Multi-source heterogeneous fair competition review data processing method

Abstract

The application provides a multi-source heterogeneous fair competition review data processing method, relates to the technical field of data processing, and comprises the following steps: structurally analyzing to-be-reviewed files and historical files, and extracting first structural feature information corresponding to the to-be-reviewed files and the historical files respectively; a competition connection structure is established according to object mapping information and constraint action information therein; the first competition connection structure and the second competition connection structure are associated and matched, and competition connection edge change information is identified; node connection path changes in the competition connection structure are analyzed, and a connection reduction object is determined; a connection drop object set is identified based on the connection reduction object, and a fair competition review result corresponding to the connection drop object set is output. According to the application, the association change of the business subject scattered in different clauses can be identified as a whole and the range of influence thereof can be traced along the original association path, so that the review result can reflect the actual influence range and the continuous decline trend of the association between the business subjects caused by the change of the clauses.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method for processing multi-source heterogeneous fair competition review data. Background Technology

[0002] When processing textual materials involving multiple business entities, it is usually necessary to compare newly generated textual materials with existing historical textual materials to determine whether the content of clauses involving business entities has changed. The processing typically involves first dividing the text into paragraphs and identifying headings, then extracting the names of business entities from the statements using named entity recognition, and finally determining the relationships between business entities in the statements through relation extraction.

[0003] Existing processing methods, after completing the above analysis, typically record the identified business entities and their functional relationships clause by clause in order. During comparison, they search for similar expressions clause by clause and determine whether the clauses have changed based on text similarity. This approach treats each functional relationship as an isolated statement, failing to link the functional relationships of the same business entity appearing in multiple clauses into a cohesive structure. When both the text to be processed and the historical text involve multiple business entities, and the relevant functional relationships are scattered across different clauses, it is difficult to determine, based solely on clause-by-clause comparison results, whether the original multiple relationships between a particular business entity and other business entities have decreased due to clause changes, or whether this decrease will further affect other business entities along the original relationships between them.

[0004] Therefore, determining the changes in the relationships between business entities between the text to be processed and historical texts, and identifying the scope of business entities affected by these changes, becomes a problem that needs to be solved. Summary of the Invention

[0005] This application provides a multi-source heterogeneous fair competition review data processing method to address the problem of how to determine changes in competitive connections between business entities and identify the scope of business entities affected by reduced connections when both the documents under review and historical documents involve multiple business entities and the limiting or exclusionary relationships between business entities are recorded in different clauses.

[0006] This application provides a method for processing multi-source heterogeneous fair competition review data, including:

[0007] The document to be reviewed and the historical documents are subjected to structured parsing to extract the first structural feature information corresponding to each of the document to be reviewed and the historical documents.

[0008] A competitive connection structure is established based on the object mapping information and constraint information in the first structural feature information;

[0009] The first competitive connection structure corresponding to the file to be reviewed and the second competitive connection structure corresponding to the historical file are associated and matched to identify the change information of the competitive connection edges.

[0010] Based on the information about changes in the competing connection edges, analyze the changes in the node connection paths in the competing connection structure to identify objects where connections are reduced.

[0011] Based on the identified connection reduction objects, a set of connection decline objects is identified, and the fair competition review results corresponding to the set of connection decline objects are output.

[0012] Optionally, in one possible implementation, the step of performing structured parsing on the document to be reviewed and the historical documents, and extracting the first structural feature information corresponding to each of the document to be reviewed and the historical documents, includes:

[0013] The document to be reviewed is parsed for chapter structure and semantic entity recognition to obtain the first file segmentation information and the first entity information corresponding to the document to be reviewed.

[0014] The historical files are parsed for chapter structure and semantic entity recognition to obtain the second file segmentation information and second entity information corresponding to the historical files.

[0015] Based on the first file segmentation information and the first entity information, an association mapping is performed to generate the first object mapping information corresponding to the file to be reviewed;

[0016] Based on the second file segmentation information and the second entity information, an association mapping is performed to generate the second object mapping information corresponding to the historical file;

[0017] The entity items in the first object mapping information and the entity items in the second object mapping information are dereferenced together, and the coreferenced entity items are merged to obtain the object mapping information and constraint information corresponding to the document to be reviewed and the historical document, respectively, after merging.

[0018] The merged object mapping information and constraint information are used as the first structural feature information.

[0019] Optionally, in one possible implementation, establishing a competing connection structure based on the object mapping information and constraint action information in the first structural feature information includes:

[0020] Extract the set of business entity nodes from the object mapping information, and treat each business entity node as a graph node in the competitive connection structure;

[0021] Extract the competitive constraint relationships between business entities from the constraint information, and treat each competitive constraint relationship as a competitive connection edge connecting two graph nodes;

[0022] The graph nodes and the competing connection edges form a competing connection graph, which is used as the first competing connection structure.

[0023] Each competing connection edge in the competing connection graph is assigned an edge identifier, which is used to distinguish different types of competing constraint relationships;

[0024] Based on the edge identifier, the competing connection edges in the competing connection graph are layered to obtain multiple competing connection subgraphs, and the multiple competing connection subgraphs are used together as the competing connection structure.

[0025] Optionally, in one possible implementation, the step of associating and matching the first competing connection structure corresponding to the file to be reviewed and the second competing connection structure corresponding to the historical file to identify competing connection edge change information includes:

[0026] Each graph node in the first competitive connection structure is paired with a graph node in the second competitive connection structure that has the same business entity identifier to obtain the node pairing result.

[0027] Based on the node pairing results, each first competitive connection edge in the first competitive connection structure is paired with the second competitive connection edge between the corresponding node pair in the second competitive connection structure to obtain the edge pairing results.

[0028] In response to the edge pairing result being that the edge identifiers of the first competing connection edge and the second competing connection edge are consistent, the first competing connection edge is marked as an invariant edge;

[0029] In response to the edge pairing result that the first competing connection edge exists while the second competing connection edge does not exist, the first competing connection edge is marked as a newly added edge;

[0030] In response to the edge pairing result that the first competing connection edge does not exist while the second competing connection edge does exist, the second competing connection edge is marked as a disappeared edge;

[0031] Collect the edge identifiers of the newly added edges and the disappearing edges, as well as the node identifiers of the connected node pairs, to generate the information on the changes in the competing connection edges.

[0032] Optionally, in one possible implementation, collecting the edge identifiers of the newly added edges and the disappeared edges, and the node identifiers of the connected node pairs, to generate the competing connection edge change information includes:

[0033] The frequency of occurrence of each business entity node connected by the newly added edge and the disappeared edge is counted separately;

[0034] The business entity nodes that appear more frequently than the first preset frequency condition in the newly added edges are marked as newly added active nodes;

[0035] The business entity nodes that appear more frequently than the second preset frequency condition in the disappearing edge are marked as disappearing active nodes;

[0036] A new change record is generated based on the node identifier of the newly added active node and the edge identifier of the newly added edge;

[0037] A disappearance change record is generated based on the node identifier of the disappearing active node and the edge identifier of the disappearing edge;

[0038] The newly added change records and the disappeared change records are combined into the competing connection edge change information.

[0039] Optionally, in one possible implementation, the step of analyzing the changes in node connection paths in the competitive connection structure based on the information about changes in the competitive connection edges to determine the objects with reduced connections includes:

[0040] Based on the edge identifier of the disappeared edge and the node identifier of the connected node pair in the information on changes in the competitive connection edge, the first node pair corresponding to the removed competitive connection edge is located in the first competitive connection structure.

[0041] Starting from each node in the first node pair, traverse the node connection paths along the reserved competitive connection edges in the first competitive connection structure to obtain the first node connection path set.

[0042] Starting from each node in the first node pair, traverse the node connection paths along all competing connection edges in the second competitive connection structure to obtain the second node connection path set;

[0043] Compare the first node connection path set and the second node connection path set, and extract the missing node connection paths that exist in the second node connection path set but do not exist in the first node connection path set;

[0044] The operating entity nodes that do not appear in the first competing connection structure are extracted from the connection path of the disappeared nodes as the connection reduction objects.

[0045] Optionally, in one possible implementation, comparing the first set of node connection paths and the second set of node connection paths to extract disappearing node connection paths that exist in the second set of node connection paths but not in the first set of node connection paths includes:

[0046] Obtain the first node connection path in the second node connection path set as the current comparison path;

[0047] Search the first node connection path set for candidate paths that match the start node identifier and end node identifier of the current comparison path;

[0048] In response to finding the candidate path, the intermediate node sequence of the currently compared path is compared with the intermediate node sequence of the candidate path;

[0049] In response to the inconsistency in the intermediate node sequence, the current alignment path is marked as a structural change path;

[0050] In response to the consistency of the intermediate node sequence, the current alignment path is marked as a structure-preserving path;

[0051] In response to the failure to find the candidate path, the current comparison path is marked as the connection path of the disappeared node;

[0052] Iterate through all node connection paths in the second node connection path set to obtain all disappeared node connection paths.

[0053] Optionally, in one possible implementation, the step of identifying a set of connection-decreasing objects based on the connection-decreasing objects and outputting the fair competition review result corresponding to the set of connection-decreasing objects includes:

[0054] Using each business entity node in the connection reduction object as a seed node, extend a preset path number of steps outward along the competitive connection edge in the second competitive connection structure to obtain the set of extended nodes corresponding to each seed node;

[0055] Take the union of the sets of extended nodes corresponding to all seed nodes, and use the union as the set of candidate connection descent nodes;

[0056] Obtain the sequence of changes in the number of competing connection edges for each business entity node in the candidate connection descent node set within a preset time window;

[0057] From the candidate set of descent nodes, select the business entity nodes whose competitive connection edge quantity change sequence shows a downward trend, and use them as the set of descent objects.

[0058] The fair competition review result is generated based on the node identifiers and the sequence of changes in the number of competing connection edges of each operating entity node in the set of connected descent objects.

[0059] Optionally, in one possible implementation, selecting the business entity nodes whose competitive connection edge quantity change sequence shows a decreasing trend from the candidate connection descent node set as the connection descent object set includes:

[0060] The sequence of changes in the number of competing connection edges of the first operating entity node in the candidate connection descent node set is differentially processed to obtain the first difference sequence;

[0061] Based on the number of consecutive negative values ​​and the proportion of the total number of negative values ​​to all difference values ​​in the first difference sequence, the first business entity node is marked as a first candidate node or a second candidate node.

[0062] The first candidate node that simultaneously satisfies the downward trend condition and the second candidate node that simultaneously satisfies the downward trend condition are selected as nodes in the set of connected downward objects.

[0063] Traverse all the business entity nodes in the candidate connection descent node set to obtain the complete connection descent object set.

[0064] Optionally, in one possible implementation, marking the first business entity node as a first candidate node or a second candidate node based on the number of consecutive negative values ​​and the proportion of the total number of negative values ​​to all difference values ​​in the first difference sequence includes:

[0065] In response to the fact that the number of consecutive negative values ​​in the first difference sequence exceeds a first preset quantity condition, the first business entity node is marked as a first candidate node that meets the downward trend condition;

[0066] In response to the fact that the proportion of negative values ​​in the first difference sequence exceeds a first preset proportion condition, the first business entity node is marked as a second candidate node that meets the downward trend condition.

[0067] The multi-source heterogeneous fair competition review data processing method provided in this application has the following beneficial effects: This application transforms the originally scattered and isolated relationships between business entities recorded in different clauses into a unified and comparable relational form. This allows changes in the relationship between the document to be processed and historical documents to no longer rely on textual similarity judgments but can be directly reflected in the specific existence or disappearance of the relationship. Consequently, a reduction in the original relationship of a particular business entity is no longer an isolated or localized judgment result but can be traced outwards along the original relationship path between that business entity and other business entities, thus determining the scope of affected business entities. Furthermore, by continuously observing the changes in the relationships of relevant business entities over a period of time, it is possible to distinguish between occasional fluctuations in relationships and continuous declines in relationships. This ensures that the final review result reflects the actual scope and trend of the impact of clause changes on the relationships between business entities, rather than merely textual differences in a single clause. Attached Figure Description

[0068] Figure 1This is a flowchart illustrating the multi-source heterogeneous fair competition review data processing method provided in the embodiments of this application;

[0069] Figure 2 This is a schematic diagram of the second competitive connection structure (historical file) of the multi-source heterogeneous fair competition review data processing method provided in the embodiments of this application;

[0070] Figure 3 This is a schematic diagram of the first competitive connection structure (document to be examined) of the multi-source heterogeneous fair competition review data processing method provided in the embodiments of this application;

[0071] Figure 4 This is a schematic diagram of the structure of the multi-source heterogeneous fair competition review data processing system provided in the embodiments of this application;

[0072] Figure 5 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0073] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0074] The technical solutions of this application will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0075] See Figure 1 This is a flowchart of the multi-source heterogeneous fair competition review data processing method provided in the embodiments of this application. Figure 1The execution entity of the method shown can be a software and / or hardware device. The execution entity of this application can include, but is not limited to, at least one of the following: user equipment, network equipment, etc. User equipment can include, but is not limited to, computers, smartphones, personal digital assistants (PDAs), and the aforementioned electronic devices. Network equipment can include, but is not limited to, a single network server, a server group consisting of multiple network servers, or a cloud based on cloud computing consisting of a large number of computers or network servers. Cloud computing is a type of distributed computing, consisting of a super virtual computer composed of a group of loosely coupled computers. This embodiment does not impose any limitations on this. The multi-source heterogeneous fair competition review data processing method provided in this application embodiment includes steps S1 to S5, as follows:

[0076] S1, perform structured parsing on the document to be reviewed and historical documents, and extract the first structural feature information corresponding to each of the document to be reviewed and historical documents.

[0077] It should be noted that, to adapt to the continuously increasing number of documents awaiting review and historical documents, and the increasingly routine review process, in some possible implementations, the methods described in this application can be deployed within a system capable of document parsing, knowledge management, intelligent review, data retrieval, and model iteration. This system can be architecturally divided into a user-interactive presentation layer, an application service layer carrying business logic, an intelligent engine layer carrying core algorithm processing, a data resource layer carrying various types of data storage, and an infrastructure layer providing the operating environment.

[0078] The presentation layer provides interactive interfaces such as the review workbench and knowledge retrieval portal; the application service layer handles business logic such as document parsing and preprocessing, review task scheduling, and knowledge base management; the intelligent engine layer carries the core processing steps such as structured parsing of documents to be reviewed and historical documents, establishing competitive connection structures, identifying changes in connection edges, and identifying connection descent objects; the data resource layer stores structured data such as the business entity naming dictionary, competitive constraint keyword library, and historical comparison archive records; and the infrastructure layer provides the computing and storage resources required for the operation of the above layers.

[0079] The processing described in steps S1 to S5 below is mainly carried out by the intelligent engine layer in the system architecture; among them, the chapter structure parsing and semantic entity recognition processing in step S1 can be performed by the service module in the system architecture that undertakes the file parsing preprocessing function, and the processing result is passed to the intelligent engine layer for use in subsequent steps.

[0080] In some possible implementations, before executing step S1, a pre-trained classification model can first determine the nature of the document to be reviewed, whether the document belongs to the category of government policy or bidding documents, and whether it involves the economic activities of market entities; for documents whose judgment result is not within the above scope, the processing described in steps S1 to S5 will not be performed; for documents whose judgment result is within the above scope, the structured parsing and competitive connection structure comparison processing described in steps S1 to S5 will be performed.

[0081] In some embodiments, step S1 includes steps S11 to S16:

[0082] Step S11: Perform chapter structure parsing and semantic entity recognition on the document to be reviewed to obtain the first document segmentation information and first entity information corresponding to the document to be reviewed;

[0083] Step S12: Perform chapter structure parsing and semantic entity recognition on the historical files to obtain the second file segment information and second entity information corresponding to the historical files;

[0084] Step S13: Based on the first file segmentation information and the first entity information, perform association mapping to generate the first object mapping information corresponding to the file to be reviewed;

[0085] Step S14: Based on the second file segmentation information and the second entity information, perform association mapping to generate the second object mapping information corresponding to the historical file;

[0086] Step S15: Perform co-reference resolution on the entity items in the first object mapping information and the entity items in the second object mapping information, merge the co-referenced entity items, and obtain the merged object mapping information and constraint action information;

[0087] Step S16: The merged object mapping information and constraint action information are used as the first structural feature information.

[0088] The terms used in the above steps are explained as follows: First document segmentation information refers to the set of paragraphs and their corresponding chapter numbers and titles after the document under review has been divided into several paragraph units; First entity information refers to the set of business entity names identified from each paragraph of the document under review and their corresponding character positions within the paragraphs; First object mapping information refers to the set of records formed by associating the business entity name with its corresponding paragraph unit number and the text describing the business entity's position or limitations within that paragraph; Constraint information refers to the set of text extracted from the object mapping information that specifically describes the limitations imposed on the business entity; Co-reference resolution refers to the process of identifying and merging multiple records with different wording but actually referring to the same business entity into one. Second document segmentation information, second entity information, and second object mapping information are similar information corresponding to historical documents, and their composition is the same as the information corresponding to the document under review.

[0089] The names of business entities and their corresponding restrictions in documents are often interspersed across different paragraphs and do not appear in a single location. Without first segmenting the text by chapter and labeling the names of business entities, subsequent processing will lack operable objects. Therefore, it is necessary to first convert the natural paragraph structure of the document into numbered paragraph units, then identify the names of business entities from the paragraph text and determine their locations, and finally establish a correspondence between the names and the restrictive text to form a structured mapping record. Since the same business entity may be written in multiple ways in different clauses, it is also necessary to merge records referring to the same business entity through coreference resolution to avoid treating the same business entity as multiple different objects in subsequent processing.

[0090] In the specific implementation plan, the documents to be reviewed are parsed according to their title font size, numbering format, and indentation level to identify the start and end positions of each title. The text content between two adjacent titles is divided into a paragraph unit, and each paragraph unit is accompanied by a corresponding chapter number and title name. Then, the text in each paragraph unit is read one by one. By comparing the pre-stored business entity naming dictionary with the strings in the paragraph text, the full name, abbreviation, or organization name with industry suffix appearing in the paragraph is identified. The identified name and its character position in the paragraph are recorded together, thus obtaining the first document segmentation information and the first entity information. Historical documents are processed in the same way to obtain the second document segmentation information and the second entity information. Then, each business entity name recorded in the first entity information is traversed. According to the character position of the name in the paragraph, the paragraph unit number to which the position belongs is found in the first document segmentation information. The business entity name, the found paragraph unit number, and the text content in the paragraph unit describing the business entity's position or limitations are combined into a mapping record. All mapping records constitute the first object mapping information. The second object mapping information for historical documents is generated in the same way. Extract the business entity name strings corresponding to each mapping record in the first and second object mapping information. Count the number of identical Chinese characters in the two strings, i.e., the size of the intersection of the Chinese characters contained in the two strings. Calculate the character overlap ratio by comparing this value with the shorter character length of the two strings. When the character overlap ratio is higher than a preset resolution threshold, it is determined that the two mapping records refer to the same business entity, and they are merged into a unified mapping record, retaining the paragraph unit number and limited text content associated with each of the two records before the merger. When the character overlap ratio is not higher than the preset resolution threshold, the two mapping records remain independent.

[0091] The preset resolution threshold is used to distinguish between different spellings of the business entity name and accidental similarities between different business entity names. Those skilled in the art can select the threshold in practice based on the common abbreviations and name duplications of business entity names in the industry to which the document to be processed belongs. This embodiment does not impose specific limitations on this.

[0092] For example, the preset resolution threshold can be preferably set to 80%, meaning that when the character overlap ratio reaches 80% or more, it is determined that two mapping records refer to the same business entity. In practical applications, for industries where abbreviations and trade names are commonly used in business entity names, the preset resolution threshold can be appropriately reduced (e.g., reduced to 70%) to improve the ability to merge and identify different spellings of names; for situations where business entity names in the same industry are similar and the risk of name duplication is high, the preset resolution threshold can be appropriately increased (e.g., increased to 85%) to avoid different business entities being mistakenly merged.

[0093] After performing the above comparison and merging process on all mapping records, the merged object mapping information is obtained. The textual content describing the restrictions imposed on the business entity in the mapping records is extracted as constraint information. The two together constitute the first structural feature information.

[0094] For example, if a clause states that a certain logistics company has the sole delivery qualification within a certain region, chapter analysis can be used to assign the paragraph containing that sentence to the corresponding paragraph unit. A dictionary of business entity names can then be used to identify the name of the logistics company, generating a mapping record containing the business entity's name, the paragraph unit number, and the limiting phrase "sole delivery qualification." If another clause in a historical document states that a certain logistics company has the sole delivery right in the region, comparing the two name strings ("certain logistics company") reveals a high degree of character overlap, indicating that they refer to the same business entity. Therefore, the two mapping records are merged, ensuring that the limiting content related to this business entity in both the document under review and the historical document can be associated with the same entity.

[0095] The beneficial effect of the above processing is that, by simply reading and comparing the text of each clause paragraph by paragraph, the name of the business entity and the restrictions it is subject to are often scattered across different paragraphs of the document. Furthermore, the same business entity may be fragmented into multiple records due to different wording, making it difficult to summarize all the restrictive relationships involving a particular business entity without omission or duplication. Step S1 extracts the name of the business entity and its restrictive text through chapter parsing and semantic entity recognition, and uses coreference resolution to merge multiple records of the same business entity generated under different wordings into one. This transforms the originally scattered and fragmented business entity information and restrictive information into structured data in a unified format that can be further processed by the program, providing a foundation for establishing a comparable business entity relationship structure.

[0096] In some embodiments, step S1 further includes steps A1 to A6:

[0097] Step A1: Parse the clause hierarchy structure of the document to be reviewed to obtain the first clause hierarchy tree of the document to be reviewed;

[0098] Step A2: Parse the clause hierarchy structure of the historical documents to obtain the second clause hierarchy tree of the historical documents;

[0099] Step A3: Extract the clause nodes containing the keywords of competing constraints from the first clause hierarchy tree and mark them as the first competing clause nodes;

[0100] Step A4: Extract the clause nodes containing the competing constraint keywords from the second clause hierarchy tree and mark them as second competing clause nodes;

[0101] Step A5: Generate a first set of constraint action paths based on the clause hierarchy path of the first competing clause node, and generate a second set of constraint action paths based on the clause hierarchy path of the second competing clause node;

[0102] Step A6: Merge the first constraint action path set and the second constraint action path set into the constraint action information.

[0103] In the above steps, the first clause hierarchy tree refers to a tree structure constructed based on the subordinate relationship of clause numbers, with each node in the tree corresponding to a specific clause; the competitive constraint keyword refers to a pre-defined set of terms used to express limiting statements; the first competing clause node refers to a clause node whose text content contains the competitive constraint keyword; the constraint action path refers to the path formed by sequentially arranging the clause numbers of each level from the root node of the clause hierarchy tree to a certain competing clause node.

[0104] The same type of limiting content may appear in clauses at different levels of a document. For example, it may be recorded in the main text of a clause or in a subordinate clause of the same clause. Simply recording the limiting text itself without recording its hierarchical position makes it difficult to distinguish the limiting relationship from different levels. Therefore, it is necessary to first establish the hierarchical relationship between clauses, then filter out the clause nodes containing limiting keywords, and record the path of that node in the hierarchical structure.

[0105] In the specific implementation plan, the clause numbers themselves have a subordinate relationship. For example, the number of a certain item belongs to the number of the clause it belongs to. Based on this, the system performs hierarchical identification on the clause numbers appearing in the document to be reviewed, and constructs the first clause hierarchy tree. Each node in the tree corresponds to a specific clause, and the parent-child relationship between nodes reflects the hierarchical affiliation of the clause in the document. Historical documents are processed in the same way to obtain the second clause hierarchy tree.

[0106] A set of keywords is predefined to represent the meaning of competitive constraints, corresponding to common restrictive expressions in documents, such as terms indicating exclusive arrangements, terms specifying specific trading partners, and terms restricting the scope of business. The specific trading partner referred to here is a specific business entity or a type of business entity explicitly named in the text of the terms.

[0107] It should be noted that this set of keywords was pre-constructed by summarizing and organizing restrictive expressions appearing in historical documents and relevant clauses involved in previous fair competition reviews, and categorized into three types according to their expressed restrictive meanings: keywords related to exclusive arrangements, corresponding to terms that exclude other business entities from participating in a certain business activity; keywords related to designated trading partners, corresponding to terms that limit trading partners to a specific business entity or a specific type of business entity; and keywords related to restricted business scope, corresponding to terms that limit the business content, operating area, or operating methods that a business entity can engage in. This set of keywords can be supplemented and updated as subsequent documents are processed.

[0108] Each clause node in the first clause hierarchy tree is checked to see if its text content contains the set of keywords. If it does, it is marked as the first competing clause node; otherwise, it is left unmarked. The second clause hierarchy tree is processed in the same way to obtain the second competing clause nodes. For each first competing clause node, the clause numbers corresponding to each level node along the path from the root node to that node in the clause hierarchy tree are recorded sequentially to form a constraint action path. All paths constitute the first constraint action path set. The second competing clause nodes are processed to obtain the second constraint action path set. The two sets are merged and written into the constraint action information.

[0109] For example, if the aforementioned limiting wording of having the sole delivery qualification is recorded in Article 3, Paragraph 2 of a document, when examining the node of that clause, if it is identified that it contains the competitive constraint keyword of "solely", the node is marked as a competitive clause node, and the numbers of Article 3, Paragraph 2 that are passed through in sequence along the path of the hierarchy tree are recorded to form the corresponding constraint action path, so that the limitation is not only recorded in textual form, but can also be traced back to its specific hierarchical position in the document.

[0110] The beneficial effect of the above processing is that simply recording the textual content of the limitation without recording its hierarchical position in the clause makes it difficult to distinguish whether similar limitations are recorded in the main clauses of the document or in subordinate clauses, and also makes it difficult to locate the specific source of the limitation in the document structure during subsequent comparison. Steps A1 to A6, through clause hierarchy parsing and generation of binding action paths, record not only the content of the limitation but also its hierarchical source, providing a more complete structural basis for subsequent processing.

[0111] S2, establish a competitive connection structure based on the object mapping information and constraint action information in the first structural feature information.

[0112] In some embodiments, step S2 includes steps S21 to S25:

[0113] Step S21: Extract the set of business entity nodes from the object mapping information, and treat each business entity node as a graph node in the competitive connection structure;

[0114] Step S22: Extract the competitive constraint relationship between business entities from the constraint information, and treat each competitive constraint relationship as a competitive connection edge connecting two graph nodes;

[0115] Step S23: Construct a competitive connection graph using graph nodes and competing connection edges, and use the competitive connection graph as the first competitive connection structure;

[0116] Step S24: Assign an edge identifier to each competing connection edge in the competing connection graph. The edge identifier is used to distinguish different types of competing constraint relationships.

[0117] Step S25: Based on the edge identifier, the competing connection edges in the competing connection graph are layered to obtain multiple competing connection subgraphs, and the multiple competing connection subgraphs are used together as the competing connection structure.

[0118] The terms used in the above steps are explained below. A graph node is a structural unit generated for each business entity in the object mapping information, carrying the name of that business entity as its identifier. A competing connection edge is a connection established between the graph nodes corresponding to two business entities when a constraint text in the constraint information involves two business entities. An edge identifier is a category label assigned to a competing connection edge based on the keyword type of the constraint text, used to distinguish whether the constraint properties represented by the connecting edges are the same. A competing connection subgraph is a local structure composed of a group of competing connection edges with the same edge identifier and the graph nodes they connect.

[0119] Although the names and limiting text of the business entities have been extracted into structured mapping records in step S1, these records are still listed separately for each business entity, failing to reflect the relationships between them or the differences between different types of limiting relationships. Without converting business entities into interconnectable nodes and limiting relationships into concrete connecting edges, it is impossible to perform a comprehensive comparison of the relationships between business entities in the document under review and historical documents. Therefore, it is necessary to convert the business entities in the object mapping information into graph nodes, convert the limiting relationships involving two business entities in the constraint information into edges connecting these two nodes, and differentiate and stratify the connecting edges according to the type of limiting relationship.

[0120] In the specific implementation plan, all entity items corresponding to the business entities are extracted from the object mapping information. Each entity item is converted into a graph node in the competitive connection structure, with the graph node carrying the business entity name as the node identifier. Each constraint text recorded in the constraint action information is traversed, and it is determined whether the number of business entities involved in the constraint text is two or more. If so, a competitive connection edge is established between the graph nodes corresponding to these two business entities. If there are fewer than two, meaning the constraint text involves only a single business entity, no competitive connection edge is generated, and the constraint text is retained in the constraint action information for use in other steps, without participating in the construction of the graph structure. All graph nodes and all competitive connection edges are combined into a graph structure, which is used as the first competitive connection structure. Based on the keyword type appearing in the constraint text corresponding to each competitive connection edge, an edge identifier is assigned to that connection edge. A specific transaction object type constraint corresponds to one edge identifier value, a business scope restriction type constraint corresponds to another edge identifier value, and an exclusive arrangement type constraint corresponds to yet another edge identifier value.

[0121] The aforementioned specific transaction object type, restricted business scope type, and exclusive arrangement type correspond to the three categories of keywords in the aforementioned competition constraint keywords: specified transaction object type, restricted business scope type, and exclusive arrangement type, respectively. The edge identifier of a competitive connection edge is determined by the category of the keyword contained in the limiting text. This achieves a one-to-one correspondence between keyword type and edge identifier value.

[0122] The competing edges in the first competing connection structure are grouped according to their edge identifiers. Edges with the same identifier are grouped together. Each group of edges and the graph nodes they connect to constitute a competing connection subgraph. All competing connection subgraphs together form the competing connection structure. The second competing connection structure corresponding to the historical files is generated in the same way. Both the first and second competing connection structures can be stored in the graph database of the aforementioned data resource layer, persistently saved as graph nodes and their connection relationships, for subsequent steps to read and compare, and for retrieval in other analysis scenarios.

[0123] For example, one clause states that there is an exclusive arrangement between business entity A and business entity B, while another clause states that there is a restriction on designated trading partners between business entity A and business entity C. Based on this, graph nodes are generated for A, B, and C respectively. An edge labeled as an exclusive arrangement is established between A and B, and an edge labeled as a designated trading partner is established between A and C. Because the edge labels of the two edges are different, they are classified into different competing connection subgraphs, thus structurally distinguishing the exclusive arrangement relationship from the designated trading partner relationship, preventing them from being confused as the same type of connection.

[0124] The beneficial effect of the above processing is that simply examining the limiting text line by line based on the mapping records only reveals what kind of limitation a particular business entity is subject to in a particular clause, but it cannot directly reveal the overall relationship between that business entity and other business entities, nor can it distinguish the structural differences between different types of limiting relationships. Steps S21 to S25 transform the business entities and their mutual limiting relationships from the form of clause text into a structure composed of graph nodes and connecting edges, and layer the connecting edges according to the limiting type. This gives the limiting relationships, which could originally only be read and understood line by line, a structural form that can be compared as a whole, providing a unified structural basis for the matching of the relationship between the document to be reviewed and historical documents.

[0125] In some embodiments, step S2 further includes steps B1 to B6:

[0126] Step B1: Obtain the business category label for each business entity node from the object mapping information;

[0127] Step B2: Obtain the action type label for each competing constraint relationship from the constraint action information;

[0128] Step B3: Group the graph nodes based on the business category labels to obtain multiple business category node groups;

[0129] Step B4: Group the competing connection edges based on the action type label to obtain multiple action type edge groups;

[0130] Step B5: For each business category node group, calculate the intra-group connection density of the business category node group based on the number of competing connection edges between graph nodes within the business category node group.

[0131] Step B6: Mark the business category node group with a connection density higher than the first density condition as a high-density competition group, and use the information of the high-density competition group as the group structure feature of the competition connection structure.

[0132] In the above steps, the business category label refers to the category marker assigned to the business entity node based on industry terms contained in the business entity's name or the business scope mentioned in the context of its paragraph. The function type label corresponds to the limited keyword type upon which the aforementioned edge identifier is based. The intra-group connection density refers to the ratio between the actual number of competing connection edges within a business category node group and the total number of connection edges that can be formed by pairwise combinations of graph nodes within that group. A high-density competitive group refers to a business category node group whose intra-group connection density reaches a preset condition.

[0133] The density of binding relationships between business entities in different business categories may vary. Some business categories may have denser binding relationships, while others may have more dispersed relationships. Simply counting the number of edges without distinguishing their respective business categories fails to reflect the clustering of binding relationships across different business areas. Therefore, it is necessary to first group graph nodes by business category, and then calculate the proportion of edges within each group relative to the total number of possible edges. This will allow us to measure the density of binding relationships within that business category.

[0134] In the specific implementation plan, a business category label is assigned to each business entity node based on industry terms contained in the business entity's name or the business scope mentioned in the context of the paragraph in the object mapping information. The action type label corresponding to each competing constraint relationship is read from the constraint action information. Graph nodes are grouped according to their business category labels, with nodes having the same business category label grouped into the same group, resulting in multiple business category node groups. Competing connection edges are grouped according to their action type labels, resulting in multiple action type edge groups.

[0135] For each business category node group, the number of actual competing edges between nodes in the graph within that group is counted. Then, the total number of edges that can be formed by pairwise combinations of nodes within that group is calculated. The actual number of edges is compared to the total number of possible edges to obtain a ratio, which is determined as the intra-group connection density for the corresponding business category node group. Business category node groups with an intra-group connection density higher than a preset first density condition are marked as high-density competing groups; business category node groups with a density lower than this condition are not marked.

[0136] The preset first density condition is used to distinguish between situations where the limiting relationships between business entities within a group are significantly denser than those outside the group, and situations where the limiting relationships are more evenly distributed. Those skilled in the art can select this density condition based on the general number of business entities in the industry involved in the processed document; this embodiment does not impose specific limitations on this. For example, the preset first density condition can preferably be set to 70%. In practical applications, for industries with a large number of business entities and a large total number of possible connecting edges within the group, the preset first density condition can be appropriately reduced, such as to 60%; for industries with a small number of business entities and a small total number of possible connecting edges within the group, the preset first density condition can be appropriately increased, such as to 80%. The scope of business entities and the connecting density value corresponding to the high-density competitive group are recorded as group structure features of the competitive connection structure. This group structure feature, as additional grouping dimension information in the competitive connection structure, together with graph nodes and competitive connecting edges, constitutes a component of the competitive connection structure, used to reflect the aggregation of limiting relationships within a specific business category in the final generated fair competition review result. The recorded content of this group structure feature does not participate in the core judgment processing regarding changes in connecting edges and objects of connection decline in steps S3 to S5.

[0137] For example, a business category node group contains five business entity nodes. The total number of possible connection edges formed by pairwise combinations of these five nodes is ten, of which eight are actually competing connection edges. Based on this, the connection density within the group is calculated to be high. If this density is higher than the preset first density condition, the business category node group is marked as a high-density competitive group, indicating that the limiting relationships between the business entities within the business category are relatively concentrated.

[0138] The beneficial effect of the above processing is that simply counting the total number of edges cannot reflect the differences in the degree of clustering of constraint relationships within different business categories, nor can it distinguish the business areas where constraint relationships are more concentrated in subsequent analysis. Steps B1 to B6, by grouping by business category and calculating the connection density within each group, enable the competitive connection structure to further possess group structure characteristics that can reflect the clustering of constraint relationships within different business areas, based on graph nodes and edges.

[0139] Step S3: Perform association matching between the first competing connection structure corresponding to the document to be reviewed and the second competing connection structure corresponding to the historical documents to identify the information on changes in competing connection edges.

[0140] In some embodiments, step S3 includes steps S31 to S36:

[0141] Step S31: Pair each graph node in the first competitive connection structure with the graph nodes in the second competitive connection structure that have the same business entity identifier to obtain the node pairing result;

[0142] Step S32: Based on the node pairing results, pair each first competitive connection edge in the first competitive connection structure with the second competitive connection edge between the corresponding node pairs in the second competitive connection structure to obtain the edge pairing results;

[0143] Step S33: In response to the edge pairing result that the edge identifiers of the first competing connection edge and the second competing connection edge are consistent, mark the first competing connection edge as an invariant edge;

[0144] Step S34: In response to the edge pairing result that the first competing connection edge exists while the second competing connection edge does not exist, the first competing connection edge is marked as a newly added edge;

[0145] Step S35: In response to the edge pairing result that the first competing connection edge does not exist while the second competing connection edge does exist, mark the second competing connection edge as a disappeared edge;

[0146] Step S36: Collect the edge identifiers of newly added edges and disappearing edges, as well as the node identifiers of the connected node pairs, and generate information on changes in competing connection edges.

[0147] In the above steps, node pairing refers to the process of matching graph nodes in the first and second competing connection structures that identify the same business entity. Edge pairing refers to the process of matching connecting edges in the first competing connection structure with connecting edges in the second competing connection structure that connect the same pair of nodes, based on the node pairing results. Unchanging edges are connecting edges whose edge identifiers are confirmed to be consistent through edge pairing, indicating that the corresponding limiting relationships remain consistent in both documents. Newly added edges are connecting edges that exist only in the first competing connection structure, indicating that the corresponding limiting relationships are new content appearing in the document under review. Disappearing edges are connecting edges that exist only in the second competing connection structure, indicating that the corresponding limiting relationships no longer appear in the document under review.

[0148] After the competitive connection structure is established, the document under review and the historical documents correspond to two independent graph structures. Without mapping nodes identifying the same business entity in both structures, it's impossible to determine whether a certain limiting relationship has changed in the two documents. Simply comparing the similarity of clause text only yields conclusions about the similarity of individual clauses, failing to accurately reflect whether the limiting relationships between business entities maintain consistency in the overall structure. Therefore, it's necessary to first establish the mapping between nodes, and then, based on this mapping, compare the existence of connecting edges and the consistency of edge identifiers, thereby converting changes at the clause level into clearly defined edge states at the structural level.

[0149] In the specific implementation plan, all graph nodes in the first competitive connection structure are traversed. For each graph node, a graph node with the same business entity identifier is searched in the second competitive connection structure. When found, these two graph nodes are paired together, and all search results constitute the node pairing result. If no graph node with the same identifier is found in the second competitive connection structure, the graph node is not paired temporarily, and this does not affect the generation of the remaining node pairing results. Based on the node pairing results, edge pairing processing is performed on each competitive connection edge in the first competitive connection structure. The two graph nodes connected by the edge are extracted, and the corresponding paired nodes in the second competitive connection structure are found according to the node pairing results. It is determined whether there is a competitive connection edge between these two paired nodes in the second competitive connection structure. If there is, the edge in the first competitive connection structure and the edge found in the second competitive connection structure are combined to form an edge pairing record, and all edge pairing records constitute the edge pairing result. Each edge pairing record is compared. If the edge in the first competitive connection structure and the corresponding edge in the second competitive connection structure have the same edge identifier, the edge in the first competitive connection structure is marked as an invariant edge. If a competing connection edge exists in the first competing connection structure but no connecting edge is found between the corresponding node pairs in the second competing connection structure, then that connecting edge in the first competing connection structure is marked as a newly added edge. If a competing connection edge exists in the second competing connection structure but no connecting edge is found between the corresponding node pairs in the first competing connection structure, then that connecting edge in the second competing connection structure is marked as a disappeared edge.

[0150] In some embodiments, step S36 includes steps S361 to S366:

[0151] Step S361: Count the frequency of occurrence of each business entity node connected by the newly added edge and the disappearing edge, respectively;

[0152] Step S362: Mark the nodes of the operating entities that appear more frequently than the first preset frequency condition in the newly added edges as newly active nodes;

[0153] Step S363: Mark the business entity nodes in the disappearing edges that appear more frequently than the second preset frequency condition as disappearing active nodes;

[0154] Step S364: Generate a new change record based on the node identifier of the newly added active node and the edge identifier of the newly added edge;

[0155] Step S365: Generate a disappearance change record based on the node identifier of the disappearing active node and the edge identifier of the disappearing edge;

[0156] Step S366: Combine the newly added change records and the disappeared change records into the competing connection edge change information.

[0157] Statistical analysis is performed on all newly added edges and all disappeared edges separately, recording the number of times each business entity node is connected to a newly added edge and the number of times it is connected to a disappeared edge, thus obtaining the frequency of newly added edges and the frequency of disappeared edges for each business entity node. Business entity nodes whose frequency of newly added edges exceeds a preset first frequency condition are marked as newly active nodes, while those below this condition are not marked. Similarly, business entity nodes whose frequency of disappeared edges exceeds a preset second frequency condition are marked as disappeared active nodes, while those below this condition are not marked. The preset first and second frequency conditions are used to distinguish between a business entity node that involves only a few changes in limiting relationships and a business entity node that changes in multiple limiting relationships. Those skilled in the art can select the above two frequency conditions based on the average number of limiting relationships involved in the business entity in the file to be processed. New change records are generated based on the node identifier of newly active nodes and the edge identifiers of their associated newly added edges, and disappearance change records are generated based on the node identifier of disappeared active nodes and the edge identifiers of their associated disappearing edges. The new change records and disappearance change records are combined into competitive connection edge change information.

[0158] For example, if business entity A has new limiting relationships with business entities B, C, and D in the documents under review, generating three new edges, and business entity A appears three times in these new edges, then if the preset first frequency condition is twice, and business entity A's frequency exceeds this condition, business entity A will be marked as a newly active node and included in the new change record. However, business entity E is only involved in one new edge, and its frequency does not exceed the preset first frequency condition, so it will not be marked as a newly active node.

[0159] The beneficial effect of the above processing is that relying solely on the similarity of clause text can only determine whether individual clauses have changed, but cannot determine whether the overall corresponding structure of the limiting relationships between business entities remains consistent, nor can it filter out business entities with concentrated changes from numerous changes. Step S3 compares the limiting relationships in the corresponding positions in the first and second competitive connection structures one by one through node pairing and edge pairing, converting clause changes into three distinct edge states: unchanged edges, newly added edges, and disappeared edges. By frequency statistics, it filters out business entities where limiting relationship changes occur frequently, so that the changes in the limiting relationship level of the document under review relative to the historical document are recorded with specific edge and node identifiers, providing a direct basis for subsequently determining which business entities are affected by the changes in limiting relationships.

[0160] Step S4: Analyze the changes in node connection paths in the competitive connection structure based on the information on changes in competitive connection edges, and determine the objects with reduced connections.

[0161] In some embodiments, step S4 includes steps S41 to S45:

[0162] Step S41: Based on the edge identifier of the disappeared edge and the node identifier of the connected node pair in the information on changes in the competitive connection edge, locate the first node pair corresponding to the removed competitive connection edge in the first competitive connection structure.

[0163] Step S42: Starting from each node in the first node pair, traverse the node connection path along the reserved competitive connection edge in the first competitive connection structure to obtain the first node connection path set.

[0164] Step S43: Starting from each node in the first node pair, traverse the node connection paths along all competing connection edges in the second competitive connection structure to obtain the set of second node connection paths.

[0165] Step S44: Compare the first node connection path set and the second node connection path set, and extract the missing node connection paths that exist in the second node connection path set but do not exist in the first node connection path set.

[0166] Step S45: Extract the operating entity nodes that did not appear in the first competitive connection structure from the connection path of the disappeared node as connection reduction objects.

[0167] In the above steps, the first node pair refers to the corresponding node combination of the two business entity nodes connected by the disappearing edge in the first competitive connection structure. This represents two business entities that originally had a limiting relationship in the historical file but whose limiting relationship no longer exists in the file under review. A node connection path refers to a sequence of nodes that starts from a starting node, passes through several nodes along the competitive connection edge, and reaches a certain ending node. A disappearing node connection path refers to a node connection path that exists in the second competitive connection structure but does not have a corresponding reachable path in the first competitive connection structure. A connection reduction object refers to a business entity node that appears in the disappearing node connection path but does not appear in the first competitive connection structure.

[0168] When a boundary relationship disappears, focusing only on the two business entities directly connected by that relationship makes it impossible to determine whether this change will affect other business entities indirectly related to those two entities. The second competing connection structure in the historical files contains all connections before the disappearing edge was removed, reflecting the paths that business entities could originally reach each other through several connecting edges. If a business entity was originally reachable via a node connected by a disappearing edge, but is no longer reachable through any path in the first competing connection structure after the disappearing edge is removed, it indicates that the original indirect connection between that business entity and the business entities connected by the disappearing edge has been simultaneously interrupted. Therefore, it is necessary to trace the node connection paths in both competing connection structures separately and determine the indirectly affected business entities by comparing the reachability of the paths.

[0169] In the specific implementation plan, based on the identifiers of the vanishing edges and the identifiers of the two nodes they connect, recorded in the information on changes in competing connections, the graph nodes corresponding to these two node identifiers are located in the first competing connection structure, and the node pair formed by these two graph nodes is determined as the first node pair. Starting from each node in the first node pair, the remaining competing connection edges are traversed layer by layer in the first competing connection structure, recording the sequence of nodes reachable from the starting point via several connection edges. Each node sequence constitutes a node connection path, and all node connection paths constitute the first node connection path set. Starting from each node in the first node pair, the same traversal is performed along all competing connection edges in the second competing connection structure, and the resulting node connection paths constitute the second node connection path set.

[0170] In some embodiments, step S44 includes steps S441 to S447:

[0171] Step S441: Obtain the first node connection path in the set of second node connection paths as the current comparison path;

[0172] Step S442: Search the first node connection path set for candidate paths that match the start node identifier and end node identifier of the current comparison path.

[0173] Step S443: In response to finding a candidate path, compare the intermediate node sequence of the currently compared path with the intermediate node sequence of the candidate path.

[0174] Step S444: In response to the inconsistency in the intermediate node sequence, mark the current alignment path as a structural change path;

[0175] Step S445: In response to the consistency of the intermediate node sequence, mark the current alignment path as a structure-preserving path;

[0176] Step S446: In response to the absence of a candidate path, mark the current comparison path as a path connecting a missing node.

[0177] Step S447: Traverse all node connection paths in the second node connection path set to obtain all disappeared node connection paths.

[0178] The first node connection path is retrieved from the second set of node connection paths and designated as the current comparison path. Based on the start and end node identifiers of the current comparison path, paths with identical start and end node identifiers are searched in the first set of node connection paths, and these paths are designated as candidate paths. If a candidate path is found, the sequence of nodes traversed by the current comparison path is compared with the sequence of nodes traversed by the candidate path. If the intermediate node sequences are inconsistent, the current comparison path is marked as a structural change path, indicating that although a reachable path still exists between the two business entities, the intermediate business entity traversed by the reachable path has changed. If the intermediate node sequences are consistent, the current comparison path is marked as a structural preservation path, indicating that the reachable path between the two business entities is not affected by the disappearing edge. If no candidate path is found, the current comparison path is marked as a disappearing node connection path, indicating that the two business entities connected by this path in the historical file no longer have a corresponding reachable path in the first competing connection structure. The above search and marking process is performed sequentially on each node connection path in the second set of node connection paths to obtain all disappearing node connection paths. Extract each business entity node traversed by the path from all the missing node connection paths, and determine whether each of these business entity nodes appears in the first competitive connection structure. Business entity nodes that do not appear in the first competitive connection structure are identified as connection reduction objects. Business entity nodes that have already appeared in the first competitive connection structure are not considered connection reduction objects because although the business entity is in the missing path, it still has other connection methods to connect with the outside in the first competitive connection structure.

[0179] For example, the original limiting relationship between business entity A and business entity B is marked as a vanishing edge, and A and B form the first node pair. In the second competitive connection structure, starting from A and passing through C, one can reach E, forming a node connection path. In the first competitive connection structure, a path with A as the starting point and E as the ending point is searched. If no such path is found, it is marked as a vanishing node connection path. Further, it is determined whether E appears in the first competitive connection structure. If E does not appear, E is identified as a connection reduction object, indicating that the original indirect relationship between E and A is interrupted due to the vanishing edge.

[0180] like Figure 2The image shows an example of the competitive connection edges between four business entity nodes (A, B, C, and E) in the second competitive connection structure corresponding to the historical file. There is one competitive connection edge between A and B, another between A and C, and a third between C and E. Therefore, A can reach E via C along the retained connection edge, forming a complete node connection path. This second competitive connection structure reflects the complete relationships that originally existed between the business entities in the historical file before the disappearing edges were removed, providing a benchmark for determining whether these relationships have changed in the first competitive connection structure.

[0181] like Figure 3 The diagram illustrates the connection state after the original competing connection edge between A and B is identified as a disappeared edge in the first competitive connection structure corresponding to the document under review. In the first competitive connection structure, there is no longer a competing connection edge between A and B. Simultaneously, the connection edge between C and E, which A relied upon to reach E via C, also no longer exists. This means that E, which was originally reachable via C in the historical document, is no longer reachable through any retained connection edge in the first competitive connection structure. Therefore, E is identified as a connection reduction object. The dashed circle in the diagram indicates that E no longer has a valid connection in the first competitive connection structure, visually demonstrating the result of the change in the limiting relationship radiating outwards along the original associated path, thereby determining the scope of affected business entities.

[0182] The beneficial effect of the above processing is that judging the change in the limiting relationship solely based on the two business entities directly connected by the disappearing edge cannot determine whether the change will affect other business entities related to these two business entities. Step S4 does not stop at the level of whether there is a change in the direct limiting relationship between the first node pair, but traces the node connection paths along the remaining connection edges and all connection edges in the two competing connection structures respectively. By comparing the reachability of the paths in the two structures and the intermediate nodes traversed by the paths, it identifies the business entities whose original indirect connections are simultaneously interrupted due to the disappearing edge. This makes the scope of business entities affected by the change in the limiting relationship no longer limited to the two business entities directly connected by the disappearing edge, but can be determined outward along the original connection path.

[0183] Step S5: Identify the set of objects with decreasing connectivity based on the objects with decreasing connectivity, and output the fair competition review results corresponding to the set of objects with decreasing connectivity.

[0184] In some embodiments, step S5 includes steps S51 to S55:

[0185] Step S51: Using each business entity node in the connection reduction object as a seed node, extend the preset path steps outward along the competitive connection edge in the second competitive connection structure to obtain the set of extended nodes corresponding to each seed node;

[0186] Step S52: Take the union of the sets of extended nodes corresponding to all seed nodes, and use the union as the set of candidate connection descent nodes.

[0187] Step S53: Obtain the sequence of changes in the number of competing connection edges for each business entity node in the candidate connection descent node set within a preset time window;

[0188] Step S54: Select the operating entity nodes whose competitive connection edge quantity change sequence shows a downward trend from the candidate connection descent node set, and use them as the connection descent object set;

[0189] Step S55: Generate fair competition review results based on the node identifiers of each operating entity node in the set of connected descending objects and the sequence of changes in the number of competing connection edges.

[0190] In the above steps, the seed node refers to each business entity node in the connection reduction object, serving as the starting point for outward expansion. The expansion node set refers to the set of all nodes reachable from the seed node via no more than a preset number of connecting edges. The candidate connection descent node set is the union of the expansion node sets corresponding to all seed nodes. The sequence of changes in the number of competing connecting edges refers to the numerical sequence formed by arranging the number of competing connecting edges for a certain business entity node in chronological order at multiple time points within a preset time window.

[0191] The decrease in the number of connections only reflects the business entities directly and indirectly affected by the disappearing edges. Whether the number of connections for a particular business entity is continuously decreasing requires observation of changes over a period of time. Based solely on a single comparison result, it is impossible to distinguish whether the decrease in the number of connections is a continuous trend or an occasional fluctuation. Therefore, it is necessary to expand the candidate range outward from the object of connection decrease, and within the candidate range, obtain the sequence of changes in the number of connections for each business entity's nodes over time, and then filter out the business entities that truly show a downward trend.

[0192] In the specific implementation plan, each business entity node in the connection reduction object is determined as a seed node. Starting from the seed node, the connection structure expands outward layer by layer along the competitive connection edge in the second competitive connection structure. The number of connection edge layers traversed during the expansion is the preset path step number. All nodes that can be reached from the seed node through no more than the preset path step number of connection edges are determined as the set of expansion nodes corresponding to the seed node.

[0193] The preset path steps are used to determine the size of the expansion range. A value that is too small may overlook indirectly affected business entities, while a value that is too large will include business entities with weak connections. Those skilled in the art can select the preset path steps based on the common hierarchical depth of the relationship transmission between business entities in the file to be processed; this embodiment does not impose specific limitations on this. For example, the preset path steps can preferably be set to 2 steps, meaning the expanded node set covers business entity nodes directly connected to the seed node and those indirectly connected via an intermediate node. In practical applications, for industries with deep hierarchical relationships between business entities, the preset path steps can be appropriately increased, such as to 3 steps; for cases with shallow hierarchical relationships and relatively direct connections between business entities, the preset path steps can also be set to 1 step. The union of the expanded node sets corresponding to all seed nodes is then determined as the candidate connection descent node set.

[0194] After each execution of step S2 to generate the first competitive connection structure, the generated first competitive connection structure and its corresponding comparison time point are archived. Based on the archived first competitive connection structures, for each business entity node in the candidate connection descent node set, the number of competitive connection edges connected to that business entity node at the corresponding archive time point can be extracted from each archived record.

[0195] Within a preset time window, the number of competing connection edges for each business entity node in the candidate connection descent node set is obtained in chronological order at multiple time points. The number values ​​at multiple time points constitute the sequence of changes in the number of competing connection edges for that business entity node.

[0196] The length of the preset time window is used to determine the time range covered by the observation. Those skilled in the art can select the length of the preset time window according to the update cycle of historical files, and this embodiment does not impose a specific limitation on this. For example, the preset time window can preferably be set to cover the archived records corresponding to the most recent 12 comparisons. In practical applications, for scenarios where historical files are updated frequently, the number of comparisons covered by the preset time window can be appropriately shortened, for example, shortened to 8 times; for scenarios where historical files are updated infrequently, the number of comparisons covered by the preset time window can be appropriately extended, for example, extended to 20 times, so that the change sequence can cover a sufficient number of time points to reflect the changing trend of the number of connected edges.

[0197] In some embodiments, step S54 includes steps S541 to S545:

[0198] Step S541: Perform differential processing on the sequence of changes in the number of competing connection edges of the first operating entity node in the candidate connection descent node set to obtain the first differential sequence;

[0199] Step S542: In response to the number of consecutive negative values ​​in the first difference sequence exceeding the first preset quantity condition, the first business entity node is marked as the first candidate node that meets the downward trend condition.

[0200] Step S543: In response to the fact that the proportion of the total negative value in the first difference sequence exceeds the first preset proportion condition, the first business entity node is marked as the second candidate node that meets the downward trend condition.

[0201] Step S544: Select the first candidate node and the second candidate node that simultaneously meet the downward trend condition as nodes in the set of descending objects.

[0202] Step S545: Traverse all business entity nodes in the candidate connection descent node set to obtain the complete connection descent object set.

[0203] The number of competitive connecting edges corresponding to a certain business entity node is compared one by one in the sequence of changes in the number of edges between adjacent time points to obtain the change in the number of edges between adjacent time points. The change in the number of edges between all adjacent time points constitutes the first difference sequence. The number of consecutive occurrences of items with negative change in the first difference sequence is counted. If the number of consecutive occurrences exceeds a preset first quantity condition, the corresponding business entity node is marked as the first candidate node that meets the downward trend condition, indicating that the number of connecting edges of the business entity node has been continuously decreasing over multiple consecutive time points; if the number of consecutive occurrences does not exceed the condition, it is not marked as the first candidate node. The proportion of items with negative change in the first difference sequence to the total number of items with change in the number of edges is counted. If the proportion exceeds a preset first proportion condition, the corresponding business entity node is marked as the second candidate node that meets the downward trend condition, indicating that the number of connecting edges of the business entity node has been decreasing as the main trend throughout the preset time window; if the proportion does not exceed the condition, it is not marked as the second candidate node. Business entity nodes that are marked as both first and second candidate nodes are determined as nodes in the set of connected downward objects. Business entity nodes that only meet one of the conditions are not included in the set of connected downward objects.

[0204] The preset first quantity condition and the preset first ratio condition must be satisfied simultaneously to distinguish between a situation where the number of connected edges continuously decreases and a situation where there are occasional fluctuations at individual time points. Those skilled in the art can select the above quantity and ratio conditions according to the number of time points covered by the preset time window; this embodiment does not impose specific limitations on this. For example, when the preset time window covers 12 time points, the preset first quantity condition can preferably be set to three consecutive negative quantity changes, and the preset first ratio condition can preferably be set to 60%. In practical applications, the preset first quantity condition and the preset first ratio condition can be adjusted accordingly with the number of time points covered by the preset time window. If the number of time points covered by the preset time window increases, the preset first quantity condition can be increased accordingly, for example, to four time points; if the number of time points covered by the preset time window decreases, the preset first quantity condition can be decreased accordingly, for example, to two time points.

[0205] The above differential AND judgment process is sequentially performed on all business entity nodes in the candidate connection descent node set to obtain a complete connection descent object set. Based on the node identifier of each business entity node in the connection descent object set and the corresponding sequence of changes in the number of competing connection edges, a fair competition review result is generated. The review result records the name of each business entity node and the numerical sequence of the number of connection edges of that business entity node changing over time within a preset time window.

[0206] In some embodiments, step S5 further includes steps C1 to C6:

[0207] Step C1: Obtain historical connection reduction trend records for each business entity node in the connection decline object set;

[0208] Step C2: Mark the business entity nodes in the historical connection reduction trend records whose reduction magnitude meets the preset magnitude level as first-level attention nodes;

[0209] Step C3: Mark the business entity nodes in the historical connection reduction trend records whose reduction magnitude does not meet the preset magnitude level but whose reduction duration meets the preset duration as secondary attention nodes;

[0210] Step C4: Generate a list of attention levels based on the node identifiers of the first-level attention nodes and the node identifiers of the second-level attention nodes;

[0211] Step C5: Merge the list of attention levels with the set of connection descent objects to obtain a set of connection descent objects with attention levels;

[0212] Step C6: Generate a tiered fair competition review result based on the set of connection decline objects with attention level, and output the tiered fair competition review result.

[0213] After obtaining the set of connection descent objects in each execution of step S5, the node identifier of each business entity node in the set of connection descent objects obtained this time, the change in the number of corresponding competitive connection edges, and the time point corresponding to this comparison are archived and saved. As the number of comparisons accumulates, the reduction of the same business entity node in each archive is arranged in chronological order, forming a historical connection reduction trend record of the business entity node.

[0214] Obtain the historical connection reduction trend record of each business entity node in the connection reduction object set in previous comparisons. This record reflects the extent of the reduction in the number of connection edges of the business entity node in previous comparisons between the documents to be reviewed and historical documents.

[0215] The reduction rate refers to the percentage decrease in the number of connected edges corresponding to a certain business entity node in this archive compared to the highest number of connected edges in the historical archives of that business entity node.

[0216] Nodes of business entities whose decrease in the number of connected edges reaches a preset level in the historical connection reduction trend record are marked as Level 1 attention nodes. Nodes of business entities whose decrease in the number of connected edges does not reach the preset level, but whose decrease lasts for a preset duration, are marked as Level 2 attention nodes; business entity nodes whose decrease does not reach the preset level and whose duration does not reach the preset duration are not marked with a attention level. The preset level and preset duration are used to identify situations where the decrease in the number of connected edges is large but short-term, and situations where the decrease is relatively small but long-term. These two situations correspond to different causes and both reflect the status of business entities that need attention. Those skilled in the art can select the preset level and preset duration based on the general distribution of the decrease in the number of connected edges and its duration in the historical comparison record. This embodiment does not impose specific limitations on this.

[0217] For example, the preset magnitude level can be preferably set to a decrease of 50%, and the preset duration can be preferably set to a continuous decrease in the number of connecting edges for three comparison cycles. That is, when the number of connecting edges corresponding to a certain business entity node in a certain archive decreases by 50% compared to the highest value of the number of connecting edges in its historical archive, it is marked as a first-level attention node; if the decrease rate does not reach 50%, but the number of connecting edges continues to decrease for three comparison cycles, it is marked as a second-level attention node. In practical applications, the preset magnitude level and preset duration can be adjusted according to the common scale of the number of limited relationships between business entities in different industries. For industries with a generally large number of limited relationships between business entities and a large base of connecting edges, the preset magnitude level can be appropriately increased (e.g., increased to 60%) and the preset duration can be appropriately extended (e.g., extended to four comparison cycles); for industries with a small base of connecting edges, the preset magnitude level can be appropriately decreased (e.g., decreased to 40%) and the preset duration can be appropriately shortened (e.g., shortened to two comparison cycles). A list of attention levels is generated based on the node identifiers corresponding to the first-level and second-level attention nodes. The list of attention levels is then merged with the set of connection descent objects to obtain a set of connection descent objects with attention levels. Finally, a graded fair competition review result is generated and output based on the set of connection descent objects with attention levels.

[0218] For example, if the set of objects experiencing connection reduction includes business entity A, and a two-step path is extended outward from A as the seed node, a set of candidate connection descent nodes is obtained, which includes business entities B and C. The number of connection edges of B at multiple consecutive time points within a preset time window is obtained. If the change in the number of edges at four consecutive time points in the calculated difference sequence is negative, exceeding a preset first quantity condition, B is marked as the first candidate node. Simultaneously, the proportion of negative values ​​in the difference sequence also exceeds a preset first proportion condition, so B is marked as the second candidate node, and B is therefore included in the set of objects experiencing connection reduction. Further, historical connection reduction trend records of B are obtained. If the records show that B's previous reduction has reached a preset level, B is marked as a first-level concern node and included in the tiered fair competition review results.

[0219] It should be noted that the results of the tiered fair competition review can be presented to the reviewers. The adoption or modification actions taken by the reviewers for the risk points can be recorded after being anonymized, and used as a reference for subsequent adjustments to the relevant preset conditions in the processing described in this application.

[0220] In some implementations, the results of the graded fair competition review obtained from multiple comparisons can be classified and analyzed offline. The business entity nodes that appear repeatedly and their associated limiting texts can be classified. The limiting texts with similar characteristics obtained from the classification can be fed back to supplement the aforementioned competition constraint keywords, so that the set of keywords can be updated as the processing continues.

[0221] In other implementations, association analysis can be performed on multiple competitive connection subgraphs obtained by layering by edge identifier within the same competitive connection structure. If there are contradictory restrictions on the same business entity node in different competitive connection subgraphs, the corresponding clause node identifier and its corresponding competitive connection subgraph identifier will be recorded as potential clause conflict information and output together with the results of the hierarchical fair competition review.

[0222] The beneficial effect of the above processing is that simply identifying the objects with reduced connections cannot reflect the continuous changes in the connection status of business entities over time, nor can it distinguish whether the reduction in the number of connection edges is an occasional fluctuation or a continuous trend. Step S5, based on the objects with reduced connections, expands outward along competing connection edges to identify candidate business entities indirectly affected. By performing differential processing on the sequence of changes in the number of connection edges, the simultaneous satisfaction of both continuous reduction and overall reduction conditions is used as the screening criterion. This allows the set of objects with declining connections to distinguish business entities with a continuous decline, rather than those affected by occasional fluctuations. Furthermore, by combining the connection reduction trend records formed in historical comparisons, a level of concern is classified, so that both the two different causes of decline—large reduction magnitude and long duration—can be distinguished and reflected in the review results. Overall, a complete identification process is achieved from changes in the wording of clauses to the continuous downward trend of business entity connections.

[0223] Based on the above steps, this application also includes the following embodiments:

[0224] This application also provides a multi-source heterogeneous fair competition review data processing system, including:

[0225] The parsing module is used to perform structured parsing of the documents to be reviewed and historical documents, and to extract the first structural feature information corresponding to each of the documents to be reviewed and historical documents.

[0226] A module is established to create a competing connection structure based on the object mapping information and constraint information in the first structural feature information;

[0227] The matching module is used to perform association matching between the first competitive connection structure corresponding to the file to be reviewed and the second competitive connection structure corresponding to the historical files, and to identify the change information of competitive connection edges.

[0228] The determination module is used to analyze changes in node connection paths in a competitive connection structure based on information about changes in competitive connection edges, and to determine objects where connections are reduced.

[0229] The output module is used to identify a set of connection-decreasing objects based on the connection-decreasing objects, and output the fair competition review results corresponding to the set of connection-decreasing objects.

[0230] See Figure 4 This is a schematic diagram of the structure of the multi-source heterogeneous fair competition review data processing system provided in this application embodiment. The document to be reviewed, A, and historical document B are input into the fair competition review data processing system 60 as data sources. The fair competition review data processing system 60 includes a parsing module 61, an establishment module 62, a matching module 63, a determination module 64, and an output module 65 connected in sequence. After performing structured parsing on the document to be reviewed, A, and historical document B, the parsing module 61 transmits the obtained structural feature information to the establishment module 62. The establishment module 62 establishes a first competitive connection structure and a second competitive connection structure based on the object mapping information and constraint information contained therein, and transmits the two competitive connection structures to the matching module 63.

[0231] After matching the first and second competing connection structures, the matching module 63 transmits the resulting information on changes in competing connection edges to the determination module 64. The determination module 64 analyzes the changes in node connection paths based on this information, identifies objects with reduced connections, and transmits this information to the output module 65. The output module 65 identifies a set of objects with decreased connections based on these objects, ultimately forming and outputting the fair competition review result 66. The aforementioned document A to be reviewed, historical document B, and various modules within the fair competition review data processing system 60 are merely illustrative examples and do not limit their specific quantity, name, or deployment method.

[0232] Figure 4 The system of the illustrated embodiment can be used to perform corresponding operations. Figure 1 The steps in the method embodiments shown are implemented in a similar manner and have similar technical effects, and will not be repeated here.

[0233] See Figure 5 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of this application. The electronic device 70 includes: a processor 71, a memory 72, and a computer program; wherein,

[0234] The memory 72 is used to store computer programs, and the memory may also be flash memory. Computer programs may be, for example, application programs or functional modules that implement the methods described above.

[0235] The processor 71 is used to execute the computer program stored in the memory to implement the various steps performed by the device in the above method. For details, please refer to the relevant descriptions in the preceding method embodiments.

[0236] Alternatively, the memory 72 can be either standalone or integrated with the processor 71.

[0237] When the memory 72 is a device independent of the processor 71, the device may also include:

[0238] Bus 73 is used to connect memory 72 and processor 71.

[0239] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A method for processing multi-source heterogeneous fair competition review data, characterized in that, include: The document to be reviewed and the historical documents are subjected to structured parsing to extract the first structural feature information corresponding to each of the document to be reviewed and the historical documents. A competitive connection structure is established based on the object mapping information and constraint information in the first structural feature information; The first competitive connection structure corresponding to the file to be reviewed and the second competitive connection structure corresponding to the historical file are associated and matched to identify the change information of the competitive connection edges. Based on the information about changes in the competing connection edges, analyze the changes in the node connection paths in the competing connection structure to identify objects where connections are reduced. Based on the identified connection reduction objects, a set of connection decline objects is identified, and the fair competition review results corresponding to the set of connection decline objects are output.

2. The method according to claim 1, characterized in that, The process involves performing structured parsing on the document to be reviewed and historical documents, extracting the first structural feature information corresponding to each document, including: The document to be reviewed is parsed for chapter structure and semantic entity recognition to obtain the first file segmentation information and the first entity information corresponding to the document to be reviewed. The historical files are parsed for chapter structure and semantic entity recognition to obtain the second file segmentation information and second entity information corresponding to the historical files. Based on the first file segmentation information and the first entity information, an association mapping is performed to generate the first object mapping information corresponding to the file to be reviewed; Based on the second file segmentation information and the second entity information, an association mapping is performed to generate the second object mapping information corresponding to the historical file; The entity items in the first object mapping information and the entity items in the second object mapping information are dereferenced together, and the coreferenced entity items are merged to obtain the object mapping information and constraint information corresponding to the document to be reviewed and the historical document, respectively, after merging. The merged object mapping information and constraint information are used as the first structural feature information.

3. The method according to claim 1, characterized in that, The step of establishing a competitive connection structure based on the object mapping information and constraint information in the first structural feature information includes: Extract the set of business entity nodes from the object mapping information, and treat each business entity node as a graph node in the competitive connection structure; Extract the competitive constraint relationships between business entities from the constraint information, and treat each competitive constraint relationship as a competitive connection edge connecting two graph nodes; The graph nodes and the competing connection edges form a competing connection graph, which is used as the first competing connection structure. Each competing connection edge in the competing connection graph is assigned an edge identifier, which is used to distinguish different types of competing constraint relationships; Based on the edge identifier, the competing connection edges in the competing connection graph are layered to obtain multiple competing connection subgraphs, and the multiple competing connection subgraphs are used together as the competing connection structure.

4. The method according to claim 1, characterized in that, The step of associating and matching the first competing connection structure corresponding to the file to be reviewed and the second competing connection structure corresponding to the historical file to identify competing connection edge change information includes: Each graph node in the first competitive connection structure is paired with a graph node in the second competitive connection structure that has the same business entity identifier to obtain the node pairing result. Based on the node pairing results, each first competitive connection edge in the first competitive connection structure is paired with the second competitive connection edge between the corresponding node pair in the second competitive connection structure to obtain the edge pairing results. In response to the edge pairing result being that the edge identifiers of the first competing connection edge and the second competing connection edge are consistent, the first competing connection edge is marked as an invariant edge; In response to the edge pairing result that the first competing connection edge exists while the second competing connection edge does not exist, the first competing connection edge is marked as a newly added edge; In response to the edge pairing result that the first competing connection edge does not exist while the second competing connection edge does exist, the second competing connection edge is marked as a disappeared edge; Collect the edge identifiers of the newly added edges and the disappearing edges, as well as the node identifiers of the connected node pairs, to generate the information on the changes in the competing connection edges.

5. The method according to claim 4, characterized in that, The step of collecting the edge identifiers of the newly added edges and the disappeared edges, and the node identifiers of the connected node pairs, to generate the competing connection edge change information includes: The frequency of occurrence of each business entity node connected by the newly added edge and the disappeared edge is counted separately; The business entity nodes that appear more frequently than the first preset frequency condition in the newly added edges are marked as newly added active nodes; The business entity nodes that appear more frequently than the second preset frequency condition in the disappearing edge are marked as disappearing active nodes; A new change record is generated based on the node identifier of the newly added active node and the edge identifier of the newly added edge; A disappearance change record is generated based on the node identifier of the disappearing active node and the edge identifier of the disappearing edge; The newly added change records and the disappeared change records are combined into the competing connection edge change information.

6. The method according to claim 1, characterized in that, The step of analyzing the changes in node connection paths in the competitive connection structure based on the information about changes in the competitive connection edges, and identifying objects with reduced connections, includes: Based on the edge identifier of the disappeared edge and the node identifier of the connected node pair in the information on changes in the competitive connection edge, the first node pair corresponding to the removed competitive connection edge is located in the first competitive connection structure. Starting from each node in the first node pair, traverse the node connection paths along the reserved competitive connection edges in the first competitive connection structure to obtain the first node connection path set. Starting from each node in the first node pair, traverse the node connection paths along all competing connection edges in the second competitive connection structure to obtain the second node connection path set; Compare the first node connection path set and the second node connection path set, and extract the missing node connection paths that exist in the second node connection path set but do not exist in the first node connection path set; The operating entity nodes that do not appear in the first competing connection structure are extracted from the connection path of the disappeared nodes as the connection reduction objects.

7. The method according to claim 6, characterized in that, The step of comparing the first set of node connection paths and the second set of node connection paths to extract missing node connection paths that exist in the second set of node connection paths but not in the first set of node connection paths includes: Obtain the first node connection path in the second node connection path set as the current comparison path; Search the first node connection path set for candidate paths that match the start node identifier and end node identifier of the current comparison path; In response to finding the candidate path, the intermediate node sequence of the currently compared path is compared with the intermediate node sequence of the candidate path; In response to the inconsistency in the intermediate node sequence, the current alignment path is marked as a structural change path; In response to the consistency of the intermediate node sequence, the current alignment path is marked as a structure-preserving path; In response to the failure to find the candidate path, the current comparison path is marked as the connection path of the disappeared node; Iterate through all node connection paths in the second node connection path set to obtain all disappeared node connection paths.

8. The method according to claim 1, characterized in that, The step of identifying a set of connection-decreasing objects based on the connection-decreasing objects and outputting the fair competition review results corresponding to the set of connection-decreasing objects includes: Using each business entity node in the connection reduction object as a seed node, extend a preset path number of steps outward along the competitive connection edge in the second competitive connection structure to obtain the set of extended nodes corresponding to each seed node; Take the union of the sets of extended nodes corresponding to all seed nodes, and use the union as the set of candidate connection descent nodes; Obtain the sequence of changes in the number of competing connection edges for each business entity node in the candidate connection descent node set within a preset time window; From the candidate set of descent nodes, select the business entity nodes whose competitive connection edge quantity change sequence shows a downward trend, and use them as the set of descent objects. The fair competition review result is generated based on the node identifiers and the sequence of changes in the number of competing connection edges of each operating entity node in the set of connected descent objects.

9. The method according to claim 8, characterized in that, The step of selecting business entity nodes from the candidate descent node set whose competitive connection edge quantity change sequence shows a decreasing trend, as the descent object set, includes: The sequence of changes in the number of competing connection edges of the first operating entity node in the candidate connection descent node set is differentially processed to obtain the first difference sequence; Based on the number of consecutive negative values ​​and the proportion of the total number of negative values ​​to all difference values ​​in the first difference sequence, the first business entity node is marked as a first candidate node or a second candidate node. The first candidate node that simultaneously satisfies the downward trend condition and the second candidate node that simultaneously satisfies the downward trend condition are selected as nodes in the set of connected downward objects. Traverse all the business entity nodes in the candidate connection descent node set to obtain the complete connection descent object set.

10. The method according to claim 9, characterized in that, The step of marking the first business entity node as a first candidate node or a second candidate node based on the number of consecutive negative values ​​and the proportion of the total number of negative values ​​to all difference values ​​in the first difference sequence includes: In response to the fact that the number of consecutive negative values ​​in the first difference sequence exceeds a first preset quantity condition, the first business entity node is marked as a first candidate node that meets the downward trend condition; In response to the fact that the proportion of negative values ​​in the first difference sequence exceeds a first preset proportion condition, the first business entity node is marked as a second candidate node that meets the downward trend condition.

Images