Information extraction processing method and system applied to data sharing service

Abstract

The application provides an information extraction processing method and system applied to a data sharing service, obtains a historical interaction text set from a cross-department business data sharing system, constructs an entity relationship network for the historical interaction text set, generates a multi-dimensional entity relationship network containing entity nodes, relationship edges and attribute labels, constructs a semantic topology graph based on the multi-dimensional entity relationship network, and obtains a semantic topology graph containing entity correlation strength and semantic evolution paths; a preset zero-knowledge proof protocol is called to perform access control vector generation on the semantic topology graph, and an access control vector containing a topology weight threshold and node access permission is obtained; the access control vector and the semantic topology graph are jointly processed through a space-time correlation analysis model, and a business event template containing standardized fields and correlation relationship indexes is generated. The application can improve the accuracy, security and scene adaptability of information extraction in data sharing services.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of data processing, in particular to an information extraction processing method and system applied to a data sharing service. BACKGROUND

[0002] With the deepening of business informatization construction, cross-department business data sharing services are increasingly becoming the key support for improving business efficiency. As the basic link of data sharing, information extraction processing technology plays an important role in extracting key information from massive unstructured text interaction records. Currently, the information extraction method in business data sharing usually relies on traditional natural language processing technology to extract entity information such as business terminology, organization name and business identifier from text through entity recognition and relationship extraction, and establish simple association relationships between entities. However, this method is often limited to surface analysis of text content in practical application, and it is difficult to convert scattered text interaction records into a structured entity association network, resulting in a lack of systematicness and dynamics in the organization of entity relationships. At the same time, due to the special requirements for data security and access control in cross-department data sharing scenarios, the existing method is difficult to achieve fine-grained access permission management according to the semantic association strength between entities, which is easy to cause data leakage risk. In addition, the format of text interaction records in different business scenarios is quite different, and the existing information extraction results lack a unified standardized template, resulting in low data reuse efficiency, which is difficult to meet the actual needs of cross-department business collaboration for information extraction results, and restricts the quality and efficiency improvement of business data sharing services. SUMMARY

[0003] The present application provides an information extraction processing method and system applied to a data sharing service.

[0004] In a first aspect, the present application provides an information extraction processing method applied to a data sharing service, which comprises: obtaining a set of historical interaction texts from a cross-department business data sharing system, the set of historical interaction texts containing text interaction records generated under different business scenarios with timestamp markers; performing entity relationship network construction on the set of historical interaction texts to generate a multi-dimensional entity relationship network containing entity nodes, relationship edges and attribute labels, the attribute labels being used to represent the context association attributes of the entity nodes in the text interaction records; performing semantic topology graph construction based on the multi-dimensional entity relationship network to obtain a semantic topology graph containing entity association strength and semantic evolution path; calling a preset zero-knowledge proof protocol to perform access control vector generation on the semantic topology graph to obtain an access control vector containing topology weight threshold and node access permission; and jointly processing the access control vector and the semantic topology graph through a spatio-temporal association analysis model to generate a business event template containing standardized fields and association relationship indexes.

[0005] In a second aspect, an embodiment of the present application provides a computer system, comprising: a memory, wherein a computer program is stored in the memory; and a processor configured to load the computer program to implement the information extraction processing method applied to a data sharing service as described above.

[0006] The information extraction processing method applied to a data sharing service provided by the present application can convert unstructured text data into a structured entity association topology, effectively improving the structured degree of data organization and the traceability of entity relationships. On this basis, a semantic topology graph is constructed to generate a semantic topology graph containing entity association strength and semantic evolution path, which can dynamically depict the trajectory of entity relationship changes over time and quantify the semantic dependence degree between entities, so that the information extraction result can adapt to the scene demand of dynamic evolution of entity relationship in cross-department data interaction. A zero-knowledge proof protocol is called to generate an access control vector for the semantic topology graph, and an access control vector containing a topology weight threshold and node access permission is obtained, which can realize fine access permission control based on the statistical characteristics of entity association strength, and ensure the operation safety and permission compliance of different sharing subjects in the data sharing process. Finally, the access control vector and the semantic topology graph are jointly processed by a space-time association analysis model to generate a business event template containing standardized fields and association relationship indexes, which can unify the text interaction record format in different business scenarios and accurately locate the mapping position of the entity node in the template, effectively solving the problem of low data reuse efficiency caused by inconsistent formats in cross-department data sharing, and improving the accuracy, security and scene adaptability of information extraction in data sharing services as a whole. BRIEF DESCRIPTION OF DRAWINGS

[0007] Figure 1 is a flowchart of an information extraction processing method applied to a data sharing service provided by an embodiment of the present application. The method can be executed by a computer system, and the method comprises:

[0008] Figure 2 is a schematic diagram of a computer system provided by an embodiment of the present application. DETAILED DESCRIPTION

[0009] Please refer to Figure 1 is a flowchart of an information extraction processing method applied to a data sharing service provided by an embodiment of the present application. The method can be executed by a computer system, and the method comprises:

[0010] Step S100: Obtain a set of historical interactive texts from a cross-department business data sharing system, the set of historical interactive texts containing text interaction records with timestamp labels generated under different business scenarios.

[0011] The cross-department business data sharing system is a comprehensive system for realizing data flow and sharing between different business departments. The system has data storage, management, interaction, etc. functions, and can integrate scattered data of different departments and break down data barriers. The set of historical interactive texts refers to a collection of a series of text interaction records stored in the cross-department business data sharing system, which come from different business scenarios. It covers various business data of multiple business departments, and the timestamp label is a time identifier added to each text interaction record, indicating the specific time when the record was generated. Text interaction records refer to text information exchanged between different departments or departments and users during the cross-department business data sharing process, such as email correspondence, chat records, approval opinions, etc.

[0012] In actual operation, the set of historical interactive texts can be obtained by calling the interface of the cross-department business data sharing system. For example, use the API (Application Programming Interface) provided by the system, send a request to the system, specify the time range, business scenario, etc. conditions for obtaining data, and the system will return the set of historical interactive texts that meet the conditions.

[0013] Step S200: Construct an entity relationship network for the set of historical interactive texts, generate a multi-dimensional entity relationship network containing entity nodes, relationship edges and attribute labels, and the attribute labels are used to represent the context association attributes of the entity nodes in the text interaction records.

[0014] The entity relationship network is a graph structure used to represent the relationship between entities, where entity nodes represent various entities in the text, such as business domain-specific terms, cross-department institution identifiers, business process nodes, etc.; relationship edges represent the association between entities, such as an institution participating in a business process, or a term being semantically related to another term; and attribute labels are additional information added to each entity node, used to describe the context association attributes of the entity in the text interaction record, such as the first appearance time of the entity, the business scenario it belongs to, and the role it plays in the interaction, etc.

[0015] The multi-dimensional entity relationship network means that the network not only contains the relationship between entities, but also considers the multiple attribute information of entities, describing entities and their relationships from multiple angles.

[0016] As an implementation, step S200 can include the following steps S210-S260:

[0017] Step S210: The historical interaction text set is blocked in ascending order of timestamp to obtain a text block sequence with continuous time sequence characteristics, and each text block contains text interaction records within a preset time interval, and there is a preset proportion of time overlap area between adjacent text blocks.

[0018] Ascending order of timestamp means sorting the historical interaction text set in ascending order of timestamp to ensure that the text records are arranged in chronological order. Blocking is the process of dividing the sorted historical interaction text set into multiple text blocks, each of which contains text interaction records generated within a preset time interval. The preset time interval can be set according to specific business needs and analysis purposes, such as one day, one week, or one month, etc. Continuous time sequence characteristics mean that each text block in the text block sequence is continuous in time, forming an ordered time sequence, which can better capture the changing pattern of text interaction records over time.

[0019] There is a preset proportion of time overlap area between adjacent text blocks, which is to avoid losing information in the blocking process. For example, when there is no overlap between adjacent text blocks, the associated information of a certain entity may be split at the boundary of two text blocks, resulting in incomplete information. By setting a certain proportion of time overlap area, it can be ensured that part of the same text interaction records are included in adjacent text blocks, so as to more accurately capture the relationship between entities.

[0020] Step S220: Entity recognition is performed on the text interaction records in the text block sequence to extract an entity set containing business domain-specific terms, cross-departmental agency identifiers, and business process nodes, each entity having a unique identifier and type annotation.

[0021] Entity recognition is the process of identifying entities with specific meanings from text. In this method, business domain-specific terms, cross-departmental agency identifiers, and business process nodes are mainly identified. Business domain-specific terms are professional vocabulary used in business work, cross-departmental agency identifiers are names or codes representing different departments, and business process nodes are links in the business process, such as "application submission", "audit pass", etc.

[0022] The entity set is a set composed of all identified entities, each entity having a unique identifier for distinguishing different entities in subsequent processing; at the same time, each entity also has a type annotation indicating whether it belongs to a business domain-specific term, a cross-departmental agency identifier, or a business process node.

[0023] As an implementation, step S220 can include the following steps S221-S229:

[0024] Step S221: Construct a hierarchical entity recognition model combining bidirectional long short-term memory network and conditional random field, which includes input layer, word embedding layer, bidirectional long short-term memory layer, multi-head attention mechanism layer, feature fusion layer and conditional random field output layer.

[0025] Bidirectional long short-term memory network can consider the context information of the sequence at the same time, and has good effect when processing natural language sequence. Conditional random field is a kind of probability graph model, which is used for sequence labeling task, and can predict the label of each position according to the context information of input sequence.

[0026] The hierarchical entity recognition model is a model combining bidirectional long short-term memory network and conditional random field, which realizes entity recognition through multiple levels of processing. The input layer is the entrance of the model, which is used to receive the input of the text interaction record; the word embedding layer converts the input text into word vector for subsequent neural network processing; the bidirectional long short-term memory layer models the forward and backward semantic dependence of the word vector sequence to capture the context information of the text; the multi-head attention mechanism layer can perform multi-scale semantic focusing processing on the output of the bidirectional long short-term memory layer to extract features of different semantic granularity; the feature fusion layer fuses the output of the multi-head attention mechanism layer to generate a feature sequence with semantic consistency; the conditional random field output layer predicts the entity label of each character position according to the output of the feature fusion layer.

[0027] Step S222: Input the text interaction record into the input layer of the hierarchical entity recognition model according to the character sequence, generate a character sequence vector through character-level segmentation processing, and the length of the character sequence vector is consistent with the number of characters in the text interaction record.

[0028] The character sequence refers to the sequence formed by arranging each character in the text interaction record in turn. The character-level segmentation processing is the process of dividing the text according to the characters, and each character is treated as an independent unit. The character sequence vector is the result of converting the character sequence into a vector representation, each character corresponds to an element in the vector, and the length of the vector is the same as the number of characters in the text interaction record.

[0029] In actual operation, first, the text in the text interaction record is extracted, and then it is segmented by characters to obtain a character sequence. Next, an index corresponding to each character is assigned, and the character sequence is converted into an index sequence. Finally, the corresponding word vector is found from the word vector matrix of the word embedding layer according to the index sequence, and a character sequence vector is generated. For example, for the text "business approval", it is segmented into the character sequence "industry", "business", "review", and "batch", each character is assigned an index, such as "industry" corresponding to index 1, "business" corresponding to index 2, "review" corresponding to index 3, and "batch" corresponding to index 4, and then the corresponding word vector is found from the word vector matrix according to the index to generate a character sequence vector.

[0030] Step S223: input the character sequence vector into the word embedding layer, and convert the character sequence into a word vector sequence containing context semantic information through the pre-trained business domain word vector matrix. The dimension of the word vector sequence matches the preset embedding dimension parameter.

[0031] The word embedding layer is a component of the hierarchical entity recognition model, and its function is to convert the character sequence vector into a word vector sequence containing context semantic information. The pre-trained business domain word vector matrix is obtained by training on a large amount of business text data, and each character or word can be mapped to a low-dimensional vector space, so that characters or words with similar semantics are close in the vector space.

[0032] The preset embedding dimension parameter refers to the dimension of the word vector, which can be set according to the specific task and data situation. By inputting the character sequence vector into the word embedding layer and using the pre-trained business domain word vector matrix, the character sequence can be converted into a word vector sequence with context semantic information, providing a basis for subsequent semantic analysis.

[0033] In actual operation, after the character sequence vector is input into the word embedding layer, the word embedding layer will find the corresponding word vector from the pre-trained business domain word vector matrix according to the index in the character sequence vector, arrange these word vectors in order, and obtain a word vector sequence.

[0034] Step S224: model the forward and backward semantic dependencies of the word vector sequence through the bidirectional long short-term memory layer to generate a forward hidden state sequence containing historical context information and a backward hidden state sequence containing future context information. The forward and backward hidden state sequences correspond one-to-one in time steps.

[0035] The bidirectional long short-term memory layer is composed of two independent long short-term memory networks, one of which processes from the beginning to the end of the sequence (forward processing), and the other of which processes from the end to the beginning of the sequence (backward processing). The forward processing can capture the historical context information of the sequence, and the backward processing can capture the future context information of the sequence.

[0036] The forward hidden state sequence is a sequence composed of hidden states output by the forward long short-term memory network at each time step, which contains historical context information from the beginning of the sequence to the current time step; the backward hidden state sequence is a sequence composed of hidden states output by the backward long short-term memory network at each time step, which contains future context information from the end of the sequence to the current time step. The time steps of the forward and backward hidden state sequences correspond one by one, that is, the i-th element of the forward hidden state sequence and the i-th element of the backward hidden state sequence correspond to the same character position.

[0037] In actual operation, the word vector sequence is sequentially input into the forward and backward long short-term memory networks of the bidirectional long short-term memory layer. For the forward long short-term memory network, the hidden state of each time step is calculated in turn from the first character of the sequence; for the backward long short-term memory network, the hidden state of each time step is calculated in turn from the last character of the sequence.

[0038] Step S225: concatenate the forward hidden state sequence and the backward hidden state sequence into a bidirectional hidden state sequence, input into the multi-head attention mechanism layer, and perform multi-scale semantic focusing processing on the bidirectional hidden state sequence through a preset number of attention heads to generate an attention-enhanced feature sequence containing different semantic granularities.

[0039] Concatenation is an operation of merging the forward hidden state sequence and the backward hidden state sequence in the feature dimension to obtain the bidirectional hidden state sequence. Multi-head attention mechanism is a mechanism that can simultaneously focus on different positions in the sequence, and performs different scale semantic focusing processing on the input sequence through multiple attention heads.

[0040] The preset number of attention heads refers to the number of attention heads set in the multi-head attention mechanism, and each attention head can focus on the input sequence from different angles to extract features of different semantic granularities. The attention-enhanced feature sequence is the output of the multi-head attention mechanism layer, which contains feature information extracted from different semantic granularities.

[0041] In actual operation, the forward hidden state sequence and the backward hidden state sequence are concatenated in the feature dimension to obtain the bidirectional hidden state sequence. Then, the bidirectional hidden state sequence is input into the multi-head attention mechanism layer, each attention head performs independent attention calculation on the bidirectional hidden state sequence, and finally the outputs of each attention head are concatenated to obtain the attention-enhanced feature sequence.

[0042] Step S226: input the attention-enhanced feature sequence into the feature fusion layer, integrate the output features of different attention heads through residual connection and layer normalization operation, and generate a fusion feature sequence with semantic consistency.

[0043] Residual connection can solve the problem of gradient vanishing in deep neural networks, while allowing the network to learn more complex features. Layer normalization is a normalization method that can normalize the features of each sample, so that the features have the same scale and distribution.

[0044] The role of the feature fusion layer is to integrate the features of different attention heads output by the multi-head attention mechanism layer to generate a fusion feature sequence with consistent semantics. Through residual connection, the input attention-enhanced feature sequence is directly added to the output of the fusion layer, so that the network can learn the residual between the input features and the fusion features; through layer normalization operation, the fused features are normalized to ensure the stability of the features.

[0045] In actual operation, the attention-enhanced feature sequence is input into the feature fusion layer. First, the attention-enhanced feature sequence is linearly transformed to obtain intermediate features; then, the intermediate features are connected with the input attention-enhanced feature sequence through residual connection, i.e. addition operation; finally, the result of the addition is normalized by layer normalization to obtain a fusion feature sequence with consistent semantics.

[0046] Step S227: input the fusion feature sequence into the conditional random field output layer, calculate the state transition probability matrix of the entity label set corresponding to each character position, and the entity label set includes entity start label, entity internal label and non-entity label.

[0047] The state transition probability matrix is a parameter in the conditional random field model, which represents the probability of transitioning from one entity label to another. The entity label set includes three types of labels: entity start label is used to represent the start position of the entity; entity internal label is used to represent the character position inside the entity; non-entity label is used to represent the character position that does not belong to any entity.

[0048] In actual operation, after inputting the fusion feature sequence into the conditional random field output layer, the conditional random field model will calculate the state transition probability matrix of the entity label set corresponding to each character position according to the feature information of the fusion feature sequence. Specifically, for each character position, the model will calculate the probability of transitioning from a certain label at the current position to each label at the next position. For example, for a character in a text, the model calculates the probability of transitioning from the non-entity label of the character to the entity start label, entity internal label or non-entity label of the next character, and the probability of transitioning from the entity start label to other labels, etc., to finally obtain the state transition probability matrix.

[0049] Step S228: based on the state transition probability matrix, use the Viterbi dynamic programming algorithm to search for the optimal path, determine the entity label category of each character position, and generate a label sequence containing entity boundary information.

[0050] The Viterbi dynamic programming algorithm is an algorithm for finding the optimal path in a Markov model, and in a conditional random field model, the most likely sequence of entity labels can be found according to the state transition probability matrix. The optimal path refers to the path that maximizes the probability of the entire sequence among all possible label sequences.

[0051] In actual operation, according to the state transition probability matrix, the Viterbi algorithm will start from the first character of the sequence and calculate the maximum probability path of each label at each character position in turn. Specifically, for each character position and each label, the algorithm will calculate the maximum probability of the transition from the previous character position to the label at the current character position, and record the path. Finally, starting from the last character of the sequence, the most likely label sequence is found by backtracking according to the recorded path.

[0052] Step S229: According to the label sequence, the character fragments corresponding to the continuous entity labels are intercepted from the text interaction record, and the entity collection containing the entity unique identifier, the text content, the type and the start and end position coordinates in the text interaction record is generated.

[0053] The character fragments corresponding to the continuous entity labels refer to the fragments composed of the characters corresponding to the continuous entity start labels and entity internal labels in the label sequence, and these fragments constitute a complete entity. The entity type annotation refers to the pre-defined entity type, such as business domain specific terminology, cross-departmental organization identifier, business process node, etc.

[0054] The entity unique identifier is a unique number assigned to each entity, which is used to distinguish different entities in subsequent processing; the text content is the specific text of the character fragment corresponding to the entity; the entity type is the type to which the entity belongs according to the entity type annotation; the start and end position coordinates are the coordinates of the start and end positions of the entity in the text interaction record, which are used to locate the position of the entity in the text.

[0055] In actual operation, the label sequence is traversed to find the character fragments corresponding to the continuous entity start labels and entity internal labels. For example, for the label sequence "non-entity label, entity start label, entity internal label, entity internal label, non-entity label", the corresponding text is "This is the business approval process", and the character fragment "business approval" is intercepted. Then, according to the pre-defined entity type annotation, it is determined that the type of the entity is business domain specific terminology. A unique identifier is assigned to the entity, such as "E001", and the start and end position coordinates in the text are recorded, such as the start position is 3 and the end position is 6, and finally an entity collection containing the entity unique identifier "E001", the text content "business approval", the type "business domain specific terminology" and the start and end position coordinates (3, 6) is generated.

[0056] Step S230: Based on the location coordinates of the entity set in the text interaction record and the context, calculate the spatial distance parameter and semantic association coefficient between entity pairs, and generate a co-occurrence feature matrix to characterize the potential association between entities.

[0057] Spatial distance parameter refers to the spatial distance between two entities in a text interaction record, reflecting the physical distance relationship between the entities. Semantic association coefficient refers to the degree of semantic association between two entities, reflecting the semantic similarity or relevance between the entities.

[0058] The co-occurrence feature matrix is ​​a matrix where rows and columns correspond to entities in the entity set, and matrix element values ​​represent the potential associations between entity pairs. By calculating the spatial distance parameter and semantic association coefficient between entity pairs and combining them, the co-occurrence feature matrix can be generated, providing a basis for subsequently constructing relationship edges between entities.

[0059] In one implementation, step S230 may include the following steps S231 to S237:

[0060] Step S231: Traverse the entity set in the text interaction record, extract the start and end coordinates of each entity in the text, and calculate the text coverage length parameter of the entity. The text coverage length parameter is the difference between the end coordinate and the start coordinate.

[0061] The start and end position coordinates represent the beginning and end positions of the entity in the text interaction record, respectively. The text coverage length parameter is the length occupied by the entity in the text, reflecting the size of the entity within the text.

[0062] In practice, each entity in the entity set is traversed, and its start and end coordinates in the text are extracted from the entity's information.

[0063] Step S232: Pair entities in the same text interaction record into pairs to generate a set of entity pairs. Each entity pair contains the identifiers of two different entities and their position coordinates in the text.

[0064] An entity pair set is a collection of pairs of entities from the same text interaction record. Each entity pair contains the identifiers of two different entities and their position coordinates in the text. By pairing entities, it is convenient to subsequently calculate the spatial distance parameter and semantic association coefficient between entity pairs.

[0065] In actual operation, two layers of loops are used to traverse the entity set in the same text interaction record, and each entity is paired with other entities. For example, for the entity set {E1, E2, E3}, the generated entity pair set is {(E1, E2), (E1, E3), (E2, E3)}, each entity pair contains the identifiers of the entities and their position coordinates in the text.

[0066] Step S233: Calculate the spatial distance parameter of the two entities in the entity pair, which is the weighted average of the absolute value of the difference between the start position coordinates of the two entities and the text coverage length parameter.

[0067] The calculation of the spatial distance parameter takes into account the difference between the start position coordinates of the two entities and their text coverage length. Through the weighted average, the distance relationship between the two entities in space can be more reasonably measured.

[0068] Step S234: Extract the context segment of the text interaction record where the entity pair is located, which contains a predetermined number of character sequences before and after the entity pair.

[0069] The context segment refers to a text segment containing a certain number of characters before and after the entity pair, which can provide semantic information around the entity pair and help calculate the semantic correlation coefficient between the entity pair. The predetermined number can be set according to the specific task and data situation.

[0070] Step S235: Calculate the semantic correlation coefficient of the two entities in the context segment through the pre-trained semantic similarity model, the semantic correlation coefficient ranges from 0 to 1, and the larger the value, the closer the semantic correlation.

[0071] The pre-trained semantic similarity model is a model trained on a large amount of text data, which can calculate the semantic similarity between two text segments. The semantic correlation coefficient is calculated based on the pre-trained semantic similarity model, and its value ranges from 0 to 1, 0 indicating no semantic correlation between the two entities, and 1 indicating complete semantic similarity between the two entities.

[0072] In actual operation, the two entities in the extracted context segment are input into the pre-trained semantic similarity model, and the model outputs a value between 0 and 1 as the semantic correlation coefficient of the two entities.

[0073] Step S236: Construct a co-occurrence feature matrix with entity identifiers as row and column indices, the matrix element value is the product of the reciprocal of the spatial distance parameter of the entity pair and the semantic correlation coefficient, and the diagonal element value is set to a predetermined constant.

[0074] The rows and columns of the co-occurrence feature matrix correspond to entity identifiers in the entity set respectively, and the matrix element values are used to represent the potential association between entity pairs. By multiplying the reciprocal of the spatial distance parameter and the semantic association coefficient, the spatial distance and semantic association of the entity pair can be considered comprehensively. The diagonal elements refer to the elements in the matrix where the row index and the column index are the same, which are set to a preset constant, usually 1, to avoid division by zero or uniform processing.

[0075] In actual operation, first, a matrix with entity identifiers as row indexes and column indexes is created. For each non-diagonal element in the matrix, the product of the reciprocal of the spatial distance parameter and the semantic association coefficient of the corresponding entity pair is calculated; for the diagonal element, it is set to a preset constant, such as 1.

[0076] Step S237: Perform row normalization processing on the co-occurrence feature matrix to make the sum of each row element equal to 1, and generate a standardized co-occurrence feature matrix for relationship edge construction.

[0077] Row normalization processing refers to normalizing each row of the matrix so that the sum of each row element is 1. Through row normalization processing, the scale difference between different rows of element values can be eliminated, making the matrix element values comparable and facilitating subsequent relationship edge construction.

[0078] In actual operation, for each row of the co-occurrence feature matrix, the sum of the elements in the row is calculated; then, each element in the row is divided by the sum of the elements in the row to obtain the normalized element value. Finally, a standardized co-occurrence feature matrix for relationship edge construction is obtained.

[0079] Step S240: Construct a set of directed relationship edges between entities according to the co-occurrence feature matrix, each relationship edge containing a source entity identifier, a target entity identifier, and a dynamic weight value, the dynamic weight value being determined by the product of the reciprocal of the spatial distance parameter and the semantic association coefficient.

[0080] The set of directed relationship edges is a set composed of directed edges between entities, and each directed edge represents a relationship from one entity to another. The source entity identifier refers to the identifier of the starting entity of the relationship edge, and the target entity identifier refers to the identifier of the ending entity of the relationship edge. The dynamic weight value is used to represent the strength of the relationship edge and is determined by the product of the reciprocal of the spatial distance parameter and the semantic association coefficient, reflecting the degree of association between the two entities.

[0081] In actual operation, a directed relationship edge is constructed for each non-zero element in the standardized co-occurrence feature matrix. For example, for the element m in the th row and th column of the matrix (≠ ), a directed relationship edge is constructed from entity to entity, the source entity identifier is, the target entity identifier is, and the dynamic weight value is. Finally, a set of directed relationship edges between entities is obtained.

[0082] Step S250: Assign a set of attribute tags containing time attributes, domain attributes, and interaction role attributes to each entity node, the time attribute corresponds to the timestamp of the first appearance of the entity in the text block sequence, the domain attribute is determined based on the business scenario keywords of the text interaction record where the entity is located;

[0083] The set of attribute tags is an additional information set added to each entity node, containing time attributes, domain attributes, and interaction role attributes. The time attribute records the timestamp of the first appearance of the entity in the text block sequence, which can reflect the order of the entity in time; the domain attribute is determined according to the business scenario keywords of the text interaction record where the entity is located, and is used to represent the business domain to which the entity belongs; the interaction role attribute is used to describe the role played by the entity in the text interaction record, such as initiator, participant, etc.

[0084] In actual operation, for each entity node, first find the text interaction record where the entity first appears from the text block sequence, record its corresponding timestamp as the time attribute; then, extract the business scenario keywords of the text interaction record, and determine the domain attribute of the entity according to the keywords; finally, analyze the role and function of the entity in the text interaction record to determine its interaction role attribute.

[0085] Step S260: Construct a multi-dimensional entity relationship network with the set of entities as the node layer, the set of relationship edges as the relationship layer, and the set of attribute tags as the attribute layer, the node layer and the relationship layer are associated through edge weight values, and the relationship layer and the attribute layer are associated through attribute mapping rules.

[0086] The node layer is a layer composed of the set of entities, each entity being a node; the relationship layer is a layer composed of the set of relationship edges, the relationship edge representing the association between nodes; the attribute layer is a layer composed of the set of attribute tags, the attribute tag providing additional information for the node.

[0087] The edge weight value is used to represent the strength of the relationship edge, and the node layer and the relationship layer are associated through the edge weight value, so that the degree of association between nodes can be measured according to the edge weight value. The attribute mapping rule is a rule for associating attribute tags with relationship edges or nodes, and the relationship layer and the attribute layer are associated through the attribute mapping rule, so that the characteristics of nodes and relationships can be further analyzed according to the attribute tags.

[0088] In actual operation, first, each entity in the entity set is constructed as a node to build a node layer; then, each relationship edge in the relationship edge set is connected to the corresponding node to build a relationship layer; finally, according to the attribute mapping rule, the attribute tags in the attribute tag set are assigned to the corresponding nodes or relationship edges to build an attribute layer.

[0089] Step S300: constructing a semantic topology graph based on the multi-dimensional entity relationship network to obtain a semantic topology graph containing entity correlation strength and semantic evolution path, the entity correlation strength is used to quantify the semantic dependence degree between entity nodes, and the semantic evolution path is used to describe the change trajectory of entity relationship with time stamp.

[0090] The semantic topology graph is a graph structure used to represent the semantic relationship between entities, which is constructed based on the multi-dimensional entity relationship network and contains two information of entity correlation strength and semantic evolution path. The entity correlation strength is a numerical value used to measure the semantic dependence degree between two entity nodes, and the larger the numerical value is, the closer the semantic dependence is; the semantic evolution path is the change trajectory of entity relationship with time, which can reflect the dynamic change of the relationship between entities at different time points.

[0091] By constructing the semantic topology graph, the semantic relationship between entities and its change over time can be more intuitively displayed, which provides a powerful tool for subsequent information extraction and analysis.

[0092] As an implementation manner, step S300 can include steps S310-S360 as follows:

[0093] Step S310: calculating the centrality index of the node layer of the multi-dimensional entity relationship network to generate a comprehensive centrality index set containing degree centrality, betweenness centrality and closeness centrality, the degree centrality considers the dynamic weight value of the relationship edge, and the betweenness centrality considers the number of shortest paths between entity pairs;

[0094] The centrality index is an index used to measure the importance of a node in the network. The degree centrality refers to the number of relationship edges of a node, in this method, the degree centrality considers the dynamic weight value of the relationship edge, that is, not only the number of relationship edges, but also the strength of each relationship edge; the betweenness centrality refers to the number of times a node appears in the shortest path between all entity pairs, which reflects the mediating role of the node in the network; the closeness centrality refers to the reciprocal of the average shortest path length from a node to all other nodes, which reflects the proximity of the node in the network. The comprehensive centrality index set is an index set obtained by weighting and summing the degree centrality, betweenness centrality and closeness centrality according to certain weight proportions, which can more comprehensively measure the importance of the node in the multi-dimensional entity relationship network.

[0095] As an implementation, step S310 can include steps S311-S318 as follows:

[0096] Step S311: Traverse the set of relationship edges of the multi-dimensional entity relationship network, and count the number of in-degree relationship edges and the number of out-degree relationship edges of each entity node, wherein the number of in-degree relationship edges is the total number of relationship edges with the entity as the target, and the number of out-degree relationship edges is the total number of relationship edges with the entity as the source.

[0097] The in-degree relationship edge refers to a relationship edge with a certain entity node as the target, and the out-degree relationship edge refers to a relationship edge with a certain entity node as the source. By counting the number of in-degree relationship edges and the number of out-degree relationship edges of each entity node, the connection of the node in the network can be understood.

[0098] In actual operation, the set of relationship edges of the multi-dimensional entity relationship network is traversed. For each relationship edge, the target entity and the source entity are checked. If the target entity is node Ei, the number of in-degree relationship edges of node Ei is incremented by 1; if the source entity is node Ei, the number of out-degree relationship edges of node Ei is incremented by 1. For example, for the set of relationship edges {(E1, E2), (E2, E3), (E3, E1)}, the number of in-degree relationship edges of node E1 is 1 (the relationship edge from E3), and the number of out-degree relationship edges of node E1 is 1 (the relationship edge pointing to E2).

[0099] Step S312: Calculate the weighted degree centrality index of each entity node, and the weighted degree centrality index is the average of the sum of the dynamic weight values of the in-degree relationship edges and the sum of the dynamic weight values of the out-degree relationship edges.

[0100] The weighted degree centrality index comprehensively considers the dynamic weight values of the in-degree and out-degree relationship edges of the node, and can more accurately measure the importance of the node in the network.

[0101] In actual operation, for each entity node Ei, the sum of the dynamic weight values of the in-degree relationship edges Wi in and the sum of the dynamic weight values of the out-degree relationship edges Wi out are calculated, and then the weighted degree centrality index Ci=(Wi in +Wi out ) / 2 is calculated.

[0102] Step S313: Normalize the weighted degree centrality index to generate a degree centrality component, and the normalization is realized by dividing the weighted degree centrality index by the maximum weighted degree centrality value in the network.

[0103] The normalization processing is to scale the weighted degree centrality index, so that its value range is between 0 and 1, which is convenient for subsequent comprehensive calculation. By dividing the weighted degree centrality index of each node by the maximum weighted degree centrality value in the network, the scale difference of the weighted degree centrality index between different nodes can be eliminated. In actual operation, first find the maximum value C max in the weighted degree centrality index of all nodes in the network; then, for the weighted degree centrality index C max of each node, calculate the degree centrality component D

[0104] Step S314: The Floyd-Warshall algorithm is used to calculate the shortest path set between all entity pairs in the multi-dimensional entity relationship network, and the length of each shortest path is the sum of the dynamic weight values of the relationship edges on the path.

[0105] The Floyd-Warshall algorithm is an algorithm for calculating the shortest path between all node pairs in a graph, which can handle weighted directed graphs. In this method, the algorithm is used to calculate the shortest path between all entity pairs in the multi-dimensional entity relationship network, and the length of the shortest path is the sum of the dynamic weight values of the relationship edges on the path.

[0106] In actual operation, first construct a distance matrix, and the elements of the matrix represent the initial distance between two nodes. If there is a relationship edge between two nodes, the initial distance is the dynamic weight value of the relationship edge, otherwise it is infinite. Then, use the Floyd-Warshall algorithm to update the distance matrix iteratively until the shortest path between all node pairs is found.

[0107] Step S315: Count the intermediate times of each entity node in the shortest path set to generate the intermediate centrality original value, and the intermediate times are the number of shortest paths passing through the entity node.

[0108] The intermediate times refer to the number of shortest paths passing through a certain entity node among all shortest paths between entity pairs. The intermediate centrality original value is calculated based on the intermediate times, which reflects the intermediary role of the node in the network. In actual operation, traverse the shortest path set between all entity pairs, and for each shortest path, check whether it passes through a certain entity node. If it does, the intermediate times of the node are increased by 1.

[0109] Step S316: Standardize the intermediate centrality original value to generate the intermediate centrality component, which is achieved by dividing the intermediate centrality original value by the total number of possible shortest paths in the network.

[0110] The standardized processing is to scale the original value of the betweenness centrality, so that the value range is between 0 and 1, which is convenient for subsequent comprehensive calculation. By dividing the original value of the betweenness centrality by the total number of shortest paths that may exist in the network, the scale difference of the original value of the betweenness centrality between different nodes can be eliminated.

[0111] In actual operation, first, the total number N of shortest paths that may exist in the network is calculated; then, for the original value Bi of the betweenness centrality of each node, the betweenness centrality component Mi=Bi / N is calculated.

[0112] Step S317: Calculate the average value of the shortest path length from each entity node to all other entity nodes to generate an average distance parameter, and the closeness centrality component is the reciprocal of the average distance parameter.

[0113] The average distance parameter refers to the average value of the shortest path length from an entity node to all other entity nodes, which reflects the average distance of the node in the network. The closeness centrality component is the reciprocal of the average distance parameter, which reflects the proximity of the node in the network, and the larger the value, the closer the node is to other nodes in the network.

[0114] In actual operation, for each entity node Ei, the sum Si of the shortest path lengths from it to all other entity nodes is calculated; then, the average distance parameter Ai=Si / (n-1) is calculated, where n is the total number of nodes in the network; finally, the closeness centrality component Ci=1 / Ai is calculated.

[0115] Step S318: Weighted sum the degree centrality component, the betweenness centrality component and the closeness centrality component according to the preset weight proportion to generate the comprehensive centrality index of each entity node, and the preset weight proportion is determined by the analytic hierarchy process.

[0116] The analytic hierarchy process (Analytic Hierarchy Process, AHP) is a method for determining the weight proportion, which determines the relative importance of multiple factors by pairwise comparison, so as to obtain the weight proportion.

[0117] The comprehensive centrality index is an index obtained by weighted sum of the degree centrality component, the betweenness centrality component and the closeness centrality component according to the preset weight proportion, which can more comprehensively measure the importance of nodes in the multi-dimensional entity relationship network.

[0118] In actual operation, first, the analytic hierarchy process is used to determine the weight proportions of the degree centrality component, the betweenness centrality component and the closeness centrality component, which are respectively denoted as w1, w2 and w3 (w1+w2+w3=1); then, for each entity node Ei, the comprehensive centrality index Zi is calculated as Zi=w1*Di+w2*Mi+w3*Ci, where Di is the degree centrality component, Mi is the betweenness centrality component, and Ci is the closeness centrality component.

[0119] Step S320: The entity nodes are sorted in descending order according to the set of comprehensive centrality indexes, and entity nodes with a preset proportion in the sorted order are selected as a core node set of the semantic topology graph, and the preset proportion is dynamically adjusted based on the total number of nodes of the multi-dimensional entity relationship network.

[0120] The descending order sorting refers to arranging the entity nodes in an order from large to small according to the comprehensive centrality indexes. The preset proportion refers to a proportion of selecting core nodes that is preset in advance and can be dynamically adjusted according to the total number of nodes of the multi-dimensional entity relationship network to ensure that the selected core node set can reflect the main structure of the network and will not be too large.

[0121] The core node set is a part of nodes selected from the entity nodes, and these nodes have high importance in the network and will constitute the core structure of the semantic topology graph.

[0122] Step S330: The time sequence tracking of the relationship edge changes of the core node set in the text block sequence is performed, the creation timestamp, termination timestamp and weight value change sequence of the relationship edge are extracted, and the dynamic evolution feature of the entity relationship is generated.

[0123] The time sequence tracking refers to tracking and recording the changes of the relationship edges of the core node set in the text block sequence in chronological order. The creation timestamp refers to the time when the relationship edge first appears; the termination timestamp refers to the time when the relationship edge disappears; and the weight value change sequence refers to the changes of the weight values of the relationship edge over time.

[0124] The dynamic evolution feature is a feature generated according to the creation timestamp, termination timestamp and weight value change sequence of the relationship edge, and can reflect the dynamic changes of the entity relationship over time.

[0125] In actual operation, the text block sequence is traversed, and for the relationship edges between the core node set, the creation timestamp, termination timestamp and weight value change are recorded.

[0126] Step S340: A relationship evolution prediction model is constructed based on the dynamic evolution feature, the relationship edge state of the current text block is input, the relationship edges that can be newly added or disappeared in the next text block are predicted, and a set of potential relationship edges is generated.

[0127] The relationship evolution prediction model is a model based on historical data for prediction, which can predict the relationship edges that may appear or disappear in the next text block according to the dynamic evolution characteristics of the relationship edges between the core node set.

[0128] The relationship edge state of the current text block refers to the existence and weight value of the relationship edges between the core node set in the current text block. The potential relationship edge set is a set composed of the relationship edges that may be added or disappeared in the next text block.

[0129] In actual operation, a relationship evolution prediction model is constructed using machine learning or deep learning algorithms such as recurrent neural network (RNN) or long short-term memory network (LSTM). The dynamic evolution characteristics are used as training data to train the model. After training, the relationship edge state of the current text block is input into the model, which outputs the relationship edges that may be added or disappeared in the next text block, generating the potential relationship edge set.

[0130] Step S350: Calculate the entity association strength between the core node set, which is determined by the weighted sum of the dynamic weight value of the relationship edge, the normalized element value in the co-occurrence feature matrix, the time decay factor and the domain similarity. The time decay factor decreases as the interval between the last co-occurrence timestamp of the entity pair and the current timestamp increases.

[0131] The entity association strength is an index for measuring the semantic dependence degree between the core node set, which considers the dynamic weight value of the relationship edge, the normalized element value in the co-occurrence feature matrix, the time decay factor and the domain similarity. The time decay factor is a factor that changes with time, reflecting the trend that the association between entity pairs weakens over time; the domain similarity refers to the similarity of the domains to which the two entities belong.

[0132] By weighting and combining these factors, the entity association strength between the core node set can be calculated more accurately.

[0133] As an implementation, step S350, calculating the entity association strength between the core node set, can specifically include the following steps S351-S356:

[0134] Step S351: Extract the dynamic weight value of the relationship edge of any two entity nodes in the core node set. If there is no direct relationship edge between the two entities, the dynamic weight value is set to a preset minimum value.

[0135] The dynamic weight value of the relationship edge refers to the weight value of the relationship edge between two entity nodes in the core node set, reflecting the direct association strength between the two entities. The preset minimum value is to assign a smaller value when there is no direct relationship edge between the two entities, avoiding the influence of zero value on subsequent calculation.

[0136] In actual operation, for any two entity nodes Ei and Ej in the core node set, it is checked whether there is a direct relationship edge between them. If there is, the dynamic weight value of the relationship edge is extracted; if not, the dynamic weight value is set to a preset minimum value, such as 0.01.

[0137] Step S352: Extract the normalized element value corresponding to the two entities from the co-occurrence feature matrix as the co-occurrence association component.

[0138] The normalized element value in the co-occurrence feature matrix reflects the co-occurrence and potential association degree of the two entities in the text interaction record. The co-occurrence association component is the normalized element value corresponding to the two entities extracted from the co-occurrence feature matrix, which is a component of calculating the entity association strength.

[0139] In actual operation, for entity nodes Ei and Ej in the core node set, find the normalized element value in the i-th row and j-th column from the co-occurrence feature matrix, and take it as the co-occurrence association component.

[0140] Step S353: Calculate the field similarity parameter according to the field attribute in the attribute label set of the two entities, and the field similarity parameter is determined by the cosine similarity of the field attribute keywords.

[0141] The field attribute is an attribute in the entity attribute label set, which is used to represent the business field to which the entity belongs. The field similarity parameter is obtained by calculating the cosine similarity of the field attribute keywords of the two entities, which reflects the similarity of the fields to which the two entities belong.

[0142] Cosine similarity is a commonly used method for calculating vector similarity, which measures the similarity of two vectors by calculating the cosine of the included angle.

[0143] In actual operation, for entity nodes Ei and Ej in the core node set, extract their field attribute keywords and convert the keywords into vector representation; then, calculate the cosine similarity of the two vectors to obtain the field similarity parameter.

[0144] Step S354: Obtain the timestamp of the last co-occurrence of the two entities in the text block sequence, and calculate the time decay factor, which is the exponential function of the difference between the current timestamp and the last co-occurrence timestamp.

[0145] The time decay factor is a factor that changes with time, reflecting the trend that the association between entity pairs weakens over time. The current timestamp refers to the current processing time; the last co-occurrence timestamp refers to the time when the two entities last appeared together in the text block sequence. The exponential function can make the time decay factor decrease rapidly as the time interval increases.

[0146] In actual operation, the timestamp of the last co-occurrence of the two entities is found from the sequence of text blocks, the difference between the current timestamp and the last co-occurrence timestamp is calculated, and then the difference is substituted into the exponential function to calculate the time decay factor.

[0147] Step S355: The dynamic weight value is taken as the basic component, the co-occurrence correlation component is taken as the enhancement component, the domain similarity parameter is taken as the adjustment component, and the time decay factor is taken as the correction component. The entity correlation strength is obtained by weighted summation, and the weight coefficients of the weighted summation are determined by the grid search method.

[0148] The basic component refers to the dynamic weight value of the relationship edge, which is the basis of the entity correlation strength. The enhancement component refers to the co-occurrence correlation component, which can enhance the entity correlation strength. The adjustment component refers to the domain similarity parameter, which can adjust the correlation strength according to the similarity of the domain to which the entity belongs. The correction component refers to the time decay factor, which can correct the correlation strength according to the time interval. The grid search method is a method for optimizing parameters, which finds the optimal parameter value by traversing all possible value combinations of the parameters. In this method, the grid search method is used to optimize the weight coefficients of the weighted summation to obtain more accurate entity correlation strength.

[0149] In actual operation, first, the value range of the weight coefficients of the basic component, the enhancement component, the adjustment component, and the correction component is determined. Then, the grid search method is used to traverse all possible value combinations, and the entity correlation strength under each combination is calculated. Finally, the weight combination that makes the entity correlation strength most consistent with the actual situation is selected.

[0150] Step S356: The calculated entity correlation strength is subjected to upper and lower limit truncation processing to limit the value range of the entity correlation strength to a preset numerical interval, and a standardized entity correlation strength for the edge weight of the semantic topology graph is generated.

[0151] The upper and lower limit truncation processing refers to limiting the calculated entity correlation strength within a preset numerical interval to avoid excessively large or small values. The preset numerical interval is a pre-set value range of the entity correlation strength, such as [0, 1].

[0152] The standardized entity correlation strength is the entity correlation strength after the upper and lower limit truncation processing, which will be used as the edge weight of the semantic topology graph to represent the correlation strength between the core node sets.

[0153] In actual operation, for the calculated entity correlation strength, if it is less than the lower limit of the preset numerical interval, it is set to the lower limit value; if it is greater than the upper limit of the preset numerical interval, it is set to the upper limit value.

[0154] Step S360: The core node set is taken as a vertex, the entity correlation strength is taken as an edge weight, the dynamic evolution feature is taken as a time attribute, and the potential relationship edge set is taken as a prediction attribute, so that a semantic topology graph containing static topology structure and dynamic evolution information is constructed, the static topology structure is composed of the core node set and the current relationship edge, and the dynamic evolution information is composed of the weight value change sequence and the potential relationship edge set.

[0155] The vertex is a node in the semantic topology graph and is composed of the core node set, the edge weight is a weight of an edge in the semantic topology graph and is represented by the standardized entity correlation strength, the time attribute is an attribute of an edge in the semantic topology graph and is provided by the dynamic evolution feature, and the prediction attribute is another attribute of the edge in the semantic topology graph and is provided by the potential relationship edge set, and is used for predicting a relationship edge that may appear or disappear in the future.

[0156] The static topology structure is a basic structure of the semantic topology graph and is composed of the core node set and the current relationship edge, and reflects a current state of the network, and the dynamic evolution information is a dynamic part of the semantic topology graph and is composed of the weight value change sequence and the potential relationship edge set, and reflects a dynamic change of the network.

[0157] In actual operation, first, the core node set is taken as a vertex to construct a node structure of the semantic topology graph, then, the standardized entity correlation strength is used to set an edge weight for each edge, then, the dynamic evolution feature is taken as a time attribute and added to an attribute of the edge, and finally, the potential relationship edge set is taken as a prediction attribute and added to the graph, so that the semantic topology graph containing the static topology structure and the dynamic evolution information is constructed.

[0158] Step S400: A preset zero-knowledge proof protocol is called to perform access control vector generation on the semantic topology graph, so that an access control vector containing a topology weight threshold value and node access permission is obtained, the topology weight threshold value is determined based on statistical features of the entity correlation strength, and the node access permission is used to limit an operation range of different sharing subjects on the entity node.

[0159] Zero-knowledge proof protocol is a cryptographic method that proves the truth of a statement without revealing the specific content of the information. In this method, the pre-configured zero-knowledge proof protocol is pre-configured and optimized for processing semantic topology graphs to generate access control vectors. Topology weight threshold is a parameter for controlling access, which is determined based on the statistical characteristics of entity association strength. Entity association strength reflects the closeness of the association between nodes in the semantic topology graph. By statistically analyzing these strength values, such as calculating the mean, median, and standard deviation, the topology weight threshold that can effectively divide the access permission level can be obtained. Node access permission explicitly defines the operation range of different sharing subjects, such as different business departments or users, on the entity nodes in the semantic topology graph, such as whether they can query, modify, or delete node information.

[0160] In actual operation, when calling the pre-configured zero-knowledge proof protocol, the relevant information of the semantic topology graph needs to be encrypted first to ensure data security and privacy. Then, the protocol calculates the topology weight threshold based on the statistical characteristics of entity association strength. For example, by sorting all entity association strength values in the semantic topology graph, a suitable quantile is selected as the topology weight threshold. For the determination of node access permission, the identity authentication information and business requirement description of the sharing subject need to be combined. Assuming that a sharing subject is a housing authority in a district, its business requirement is mainly to query node information related to building approval, then according to its identity and requirement, it can be allocated corresponding node access permission, such as only allowing to query information of nodes related to building approval process, but not allowing to modify or delete.

[0161] As an implementation, step S400 can include steps S410-S460 as follows:

[0162] Step S410: Topology structure preprocessing is performed on the semantic topology graph, and a topology feature vector containing a core node set, entity association strength, and semantic evolution path is extracted, and the dimension of the topology feature vector matches the size of the core node set.

[0163] Topology structure preprocessing is a process of preliminary processing of the semantic topology graph, aiming to extract useful information for subsequent access control vector generation. The core node set is a set of nodes with important status in the semantic topology graph, which plays a key role in the structure and semantic expression of the graph. Entity association strength reflects the closeness of the association between nodes, and semantic evolution path records the change of node relationship over time. The topology feature vector is a vector representation of these key information, and its dimension matches the size of the core node set, so that the relevant information of each core node can be accurately reflected in the vector.

[0164] In actual operation, when topological structure preprocessing is performed on the semantic topology graph, firstly, the structure of the graph needs to be analyzed, and the core node set is identified. According to the comprehensive centrality index calculated in the previous step, the top-ranked nodes can be selected as the core nodes. Then, the entity correlation strength and semantic evolution path information between each core node are extracted. For example, for a semantic topology graph containing multiple business approval process nodes, through analyzing the relationship between the nodes and historical data, the core nodes are determined as “application submission”, “initial review”, “re-review”, etc. Next, the entity correlation strength between these core nodes is extracted, such as the correlation strength between “application submission” and “initial review” is 0.8. At the same time, the semantic evolution path of the relationship between these nodes over time is recorded, such as the correlation strength between “initial review” and “re-review” gradually increases in a certain time period. Finally, these information is combined into a topological feature vector according to certain rules, and each dimension of the vector corresponds to the related information of a core node.

[0165] Step S420: calculating statistical feature values based on the entity correlation strength in the topological feature vector, generating a topological weight threshold for dividing access permission levels, and the statistical feature values include the mean, median and standard deviation of the entity correlation strength.

[0166] The statistical feature values are numerical values obtained by statistical analysis of the entity correlation strength data in the topological feature vector. The mean reflects the average level of the entity correlation strength, the median represents the middle position value of the data, and the standard deviation reflects the dispersion degree of the data. By calculating these statistical feature values, the distribution of the entity correlation strength can be comprehensively understood, so as to generate reasonable topological weight thresholds to divide access permission levels.

[0167] In actual operation, firstly, all the entity correlation strength values are extracted from the topological feature vector. Then, the mean, median and standard deviation are calculated using the corresponding statistical methods. According to these statistical feature values, different strategies can be used to generate topological weight thresholds. A common method is to combine the mean and the standard deviation, for example, the mean plus a certain multiple of the standard deviation is used as one topological weight threshold, and the mean minus a certain multiple of the standard deviation is used as another threshold, so that different access permission levels can be divided according to different weight thresholds.

[0168] As an implementation manner, step S420 can include steps S421-S425 as follows:

[0169] Step S421: performing outlier detection on the entity correlation strength in the topological feature vector, identifying and removing the abnormal correlation strength values that exceed the upper and lower quartile range by a preset multiple using the quartile method, and generating a purified entity correlation strength sequence.

[0170] An outlier is an extreme value in the data that deviates from most of the data and can have a significant impact on the results of statistical analysis. The quartile method is an outlier detection method that arranges the data in ascending order and then divides the data into four parts to calculate the lower quartile (Q1), median (Q2), and upper quartile (Q3). The interquartile range (IQR) is equal to Q3-Q1. The preset multiple is usually an empirical value, such as 1.5 or 3, and data points exceeding the preset multiple of the interquartile range are considered outliers. By removing these outliers, a purified entity correlation strength sequence can be obtained, making subsequent statistical analysis more accurate.

[0171] Step S422: Calculate the statistical characteristic values of the purified entity correlation strength sequence, including the arithmetic mean, geometric mean, median, first quartile, and third quartile.

[0172] The arithmetic mean is the sum of all data values divided by the number of data, reflecting the average level of the data. The geometric mean is the nth root of the continuous product of n data values, suitable for calculating the average of data with proportional relationships. The median is the value in the middle position after arranging the data in ascending order. The first quartile (Q1) and the third quartile (Q3) divide the data into four equal parts, with Q1 below 25% and Q3 above 25%.

[0173] In actual operation, for the purified entity correlation strength sequence, when calculating the arithmetic mean, add all the values in the sequence and then divide by the length of the sequence. For example, the purified sequence is [0.2, 0.3, 0.5, 0.6, 0.8], and the arithmetic mean is (0.2+0.3+0.5+0.6+0.8) / 5=0.48. When calculating the geometric mean, multiply all the values first and then take the 5th power (since the sequence length is 5). The median can be found by sorting the values and finding the value in the middle position. If the sequence length is odd, the middle value is the median; if it is even, the median is the average of the two middle values. For the first quartile and the third quartile, a similar method as in the previous outlier detection can be used to calculate them after sorting the data.

[0174] Step S423: Construct a multi-threshold division model based on the statistical characteristic values to divide the entity correlation strength into multiple intervals, each interval corresponding to a permission level reference value.

[0175] The multi-threshold division model determines multiple thresholds based on the statistical characteristic values calculated earlier, such as the arithmetic mean, median, first quartile, and third quartile, to divide the entity correlation strength into different intervals. Each interval corresponds to a permission level reference value, which is used to determine the access permission level of different sharing subjects in the subsequent process.

[0176] In actual operation, the threshold value is determined according to the statistical characteristic value. For example, the first quartile, the median and the third quartile can be taken as the threshold value, and the entity correlation strength is divided into four intervals. Assuming that the first quartile is 0.3, the median is 0.5, and the third quartile is 0.7, then the four intervals are [0, 0.3), [0.3, 0.5), [0.5, 0.7) and [0.7, 1]. Assign a permission level reference value to each interval, such as the interval [0, 0.3) corresponds to a permission level reference value of 1, [0.3, 0.5) corresponds to 2, [0.5, 0.7) corresponds to 3, and [0.7, 1] corresponds to 4. In this way, through the interval where the entity correlation strength is located, the corresponding permission level can be preliminarily determined.

[0177] Step S424: Determine the optimal threshold combination through ROC curve analysis, and the optimal threshold combination makes the misjudgment rate corresponding to different permission levels lower than the preset threshold.

[0178] The ROC curve (Receiver Operating Characteristic Curve) is a tool for evaluating the performance of a classification model, which shows the classification effect of the model by plotting the relationship curve between the true positive rate and the false positive rate. In this method, the ROC curve analysis can find the optimal threshold combination that makes the misjudgment rate corresponding to different permission levels lower than the preset threshold. The misjudgment rate includes the cases of misjudging low permission levels as high permission levels and misjudging high permission levels as low permission levels. The preset threshold is a pre-set acceptable upper limit of misjudgment rate.

[0179] In actual operation, first, the entity correlation strength needs to be classified according to different threshold combinations, and divided into different permission levels. Then, the true positive rate and the false positive rate under each threshold combination are calculated, and the ROC curve is plotted. For example, for different threshold combinations, the proportion of correctly classifying high permission levels (true positive rate) and the proportion of incorrectly classifying low permission levels as high permission levels (false positive rate) are calculated respectively. By comparing the positions of different threshold combinations on the ROC curve, the optimal threshold combination that makes the misjudgment rate lower than the preset threshold is found. Special algorithms can be used to determine the optimal threshold combination, such as the method of maximizing the Youden index, which is equal to the true positive rate minus the false positive rate.

[0180] Step S425: Take each interval boundary value in the optimal threshold combination as a topological weight threshold, generate a threshold set containing multiple hierarchical thresholds, and the number of hierarchical thresholds is consistent with the number of access permission levels.

[0181] After the optimal threshold combination is determined, the boundary values of each interval are taken as the topological weight thresholds. These thresholds constitute a threshold set containing multiple levels, and the number of level thresholds is the same as the number of access permission levels, so that different access permission levels can be accurately divided according to the entity correlation strength.

[0182] In actual operation, assuming that the optimal threshold combination obtained through ROC curve analysis divides the entity correlation strength into four intervals, the interval boundary values are 0.3, 0.5, and 0.7, then the three boundary values are taken as the topological weight thresholds, and the generated threshold set is {0.3, 0.5, 0.7}, and the corresponding access permission levels can be divided into four levels, such as low, medium-low, medium-high, and high. According to the comparison of the entity correlation strength with these thresholds, different entity nodes can be accurately assigned to the corresponding access permission levels.

[0183] Step S430: According to the identity authentication information of the sharing subject and the business requirement description, the access permission levels of the core node set are divided, and each access permission level corresponds to a different operation permission set, and the operation permission set includes query permission, modification permission, and deletion permission.

[0184] The identity authentication information of the sharing subject includes the basic identity characteristics of the subject, such as the business department to which the subject belongs and the position, etc. These information can reflect the position and role of the subject in the business system. The business requirement description clearly describes the subject's demand for the core node information in the semantic topology graph in actual business, for example, whether it needs to query the information of a specific business process, whether it needs to modify the information of some nodes, etc. By combining the identity authentication information and the business requirement description, the access permission levels of the core node set can be reasonably divided. Each access permission level corresponds to a different operation permission set, in which the query permission allows the subject to view the node information, the modification permission allows the subject to change the node information, and the deletion permission allows the subject to delete the node information.

[0185] In actual operation, first, the identity authentication information of the sharing subject is parsed to extract key information such as department identifier and role identifier. Then, the key words of the business requirement description are extracted to determine the specific business requirement. Assuming that the business requirement description is "query and audit the node information related to the financial budget approval process", then the key words "query", "audit", and "financial budget approval process" can be extracted. According to these information, combined with the attributes of the core node set and the topological weight thresholds, the access permission levels of the core node set for the sharing subject are divided. For example, for the core nodes related to the financial budget approval process, according to its business requirement and identity, it is assigned to a medium-high access permission level, and the corresponding operation permission set is that it can query and modify the related node information, but cannot delete.

[0186] As an implementation, step S430 can include steps S431-S435 as follows:

[0187] Step S431: Extract the role identifier, department identifier and security level identifier in the identity authentication information of the sharing subject, and generate a subject feature vector.

[0188] The role identifier is used to clearly define the role played by the sharing subject in the business system, such as an approval personnel, a query personnel, etc.; the department identifier indicates the business department to which the sharing subject belongs; and the security level identifier reflects the security permission level possessed by the sharing subject. The subject feature vector is a vector representation of integrating these key information together, which can be used as a basis for subsequent division of access permission levels.

[0189] Step S432: Perform keyword extraction on the business requirement description of the sharing subject, and generate a requirement keyword set containing business scenario words, operation type words and data type words.

[0190] The business scenario word clearly defines the specific scenario involved in the business requirement of the sharing subject, such as "financial approval"; the operation type word indicates the operation that the subject needs to perform, such as "query", "modify", "delete", etc.; and the data type word indicates the data type involved in the subject requirement, such as "budget data", "student information", etc. By extracting these keywords, the business requirement of the sharing subject can be accurately grasped. In actual operation, natural language processing technology is used to process the business requirement description.

[0191] Step S433: Query the pre-set role permission mapping table based on the subject feature vector to obtain the basic permission level set corresponding to the role.

[0192] The pre-set role permission mapping table is a table that is established in advance, recording the basic permission level set corresponding to different roles. The basic permission level set contains the access permission level range that the role has in general cases. Through the role identifier in the subject feature vector, the corresponding basic permission level set can be found in the mapping table.

[0193] In actual operation, the role identifier in the subject feature vector is extracted and queried in the pre-set role permission mapping table. Assuming that the role identifier is "approval officer", the basic permission level set corresponding to this role is found in the mapping table to be {medium, high}, which indicates that the role has medium level and high level access permissions in general cases.

[0194] Step S434: Perform semantic matching between the requirement keyword set and the attribute tag set of the core node set, calculate the matching degree score of the requirement and the node, and the matching degree score is determined by weighted sum of the business scenario word matching degree, the operation type word matching degree and the data type word matching degree.

[0195] Semantic matching refers to comparing the semantic similarity between the set of requirement keywords and the set of attribute labels of the core nodes. The business scenario word matching degree measures the similarity between the business scenario in the requirement and the business scenario of the node; the operation type word matching degree reflects the matching of the operation type in the requirement and the allowed operation type of the node; and the data type word matching degree reflects the consistency between the data type in the requirement and the data type of the node. The three matching degrees are integrated by weighted summation to obtain the matching degree score of the requirement and the node.

[0196] In actual operation, for each core node, the business scenario word matching degree, the operation type word matching degree and the data type word matching degree are calculated respectively. A semantic similarity algorithm such as cosine similarity can be used to calculate the matching degree. For example, for the business scenario word, the business scenario words in the requirement and the business scenario words in the attribute labels of the node are converted into vector representation, and then the cosine similarity is calculated as the business scenario word matching degree.

[0197] Step S435: Based on the set of basic permission levels and the matching degree score, the access permission level of each core node to the sharing subject is determined by fuzzy comprehensive evaluation method, and the access permission level contains multiple levels, and different levels correspond to different operation permission ranges.

[0198] Fuzzy comprehensive evaluation method is a method of considering multiple factors for evaluation, which can handle information with fuzziness. In this method, the set of basic permission levels and the matching degree score are combined to determine the access permission level of each core node to the sharing subject by using fuzzy comprehensive evaluation method. The access permission level contains multiple levels, such as low, low, medium, high, and each level corresponds to a different operation permission range, such as low level may only allow query, high level may allow query, modification and deletion.

[0199] In actual operation, first, the membership functions of the set of basic permission levels and the matching degree score are determined, which are used to describe their membership relationship with different access permission levels. Then, according to the algorithm of fuzzy comprehensive evaluation, the set of basic permission levels and the matching degree score are calculated comprehensively. For example, for a core node, the set of basic permission levels shows that the role may have medium level and high level access permissions, and the matching degree score is high. After calculation by fuzzy comprehensive evaluation method, it is determined that the access permission level of the sharing subject to the core node is high, i.e. query and modification operations are allowed, but deletion is not allowed.

[0200] Step S440: Map and associate the topology weight threshold with the access permission level to generate a node access permission matrix, where the rows of the node access permission matrix correspond to the sharing subject identifiers, the columns correspond to the core node identifiers, and the element values are the access permission levels.

[0201] The topological weight threshold is used to divide the entity correlation strength interval, thereby determining the approximate range of access permission levels. By mapping the topological weight threshold and the access permission level, it can be determined that different entity correlation strength intervals correspond to which access permission level. The node access permission matrix is a two-dimensional matrix that clearly shows the access permission level of each sharing subject to each core node. The rows of the matrix correspond to different sharing subject identifiers, which are used to uniquely identify each sharing subject; the columns correspond to core node identifiers, which are used to uniquely identify each core node; and the element values in the matrix are the access permission levels of the sharing subject to the core node.

[0202] In actual operation, first, a mapping table is established according to the correspondence between the topological weight threshold and the access permission level determined in the previous steps. Then, for each sharing subject and each core node, the access permission level of the sharing subject to the core node is determined according to the matching score of the subject feature vector of the sharing subject, the business requirement and the core node, and the topological weight threshold, and is filled into the node access permission matrix.

[0203] Step S450: Call the preset zero-knowledge proof protocol to encrypt the node access permission matrix, and generate an encrypted permission vector containing permission proof parameters, which are used to verify the legality of the sharing subject's permission.

[0204] The preset zero-knowledge proof protocol has been used to generate the access control vector in the previous steps, and it is used again here to encrypt the node access permission matrix to ensure that the information in the matrix is not leaked. The encrypted permission vector is a vector representation of the encrypted node access permission matrix, containing permission proof parameters. The permission proof parameters are key information for verifying the legality of the sharing subject's permission. When the sharing subject performs an access operation, the system can verify these parameters to determine whether it has the corresponding access permission.

[0205] In actual operation, the information of the node access permission matrix is input into the preset zero-knowledge proof protocol. The protocol will encrypt the matrix, using advanced encryption algorithms such as the Elliptic Curve Cryptography (ECC) to convert the elements in the matrix into ciphertext form. At the same time, generate permission proof parameters, which can be hash values, signatures, etc. generated during the encryption process. For example, use the Elliptic Curve Cryptography to encrypt the node access permission matrix and generate an encrypted permission vector, which contains digital signatures for verifying the legality of the permission. When a sharing subject requests access to a node, the system will require it to provide permission proof parameters, which are used to verify the legality of its access permission.

[0206] Step S460: fuse the topology weight threshold, the encrypted permission vector, and the node identification information of the semantic topology graph to generate an access control vector containing multi-layer permission control logic, which includes access filtering based on the weight threshold, operation restriction based on the permission level, and legality verification based on zero-knowledge proof.

[0207] Fusion is the process of integrating the topology weight threshold, the encrypted permission vector, and the node identification information of the semantic topology graph. The topology weight threshold is used to preliminarily filter access requests that do not meet the requirements. Only when the entity association strength meets the threshold condition, further access operations are allowed. The encrypted permission vector ensures the security and privacy of access permission information. The node identification information of the semantic topology graph explicitly identifies the specific identity of each node, facilitating accurate access control. Multi-layer permission control logic is the core of the access control vector, including access filtering based on the weight threshold, deciding whether to allow access by comparing the entity association strength with the topology weight threshold; operation restriction based on the permission level, limiting the operations that the sharing subject can perform on the node according to the access permission level of the sharing subject; legality verification based on zero-knowledge proof, ensuring that the sharing subject's permission is legal and effective by verifying the permission proof parameter.

[0208] In actual operation, first, the topology weight threshold, the encrypted permission vector, and the node identification information of the semantic topology graph are combined according to certain rules. For example, the topology weight threshold can be used as part of the access control vector to indicate different permission levels; the encrypted permission vector can be used as the key permission information part; and the node identification information of the semantic topology graph can be used as the node identification part. Then, multi-layer permission control logic is constructed. When a sharing subject initiates an access request, the system first filters the request according to the topology weight threshold. If the entity association strength does not meet the threshold requirement, access is denied. Then, according to the permission level information in the encrypted permission vector, the operation of the sharing subject is restricted. Finally, the zero-knowledge proof is used to verify the permission proof parameter to ensure the legality of the sharing subject's permission.

[0209] Step S500: jointly process the access control vector and the semantic topology graph through the spatio-temporal association analysis model to generate a business event template containing standardized fields and association relationship indexes. The standardized fields are used to unify the text interaction record format in different business scenarios, and the association relationship indexes are used to locate the mapping position of the entity node in the business event template.

[0210] The spatio-temporal correlation analysis model is a model that can comprehensively consider time and space factors for data analysis. In the method, the access control vector and the semantic topology graph are jointly processed to mine the spatio-temporal correlation information therein. The access control vector contains multi-layer permission control logic and access permission information, and the semantic topology graph shows the correlation relationship and semantic evolution between entity nodes. By combining the two, a business event template can be generated. The standardized field is a component in the business event template, and its role is to unify the format of text interaction records in different business scenarios, so that text information of different sources and types can be stored and processed in a consistent manner. The correlation relationship index is used to accurately locate the mapping position of the entity node in the business event template.

[0211] In actual operation, the access control vector and the semantic topology graph are input into the spatio-temporal correlation analysis model. The model first parses the permission information in the access control vector to determine the access permissions of different sharing subjects to the nodes in the semantic topology graph. Then, the spatio-temporal correlation relationship of the nodes in the semantic topology graph is analyzed, such as the change of the correlation strength between the nodes over time. According to the analysis results, the standardized field is generated. For example, for text interaction records in different business scenarios, the unified field name, data type, and length limit are determined. At the same time, according to the correlation relationship of the nodes in the semantic topology graph, the correlation relationship index is generated. Assuming that there are nodes A, B, and C in the semantic topology graph, and there is a certain correlation relationship between them, by analyzing, the mapping position of node A in the business event template and its correlation index with other nodes can be determined. Finally, the standardized field and the correlation relationship index are integrated together to generate the business event template.

[0212] As an implementation, step S500 can include steps S510-S570 as follows:

[0213] Step S510: Perform permission filtering processing on the access control vector, filter a core node subset and a relationship edge subset that meet the permission requirements in the semantic topology graph based on a topology weight threshold, and generate a permission-filtered sub-topology graph.

[0214] The permission filtering processing is a process of filtering the semantic topology graph according to the topology weight threshold in the access control vector. The topology weight threshold is used to divide different access permission levels, and only those core nodes and relationship edges with entity correlation strength meeting the threshold requirement are retained. The core node subset is a node set meeting the permission requirements filtered from the core node set of the semantic topology graph; the relationship edge subset is a corresponding relationship edge set meeting the permission requirements. The permission-filtered sub-topology graph is a new topology graph that only contains the filtered core nodes and relationship edges, and more accurately reflects the part that the sharing subject has permission to access.

[0215] In actual operation, first, the topology weight threshold is extracted from the access control vector. Then, the core nodes and the relationship edges in the semantic topology graph are traversed, and for each core node and relationship edge, the corresponding entity correlation strength is calculated. The entity correlation strength is compared with the topology weight threshold, and if the threshold requirement is met, the core node and the relationship edge are retained in the subset. Finally, according to the filtered core node subset and the relationship edge subset, a sub-topology graph after permission filtering is generated.

[0216] Step S520: Extracting the attribute label set of the entity node, the entity correlation strength of the relationship edge and the semantic evolution path from the sub-topology graph, generating a topology feature set containing spatio-temporal features, and the spatio-temporal features contain evolution features in the time dimension and correlation features in the space dimension.

[0217] The attribute label set of the entity node contains various attribute information of the node, such as time attribute, domain attribute, interaction role attribute, etc., which can reflect the basic characteristics of the node and the role in the business scenario. The entity correlation strength of the relationship edge reflects the correlation closeness between nodes, and the semantic evolution path records the change of node relationship over time. The spatio-temporal features are the core of the topology feature set, including evolution features in the time dimension, such as the creation time, termination time and weight value change sequence of the relationship edge, reflecting the dynamic change of node relationship over time; and correlation features in the space dimension, such as the entity correlation strength between nodes, reflecting the association structure of nodes in space.

[0218] As an implementation, step S520 can include steps S521-S525 as follows:

[0219] Step S521: Traversing the core node subset of the sub-topology graph, collecting the attribute label set of each node, and generating a node attribute sequence in ascending order of time attribute.

[0220] Traversing the core node subset of the sub-topology graph means sequentially accessing each core node in the sub-topology graph. The attribute label set contains various attribute information of the node, such as the time attribute, domain attribute, and interaction role attribute mentioned above. Arranging in ascending order of time attribute is to arrange the collected node attribute label set in the order of the first appearance time of the node, which can clearly show the sequence of the node in the time dimension, and the generated node attribute sequence can provide a time reference for subsequent analysis.

[0221] In actual operation, a loop structure is used to traverse the core node subset of the sub-topology graph. For each core node, the attribute tag set is extracted from its attribute information. After collecting the attribute tag sets of all core nodes, they are sorted according to the time attribute. The attribute tag set can be sorted according to the time attribute in order using a sorting algorithm such as the quicksort algorithm, and finally a node attribute sequence is generated.

[0222] Step S522: Extract the entity association strength of the relationship edge subset in the sub-topology graph, and generate a directed weight matrix combining the direction information of the relationship edge. The rows of the directed weight matrix correspond to the source nodes, and the columns correspond to the target nodes.

[0223] The entity association strength reflects the strength of the relationship edge in the sub-topology graph, and is an index for measuring the closeness of the association between nodes. The direction information of the relationship edge indicates the flow direction of the relationship, i.e., which node points to which node. The directed weight matrix is a two-dimensional matrix, with rows corresponding to source nodes and columns corresponding to target nodes. The element value in the matrix is the entity association strength of the corresponding relationship edge. Through the directed weight matrix, the directed association structure between nodes in the sub-topology graph can be clearly displayed.

[0224] In actual operation, the relationship edge subset of the sub-topology graph is traversed. For each relationship edge, its entity association strength and direction information are extracted. For example, for a relationship edge from node A pointing to node B with an entity association strength of 0.7, the element value in the directed weight matrix corresponding to the row where node A is located and the column where node B is located is 0.7. If there is no relationship edge between two nodes, the corresponding element value in the matrix can be set to 0. In this way, the information of all relationship edges is integrated into the directed weight matrix to generate a complete directed weight matrix.

[0225] Step S523: Time sequence segmentation processing is performed on the semantic evolution path, which is divided into multiple evolution periods according to the timestamp interval of the text block sequence, and each evolution period contains the relationship edge change record in that period.

[0226] The semantic evolution path records the changes of node relationships over time. Time sequence segmentation processing divides the semantic evolution path into multiple evolution periods according to the timestamp interval of the text block sequence. Each evolution period contains the change record of the relationship edge in that period, such as the creation, termination, and weight value change of the relationship edge, which allows more detailed analysis of the dynamic changes of node relationships in different time periods.

[0227] In actual operation, first, the timestamp interval of the text block sequence is determined. For example, the text block sequence is divided by month, so the timestamp interval is one month. Then, according to the time information in the semantic evolution path, it is divided into multiple evolution periods. For each evolution period, the change record of the relationship edge in that period is collected.

[0228] Step S524: Calculate the weight change rate of each relationship edge in each evolution period to generate an evolution rate sequence, the weight change rate being the ratio of the difference between the current period weight value and the previous period weight value to the previous period weight value.

[0229] The weight change rate reflects the degree of change of the weight value of the relationship edge in each evolution period. By calculating the weight change rate of each relationship edge in each evolution period, the change speed and trend of the node relationship can be understood. The evolution rate sequence is a sequence composed of the weight change rate of each relationship edge in each evolution period, which can intuitively show the change of the weight value of the relationship edge over time.

[0230] In actual operation, for each evolution period, for each relationship edge, its weight change rate is calculated. Assuming that in the i th evolution period, the weight value of the relationship edge is W i , in the i-1 th evolution period, the weight value of the relationship edge is W i-1 , then the weight change rate r=(W i -W i-1 ) / W i-1 . After calculating the weight change rate of all relationship edges in each evolution period, the evolution rate sequence is arranged.

[0231] Step S525: Feature fusion is performed on the node attribute sequence, the directed weight matrix, the evolution period division result, and the evolution rate sequence to generate a topology feature set containing time dimension features and space dimension features, the time dimension features corresponding to the evolution rate sequence and the space dimension features corresponding to the directed weight matrix.

[0232] Feature fusion is the process of integrating the node attribute sequence, the directed weight matrix, the evolution period division result, and the evolution rate sequence together. The node attribute sequence provides the order and basic attribute information of the nodes in time; the directed weight matrix shows the spatial association structure between the nodes; the evolution period division result and the evolution rate sequence reflect the dynamic changes of the node relationship in the time dimension. Through feature fusion, a comprehensive topology feature set can be generated, in which the time dimension features are embodied by the evolution rate sequence, reflecting the change speed of the weight value of the relationship edge over time; the space dimension features are embodied by the directed weight matrix, showing the spatial association strength between the nodes.

[0233] In actual operation, first, the node attribute sequence, the directed weight matrix, the evolution period division result, and the evolution rate sequence are arranged and aligned. For example, it is ensured that each value in the evolution rate sequence corresponds to the correct evolution period and relationship edge. Then, these information is fused according to certain rules. They can be combined into a multi-dimensional array or data structure, so that the time dimension features and the space dimension features can be clearly separated. Finally, a topology feature set containing time dimension features and space dimension features is generated.

[0234] Step S530: input the topology feature set into the feature encoding layer of the spatio-temporal association analysis model, vectorize and encode the topology features through the graph neural network, and generate a graph embedding vector containing node association information.

[0235] The feature encoding layer is a component of the spatio-temporal association analysis model, and its main function is to vectorize and encode the topology feature set. The graph neural network (GNN) is a neural network specially designed for processing graph structure data, which can effectively capture the feature information of nodes and edges in the graph. The graph embedding vector is a vector representation of the encoded topology features, containing the association information between nodes, and can be more efficiently used for subsequent analysis and processing.

[0236] In actual operation, the topology feature set is input into the feature encoding layer of the spatio-temporal association analysis model. The graph neural network will process the topology features. First, the node attribute sequence, the directed weight matrix and other information are converted into the node features and edge features of the graph. For example, the attribute information in the node attribute sequence is converted into the feature vector of the node, and the element value in the directed weight matrix is used as the feature of the edge. Then, the graph neural network extracts and converts the features of the graph through multiple layers of convolution or aggregation operations. In each layer, the node is updated according to the information of its neighbor nodes, so as to capture the association information between nodes. Finally, a graph embedding vector containing node association information is generated. Common graph neural network models such as Graph Convolutional Network (GCN) or Graph Attention Network (GAT) can be used to implement this process.

[0237] Step S540: through the association rule mining layer of the spatio-temporal association analysis model, the graph embedding vector is mined for frequent item sets, and a set of strong association rules between entity nodes is extracted, which contains the antecedent entity, the consequent entity and the confidence value.

[0238] The association rule mining layer is a functional module in the spatiotemporal association analysis model, and the main task is to mine the association rules between entity nodes from the graph embedding vectors. Frequent item set mining is a data mining technique that finds frequently occurring item sets by counting the frequency of item sets in data. Strong association rules refer to association rules with high confidence, and the confidence value represents the probability of the occurrence of the consequent entity under the condition that the antecedent entity is met. The antecedent entity and the consequent entity are the premise and conclusion parts of the association rule, respectively.

[0239] In actual operation, the graph embedding vectors are input to the association rule mining layer of the spatiotemporal association analysis model. First, a frequent item set mining algorithm such as the Apriori algorithm or the FP-growth algorithm is used to process the graph embedding vectors to find frequently occurring node combinations. For example, it is found that node A and node B often appear at the same time to form a frequent item set. Then, association rules are generated based on the frequent item sets, and the confidence value of each rule is calculated. Assuming that an association rule is “if node A appears, then node B appears”, the confidence value of the rule is obtained by counting the proportion of the simultaneous occurrence of node B under all conditions of the occurrence of node A. Strong association rules are selected by filtering rules with high confidence values.

[0240] Step S550: Construct a standardized field mapping table based on the strong association rule set, and the standardized field mapping table includes field name, data type, length limit, and source entity identifier.

[0241] The standardized field mapping table is used to map the attribute information of the entity nodes to the standardized fields. The field name is the specific name of the standardized field, which is used to identify the meaning of the field; the data type specifies the type of data stored in the field, such as integer, string, date, etc.; the length limit restricts the data length of the field, ensuring the consistency and standardization of the data; and the source entity identifier specifies the data source of the field, which is the entity node.

[0242] As an implementation, step S550 can include the following steps S551-S557:

[0243] Step S551: Perform confidence level processing on the strong association rule set, and divide it into multiple rule levels from high to low according to the confidence value, and each rule level corresponds to a different association reliability level.

[0244] The confidence level processing is a process of sorting and dividing the strong association rule set according to the size of the confidence value. The confidence value reflects the reliability of the association rule, and the higher the confidence value, the stronger the reliability of the rule. The rules are divided into multiple rule levels from high to low according to the confidence value, and each level corresponds to a different association reliability level, which can more detailedly manage and apply the association rules.

[0245] Step S552: Rule clustering processing is performed on the strong association rules within each rule hierarchy, and a rule cluster set with business scenario commonality is generated by calculating the attribute similarity between the antecedent entity and the consequent entity of the rules. Each rule cluster contains semantically associated strong association rules.

[0246] Rule clustering processing is a process of grouping strong association rules within each rule hierarchy, aiming to gather rules with similar business scenarios and semantic associations. By calculating the attribute similarity between the antecedent entity and the consequent entity of the rules, the similarity between rules can be determined. Attribute similarity can be calculated by comparing the attribute label set of entities, such as domain attributes, time attributes, etc., using similarity algorithms such as cosine similarity or edit distance. The rule cluster set is composed of multiple rule clusters, each containing semantically associated strong association rules that share common business scenarios.

[0247] Step S553: Extract the attribute label set of entity nodes in each rule cluster, calculate the prediction contribution of each attribute label to the business field using the information gain algorithm, and generate an attribute importance ranking sequence. The information gain algorithm is an algorithm for evaluating feature importance, which measures the prediction contribution of attribute labels by calculating the information gain of attribute labels to business fields. Business fields refer to fields that need to be focused on in business data sharing and processing, such as approval results, business process status, etc. The attribute importance ranking sequence is obtained by sorting attribute labels according to their prediction contribution, which can help determine which attribute labels are more important for predicting business fields.

[0248] Step S554: Construct a dynamic threshold screening model based on the attribute importance ranking sequence, adjust the attribute screening threshold according to the association reliability level of the rule cluster, and retain the core attribute labels above the threshold.

[0249] The dynamic threshold screening model is a model that screens attribute labels based on the attribute importance ranking sequence and the association reliability level of the rule cluster. The attribute screening threshold is a critical value for determining whether an attribute label is a core attribute label. According to the association reliability level of the rule cluster, the attribute screening threshold can be adjusted. For rule clusters with high association reliability level, the threshold can be appropriately increased to retain only the most core attribute labels; for rule clusters with low association reliability level, the threshold can be reduced to retain more attribute labels.

[0250] Step S555: Bidirectional matching of core attribute labels with metadata requirements in business data sharing standard specifications is performed, semantic similarity and structural matching degree between attributes and metadata are calculated, and a matching score matrix is generated.

[0251] The metadata requirement in the business data sharing standard specification specifies the standards and norms that business data needs to follow during sharing and processing, including field name, data type, length limit, etc. Bidirectional matching refers to the mutual comparison between core attribute labels and metadata requirements, considering both semantic similarity, i.e., the similarity in meaning, and structural matching degree, i.e., the matching degree in data type, length limit, etc. The matching score matrix is a two-dimensional matrix, with rows corresponding to core attribute labels and columns corresponding to metadata requirements. The element value in the matrix is the matching score of the corresponding attribute label and metadata requirement.

[0252] In actual operation, for each core attribute label and each metadata requirement, their semantic similarity and structural matching degree are calculated respectively. Semantic similarity can be calculated using semantic similarity algorithms such as WordMover's Distance (WMD) or similarity calculation methods based on pre-trained language models. Structural matching degree is calculated by comparing the data type, length limit, and other structural information of the attribute label and metadata requirement using matching rules. For example, if the data type of the attribute label and the data type of the metadata requirement are the same, the structural matching degree is 1, otherwise it is 0. The semantic similarity and structural matching degree are weighted and summed according to certain weights to obtain the matching score. Finally, all the matching scores of the core attribute labels and metadata requirements are arranged into a matching score matrix.

[0253] Step S556: Optimal matching of the matching score matrix is solved by the Hungarian algorithm to determine the mapping relationship between each standardized field and core attribute label, which includes attribute source entity identifier, attribute name, and matching confidence.

[0254] The Hungarian algorithm is a classic algorithm for solving the maximum weight matching problem of bipartite graphs, which can find the optimal matching scheme in the matching score matrix to maximize the total matching score. The standardized field is a standard field in the business event template, and the core attribute label is an important attribute label selected from the rule cluster. The mapping relationship clarifies the correspondence between each standardized field and core attribute label, including attribute source entity identifier, which identifies the attribute from which entity node; attribute name, i.e., the specific name of the attribute label; and matching confidence, i.e., the reliability of the match, which can be calculated according to the matching score.

[0255] In actual operation, the matching score matrix is input into the Hungarian algorithm. The algorithm finds the optimal matching scheme so that the total matching score is maximum. For each normalized field, according to the output result of the algorithm, the core attribute label matched therewith is determined. For example, for the normalized field "department name", the core attribute label "institution name" matched therewith is found by the Hungarian algorithm, the attribute source entity identifier thereof is node A, the attribute name is "institution name", and the matching confidence can be calculated according to the corresponding element value in the matching score matrix, for example, the higher the matching score, the higher the matching confidence. Finally, these mapping relationships are sorted into a mapping relationship set containing the attribute source entity identifier, the attribute name and the matching confidence.

[0256] Step S557: Cross-rule cluster consistency verification is performed on the mapping relationships. If the mapping relationships of the same normalized field corresponding to different rule clusters conflict, the final mapping relationship is determined by weighted voting according to the rule level priority and the matching confidence, and a normalized field mapping table containing the field name, the data type, the source attribute and the verification rule is generated.

[0257] The cross-rule cluster consistency verification is a process of checking the mapping relationships, and the purpose is to ensure that the mapping relationships of the same normalized field corresponding to different rule clusters are consistent. If there is a conflict, that is, different rule clusters provide different mapping relationships for the same normalized field, the rule level priority and the matching confidence weighted voting are needed to solve it. The rule level priority is determined according to the associated reliability level of the rule cluster, and the rule cluster with a higher associated reliability level has a higher priority. The matching confidence reflects the reliability of the mapping relationship. The normalized field mapping table is the final output result, which contains the field name, the data type, the source attribute and the verification rule, etc., and provides a basis for the standardization processing of business data.

[0258] In actual operation, all normalized fields are traversed, and for each normalized field, the mapping relationships provided by different rule clusters are checked. If there is a conflict, for example, the mapping relationship provided by rule cluster 1 for the normalized field "department name" is attribute A, and the mapping relationship provided by rule cluster 2 is attribute B, then weighted voting is performed according to the rule level priority and the matching confidence. Assuming that the associated reliability level of rule cluster 1 is high and the matching confidence is 0.8, the associated reliability level of rule cluster 2 is medium and the matching confidence is 0.6, and the weighted calculation is performed according to the rule level priority and the matching confidence, and finally it is determined that attribute A is selected as the mapping relationship of the normalized field. The final mapping relationships of all normalized fields are sorted into a normalized field mapping table containing the field name, the data type, the source attribute and the verification rule.

[0259] Step S560: generating a correlation relationship index according to the standardized field mapping table and the entity correlation strength, the correlation relationship index being used to record the mapping position and the correlation strength between the entity nodes and the standardized fields.

[0260] The correlation relationship index is a data structure used to record the correlation information between the entity nodes and the standardized fields. The standardized field mapping table explicitly shows the mapping relationship between the standardized fields and the core attribute labels, and the entity correlation strength reflects the correlation closeness between the entity nodes. By combining the two, the correlation relationship index can be generated, which records the mapping position between the entity nodes and the standardized fields, i.e., how the attributes of the entity nodes are mapped to the standardized fields, and the correlation strength therebetween.

[0261] In actual operation, first, the mapping relationship between each standardized field and the core attribute label is obtained from the standardized field mapping table. For example, the standardized field "department name" is mapped to the attribute "institution name" of entity node A. Then, according to the entity correlation strength, the correlation strength between the entity nodes and the standardized fields is determined. If the correlation strength of entity node A with other related nodes is high, and its attribute is mapped to the standardized field "department name", then it can be considered that the correlation strength of entity node A with the standardized field is also high. The mapping position and the correlation strength between the entity nodes and the standardized fields are recorded to form the correlation relationship index. Data structures such as dictionaries or lists can be used to store this information, facilitating subsequent queries and use.

[0262] Step S570: fusing the standardized field mapping table and the correlation relationship index to generate a business event template containing a field definition layer and a relationship index layer, the field definition layer specifying the unified field format under different business scenarios, and the relationship index layer recording the entity correlation path between the fields.

[0263] Fusing the standardized field mapping table and the correlation relationship index is the process of integrating the two data structures together, and the purpose is to generate a complete business event template. The field definition layer is part of the business event template, which specifies the unified field format under different business scenarios based on the standardized field mapping table, including field name, data type, length limit, etc., to ensure that the business data has a consistent representation form under different business scenarios. The relationship index layer is based on the correlation relationship index and records the entity correlation path between the fields, i.e., the correlation relationship between different standardized fields through entity nodes, which helps to understand the internal relationship and dependency between the fields when processing business data.

[0264] Please refer to Figure 2 , Figure 2A structural schematic diagram of a computer system provided by the embodiment of the present application is shown in FIG. 1. The computer system includes at least a processor 101, a communication interface 102 and a memory 103. The processor 101, the communication interface 102 and the memory 103 can be connected through a bus or other means. The processor 101 (also called a central processing unit (CPU)) is the computing core and control core of the computer system, which can parse various instructions in the computer system and process various data of the computer system. The communication interface 102 can optionally include a standard wired interface, a wireless interface (such as WI-FI, a mobile communication interface, etc.), and can be used for transmitting and receiving data under the control of the processor 101; the communication interface 102 can also be used for transmitting and interacting data within the computer system. The memory 103 is a memory device in the computer system, which is used for storing programs and data. It can be understood that the memory 103 can include a built-in memory of the computer system, and of course can also include an extended memory supported by the computer system. The memory 103 provides a storage space, which stores an operating system of the computer system, and the present application does not limit this.

[0265] In one embodiment, the processor 101 executes the information extraction processing method applied to the data sharing service provided by the above embodiment of the present application by running a computer program in the memory 103.

Claims

1. An information extraction processing method applied to a data sharing service, characterized by, The method comprises: obtaining a historical interaction text set from a cross-department business data sharing system, the historical interaction text set containing text interaction records generated under different business scenarios and marked with timestamps; performing entity relationship network construction on the historical interaction text set to generate a multi-dimensional entity relationship network containing entity nodes, relationship edges and attribute labels, the attribute labels being used to represent the context association attributes of the entity nodes in the text interaction records; performing semantic topology graph construction based on the multi-dimensional entity relationship network to obtain a semantic topology graph containing entity association strength and semantic evolution paths; calling a preset zero-knowledge proof protocol to perform access control vector generation on the semantic topology graph to obtain an access control vector containing a topology weight threshold and node access permissions, wherein the topology weight threshold is determined based on statistical characteristics of the entity association strength, the entity association strength reflecting the association closeness between nodes in the semantic topology graph, and the node access permissions being used to limit the operation range of different sharing subjects on the entity nodes; jointly processing the access control vector and the semantic topology graph through a spatio-temporal association analysis model to generate a business event template containing standardized fields and association relationship indexes.

2. The method of claim 1, wherein, The entity relationship network construction on the historical interaction text set to generate a multi-dimensional entity relationship network containing entity nodes, relationship edges and attribute labels comprises: blocking the historical interaction text set in ascending order of timestamps to obtain a text block sequence with continuous time sequence characteristics, each text block containing text interaction records within a preset time interval, and there being a preset proportion of time overlap areas between adjacent text blocks; performing entity recognition on the text interaction records in the text block sequence to extract an entity set containing business domain-specific terms, cross-departmental institution identifiers and business process nodes, each entity having a unique identifier and a type label; based on the position coordinates and context of the entities in the text interaction records, calculating the spatial distance parameters and semantic correlation coefficients between entity pairs to generate a co-occurrence feature matrix representing the potential association between entities; constructing a directed relationship edge set between entities according to the co-occurrence feature matrix, each relationship edge containing a source entity identifier, a target entity identifier and a dynamic weight value, the dynamic weight value being determined by the product of the reciprocal of the spatial distance parameter and the semantic correlation coefficient; allocating an attribute label set containing time attributes, domain attributes and interaction role attributes to each entity node, the time attribute corresponding to the timestamp of the first appearance of the entity in the text block sequence, and the domain attribute being determined based on the business scenario keywords of the text interaction record in which the entity is located; constructing a multi-dimensional entity relationship network by taking the entity set as a node layer, the relationship edge set as a relationship layer and the attribute label set as an attribute layer, the node layer being associated with the relationship layer through edge weight values, and the relationship layer being associated with the attribute layer through attribute mapping rules.

3. The method of claim 2, wherein, The entity recognition on the text interaction records in the text block sequence to extract an entity set containing business domain-specific terms, cross-departmental institution identifiers and business process nodes comprises: A hierarchical entity recognition model is constructed by fusing a bidirectional long short-term memory network and a conditional random field, and the hierarchical entity recognition model comprises an input layer, a word embedding layer, a bidirectional long short-term memory layer, a multi-head attention mechanism layer, a feature fusion layer, and a conditional random field output layer; The text interaction record is input into the input layer of the hierarchical entity recognition model according to a character sequence, and a character sequence vector is generated through character-level segmentation processing, the length of the character sequence vector being consistent with the number of characters of the text interaction record; The character sequence vector is input into the word embedding layer, and the character sequence is converted into a word vector sequence containing context semantic information through a pre-trained business domain word vector matrix, the dimension of the word vector sequence matching a preset embedding dimension parameter; The word vector sequence is subjected to forward and backward semantic dependency modeling through the bidirectional long short-term memory layer, to generate a forward hidden state sequence containing historical context information and a backward hidden state sequence containing future context information, the forward and backward hidden state sequences corresponding in time steps; The forward hidden state sequence and the backward hidden state sequence are spliced into a bidirectional hidden state sequence, which is input into the multi-head attention mechanism layer, and the bidirectional hidden state sequence is subjected to multi-scale semantic focusing processing through a preset number of attention heads to generate an attention-enhanced feature sequence containing different semantic granularities; The attention-enhanced feature sequence is input into the feature fusion layer, and the output features of different attention heads are integrated through residual connection and layer normalization operation to generate a fusion feature sequence with semantic consistency; The fusion feature sequence is input into the conditional random field output layer, and a state transition probability matrix of an entity label set corresponding to each character position is calculated, the entity label set including an entity start label, an entity internal label, and a non-entity label; An optimal path is searched based on the state transition probability matrix to determine the entity label category of each character position, and a label sequence containing entity boundary information is generated; According to the label sequence, a character fragment corresponding to a continuous entity label is intercepted from the text interaction record, and an entity set is generated in combination with entity type labeling.

4. The method of claim 3, wherein, Based on the position coordinates and context of the entity set in the text interaction record, a spatial distance parameter and a semantic correlation coefficient between entity pairs are calculated to generate a co-occurrence feature matrix representing the potential association between entities, including: The starting position coordinates and ending position coordinates of each entity in the text are extracted by traversing the entity set in the text interaction record, and a text coverage length parameter of the entity is calculated, the text coverage length parameter being the difference between the ending position coordinates and the starting position coordinates; Entities in the same text interaction record are paired two by two to generate an entity pair set, each entity pair including identifiers of two different entities and position coordinates in the text; The spatial distance parameter of the two entities in the entity pair is calculated, the spatial distance parameter being a weighted average of the absolute value of the difference between the starting position coordinates of the two entities and the text coverage length parameter; The context fragment of the text interaction record where the entity pair is located is extracted, the context fragment including a preset number of character sequences before and after the entity pair. Calculate a semantic correlation coefficient of the two entities in the context snippet by a pre-trained semantic similarity model; Construct a co-occurrence feature matrix with entity identifiers as row and column indexes, and the matrix element value is the product of the reciprocal of the spatial distance parameter of the entity pair and the semantic correlation coefficient, and the diagonal element value is a preset constant; Perform row normalization processing on the co-occurrence feature matrix to make the sum of each row element 1, and generate a standardized co-occurrence feature matrix for relationship edge construction.

5. The method of claim 1, wherein, The semantic topology graph construction based on the multi-dimensional entity relationship network includes: Calculate the center index of the node layer of the multi-dimensional entity relationship network to generate a comprehensive center index set including degree centrality, betweenness centrality and closeness centrality; Sort the entity nodes in descending order according to the comprehensive center index set, and select the entity nodes of a preset proportion as the core node set of the semantic topology graph, and the preset proportion is dynamically adjusted based on the total number of nodes of the multi-dimensional entity relationship network; Track the relationship edge changes of the core node set in the text block sequence in time sequence, extract the creation timestamp, termination timestamp and weight value change sequence of the relationship edge, and generate dynamic evolution features of entity relationship; Construct a relationship evolution prediction model based on the dynamic evolution features, input the relationship edge state of the current text block, predict the possible newly added or disappeared relationship edges in the next text block, and generate a potential relationship edge set; Calculate the entity association strength between the core node set, which is determined by the weighted sum of the dynamic weight value of the relationship edge, the normalized element value in the co-occurrence feature matrix, the time decay factor and the domain similarity, and the time decay factor decreases as the interval between the last co-occurrence timestamp and the current timestamp of the entity pair increases; The core node set is taken as the vertex, the entity association strength is taken as the edge weight, the dynamic evolution features are taken as the time attribute, and the potential relationship edge set is taken as the prediction attribute, to construct a semantic topology graph containing static topology structure and dynamic evolution information.

6. The method of claim 5, wherein, The center index calculation of the node layer of the multi-dimensional entity relationship network includes: Traverse the relationship edge set of the multi-dimensional entity relationship network, and count the in-degree relationship edge number and out-degree relationship edge number of each entity node, wherein the in-degree relationship edge number is the total number of relationship edges with the entity as the target, and the out-degree relationship edge number is the total number of relationship edges with the entity as the source; Calculate the weighted degree centrality index of each entity node, which is the average value of the sum of the dynamic weight values of the in-degree relationship edges and the sum of the dynamic weight values of the out-degree relationship edges; Normalize the weighted degree centrality index to generate a degree centrality component; Calculate the shortest path set between all entity pairs in the multi-dimensional entity relationship network, and the length of each shortest path is the sum of the dynamic weight values of the relationship edges on the path; counting a number of intermediation times of each entity node in the set of shortest paths, the intermediation times being a number of shortest paths passing through the entity node, to generate an original value of betweenness centrality; performing standardization processing on the original value of betweenness centrality to generate a betweenness centrality component; calculating an average value of a shortest path length of each entity node to all other entity nodes to generate an average distance parameter, the proximity centrality component being an inverse of the average distance parameter; performing weighted summation on the degree centrality component, the betweenness centrality component and the proximity centrality component according to a preset weight ratio to generate a comprehensive centrality index of each entity node.

7. The method of claim 6, wherein, the calculating the entity association strength between the set of core nodes comprises: extracting a relationship edge dynamic weight value of any two entity nodes in the set of core nodes, and setting the dynamic weight value as a preset minimum value if there is no direct relationship edge between the two entities; extracting a normalized element value corresponding to the two entities from the co-occurrence feature matrix as a co-occurrence association component; calculating a field similarity parameter according to a field attribute in the attribute label set of the two entities, the field similarity parameter being determined by a cosine similarity of field attribute keywords; obtaining a time stamp of the last co-occurrence of the two entities in the sequence of text blocks to calculate a time decay factor, the time decay factor being an exponential function of a difference between a current time stamp and the last co-occurrence time stamp; performing weighted summation on the dynamic weight value as a basic component, the co-occurrence association component as an enhanced component, the field similarity parameter as an adjustment component and the time decay factor as a correction component to obtain the entity association strength; performing upper and lower limit truncation processing on the calculated entity association strength to limit the value range of the entity association strength to a preset numerical interval to generate a standardized entity association strength used for edge weight of the semantic topology graph.

8. The method of claim 1, wherein, the calling the preset zero-knowledge proof protocol to perform access control vector generation on the semantic topology graph to obtain an access control vector containing a topology weight threshold and node access permission comprises: performing topology structure preprocessing on the semantic topology graph to extract a topology feature vector containing the set of core nodes, the entity association strength and the semantic evolution path, the dimension of the topology feature vector matching the size of the set of core nodes; calculating statistical characteristic values based on the entity association strength in the topology feature vector to generate a topology weight threshold used for dividing access permission levels, the statistical characteristic values including a mean value, a median and a standard deviation of the entity association strength; dividing access permission levels of the set of core nodes according to identity authentication information and business requirement description of a sharing subject, each access permission level corresponding to a different set of operation permissions; mapping and associating the topology weight threshold and the access permission levels to generate a node access permission matrix, a row of the node access permission matrix corresponding to a sharing subject identifier, a column corresponding to a core node identifier, and an element value being an access permission level; calling the preset zero-knowledge proof protocol to perform encryption processing on the node access permission matrix to generate an encrypted permission vector containing a permission proof parameter, the permission proof parameter being used to verify the legality of the permission of the sharing subject; The topology weight threshold, the encryption permission vector and the node identification information of the semantic topology graph are fused to generate an access control vector containing multi-layer permission control logic.

9. The method of claim 8, wherein, The statistical characteristic value is calculated based on the entity correlation strength in the topology feature vector to generate a topology weight threshold for dividing access permission levels, including: Outlier detection is performed on the entity correlation strength in the topology feature vector to eliminate abnormal correlation strength values exceeding a preset multiple of the upper and lower quartile ranges to generate a purified entity correlation strength sequence; The statistical characteristic value of the purified entity correlation strength sequence is calculated; A multi-threshold division model is constructed based on the statistical characteristic value to divide the entity correlation strength into multiple intervals, each interval corresponding to a permission level reference value; An optimal threshold combination is determined through ROC curve analysis, the optimal threshold combination making the false negative rate corresponding to different permission levels lower than a preset threshold; The interval boundary values in the optimal threshold combination are taken as the topology weight threshold to generate a threshold set containing multiple hierarchical thresholds.

10. A computer system, characterized by including: a memory in which a computer program is stored; a processor configured to load the computer program to implement the information extraction processing method for data sharing services according to any one of claims 1-9.

Images