Enterprise knowledge graph automatic construction and intelligent retrieval method

Abstract

The invention provides an enterprise knowledge graph automatic construction and intelligent retrieval method, which comprises the following steps: collecting multi-source data from a heterogeneous enterprise information system, and carrying out data cleaning and standardization processing; on the basis of a comprehensive scoring mechanism of field similarity and behavior semantic vectors, entities from different systems are merged, and a standard entity set with a unique identifier is generated; based on the standard entity set, in combination with a scoring mechanism of a task-type relationship and a collaborative relationship, extracting a semantic relationship from a behavior record, and constructing an enterprise knowledge graph structure; extracting representative semantic paths from the knowledge graph structure, screening high-quality paths through a path scoring model, and organizing the high-quality paths into a structured path index set; and receiving a natural language query statement, encoding the natural language query statement into a semantic vector, matching the semantic vector with the path index set, executing query in the atlas in combination with an authority control mechanism, and returning a result.

Description

Technical Field

[0001] This invention belongs to the field of data analysis, and in particular relates to a method for automatic construction and intelligent retrieval of enterprise knowledge graphs. Background Technology

[0002] As enterprises accelerate their digital transformation, internal information systems are becoming increasingly complex, involving heterogeneous environments such as meeting systems, operation and maintenance log systems, work order platforms, and personnel master data systems. These systems each store key information such as personnel, tasks, and behavior records. However, in actual business operations, the same object often appears with different identifiers. For example, the same employee might be recorded as the speaker's name in a meeting system, as the login account in the logs, and as the employee ID or email address in the work order platform. This multi-source heterogeneity and identity fragmentation leads to severe fragmentation and redundancy in the structured representation of enterprise knowledge. Furthermore, internal tasks and behaviors are often event-driven and chain-like. The completion of a single business transaction often involves multiple roles, multiple stages, and multiple interactions. Traditional knowledge organization methods based solely on static entities and single-relationships cannot fully reflect the business flow logic, resulting in a lack of dynamism and semantic completeness in knowledge graphs. Even with attempts to introduce relation extraction or knowledge indexing into some existing systems, most remain at the static template matching stage. This fails to effectively address issues such as irregular semantic chains, severe path redundancy, and complex permission isolation in enterprise scenarios, resulting in insufficient accuracy, low response efficiency, and an inability to meet the requirements of actual business applications. Therefore, there is an urgent need for a complete method that can achieve multi-source entity unification, dynamic relation modeling, semantic path organization, and natural language-based query capabilities, addressing the shortcomings of existing technologies such as entity fragmentation, singular relationships, path redundancy, and uncontrollable retrieval. Summary of the Invention

[0003] The purpose of this invention is to design an automatic construction and intelligent retrieval method for enterprise knowledge graphs, which can realize unified entity modeling in a multi-source heterogeneous environment, extract task behavior relationships to construct a complete enterprise knowledge graph structure, and then form a path organization method that is highly aligned with natural language queries through a semantic path indexing mechanism. Finally, the user's query statement is accurately mapped to the graph path and the results that meet the permission and semantic constraints are returned.

[0004] To achieve the above objectives, this invention provides a method for automatic construction and intelligent retrieval of enterprise knowledge graphs, the method comprising: Data from multiple sources is collected from heterogeneous enterprise information systems, and then cleaned and standardized. The multi-source data includes meeting minutes, operation and maintenance logs, and personnel master data. Based on a comprehensive scoring mechanism of field similarity and behavioral semantic vectors, entities from different systems are merged to generate a set of standard entities with unique identifiers. Based on the aforementioned standard entity set, and combined with the scoring mechanism for task-based and collaborative relationships, semantic relationships are extracted from behavioral records to construct an enterprise knowledge graph. Representative semantic paths are extracted from the enterprise knowledge graph, high-quality paths are selected through a path scoring model, and organized into a structured set of path indexes. The system receives natural language query statements, encodes them into semantic vectors, matches them with the path index set, executes the query in the enterprise knowledge graph in conjunction with the access control mechanism, and returns the results.

[0005] Furthermore, the steps for merging entities from different systems include: constructing matching pairs between entity identifiers in different systems; calculating the structural similarity of each pair of entities in the static fields of name, ID, and email; extracting the three most recent action texts of each entity, generating action semantic vectors through word-level attention and sentence-level aggregation network encoding, and calculating their similarity; weighted summing of structural similarity and action semantic similarity to obtain a comprehensive matching score; if the score exceeds a preset threshold, the entities are merged and assigned a unified ID and source identifier.

[0006] Furthermore, the behavioral semantic vector is generated as follows: the behavioral text is encoded using a word-level attention network to output a vector with a dimension of 128; then it is compressed through a sentence-level aggregation layer to generate a semantic vector with a dimension of 64; finally, the similarity between the behavioral semantic vectors of two entities is calculated using cosine similarity.

[0007] Furthermore, the scoring of task-based relationships specifically includes: the normalized value of the co-occurrence frequency of entities in the original data; the reciprocal of the temporal distance between the behavioral records of entities; the confidence of the word order directionality in the behavioral text; and the penalty for differences in the role categories between entities.

[0008] Furthermore, the scoring of collaborative relationships specifically includes: Co-occurrence frequency, time overlap ratio of joint speeches in meeting minutes, and current shortest path distance between two entities in the graph; by multiplying the co-occurrence frequency and time overlap ratio, dividing by a product of distance penalty term and shortest path distance, the cooperative edge retention value is obtained, and a threshold is used to determine whether to retain the edge.

[0009] Furthermore, the path scoring model specifically includes: The semantic weights of each edge in the path; the normalized redundancy of the number of times each edge appears in the graph structure; the semantic directional weights of the edges; the frequency of business recurrence of the path in historical real task data; the path length; and the sparse regularization term used to penalize high-frequency paths.

[0010] Furthermore, each index record in the path index set includes: An ordered sequence of semantic labels for each edge in the path; a unique identifier for the starting entity of the path; a unique identifier for the ending entity of the path; and a comprehensive score for the path.

[0011] Furthermore, the semantic vector is encoded as follows: The first-layer attention network is used to extract the contextual dependencies between query keywords; the second-layer attention network is used to perform sentence-level semantic combination; and a 64-dimensional vector is output to represent the overall semantic intent of the query.

[0012] Further, the matching process with the path index set involves calculating a path matching score, the calculation elements of which include: The query calculates the cosine similarity between the semantic vector and the path structure vector; the path's own score; the offset between the actual path length and the system's average optimal path length; and the offset penalty coefficient used to penalize excessively long paths.

[0013] Furthermore, the specific implementation of the access control mechanism is as follows: Set an access control mask for each edge in the knowledge graph; during query execution, determine whether the two entity nodes connected to the edge are visible to the current user based on the current user's access permissions; the edge is considered a valid edge for query matching only when both edge nodes are visible.

[0014] The beneficial technical effects of the present invention are at least as follows: To address the aforementioned issues, this invention provides an automatic knowledge graph construction and intelligent retrieval method. By introducing a comprehensive scoring mechanism based on field similarity and behavioral semantic vectors, it merges the same entity across different systems, generating a unique and traceable entity set. Then, it designs a relationship scoring method combining frequency of occurrence, temporal distance, and role differences, along with a collaboration relationship density control mechanism, to construct a graph structure representing both enterprise tasks and collaborations. Based on this, it proposes a path scoring model integrating edge semantic weights, redundancy penalties, and business recurrence frequency, compressing multi-hop task chains into representative semantic path indexes, forming a structured index set that can be directly invoked for queries. Finally, through a joint matching mechanism of natural language query semantic vectors and path index structures, it introduces path length offset penalties and access control masks to achieve a closed loop from natural language intent to graph query execution. Through these steps, this invention significantly improves the uniformity of enterprise knowledge, the accuracy of business relationship expression, the efficiency of path retrieval, and the controllability of query results, constructing an automatic knowledge graph construction and intelligent retrieval system that conforms to the actual business semantics of enterprises and supports natural language usage. Attached Figure Description

[0015] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0016] Figure 1 This is a flowchart of an automatic construction and intelligent retrieval method for enterprise knowledge graphs according to the present invention. Detailed Implementation

[0017] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0018] In one or more embodiments, such as Figure 1 As shown, a method for automatic construction and intelligent retrieval of enterprise knowledge graphs is disclosed, the method comprising the following: S1: Collect multi-source data from heterogeneous enterprise information systems, and perform data cleaning and standardization processing; the multi-source data includes meeting record data, operation and maintenance log data, and personnel master data; and based on a comprehensive scoring mechanism of field similarity and behavioral semantic vectors, merge entities from different systems to generate a set of standard entities with unique identifiers; Specifically, in an enterprise environment where multiple systems operate in parallel, the same employee, the same task, or the same organizational unit is often represented differently in different business systems. For example, in a meeting system, employee Zhang San might appear as "Speaker: Zhang San"; in a log system, it might be recorded as "User ID: U3856"; and in an operations and maintenance system, it might be identified by the operator's email address or login account. If these entities are used directly for graph construction without differentiation, it can easily lead to fragmented nodes in the graph structure and duplicate modeling of the same entity, affecting the accurate representation and retrieval of knowledge. Therefore, this step aims to merge synonymous entities appearing in different systems through field alignment and behavioral semantic modeling, generating a unified set of standard entities to provide accurate and unique node inputs for subsequent graph structure construction.

[0019] This step involves input from three main data sources. First, meeting minutes data, from the structured interface of the meeting system, includes fields such as name, speaking tag, and organization. Second, log data, from the structured log collection tool of the operations and maintenance system, includes fields such as operator account, operation type, and login terminal. Third, personnel file data from the master data system, includes fields such as employee number, unified account, organizational level, and on-duty status. After being collected uniformly through a pre-configured interface, this data enters the data cleaning process, completing field unification, encoding standardization, and character format normalization. Subsequently, based on each original entity representation, the system constructs all possible entity matching pairs and calculates their similarity in turn.

[0020] To determine whether two entities can be merged into the same actual object, the following scoring function is designed: ; in: For the first The entity and the first A comprehensive matching score between entities; For the first The and the first The structural similarity of entities on static fields (such as name, ID, email), with a value range of [value range missing]. ; For the first The and the first The similarity of entities in their behavioral semantic representations, with a value range of . The specific calculation is shown in the following formula; This is the weighted coefficient between structural field similarity and behavioral semantic similarity, ranging from... The value is usually taken as to , is preset by the system.

[0021] The formula for calculating behavioral semantic similarity is as follows: ; in: For entities The context semantic vector is formed by passing the text of its three most recent operational actions (such as meeting speeches, operation and maintenance operation descriptions, and work order submission instructions) through a word-level attention network (first layer, vector dimension ). ) and sentence-level aggregation layer (second layer, vector dimension is ) Encoded to obtain; For entities The context semantic vector, obtained in the same way as The similarity is the same; the numerator is the dot product of the two semantic vectors, the denominator is the product of their respective moduli, and the result is the cosine similarity with a range of [value missing]. However, after vector normalization, the actual output is limited to... Interval.

[0022] when Time (e.g., threshold) ), consider entities With entity They should be merged into a single object. During the merge, the personnel master data system's ID number should be used as the unified identification number first; if none exists, a unique identifier generation algorithm should be used to automatically generate an ID number. The new entity formed after the merge is denoted as... Its fields include: Uniform ID; Original ID set (used to trace the source of the merge); Department affiliation path; Source system ID; Recent behavior density level, etc.

[0023] The final output is a set of deduplicated, structurally consistent standard entities: ; Each of them This represents a merged entity node with a unique number and complete attribute structure, which will be directly used as a graph node in the subsequent graph structure construction.

[0024] This step addresses the issue of entity inconsistency across multiple enterprise systems by constructing a unified identity recognition mechanism that combines structural field matching with behavioral semantic analysis. This solution not only relies on traditional static fields such as name, ID, and email for comparison but also incorporates the entity's operational behavior in actual business processes to extract its behavioral semantic features, serving as a reference for entity merging. This significantly improves matching accuracy and robustness. Through a behavior-driven entity merging method, a unified, unique, and context-rich set of entity nodes is established.

[0025] S2: Based on the aforementioned standard entity set, and combined with the scoring mechanism of task-based and collaborative relationships, semantic relationships are extracted from behavioral records to construct an enterprise knowledge graph; Specifically, this step will use this set as a basis and combine it with the records of tasks, behaviors, and operations in the actual enterprise operation scenario to automatically construct a structured graph relationship that reflects the real business semantics of the enterprise.

[0026] Unlike typical text relation extraction tasks, this step is specifically designed to address the characteristics of knowledge flow within enterprises. First, enterprise behavioral data is characterized by being "event-driven," "incompletely structured," and "unstable in roles." For example, speaker roles are unclear in meeting minutes, responsibility attribution is ambiguous in work order processes, and task operations lack subjects in logs. Second, behavioral relationships in enterprise scenarios are often not of a single type, but rather a coexistence of "task-based relationships" and "collaborative relationships." Therefore, this step needs to semantically capture both types of relationships simultaneously and model them separately within the graph structure.

[0027] The system first models task-oriented semantic relations as triples. ,in This indicates directional relationships such as "responsible," "executed," and "dispatched." To ensure the credibility of the triples, the following triple scoring function is designed, combining three factors: structural matching, behavioral temporal density, and role semantic consistency: ; in: Representing entities and Building relationships between The overall credibility score; This indicates that in the original data, and The frequency normalization result of co-occurrence; It represents the reciprocal of the time distance between the two behavior records, used to determine whether the behavior has the potential for actual interaction; This indicates the directional confidence of the word order structure of the two in the original behavioral text. For example, in "Zhang San reported to Li Si", Zhang San appears first, indicating a higher directional confidence. Representing entities The corresponding role category ID (e.g., department ID, job type) is used as a role difference penalty to suppress unreasonable cross-level edge construction; coefficient. For weighted parameters, satisfying , This is a role difference penalty factor, typically with a value no greater than [value missing]. .

[0028] The last item in the above scoring function is the role difference item: Traditional graph construction often only focuses on the degree of behavioral correlation, ignoring the impact of phenomena such as "cross-level authorization" and "misaligned responsibility" on the credibility of relationships in actual enterprise operations. This is addressed through the role numbering function. Introducing a penalty mechanism can significantly improve the accuracy of graph structures in enterprise process modeling.

[0029] For collaborative relationships (e.g., "joint processing," "collaborative discussion during meetings," "joint operation of a system"), the system uses undirected edge modeling and introduces a relationship density adjustment function to ensure that an overly dense and redundant graph is not formed. For all candidate collaborative relationship triples... ,like If it belongs to a set of collaborative relationships, then the following determination value is calculated: ; in: Representing cooperative edges The structural retention value; Consistent with the previous formula, this indicates the co-occurrence frequency; This indicates the overlap in the time spent speaking by both individuals in the meeting minutes (value). ); This represents the current shortest path distance between two elements in the graph structure. This is a distance penalty term to avoid a large number of redundant cooperative edges within short paths. An undirected edge will only be retained in the graph if the system sets a threshold; otherwise, it will be filtered out to prevent excessive graph connectivity.

[0030] The system organizes all task-based and collaborative relationships that meet the scoring criteria into a set of edges. The final map structure is as follows: ; in, This is the standard entity set output in step one. This is the set of task-oriented and collaborative semantic edges selected based on the scoring function in this step. Each edge is accompanied by a relationship label (such as "execution", "collaboration", "chain of responsibility") and a score value. or This is to support subsequent path analysis and semantic queries.

[0031] S3: Extract representative semantic paths from the enterprise knowledge graph, filter high-quality paths through a path scoring model, and organize them into a structured path index set; Specifically, the goal of this step is to [address the issue in the graph structure]. Based on this, a representative set of semantically mappable paths is extracted, and these paths are organized into an index structure that can be directly used for natural language queries. Unlike general-purpose graphs, enterprise business graphs have two main characteristics: Irregularity of paths: Business processes often experience significant variations in path depth and structure due to factors such as organizational structure, event dynamics, and work order assignment. Strong semantic chain dependency of the path: Some query intents (such as "Did Zhang San participate in handling customer A's problem") must depend on the "task-task-customer" chain, otherwise a match cannot be made.

[0032] Therefore, this step does not enumerate all paths from the graph, but rather from the graph itself. High-quality semantic chains are selectively extracted and organized into a queryable index set. .

[0033] The system first in the diagram Perform multi-hop path scanning. For each pair of entities... Within the number of jumps Within the range (generally set to) ), enumerate all its path sets Each path It represents a directed, sequential sequence of nodes.

[0034] To assess the retention value of these paths, the system designed a path scoring function that combines "structural stability" and "semantic consistency." This function comprehensively considers the semantic weight of each edge in the path, the redundancy of the path, and the frequency of its recurrence in historical business processes, and is calculated as follows: ; in: Representing a path The rating value; It is the edge The semantic weights are derived from the scoring function in step two; It is the edge The normalized value of the frequency of occurrence in a graph structure represents its redundancy; the higher the value, the more common it is, but the lower the information content. It is the semantic directionality weight of the edge (e.g., "report" is more causal than "notify"); It is the path length; Representing a path The number of times the task was reproduced by business records in historical real-world task data; It is the coefficient of the path sparsity regularization term, which controls the penalty strength for high-frequency paths, and is usually taken as... .

[0035] This formula integrates three mechanisms: edge-level semantic representation, redundancy penalty, and historical recurrence. In particular, through... The project avoids "path drift" (i.e., the proliferation of low-quality paths due to frequent connections) and combines... Prioritize the retention of real business processes.

[0036] The system will assign all scores above a set threshold. The paths are included in the semantic path index. Each path will be transformed into a semantic template, which is a sequence of ordered combinations of relation labels: ; in, Indicates the first in the path Strip edge The semantic tags they represent include, for example, “reporting,” “reviewing,” “deploying,” and “archiving.”

[0037] The system will provide each path template Together with its origin and destination entity numbers, they are organized into a structured path index, forming the final index set: ; This index set supports fast query path matching via "semantic path → start and end entities → reachability → rating", and is a necessary interface for the natural language parsing system to map user intent to structural path in the next step.

[0038] Output It is stored in an intermediate service component and supports mechanisms such as high-frequency path priority caching and low-frequency path periodic refresh to achieve a balance between query efficiency and system load.

[0039] S4: Receive a natural language query statement, encode the natural language query statement into a semantic vector, match it with the path index set, and execute the query in the enterprise knowledge graph in conjunction with the access control mechanism and return the result.

[0040] Specifically, the task of this step is to transform the user's natural language query request into path matching and entity retrieval behavior based on graph structure, thereby realizing a complete closed loop from language intent to structured knowledge return.

[0041] User submits natural language query (For example, "Has Zhang San participated in handling customer A's problem?"), the system first performs semantic vector encoding on the query. A two-layer attention mechanism network is used: the first layer extracts the contextual dependencies between keywords, and the second layer performs sentence-level semantic combination, ultimately outputting a... dimensional query vector , which represents the overall structural semantics of the query intent.

[0042] The system will then Structure vector of all paths in the path index set Perform matching, where each It is a path The path matching score is the weighted average of all semantic label embedding vectors. The path matching score is calculated using the following formula: ; in: Representing natural language queries With path Match score; Represents the semantic vector of the query statement; Represents the path structure vector; The path score is derived from the path index structure; This is the path length offset value, which is equal to the actual path length minus the average optimal path length. The offset penalty coefficient controls the degree of penalty for ultra-deep paths in the structure, and is usually set to... .

[0043] After the path score is calculated, the system selects the previous path. The system retrieves the index of the path with the highest matching score and proceeds to the graph query phase. The system traverses the graph structure. Find all matching path structures start and end node pairs And construct an entity chain for returning the result.

[0044] To ensure the compliance of query results permissions in multi-organizational scenarios, the system introduces a graph edge access control mechanism during the query execution phase. Each graph edge... An access control mask will be attached, indicating whether the edge is open to the current querying user. The determination formula is as follows: ; in: Representing an edge Validity in the current session; Indicates whether the original edge of the graph exists (1 or 0); Indicates whether two entity nodes can be accessed by the current user (access control identifier).

[0045] This mechanism is highly adaptable to scenarios involving enterprise tenant isolation and departmental data restrictions, ensuring that the results are reasonable and compliant.

[0046] Specifically, when a user queries "Did Zhang San participate in the decision-making process for customer A's problem?", the system determines the semantic intent as a "personnel-task-decision-customer" path through semantic encoding. The system then finds the structural template in the path index. The starting point is Zhang San's entity number. The endpoint is the entity number of customer A. The path score is as high as And matching score The system ranks in the top three. It then enters the graph traversal phase, searching for all path chains that match the template, and finally returns the results. The response is sent as the hit client entity, along with the complete path structure.

[0047] This invention also provides an automatic enterprise knowledge graph construction and intelligent retrieval device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the steps described in the above embodiment of the automatic enterprise knowledge graph construction and intelligent retrieval method, for example... Figure 1 The steps S1 to S4 described above; or, when the processor executes the computer program, it implements the functions of each module in the above system embodiments.

[0048] For example, the computer program may be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the enterprise knowledge graph automatic construction and intelligent retrieval device.

[0049] The aforementioned automatic enterprise knowledge graph construction and intelligent retrieval device can be a desktop computer, laptop, handheld computer, or cloud server, among other computing devices. This device may include, but is not limited to, processors and memory. Those skilled in the art will understand that the device may also include input / output devices, network access devices, buses, etc.

[0050] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASACs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the enterprise knowledge graph automatic construction and intelligent retrieval device, connecting all parts of the device via various interfaces and lines.

[0051] The memory can be used to store the computer programs and / or modules. The processor, by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory, realizes various functions of the enterprise knowledge graph automatic construction and intelligent retrieval device. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function, etc.; the data storage area may store data created based on the operation of the air conditioner controller, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage devices.

[0052] The module integrating automatic enterprise knowledge graph construction and intelligent retrieval equipment, if implemented as a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0053] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0054] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for automatic construction and intelligent retrieval of enterprise knowledge graphs, characterized in that, The method includes: Data from multiple sources is collected from heterogeneous enterprise information systems, and then cleaned and standardized. The multi-source data includes meeting minutes, operation and maintenance logs, and personnel master data. Based on a comprehensive scoring mechanism of field similarity and behavioral semantic vectors, entities from different systems are merged to generate a set of standard entities with unique identifiers. Based on the aforementioned standard entity set, and combined with the scoring mechanism for task-based and collaborative relationships, semantic relationships are extracted from behavioral records to construct an enterprise knowledge graph. Representative semantic paths are extracted from the enterprise knowledge graph, high-quality paths are selected through a path scoring model, and organized into a structured set of path indexes. The system receives natural language query statements, encodes them into semantic vectors, matches them with the path index set, executes the query in the enterprise knowledge graph in conjunction with the access control mechanism, and returns the results.

2. The method for automatic construction and intelligent retrieval of enterprise knowledge graphs according to claim 1, characterized in that, The steps for merging entities from different systems include: constructing matching pairs between entity identifiers in different systems; calculating the structural similarity of each pair of entities in the static fields of name, ID, and email; extracting the three most recent action texts of each entity, generating action semantic vectors through word-level attention and sentence-level aggregation network encoding, and calculating their similarity; weighted summing of structural similarity and action semantic similarity to obtain a comprehensive matching score; if the score exceeds a preset threshold, the entities are merged and assigned a unified ID and source identifier.

3. The method for automatic construction and intelligent retrieval of enterprise knowledge graphs according to claim 1, characterized in that, The behavioral semantic vector is generated as follows: the behavioral text is encoded using a word-level attention network to output a vector with a dimension of 128; then it is compressed through a sentence-level aggregation layer to generate a semantic vector with a dimension of 64; finally, the similarity between the behavioral semantic vectors of the two entities is calculated using cosine similarity.

4. The method for automatic construction and intelligent retrieval of enterprise knowledge graphs according to claim 1, characterized in that, The scoring of the task-based relationship specifically includes: the normalized value of the co-occurrence frequency of the entity in the original data; the reciprocal of the time distance between the entity's behavior records; the confidence of the word order directionality in the behavior text; and the penalty for the difference in the role category between the entities.

5. The method for automatic construction and intelligent retrieval of enterprise knowledge graphs according to claim 1, characterized in that, The scoring of the collaborative relationship specifically includes: Co-occurrence frequency, time overlap ratio of joint speeches in meeting minutes, and current shortest path distance between two entities in the graph; by multiplying the co-occurrence frequency and time overlap ratio, dividing by a product of distance penalty term and shortest path distance, the cooperative edge retention value is obtained, and a threshold is used to determine whether to retain the edge.

6. The method for automatic construction and intelligent retrieval of enterprise knowledge graphs according to claim 1, characterized in that, The path scoring model specifically includes: The semantic weights of each edge in the path; the normalized redundancy of the number of times each edge appears in the graph structure; the semantic directional weights of the edges; the frequency of business recurrence of the path in historical real task data; the path length; and the sparse regularization term used to penalize high-frequency paths.

7. The method for automatic construction and intelligent retrieval of enterprise knowledge graphs according to claim 1, characterized in that, Each index record in the path index set includes: An ordered sequence of semantic labels for each edge in the path; a unique identifier for the starting entity of the path; a unique identifier for the ending entity of the path; and a comprehensive score for the path.

8. The method for automatic construction and intelligent retrieval of enterprise knowledge graphs according to claim 1, characterized in that, The encoding method of the semantic vector is as follows: The first-layer attention network is used to extract the contextual dependencies between query keywords; the second-layer attention network is used to perform sentence-level semantic combination; and a 64-dimensional vector is output to represent the overall semantic intent of the query.

9. The method for automatic construction and intelligent retrieval of enterprise knowledge graphs according to claim 1, characterized in that, The process of matching with the path index set involves calculating a path matching score, the elements of which include: The query calculates the cosine similarity between the semantic vector and the path structure vector; the path's own score; the offset between the actual path length and the system's average optimal path length; and the offset penalty coefficient used to penalize excessively long paths.

10. The method for automatic construction and intelligent retrieval of enterprise knowledge graphs according to claim 1, characterized in that, The specific implementation of the access control mechanism is as follows: Set an access control mask for each edge in the enterprise knowledge graph; during query execution, determine whether the two entity nodes connected to the edge are visible to the current user based on the current user's access permissions; the edge is considered a valid edge for query matching only when both edge nodes are visible.

Images