identify the sinks. The screened associated entities are then identified as sinks. The associated entity set and the entity risk mitigation configuration set are then subjected to combustion initialization processing to obtain the unburned entity set and the initial risk mitigation configuration set.

[0010] Optionally, the script configuration calculation module includes: The script vector generation process is based on the directed risk topology matrix, sink, and unburned entity set. The matrix rows and columns corresponding to the sink are removed from the directed risk topology matrix to obtain the reduced directed risk topology matrix corresponding to the unburned entity set. Positive integer script components are set for each unburned entity according to the order of the unburned entities. The topological action operation is performed on each script component using the transpose of the reduced directed risk topology matrix. The topological action operation is re-executed based on all updated script components until all script action components are greater than zero. The current script components are combined into a script vector according to the order of the unburned entities, and the corresponding script action components are combined into a script action. Based on the initial risk mitigation configuration set and script action quantity, configuration superposition processing is performed to obtain the configuration verification quantity corresponding to each unburned entity. All configuration verification quantities are arranged into a configuration verification quantity set according to the order of the unburned entities. The configuration superposition processing is to superimpose the entity risk mitigation configuration quantity of each unburned entity with the corresponding script action component according to the one-to-one correspondence of the unburned entities, and the obtained superposition result is determined as the configuration verification quantity of the corresponding unburned entity. Based on the set of unburned entities and the reduced directed risk topology matrix, internal correlation quantities are extracted to obtain the directed internal correlation quantities corresponding to each unburned entity. All directed internal correlation quantities are then arranged into a set of directed internal correlation quantities according to the order of the unburned entities.

[0011] Optionally, the combustion entity screening module includes: Read the current set of unburned entities, the set of configuration check quantities, and the set of directed internal correlation quantities. Perform Dhar combustion condition determination on each unburned entity. Determine unburned entities whose configuration check quantities are not less than the corresponding directed internal correlation quantities as candidate combustion entities, and form a candidate combustion entity set from all candidate combustion entities. The sink and the burning entities that have been removed from the unburned entity set before the current Dhar burning test round are combined into a burning entity set. For each candidate burning entity in the candidate burning entity set, the directed risk dependency relationship from the technology achievement transformation risk knowledge graph to each entity in the burning entity set is retrieved. The retrieved directed risk dependency relationships are combined into a candidate parent relationship set for the corresponding candidate burning entity. Perform relation ordering on the candidate parent relations to obtain the relation ordering value corresponding to each candidate parent relation; The relational ordinal values ​​of the candidate parent relations corresponding to each candidate combustion entity are compared. The candidate parent relation with the largest relational ordinal value is selected as the undetermined parent relation of the corresponding candidate combustion entity, and the relational ordinal value of the undetermined parent relation is determined as the entity ordinal value of the corresponding candidate combustion entity. The entity ordinal values ​​of each candidate combustion entity in the candidate combustion entity set are compared. The candidate combustion entity with the largest entity ordinal value is determined as the combustion entity of the current Dhar combustion test round, and the undetermined parent relation corresponding to the current combustion entity is determined as the parent relation. When the entity ordinal values ​​of different candidate combustion entities are the same, the difference between the configuration test quantity and the directed internal association quantity of each candidate combustion entity is compared. The candidate combustion entity with the largest difference is determined as the combustion entity. When the difference is still the same, the combustion entity is determined according to the preset arrangement order of the associated entities. Remove the burning entity from the current unburned entity set to obtain an updated unburned entity set. Write the parent relation and the corresponding burning entity into the parent relation set, and perform internal association quantity update processing based on the updated unburned entity set.

[0012] Optionally, the risk lockout identification module includes: Based on the updated set of unburned entities and the updated directed internal correlation, perform the Dhar combustion test again until there are no entities in the set of unburned entities that meet the Dhar combustion condition. Connect the sink node and the entity that has completed combustion according to the parent relationship corresponding to the burning entity, set the sink node as the root node, set the entity that has completed combustion as the tree node, and set the parent relationship as the tree edge to generate a dominant risk dependency tree with the sink node as the root node. The entities that remain in the set of unburned entities when the Dhar combustion test is terminated are defined as the risk-locked entity set.

[0013] Optionally, the critical risk inversion module includes: The risk-locked entities in the risk-locked entity set are identified as the starting nodes for critical risk inversion. The directed risk topology weights between the risk-locked entities and the combustion entities in the dominant risk dependency tree are read. The combustion entities with non-zero directed risk topology weights are identified as path access entities. The parent relationships corresponding to the path access entities are traced layer by layer to the sink. The risk-locked entities, path access entities, combustion entities connected by parent relationships, and corresponding parent relationships are defined as the critical risk inversion range. The range identifier vector is generated according to the arrangement order of the associated entities in the directed risk topology matrix. Extract the matrix rows and columns corresponding to the associated entities within the critical risk inversion range from the directed risk topology matrix according to the range identifier vector, construct the inversion topology submatrix, determine the inversion internal correlation quantity corresponding to each associated entity within the critical risk inversion range based on the inversion topology submatrix, obtain the boundary gap quantity of the corresponding associated entity based on the inversion internal correlation quantity, establish inversion counting constraints according to the boundary gap quantity corresponding to each associated entity, and find the critical inversion counting vector that minimizes the sum of the inversion counting components, and whose configuration verification quantity after the action of the inversion topology submatrix is ​​not less than the corresponding inversion internal correlation quantity, and whose entity risk resolution configuration quantity after the action of the inversion topology submatrix is ​​not less than zero. The topological action operation is performed on the critical inversion counting vector using the transpose relation of the inversion topological submatrix to obtain the local configuration adjustment amount corresponding to each associated entity within the critical risk inversion range. The local configuration adjustment amount is written into the corresponding position of the associated entity according to the range identifier vector. The corresponding position of the associated entity outside the critical risk inversion range is set to zero to generate the critical configuration adjustment vector. The critical configuration adjustment vector is superimposed with the entity risk resolution configuration amount item by item to generate the critical risk configuration.

[0014] Optionally, the risk prediction output module includes: Topological action processing is performed using the reduced directed risk topology matrix and script vector. The topological action results are then superimposed on the critical risk configurations item by item to obtain the second configuration test value for each associated entity. Replace the configuration check quantity with the second configuration check quantity, and sequentially perform candidate combustion entity screening, combustion entity and parent relationship determination, combustion entity removal and directed internal association quantity update to complete the second Dhar combustion check until there are no entities in the unburned entity set that meet the Dhar combustion conditions; The risk lockout ratio is determined based on the number of entities at the start of the second Dhar combustion test and the number of remaining entities at the end of the test. The critical configuration adjustment vector is determined as the critical boundary of the technology transfer risk. The entities remaining in the set of unburned entities at the end of the test are determined as residual risk entities. The residual configuration gap is determined based on the difference between the directed internal correlation quantity corresponding to the residual risk entity and the second configuration test quantity. The critical boundary of the technology transfer risk, the risk lockout ratio, the residual risk entities, and the residual configuration gap are written into the technology transfer risk prediction results.

[0015] The beneficial effects of this invention are: This invention constructs a knowledge graph of technology transfer risks, unifying the expression of scientific and technological achievements, participating entities, resource conditions, risk factors, and directed risk dependencies. It also transforms relationship confidence and risk direction into a directed risk topology matrix, enabling dispersed technology transfer-related data to form a topological structure with a clear risk transmission direction and risk dependency strength. The Dhar burning algorithm is employed because it can progressively identify entities that can be released from risk dependencies based on the comparison between entity configuration states and internal association constraints, while retaining entities that cannot release internal risk dependency constraints as risk-locked entities. This overcomes the shortcomings of traditional risk scoring methods, which can only output risk levels and struggle to identify risk-locked structures, thus improving the accuracy and structural interpretability of technology transfer risk identification.

[0016] This invention addresses the problem that conventional Dhar combustion algorithms, primarily used for graph structure state determination, are difficult to directly adapt to risk prediction in technology transfer. It introduces script vectors, entity risk resolution configuration quantities, configuration verification quantities, and directed internal association quantities to map the risk state of technology transfer to a configuration state capable of performing combustion verification. Furthermore, it utilizes preset relational ordering parameters to determine candidate combustion entities and their parent relations, ensuring that the combustion order is simultaneously constrained by relational semantics, relational confidence, and directed risk topological weights. This improvement not only determines whether associated entities can resolve internal risk dependencies but also generates a dominant risk dependency tree rooted at the sink node, clarifying the structural relationship between risk-locked entities and dominant risk transmission paths, thus enhancing the stability and traceability of risk prediction results.

[0017] Furthermore, this invention uses the risk-locked entity set and the dominant risk dependency tree to jointly limit the critical risk inversion range. It obtains the critical configuration adjustment vector by using the inversion topological submatrix, boundary gap quantity, and critical inversion counting vector, reducing the computational range unrelated to risk lock-out release and determining the minimum configuration adjustment requirement needed for the technology transfer state to reach the critical combustion condition. After generating the critical risk configuration, a second Dhar combustion test is performed based on the script vector to verify the critical configuration adjustment results. The invention outputs the critical boundary of the technology transfer risk, the risk lock-out ratio, residual risk entities, and residual configuration gap quantity, thereby avoiding judgment bias caused by a single risk calculation and providing a more accurate and targeted basis for adjusting technology transfer schemes, allocating resources, and managing risks. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a knowledge graph-based risk prediction system for technology transfer proposed in this invention. Figure 2 This is a schematic diagram illustrating the critical risk inversion based on the joint constraints of Dhar burning, risk-locked entity set, and dominant risk dependency tree in a knowledge graph-based risk prediction system for technology transfer proposed in this invention. Detailed Implementation

[0019] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0020] refer to Figures 1-2 A knowledge graph-based risk prediction system for technology transfer includes: The knowledge graph construction module is used to acquire data related to the transformation of scientific and technological achievements, and to perform entity recognition, relationship extraction and entity alignment to construct a knowledge graph of risks in the transformation of scientific and technological achievements. The risk topology generation module is used to construct a directed risk topology matrix based on the knowledge graph of technology transfer risks, generate entity risk mitigation configuration quantities, and determine the sink point and the set of unburned entities; The script configuration calculation module is used to generate script vectors from the directed risk topology matrix, determine the set of configuration inspection quantities by combining the entity risk mitigation configuration quantity, and determine the set of directed internal correlation quantities based on the set of unburned entities. The combustion entity screening module is used to compare the configuration check quantity and the directed internal correlation quantity using the Dhar combustion algorithm to screen out candidate combustion entities, determine the combustion entity and parent relationship according to the preset relational ordering parameters, remove combustion entities from the set of unburned entities and update the directed internal correlation quantity. The risk lockout identification module is used to repeatedly perform the Dhar burning test based on the updated directed internal correlation quantity until no entity meets the Dhar burning condition. Based on the parent relationship, a dominant risk dependency tree with the sink node as the root node is generated, and the remaining entities are identified as the risk lockout entity set. The critical risk inversion module is used to limit the critical risk inversion range by the risk-locked entity set and the dominant risk dependency tree, and to obtain the critical configuration adjustment vector by combining the directed risk topology matrix to generate the critical risk configuration. The risk prediction output module is used to perform a second Dhar combustion test on the critical risk configuration based on the script vector, and generate risk prediction results for the transformation of scientific and technological achievements.

[0021] In this embodiment, the knowledge graph construction module includes: The process involves acquiring relevant data on the transformation of target scientific and technological achievements, performing association identification binding and data standardization on this data, and then aggregating the processed data to generate standardized data. This data is a collection of data generated during the formation, evaluation, transaction, and industrialization of the target scientific and technological achievement, used to characterize the achievement's own state, resource input state, external demand state, and transformation implementation conditions. The association identification binding process establishes a binding relationship between the data source identifier, data generation time identifier, target scientific and technological achievement identifier, and data content identifier and the corresponding data. The data standardization process performs format unification, field mapping, unit of measurement conversion, time format unification, missing data marking, duplicate data elimination, and conflict data retention on the data. The data aggregation process uses the target scientific and technological achievement identifier as an index to group data records belonging to the same target scientific and technological achievement into the same data set. In the duplicate data elimination process, data records with a data content similarity of not less than 0.95 and the same target scientific and technological achievement identifier are identified as duplicate data, and data records with a field missing ratio of not less than 0.20 are set as missing data records. Entity parsing and entity alignment processing are performed on the standardized scientific and technological achievement transformation-related data to obtain a set of related entities and a set of entity attributes. The entity parsing process involves identifying candidate entities from the standardized scientific and technological achievement transformation-related data and extracting the corresponding candidate entity attribute sets. The entity alignment process involves determining the entity alignment confidence based on the matching results of the names, attributes, contextual semantics, and data sources between candidate entities. The candidate entities and candidate entity attribute sets whose entity alignment confidence reaches a preset entity alignment threshold are merged to obtain a set of related entities and a set of entity attributes. The preset entity alignment threshold is a confidence limit used to determine whether two candidate entities point to the same object. The entity alignment confidence is obtained by weighting name matching degree, attribute matching degree, contextual semantic matching degree, and data source matching degree. The weights of name matching degree, attribute matching degree, contextual semantic matching degree, and data source matching degree are set to 0.35, 0.30, 0.20, and 0.15, respectively, and the preset entity alignment threshold is set to 0.85. Based on standardized data related to technology transfer, sets of related entities, and sets of entity attributes, a knowledge graph of technology transfer risks is obtained through relationship parsing and graph construction. The relationship parsing process identifies the semantics of relationships between related entities, the direction of risk impact, and relationship evidence information. It calculates the semantic matching degree, the consistency of risk impact direction, the credibility of relationship evidence, and the timeliness of relationships between related entities. The calculation results are normalized and weighted according to preset weights to obtain the relationship confidence level. Entity relationships with a relationship confidence level reaching a preset relationship confidence threshold are identified as directed risk dependencies. The preset relationship confidence threshold is the confidence limit used to determine whether a relationship between two related entities constitutes a directed risk dependency. The graph construction process uses related entities as nodes and directed risk dependencies as directed edges. Entity attributes are written into the corresponding nodes, and relationship semantics, risk impact direction, relationship evidence information, and relationship confidence levels are written into the corresponding directed edges to construct the technology transfer risk knowledge graph.

[0022] In this embodiment, the risk topology generation module includes: This process involves reading the associated entities, directed risk dependencies, and relationship confidence levels from a knowledge graph of technology transfer risks. The directed risk dependencies are then topologically quantified, and a matrix construction process is performed based on the quantization results to obtain a directed risk topology matrix. The associated entities are data nodes in the knowledge graph representing technological achievements, participating entities, resource conditions, and risk factors. The directed risk dependencies are directional relationships where a change in the state of one associated entity affects another. The relationship confidence level is a numerical value representing the degree of credibility of a corresponding directed risk dependency between two associated entities. The quantization process determines the directed risk topology weights between corresponding associated entities based on the relationship confidence levels of the directed risk dependencies, and sets the directed risk topology weights between associated entities without a directed risk dependency relationship to zero. The matrix construction process represents the directed connections between associated entities in a matrix form based on the directed risk topology weights and risk action directions corresponding to each associated entity. The risk action direction is a directional relationship where a change in the risk state of one associated entity is transmitted to another, generating a risk impact. The entity attributes in the knowledge graph of technology transfer risk are read, and risk configuration mapping is performed on the entity attributes corresponding to each associated entity to obtain the entity risk mitigation configuration quantity corresponding to each associated entity, and form a set of entity risk mitigation configuration quantities. The risk configuration mapping process is to extract risk attributes that characterize the risk status and risk mitigation ability of associated entities from entity attributes, perform scale unification and weighted summarization on different risk attributes, and convert the weighted summarization result into discrete configuration quantities. The entity risk mitigation configuration quantity is an integer value used to characterize the risk mitigation configuration status of associated entities before the start of the Dhar combustion test. Based on the directed risk topology matrix, the associated entities are screened for sink points. The screened associated entities are determined as sink points. The associated entity set and the entity risk mitigation configuration quantity set are then subjected to burning initialization processing to obtain an unburned entity set and an initial risk mitigation configuration quantity set. The sink point screening process involves selecting associated entities that can receive the directed risk effects of other associated entities from the associated entities, and determining the sink points based on the directed risk topology weights borne by the associated entities. The burning initialization process involves removing the sink points from the associated entity set, forming an unburned entity set from the remaining associated entities, and removing the entity risk mitigation configuration quantity corresponding to the sink points from the entity risk mitigation configuration quantity set to obtain an initial risk mitigation configuration quantity set that corresponds one-to-one with the unburned entity set. The burning operation is a state determination operation in the Dhar burning algorithm. It refers to comparing the configuration check quantity corresponding to the associated entity with its directed internal association quantity in the current unburned entity set. When the configuration check quantity meets the Dhar burning condition, the current associated entity is determined as a burning entity and removed from the unburned entity set, so as to indicate that the associated entity can be freed from internal risk dependence under the current topological constraints.

[0023] This invention transforms associated entities, risk action directions, and relationship confidence into a directed risk topology matrix, and maps entity attributes into discrete entity risk mitigation configuration quantities. This enables the transformation of dispersed scientific and technological achievements into risk information to form a risk topology structure with clear direction, quantifiable weights, and computable states. At the same time, by using sink selection and combustion initialization, the consistency of inputs for subsequent Dhar combustion tests is ensured, thereby improving the accuracy of risk dependency expression, the stability of risk state determination, and the interpretability of risk prediction results.

[0024] In this embodiment, the script configuration calculation module includes: Script vector generation is performed based on a directed risk topology matrix, sinks, and a set of unburned entities. The matrix rows and columns corresponding to the sinks are removed from the directed risk topology matrix to obtain a reduced directed risk topology matrix corresponding to the set of unburned entities. Positive integer script components are assigned to each unburned entity according to their order. A topological action operation is performed on each script component using the transpose of the reduced directed risk topology matrix. This topological action operation is then re-executed based on all updated script components until all script action components are greater than zero. The current script components are then arranged into a script vector according to the order of the unburned entities, and the corresponding script action components are combined into a script action quantity. The topological action operation includes obtaining the script action components corresponding to each unburned entity, detecting whether all script action components are greater than zero, and, if there is a script action component not greater than zero, increasing the script component corresponding to the current script action component. The script component is the positive integer number of times the corresponding unburned entity participates in the directed Dhar combustion test. The script action component is the configuration increment applied to the corresponding unburned entity after the script vector is processed by the reduced directed risk topology matrix. Based on the initial risk mitigation configuration set and script action quantity, configuration superposition processing is performed to obtain the configuration verification quantity corresponding to each unburned entity. All configuration verification quantities are then arranged into a configuration verification quantity set according to the order of the unburned entities. The configuration superposition processing is performed by superimposing the entity risk mitigation configuration quantity of each unburned entity with the corresponding script action component according to the one-to-one correspondence of the unburned entities. The superposition result is determined as the configuration verification quantity of the corresponding unburned entity. The configuration verification quantity is the entity configuration state used to compare the Dhar combustion conditions with the directed internal correlation quantity after the script vector action. The configuration verification quantity set maintains the same arrangement order as each unburned entity in the unburned entity set. Based on the set of unburned entities and the reduced directed risk topology matrix, internal correlation quantity extraction processing is performed to obtain the directed internal correlation quantity corresponding to each unburned entity. All directed internal correlation quantities are then arranged into a set of directed internal correlation quantities according to the order of the unburned entities. The internal correlation quantity extraction processing is as follows: for each unburned entity in the set of unburned entities, the directed risk topology weights pointing from the unburned entity to other unburned entities are read from the reduced directed risk topology matrix. All directed risk topology weights between the current unburned entity and other unburned entities in the set of unburned entities are accumulated, and the accumulated result is determined as the directed internal correlation quantity of the current unburned entity. The directed internal correlation quantity is a numerical value that characterizes the strength of the directed risk dependency constraint on the corresponding unburned entity within the current set of unburned entities.

[0025] In this embodiment, the combustion entity screening module includes: The algorithm reads the current set of unburned entities, the set of configuration check quantities, and the set of directed internal association quantities. It performs Dhar burning condition judgment on each unburned entity, determines the unburned entities whose configuration check quantity is not less than the corresponding directed internal association quantity as candidate burning entities, and forms a candidate burning entity set by combining all candidate burning entities. The Dhar burning condition is to compare the configuration check quantity formed by the script vector of each unburned entity with the directed internal association quantity corresponding to the current unburned entity item by item according to the order of the unburned entities. The burning is the entity state judgment operation in the Dhar burning algorithm. The candidate burning entity is the associated entity that meets the Dhar burning condition but has not yet been removed from the current set of unburned entities. The sink and the burning entities that have been removed from the unburned entity set before the current Dhar burning test round are combined to form the burned entity set. For each candidate burning entity in the candidate burning entity set, the directed risk dependency relationship from the technology achievement transformation risk knowledge graph is retrieved from the candidate burning entity to each entity in the burned entity set. The retrieved directed risk dependency relationships are combined to form the candidate parent relationship set of the corresponding candidate burning entity. The candidate parent relationship is the directed risk dependency relationship that connects the candidate burning entity and the burned entity and points to the burned entity. When there is a directed risk dependency relationship between the candidate burning entity and multiple entities in the burned entity set, all the corresponding directed risk dependency relationships are written into the candidate parent relationship set. Relationship ordering is performed on candidate parent relations to obtain the relationship ordering value corresponding to each candidate parent relation. The relationship ordering process involves determining the semantic order of candidate parent relations according to the preset risk relationship type order, unifying the scale of relation confidence and directed risk topology weight, weighting the relation semantic order, relation confidence, and directed risk topology weight according to preset relationship ordering parameters, and summing the weighting results to obtain the relationship ordering value of the corresponding candidate parent relation. The preset relationship ordering parameters are parameters that limit the proportion of relation semantic order, relation confidence, and directed risk topology weight in the relationship ordering process. The sum of all preset relationship ordering parameters is one. The relation semantic order, relation confidence, and directed risk topology weight are all normalized to 0 to 1. The preset relationship ordering parameters corresponding to the three are set to 0.40, 0.35, and 0.25, respectively. The relationship ordering value is retained to four decimal places. The relationship ordinal values ​​of the candidate parent relations corresponding to each candidate combustion entity are compared. The candidate parent relation with the largest relationship ordinal value is selected as the undetermined parent relation of the corresponding candidate combustion entity, and the relationship ordinal value of the undetermined parent relation is determined as the entity ordinal value of the corresponding candidate combustion entity. The entity ordinal values ​​of each candidate combustion entity in the candidate combustion entity set are compared. The candidate combustion entity with the largest entity ordinal value is determined as the combustion entity of the current Dhar combustion test round, and the undetermined parent relation corresponding to the current combustion entity is determined as the parent relation. When the entity ordinal values ​​of different candidate combustion entities are the same, the difference between the configuration test quantity and the directed internal association quantity of each candidate combustion entity is compared. The candidate combustion entity with the largest difference is determined as the combustion entity. When the difference is still the same, the combustion entity is determined according to the preset arrangement order of the associated entities. When the absolute difference between two relationship ordinal values ​​is less than 0.0001, the two relationship ordinal values ​​are determined to be the same. When the absolute difference between two configuration test quantities and the directed internal association quantity is less than 0.0001, the two differences are determined to be the same. Remove the burning entity from the current unburned entity set to obtain an updated unburned entity set. Write the parent relation and the corresponding burning entity into the parent relation set. Perform internal association update processing based on the updated unburned entity set. The internal association update processing is as follows: for each unburned entity in the updated unburned entity set, read the directed risk topology weight of the current unburned entity pointing to the burning entity in this round from the directed risk topology matrix, subtract the read directed risk topology weight from the directed internal association quantity of the current unburned entity before the update, obtain the corresponding unburned entity's updated directed internal association quantity, and form the updated directed internal association quantity set according to the arrangement order of the updated unburned entity set.

[0026] This invention combines the Dhar combustion conditions with relational semantic order, relational confidence, and directed risk topology weights to determine combustion entities and parent relations. This allows the combustion order to simultaneously reflect the entity configuration state and risk dependence strength. Furthermore, it eliminates ordering conflicts through difference comparison and preset arrangement order, thereby improving the stability of combustion entity screening, the accuracy of dominant risk path identification, and the interpretability of the risk transmission process.

[0027] In this embodiment, the risk lockout identification module includes: Based on the updated set of unburned entities and the updated directed internal correlation, perform the Dhar combustion test again until there are no entities in the set of unburned entities that meet the Dhar combustion condition. Connect the sink and the completed combustion entity according to the parent relationship corresponding to the combustion entity. Set the sink as the root node, the completed combustion entity as the tree node, and the parent relationship as the tree edge to generate a dominant risk dependency tree with the sink as the root node. The parent relationship is a directed risk dependency relationship between the combustion entity and the sink and the completed combustion entity whose combustion order is earlier than the combustion entity. The dominant risk dependency tree is a tree structure that forms and represents the dominant dependency path of the risk of scientific and technological achievement transformation according to the risk action direction and combustion order of the parent relationship. When the Dhar combustion test is terminated, the entities that remain in the set of unburned entities are identified as the risk-locked entity set. The risk-locked entity set is a set of entities whose configuration test quantities are all less than the corresponding directed internal correlation quantities and whose internal risk dependency constraints cannot be released under the current directed risk topology.

[0028] In this embodiment, the critical risk inversion module includes: The risk-locked entities in the risk-locked entity set are identified as the starting nodes for critical risk inversion. The directed risk topology weights between the risk-locked entities and the burning entities in the dominant risk dependency tree are read. Burning entities with non-zero directed risk topology weights are identified as path access entities. The parent relationships corresponding to the path access entities are traced layer by layer to the sink. The risk-locked entities, path access entities, burning entities connected by parent relationships, and corresponding parent relationships are defined as the critical risk inversion range. A range identifier vector is generated according to the order of the associated entities in the directed risk topology matrix. The path access entities are burning entities in the dominant risk dependency tree that have a directed risk dependency relationship with the risk-locked entities. The critical risk inversion range is the range of entities and parent relationships jointly defined by the risk-locked entities, path access entities, and the dominant risk dependency path from the path access entities to the sink. The range identifier vector is a vector used to mark whether the associated entities are within the critical risk inversion range. The corresponding components of associated entities within the critical risk inversion range are set to one, and the corresponding components of associated entities outside the critical risk inversion range are set to zero. Extract the matrix rows and columns corresponding to associated entities within the critical risk inversion range from the directed risk topology matrix according to the range identifier vector, construct the inversion topology submatrix, determine the inversion internal correlation quantity corresponding to each associated entity within the critical risk inversion range based on the inversion topology submatrix, and obtain the boundary gap quantity of the corresponding associated entity based on the inversion internal correlation quantity. Specifically, the boundary gap quantity of the corresponding associated entity is obtained by subtracting the corresponding configuration check quantity from the inversion internal correlation quantity, and the positive difference value is determined as the boundary gap quantity of the corresponding associated entity. The boundary gap quantity corresponding to the non-positive difference value is set to zero. Establish inversion counting constraints according to the boundary gap quantity corresponding to each associated entity. Under the condition that all inversion count components are non-negative integers, find the value that minimizes the sum of the inversion count components, and the configuration check quantity after the action of the inversion topology submatrix is ​​not less than the corresponding inversion internal correlation quantity, and the actual value after the action of the inversion topology submatrix is ​​less than the corresponding internal correlation quantity. The critical inversion count vector is defined as follows: the inversion topology submatrix is ​​a local matrix in the directed risk topology matrix corresponding to the critical risk inversion range; the inversion internal correlation quantity is the sum of the directed risk topology weights of each correlated entity within the critical risk inversion range pointing to other correlated entities within the critical risk inversion range; the boundary gap quantity is the configuration quantity missing by the correlated entity to reach the critical combustion condition; the inversion count constraint is a constraint condition that limits the number of configuration adjustments of each correlated entity during the critical risk inversion process, the relationship between the configuration verification quantity and the inversion internal correlation quantity, and the non-negative relationship of the entity risk mitigation configuration quantity; the inversion count component is the number of times a single correlated entity participates in topology configuration adjustment within the critical risk inversion range, and its value is a non-negative integer; the critical inversion count vector is a non-negative integer vector that satisfies the boundary gap constraint and minimizes the total number of configuration adjustments. Topological action operations are performed on the critical inversion counting vector using the transpose of the inversion topological submatrix to obtain the local configuration adjustment amounts corresponding to each associated entity within the critical risk inversion range. The local configuration adjustment amounts are written to the corresponding positions of the associated entities according to the range identifier vector. The corresponding positions of associated entities outside the critical risk inversion range are set to zero to generate a critical configuration adjustment vector. The critical configuration adjustment vector is superimposed on the entity risk resolution configuration amount item by item to generate a critical risk configuration. The critical configuration adjustment vector is the minimum configuration adjustment result required for the associated entities within the critical risk inversion range to reach the critical combustion state under the joint constraints of the risk-locked entity set and the dominant risk dependency tree. The critical risk configuration is the entity risk resolution configuration state after the critical configuration adjustment is completed. The critical risk configuration and the script vector are used as inputs for the second Dhar combustion test.

[0029] This invention defines the critical risk inversion range by combining the risk-locked entity set and the dominant risk dependency tree, focusing risk calculation on key risk entities and dominant dependency paths. It also uses the inversion topology submatrix, boundary gap quantity, and critical inversion counting vector to determine the minimum configuration adjustment result, thereby reducing the interference of irrelevant topological relationships on the inversion results, improving the accuracy, computational efficiency, and interpretability of critical risk boundary identification, and providing a quantitative basis for resource adjustment and risk management in the transformation of scientific and technological achievements.

[0030] In this embodiment, the risk prediction output module includes: Topological action processing is performed using the reduced directed risk topology matrix and script vector. The topological action result is superimposed on the critical risk configuration item by item to obtain the second configuration check quantity corresponding to each associated entity. The second configuration check quantity is the entity configuration state formed after the critical risk configuration is processed by the script vector. The configuration check quantity is replaced with the second configuration check quantity. Candidate combustion entity screening, combustion entity and parent relationship determination, combustion entity removal and directed internal association quantity update are performed in sequence to complete the second Dhar combustion check until there are no entities in the unburned entity set that meet the Dhar combustion conditions. The combustion entity and parent relationship are determined by determining the combustion entity from the candidate combustion entities according to the preset relationship ordering parameters and determining the parent relationship corresponding to the combustion entity. The risk locking ratio is determined based on the number of entities at the start of the second Dhar combustion test and the number of remaining entities at the end of the test. The critical configuration adjustment vector is determined as the critical boundary of the technology transfer risk. Entities remaining in the unburned entity set at the end of the test are determined as residual risk entities. The residual configuration gap is determined based on the difference between the directed internal correlation quantity corresponding to the residual risk entity and the second configuration test quantity. The critical boundary of the technology transfer risk, the risk locking ratio, the residual risk entities, and the residual configuration gap are written into the technology transfer risk prediction result. The risk locking ratio is the ratio of the number of remaining entities in the unburned entity set at the end of the second Dhar combustion test to the number of entities in the unburned entity set at the start of the second Dhar combustion test. The technology transfer risk prediction result is an output that characterizes the risk status and risk adjustment needs of the target technology transfer. It is used to determine the configuration adjustment range required for the target technology to reach the critical risk status, identify related entities still constrained by internal risk dependence, quantify the degree of residual risk, and provide a basis for adjusting the technology transfer plan and handling risks.

[0031] Example 1: To verify the feasibility of this invention in practice, it was applied to a set of anonymized data that had undergone verification of the technology transfer status. The selected data covered the stages of achievement research and development, technology evaluation, pilot-scale verification, resource input, market demand, and industrialization implementation, encompassing a total of 80 technological achievements. For each achievement, the data recorded its technological maturity status, funding continuity status, equipment compatibility status, raw material supply status, market demand status, and the capabilities of the implementing entity. After manual verification, 31 of the 80 technological achievements encountered problems during the transfer process, such as pilot-scale stagnation, funding interruption, equipment incompatibility, insufficient supply, or unmet demand. The remaining 49 achievements were able to complete the transfer according to the predetermined plan. Traditional risk assessment methods mainly calculate a comprehensive score based on the weight of each risk indicator, making it difficult to determine the transmission direction between risk factors and to identify the risk-locking entities and minimum configuration adjustment requirements that led to the transfer stagnation.

[0032] The knowledge graph construction module read 6240 pieces of data related to technology transfer. After standardization, field mapping, missing data marking, and duplicate data removal, 5976 standardized data pieces were retained. Entity alignment confidence was obtained by weighting name matching, attribute matching, contextual semantic matching, and data source matching, with corresponding weights set to 0.35, 0.30, 0.20, and 0.15, respectively. The preset entity alignment threshold was set to 0.85. After entity parsing and alignment, a total of 846 related entities were formed. The weights for relation semantic matching, consistency of risk action direction, credibility of relation evidence, and timeliness were set to 0.30, 0.25, 0.30, and 0.15, respectively. The preset relation confidence threshold was set to 0.78. A total of 2518 directed risk dependencies were retained. A knowledge graph of technology transfer risk was constructed from the related entities and directed risk dependencies.

[0033] The risk topology generation module determines the directed risk topology weights based on the relationship confidence level. Weights between 0.78 and 0.85 are set to 1, between 0.85 and 0.90 to 2, between 0.90 and 0.94 to 3, between 0.94 and 0.97 to 4, and not less than 0.97 to 5. The risk configuration mapping weights corresponding to the technology maturity status, resource input status, external demand status, implementing entity capabilities, and conversion implementation conditions are set to 0.30, 0.25, 0.20, 0.15, and 0.10, respectively, mapping to entity risk mitigation configuration quantities ranging from 0 to 10.

[0034] A scientific and technological achievement containing 10 non-sinking related entities is selected from the validation data for illustration. An initial risk mitigation configuration set [3, 2, 4, 1, 2, 3, 2, 1, 4, 2] is obtained according to a preset arrangement. The reduced directed risk topology matrix is ​​applied to generate script vectors [1, 2, 1, 1, 2, 1, 1, 2, 1, 1] and script action quantities [2, 1, 2, 1, 2, 2, 1, 2, 1, 2], thus obtaining the configuration verification set [5, 3, 6, 2, 4, 5, 3, 3, 5, 4]. The corresponding directed internal correlation set is [4, 5, 6, 4, 5, 3, 4, 5, 3, 4].

[0035] The combustion entity screening module sets the preset relation ordering parameters corresponding to relation semantic order, relation confidence, and directed risk topology weight to 0.40, 0.35, and 0.25, respectively, and determines the combustion entities and parent relations based on the relation ordering values. After completing the first Dhar combustion test, the pilot-scale stability risk, funding continuity risk, and market demand fulfillment risk are retained in the unburned entity set, with a risk-locked entity count of 3 and a risk-locked ratio of 0.3000. A dominant risk dependency tree is generated based on the combustion entities and their parent relations. It is determined that the pilot-scale stability risk is accessed through the core equipment adaptation risk via the dominant risk dependency path, the funding continuity risk is accessed through the technology maturity risk via the dominant risk dependency path, and the market demand fulfillment risk is accessed through the raw material supply risk via the dominant risk dependency path.

[0036] The critical risk inversion module constructs an inversion topology submatrix based on the risk-locked entities and the dominant risk dependency path, calculates the boundary gap quantity [0, 0, 0, 2, 1, 0, 1, 0, 2, 0], and obtains the critical configuration adjustment vector [0, 0, 0, 2, 1, 1, 1, 0, 2, 1]. The critical configuration adjustment vector is then superimposed item by item with the entity risk resolution configuration quantity to generate the critical risk configuration. After the second Dhar combustion test, all 10 associated entities completed combustion, leaving 0 entities and a risk lock-in ratio of 0, indicating that the critical configuration adjustment vector can release the internal risk dependency constraints in the current risk topology.

[0037] To verify the beneficial effects of this invention, it was compared with traditional weighted scoring methods, knowledge graph threshold analysis methods, and the Dhar burning method without relation ordering and critical risk inversion. Traditional weighted scoring methods assign weights to risk indicators such as technology maturity, resource input, market demand, and implementation capability, and obtain a comprehensive risk score through weighted aggregation. A preset threshold is then used to determine the risk status of technological achievement transformation. Knowledge graph threshold analysis methods construct a knowledge graph from technological achievements, participating entities, resource conditions, and risk factors. High-risk entities and risk relationships are identified based on entity attributes, relation confidence, and preset thresholds, but Dhar burning tests and critical risk inversion are not performed. The Dhar burning method without relation ordering and critical risk inversion directly determines the burning conditions based on configuration check quantities and directed internal correlation quantities, identifying risk-locked entities. However, it does not combine relation semantics, relation confidence, and directed risk topology weights to determine the burning order, nor does it calculate the critical configuration adjustment vector. The results are shown in Table 1. Table 1. Performance Comparison Results of Different Risk Prediction Methods for Technology Transfer

[0038] As shown in Table 1, among the 80 scientific and technological achievements verification data, this invention correctly predicted 74 scientific and technological achievements, achieving a prediction accuracy of 92.50%, which is 15.00 percentage points, 10.00 percentage points, and 6.25 percentage points higher than the traditional weighted scoring method, knowledge graph threshold analysis method, and unimproved Dhar burning method, respectively. The risk sample recall rate of this invention reached 90.32%, and the risk miss rate was reduced to 9.68%, indicating that this invention can more fully identify the risk status in the process of scientific and technological achievement transformation. The consistency rate of the dominant risk path of this invention reached 93.55%, which is 9.68 percentage points higher than the unimproved Dhar burning method, and the critical configuration verification pass rate reached 96.77%, which is 16.12 percentage points higher than the unimproved Dhar burning method. This demonstrates that by using relational ordering, risk lock-in identification, critical risk inversion, and a second Dhar burning test, the accuracy of risk transmission path identification and the reliability of critical configuration adjustment results can be improved.

[0039] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A technology transfer risk prediction system based on knowledge graphs, characterized in that, include: The knowledge graph construction module is used to acquire data related to the transformation of scientific and technological achievements, and to perform entity recognition, relationship extraction and entity alignment to construct a knowledge graph of risks in the transformation of scientific and technological achievements. The risk topology generation module is used to construct a directed risk topology matrix based on the knowledge graph of technology transfer risks, generate entity risk mitigation configuration quantities, and determine the sink point and the set of unburned entities; The script configuration calculation module is used to generate script vectors from the directed risk topology matrix, determine the set of configuration inspection quantities by combining the entity risk mitigation configuration quantity, and determine the set of directed internal correlation quantities based on the set of unburned entities. The combustion entity screening module is used to compare the configuration check quantity and the directed internal correlation quantity using the Dhar combustion algorithm to screen out candidate combustion entities, determine the combustion entity and parent relationship according to the preset relational ordering parameters, remove combustion entities from the set of unburned entities and update the directed internal correlation quantity. The risk lockout identification module is used to repeatedly perform the Dhar burning test based on the updated directed internal correlation quantity until no entity meets the Dhar burning condition. Based on the parent relationship, a dominant risk dependency tree with the sink node as the root node is generated, and the remaining entities are identified as the risk lockout entity set. The critical risk inversion module is used to limit the critical risk inversion range by the risk-locked entity set and the dominant risk dependency tree, and to obtain the critical configuration adjustment vector by combining the directed risk topology matrix to generate the critical risk configuration. The risk prediction output module is used to perform a second Dhar combustion test on the critical risk configuration based on the script vector, and generate risk prediction results for the transformation of scientific and technological achievements.

2. The technology transfer risk prediction system based on knowledge graphs according to claim 1, characterized in that, The knowledge graph construction module includes: Obtain the technology transfer-related data corresponding to the target scientific and technological achievement, perform association identification binding and data standardization processing on the technology transfer-related data, and collect the processed technology transfer-related data to generate standardized technology transfer-related data. Entity parsing and entity alignment are performed on the standardized scientific and technological achievement transformation-related data to obtain a set of related entities and a set of entity attributes. Based on standardized data related to the transformation of scientific and technological achievements, sets of related entities, and sets of entity attributes, relationship parsing and graph construction are performed to obtain a knowledge graph of risks in the transformation of scientific and technological achievements.

3. The technology transfer risk prediction system based on knowledge graphs according to claim 1, characterized in that, The risk topology generation module includes: Read the associated entities, directed risk dependencies, and relationship confidence in the knowledge graph of technology transfer risks, perform topological quantification on the directed risk dependencies, and perform matrix construction based on the topological quantification results to obtain the directed risk topological matrix; Read the entity attributes in the knowledge graph of technology transfer risk, perform risk configuration mapping on the entity attributes corresponding to each associated entity, obtain the entity risk mitigation configuration quantity corresponding to each associated entity, and form a set of entity risk mitigation configuration quantities. Based on the directed risk topology matrix, the associated entities are screened to identify the sinks. The screened associated entities are then identified as sinks. The associated entity set and the entity risk mitigation configuration set are then subjected to combustion initialization processing to obtain the unburned entity set and the initial risk mitigation configuration set.

4. The technology transfer risk prediction system based on knowledge graphs according to claim 1, characterized in that, The script configuration calculation module includes: The script vector generation process is based on the directed risk topology matrix, sink, and unburned entity set. The matrix rows and columns corresponding to the sink are removed from the directed risk topology matrix to obtain the reduced directed risk topology matrix corresponding to the unburned entity set. Positive integer script components are set for each unburned entity according to the order of the unburned entities. The topological action operation is performed on each script component using the transpose of the reduced directed risk topology matrix. The topological action operation is re-executed based on all updated script components until all script action components are greater than zero. The current script components are combined into a script vector according to the order of the unburned entities, and the corresponding script action components are combined into a script action. Based on the initial risk mitigation configuration set and script action quantity, configuration superposition processing is performed to obtain the configuration verification quantity corresponding to each unburned entity. All configuration verification quantities are arranged into a configuration verification quantity set according to the order of the unburned entities. The configuration superposition processing is to superimpose the entity risk mitigation configuration quantity of each unburned entity with the corresponding script action component according to the one-to-one correspondence of the unburned entities, and the obtained superposition result is determined as the configuration verification quantity of the corresponding unburned entity. Based on the set of unburned entities and the reduced directed risk topology matrix, internal correlation quantities are extracted to obtain the directed internal correlation quantities corresponding to each unburned entity. All directed internal correlation quantities are then arranged into a set of directed internal correlation quantities according to the order of the unburned entities.

5. The technology transfer risk prediction system based on knowledge graphs according to claim 1, characterized in that, The combustion entity screening module includes: Read the current set of unburned entities, the set of configuration check quantities, and the set of directed internal correlation quantities. Perform Dhar combustion condition determination on each unburned entity. Determine unburned entities whose configuration check quantities are not less than the corresponding directed internal correlation quantities as candidate combustion entities, and form a candidate combustion entity set from all candidate combustion entities. The sink and the burning entities that have been removed from the unburned entity set before the current Dhar burning test round are combined into a burning entity set. For each candidate burning entity in the candidate burning entity set, the directed risk dependency relationship from the technology achievement transformation risk knowledge graph to each entity in the burning entity set is retrieved. The retrieved directed risk dependency relationships are combined into a candidate parent relationship set for the corresponding candidate burning entity. Perform relation ordering on the candidate parent relations to obtain the relation ordering value corresponding to each candidate parent relation; The relational ordinal values ​​of the candidate parent relations corresponding to each candidate combustion entity are compared. The candidate parent relation with the largest relational ordinal value is selected as the undetermined parent relation of the corresponding candidate combustion entity, and the relational ordinal value of the undetermined parent relation is determined as the entity ordinal value of the corresponding candidate combustion entity. The entity ordinal values ​​of each candidate combustion entity in the candidate combustion entity set are compared. The candidate combustion entity with the largest entity ordinal value is determined as the combustion entity of the current Dhar combustion test round, and the undetermined parent relation corresponding to the current combustion entity is determined as the parent relation. When the entity ordinal values ​​of different candidate combustion entities are the same, the difference between the configuration test quantity and the directed internal association quantity of each candidate combustion entity is compared. The candidate combustion entity with the largest difference is determined as the combustion entity. When the difference is still the same, the combustion entity is determined according to the preset arrangement order of the associated entities. Remove the burning entity from the current unburned entity set to obtain an updated unburned entity set. Write the parent relation and the corresponding burning entity into the parent relation set, and perform internal association quantity update processing based on the updated unburned entity set.

6. The technology transfer risk prediction system based on knowledge graphs according to claim 1, characterized in that, The risk lockout identification module includes: Based on the updated set of unburned entities and the updated directed internal correlation, perform the Dhar combustion test again until there are no entities in the set of unburned entities that meet the Dhar combustion condition. Connect the sink node and the entity that has completed combustion according to the parent relationship corresponding to the burning entity, set the sink node as the root node, set the entity that has completed combustion as the tree node, and set the parent relationship as the tree edge to generate a dominant risk dependency tree with the sink node as the root node. The entities that remain in the set of unburned entities when the Dhar combustion test is terminated are defined as the risk-locked entity set.

7. The technology transfer risk prediction system based on knowledge graphs according to claim 1, characterized in that, The critical risk inversion module includes: The risk-locked entities in the risk-locked entity set are identified as the starting nodes for critical risk inversion. The directed risk topology weights between the risk-locked entities and the combustion entities in the dominant risk dependency tree are read. The combustion entities with non-zero directed risk topology weights are identified as path access entities. The parent relationships corresponding to the path access entities are traced layer by layer to the sink. The risk-locked entities, path access entities, combustion entities connected by parent relationships, and corresponding parent relationships are defined as the critical risk inversion range. The range identifier vector is generated according to the arrangement order of the associated entities in the directed risk topology matrix. Extract the matrix rows and columns corresponding to the associated entities within the critical risk inversion range from the directed risk topology matrix according to the range identifier vector, construct the inversion topology submatrix, determine the inversion internal correlation quantity corresponding to each associated entity within the critical risk inversion range based on the inversion topology submatrix, obtain the boundary gap quantity of the corresponding associated entity based on the inversion internal correlation quantity, establish inversion counting constraints according to the boundary gap quantity corresponding to each associated entity, and find the critical inversion counting vector that minimizes the sum of the inversion counting components, and whose configuration verification quantity after the action of the inversion topology submatrix is ​​not less than the corresponding inversion internal correlation quantity, and whose entity risk resolution configuration quantity after the action of the inversion topology submatrix is ​​not less than zero. The topological action operation is performed on the critical inversion counting vector using the transpose relation of the inversion topological submatrix to obtain the local configuration adjustment amount corresponding to each associated entity within the critical risk inversion range. The local configuration adjustment amount is written into the corresponding position of the associated entity according to the range identifier vector. The corresponding position of the associated entity outside the critical risk inversion range is set to zero to generate the critical configuration adjustment vector. The critical configuration adjustment vector is superimposed with the entity risk resolution configuration amount item by item to generate the critical risk configuration.

8. The technology transfer risk prediction system based on knowledge graphs according to claim 1, characterized in that, The risk prediction output module includes: Topological action processing is performed using the reduced directed risk topology matrix and script vector. The topological action results are then superimposed on the critical risk configurations item by item to obtain the second configuration test value for each associated entity. Replace the configuration check quantity with the second configuration check quantity, and sequentially perform candidate combustion entity screening, combustion entity and parent relationship determination, combustion entity removal and directed internal association quantity update to complete the second Dhar combustion check until there are no entities in the unburned entity set that meet the Dhar combustion conditions; The risk lockout ratio is determined based on the number of entities at the start of the second Dhar combustion test and the number of remaining entities at the end of the test. The critical configuration adjustment vector is determined as the critical boundary of the technology transfer risk. The entities remaining in the set of unburned entities at the end of the test are determined as residual risk entities. The residual configuration gap is determined based on the difference between the directed internal correlation quantity corresponding to the residual risk entity and the second configuration test quantity. The critical boundary of the technology transfer risk, the risk lockout ratio, the residual risk entities, and the residual configuration gap are written into the technology transfer risk prediction results.

Images