Production command knowledge base management system and method based on graph database

Abstract

The invention discloses a graph database-based production command knowledge base management system and method, and the method comprises the steps: constructing a production command knowledge graph database through employing a Neo4j graph database platform, extracting node semantic features and edge semantic features in the graph database through employing a deep learning technology, and meanwhile, carrying out the extraction of the node semantic features and the edge semantic features in the graph database; semantic optimization is performed on node features by taking significant semantic information of all edges connected with nodes as guidance so as to improve the expression ability of the node semantic features, and then fuzzy semantic matching is performed on user query intention description and the optimized node semantic features so as to improve the user query intention description and the user query intention description. Therefore, the node information related to the query intention of the user is intelligently returned as a query result. In this way, efficient management and intelligent query of production command knowledge can be achieved, the accuracy and correlation of query results are effectively improved, and query requirements in a production command scene can be better met.

Description

Technical Field

[0001] The present application relates to the field of intelligent management, and more specifically, to a production command knowledge base management system and method based on a graph database. Background Art

[0002] With the continuous advancement of enterprise informatization, production command systems have become an indispensable component of modern manufacturing. They not only monitor the production process in real time, but also assist decision makers in responding quickly to complex market changes. However, traditional production command systems primarily rely on relational databases to store data. While this structured data model is suitable for processing fixed tabular data, in complex production command scenarios, knowledge information often exhibits a high degree of correlation, hierarchy, and dynamics. For example, process connections within the production process, dependencies between equipment, and the allocation of personnel and tasks make this unstructured or semi-structured information difficult to efficiently manage and effectively utilize in relational databases.

[0003] In recent years, graph databases have demonstrated significant advantages in processing complex network data due to their powerful ability to express relationships. Graph databases, based on nodes (representing entities) and edges (representing relationships between entities), are able to intuitively represent and store complex relationships between entities, making them ideal for building knowledge graphs in the production control field.

[0004] However, most current knowledge management systems based on graph databases often only return query results through simple string matching or keyword retrieval when processing user queries. They lack effective use of the contextual semantic information of knowledge nodes, resulting in low relevance of query results and difficulty in meeting query requirements in complex production command scenarios.

[0005] Therefore, an optimized production command knowledge base management system and method based on a graph database is expected. Summary of the Invention

[0006] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a production command knowledge base management system and method based on a graph database, which constructs a production command knowledge graph database by using the Neo4j graph database platform, and uses deep learning technology to extract the node semantic features and edge semantic features in the graph database. At the same time, the node features are semantically optimized with the significant semantic information of all edges connected to the node as a guide to improve the expressive power of the node semantic features, and then, by fuzzy semantic matching of the user query intent description with the optimized node semantic features, the node information related to the user query intent is intelligently returned as the query result. In this way, efficient management and intelligent query of production command knowledge can be achieved, which effectively improves the accuracy and relevance of the query results, and helps to better meet the query needs in the production command scenario.

[0007] According to one aspect of the present application, a production command knowledge base management system based on a graph database is provided, which includes:

[0008] A graph database construction module is used to construct a production command knowledge graph database using the Neo4j graph database platform, wherein the production command knowledge graph database includes a series of nodes and a series of edges between nodes;

[0009] a node information extraction module, configured to extract a natural language description of a first node from the production command knowledge graph database, and extract natural language descriptions of all edges connected to the first node to obtain a set of edge natural language descriptions;

[0010] a semantic encoding module, configured to semantically encode the natural language description of the first node and each edge natural language description in the set of edge natural language descriptions to obtain a first node semantic encoding feature vector and a set of edge semantic encoding feature vectors;

[0011] a node feature optimization module, configured to perform query response optimization on the first node semantic coding feature vector based on the set of edge semantic coding feature vectors to obtain a first node optimized semantic coding feature vector;

[0012] A user query intention acquisition module is used to obtain the query intention description input by the user;

[0013] A semantic matching metric module is used to determine whether to return a natural language description of the first node as a query result based on the semantic matching between the semantic coding features of the query intent description and the optimized semantic coding feature vector of the first node.

[0014] According to another aspect of the present application, a method for managing a production command knowledge base based on a graph database is provided, comprising:

[0015] A production command knowledge graph database is constructed using the Neo4j graph database platform, wherein the production command knowledge graph database includes a series of nodes and a series of edges between nodes;

[0016] Extracting a natural language description of a first node from the production command knowledge graph database, and extracting natural language descriptions of all edges connected to the first node to obtain a set of edge natural language descriptions;

[0017] Semantically encoding the natural language description of the first node and each edge natural language description in the set of edge natural language descriptions to obtain a set of first node semantic encoding feature vectors and edge semantic encoding feature vectors;

[0018] Based on the set of edge semantic coding feature vectors, performing query response optimization on the first node semantic coding feature vector to obtain a first node optimized semantic coding feature vector;

[0019] Obtain the query intent description entered by the user;

[0020] Based on the semantic matching between the semantic coding features of the query intent description and the optimized semantic coding feature vector of the first node, it is determined whether to return the natural language description of the first node as a query result.

[0021] Compared with the existing technology, the present application provides a production command knowledge base management system and method based on a graph database. It uses the Neo4j graph database platform to build a production command knowledge graph database, and adopts deep learning technology to extract the node semantic features and edge semantic features in the graph database. At the same time, it uses the significant semantic information of all edges connected to the node as a guide to semantically optimize the node features to improve the expressive power of the node semantic features. Then, by fuzzy semantic matching between the user query intent description and the optimized node semantic features, it intelligently returns the node information related to the user query intent as the query result. In this way, efficient management and intelligent query of production command knowledge can be achieved, effectively improving the accuracy and relevance of query results, which helps to better meet the query needs in the production command scenario. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are intended to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0023] Figure 1 is a block diagram of a production command knowledge base management system based on a graph database according to an embodiment of the present application;

[0024] Figure 2 Schematic diagram of data flow of a production command knowledge base management system based on a graph database according to an embodiment of the present application;

[0025] Figure 3 1 is a block diagram of a node feature optimization module in a production command knowledge base management system based on a graph database according to an embodiment of the present application;

[0026] Figure 4 This is a flowchart of a production command knowledge base management method based on a graph database according to an embodiment of the present application. DETAILED DESCRIPTION

[0027] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described herein.

[0028] As used in this application and the claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not intended to refer to the singular but may include the plural. Generally speaking, the terms "comprises" and "include" only indicate the inclusion of the steps and elements specifically identified, and these steps and elements do not constitute an exclusive list. A method or apparatus may also include other steps or elements.

[0029] Although the present application makes various references to certain modules in the system according to embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are illustrative only, and different aspects of the system and method can use different modules.

[0030] Flowcharts are used in this application to illustrate the operations performed by the systems according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, the various steps may be processed in reverse order or simultaneously, as needed. Furthermore, other operations may be added to these processes, or one or more operations may be removed from these processes.

[0031] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described herein.

[0032] Most current graph-based knowledge management systems, when processing user queries, often return results through simple string matching or keyword searches. These systems lack effective utilization of the contextual semantics of knowledge nodes, resulting in low relevance and difficulty meeting the query requirements of complex production command scenarios. Therefore, an optimized graph-based production command knowledge base management system and method are desired.

[0033] In the technical solution of this application, a production command knowledge base management system based on a graph database is proposed. Figure 1 This is a block diagram of a production command knowledge base management system based on a graph database according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of production command knowledge base management system based on graph database according to an embodiment of the present application. Figure 1 and Figure 2 As shown, according to an embodiment of the present application, a production command knowledge base management system 300 based on a graph database includes: a graph database construction module 310 for using a Neo4j graph database platform to construct a production command knowledge graph database, wherein the production command knowledge graph database includes a series of nodes and a series of edges between nodes; a node information extraction module 320 for extracting a natural language description of a first node from the production command knowledge graph database, and extracting natural language descriptions of all edges connected to the first node to obtain a set of edge natural language descriptions; a semantic encoding module 330 for encoding the natural language description of the first node and the set of edge natural language descriptions. The natural language descriptions of each edge in the combination are semantically encoded respectively to obtain a first node semantic encoding feature vector and a set of edge semantic encoding feature vectors; a node feature optimization module 340 is used to optimize the query response of the first node semantic encoding feature vector based on the set of edge semantic encoding feature vectors to obtain a first node optimized semantic encoding feature vector; a user query intention acquisition module 350 is used to obtain the query intention description input by the user; a semantic matching metric module 360 is used to determine whether to return the natural language description of the first node as a query result based on the semantic matching between the semantic encoding features of the query intention description and the first node optimized semantic encoding feature vector.

[0034] In particular, the graph database construction module 310 is used to use the Neo4j graph database platform to construct a production command knowledge graph database, wherein the production command knowledge graph database includes a series of nodes and a series of edges between nodes. It should be understood that the graph database uses nodes and edges as basic units, and can intuitively represent and store entities and the complex relationships between them. Through the Neo4j graph database platform, various elements in production command (such as equipment, personnel, processes, etc.) are abstracted into nodes, and the interactions or influences between these elements (such as process connections, equipment dependencies, task allocation, etc.) are abstracted into edges, so as to construct a complete and intuitive production command knowledge graph, thereby providing a highly organized and associated data structure for subsequent information queries.

[0035] In particular, the node information extraction module 320 is used to extract the natural language description of the first node from the production command knowledge graph database, and extract the natural language descriptions of all edges connected to the first node to obtain a set of edge natural language descriptions. In the technical solution of the present application, the nodes in the production command knowledge graph database and the edges between the nodes exist in the form of natural language descriptions, which can provide users with clear and intuitive information understanding. In order to facilitate user query and retrieval, it is necessary to further extract the semantic information of the nodes and edges in the production command knowledge graph database. Based on this, first, the natural language description of the first node is extracted from the production command knowledge graph database, and the natural language descriptions of all edges connected to the first node are extracted to obtain a set of edge natural language descriptions. To simplify the description, in the technical solution of the present application, only the data processing process of a single node and all edges connected to it is elaborated in detail, but it should be understood that in actual application, this method can be applied to each node in the production command knowledge graph database, thereby realizing traversal query of the production command knowledge graph database.

[0036] In particular, the semantic encoding module 330 is used to semantically encode the natural language description of the first node and each edge natural language description in the set of edge natural language descriptions to obtain a first node semantic encoding feature vector and a set of edge semantic encoding feature vectors. In a specific example of the present application, the natural language description of the first node and each edge natural language description in the set of edge natural language descriptions are respectively input into a semantic encoder based on the Bert model to obtain the first node semantic encoding feature vector and the set of edge semantic encoding feature vectors. In order to fully capture the semantic information of the natural language description of the first node and each edge natural language description, the present application further uses a pre-trained language model to semantically encode them to convert the natural language description into a numerical feature vector representation with semantic information. In a specific example of the present application, a semantic encoder based on the Bert model is used to semantically encode the natural language description of the first node and each edge natural language description in the set of edge natural language descriptions, so as to utilize the bidirectional semantic encoding capability of the Bert model to deeply understand the contextual information in the text and extract more accurate semantic feature representations, thereby obtaining a set of first node semantic encoding feature vectors and edge semantic encoding feature vectors, so as to fully reflect the semantic content of nodes and edges, and provide a useful data basis for subsequent semantic optimization and query matching. It is worth mentioning that, unlike traditional unidirectional RNN or LSTM, BERT uses the encoder part of the Transformer model. This structure allows the model to understand the text from left to right and from right to left at the same time during training, thereby obtaining richer contextual information.

[0037] In particular, the node feature optimization module 340 is used to optimize the query response of the first node semantic encoding feature vector based on the set of edge semantic encoding feature vectors to obtain a first node optimized semantic encoding feature vector. Considering that in the production command knowledge graph database, the semantic information of nodes and edges are interrelated, for example, a device node not only has its own attributes (such as model, function), but also has associated attributes with other nodes (such as raw materials, process flow). Therefore, in order to strengthen the contextual perception of node features, in the technical solution of the present application, the node and all edges connected to it are further fused to achieve contextual optimization of the node semantic features. It is worth mentioning that, considering that when processing large-scale edge information, the set of edge semantic encoding feature vectors may contain some irrelevant information, making it difficult to efficiently focus on useful edge information, thereby affecting the effect of node semantic feature optimization. For this reason, the present application proposes a semantic query optimization coding method based on significant feature guidance. In a specific example of the present application, such as Figure 3As shown, the node feature optimization module 340 includes: a prompt template construction unit 341, used to extract the significant features of the set of edge semantic coding feature vectors as a prompt template; a cross-domain optimization query unit 342, used to perform cross-domain optimization query encoding on the set of the first node semantic coding feature vector and the edge semantic coding feature vector based on the prompt template to obtain the first node optimized semantic coding feature vector.

[0038] Specifically, the prompt template construction unit 341 is used to extract the salient features of the set of edge semantic coding feature vectors as a prompt template. In a specific example of the present application, a key matrix is first constructed based on the set of edge semantic coding feature vectors; specifically, the process of constructing a key matrix based on the set of edge semantic coding feature vectors includes: first, using a key embedding matrix to perform linear embedding coding on each edge semantic coding feature vector in the set of edge semantic coding feature vectors to obtain a set of edge semantic coding feature vectors after linear transformation; that is, using a key embedding matrix to perform linear transformation on each edge semantic coding feature vector in the set of edge semantic coding feature vectors, mapping the original features to a new latent feature space, so that it is suitable for subsequent attention mechanism calculations. Then, using the edge semantic coding feature vector after linear transformation as the key vector, the set of edge semantic coding feature vectors after linear transformation is matrix-arranged to obtain the key matrix. Here, by using each edge semantic coding feature vector after linear transformation as a key vector, it is arranged into a key matrix to facilitate subsequent parallel processing. Furthermore, the maximum value of each key vector in the key matrix is extracted to obtain a key matrix significant feature vector as the prompt template; by selecting the maximum value of each key vector from the key matrix to combine into a key matrix significant feature vector, important side information is captured and used as a prompt template, thereby guiding the subsequent node semantic optimization process to focus more on the key parts of the side information and reduce noise interference.

[0039] Specifically, the cross-domain optimization query unit 342 is used to perform cross-domain optimization query encoding on the set of the first node semantic encoding feature vector and the edge semantic encoding feature vector based on the prompt template to obtain the first node optimized semantic encoding feature vector. In a specific example of the present application, the query embedding matrix and the value embedding matrix are first used to perform linear embedding encoding on the first node semantic encoding feature vector to obtain a value vector and a query vector; further, the query vector, the value vector, each key vector in the key matrix and the prompt template are respectively input into the heterogeneous converter structure based on template prompt optimization to obtain a set of first node cross-domain optimized query encoding feature vectors; that is, the query embedding matrix and the value embedding matrix are respectively used to act on the first node semantic encoding feature vector to generate the corresponding query vector and value vector, and the query vector, the value vector, each key vector in the key matrix and the prompt template are interactively encoded through a specially designed heterogeneous converter structure, and the weight distribution strategy is automatically adjusted by introducing an additional prompt template into the traditional attention mechanism, thereby optimizing the query response of the node semantic feature, so as to generate a more accurate and relevant node semantic feature output. Specifically, the query vector, the value vector, each key vector in the key matrix, and the hint template are respectively input into a heterogeneous transformer structure based on template hint optimization to obtain a set of first-node cross-domain optimized query encoding feature vectors, including: multiplying the query vector by the transposed vector of the key vector and then dividing it by the bi-norm of the hint template to obtain an attention score matrix; passing the attention score matrix through a softmax function and then multiplying it by the hint template to obtain a template hint optimized attention weight vector; calculating the positional point multiplication between the template hint optimized attention weight vector and the value vector to obtain the first-node cross-domain optimized query encoding feature vector. Finally, calculating the positional mean vector of the set of the first-node cross-domain optimized query encoding feature vectors to obtain the first-node optimized semantic encoding feature vector. Here, the cross-domain optimized query results of the node and each edge are fused by positional mean to integrate all edge information and generate the first-node optimized semantic encoding feature vector, thereby achieving a global understanding of the first-node information.

[0040] In summary, in the above embodiment, based on the set of edge semantic coding feature vectors, query response optimization is performed on the first node semantic coding feature vector to obtain a first node optimized semantic coding feature vector, including: processing the set of edge semantic coding feature vectors and the first node semantic coding feature vector using the following cross-domain optimization query formula to obtain the first node optimized semantic coding feature vector, wherein the cross-domain optimization query formula is:

[0041] K={k1,k2,...,kn}

[0042] M o ={k1′;k2′;...;k n ′}

[0043] k i ′=f(k i , W k )=k i W k +b k

[0044] v tip ={max(k1′);max(k2′);..;max(k n ′)}

[0045] v q =v1W q +b q

[0046] v v =v1W v +b v

[0047]

[0048] Where K represents the set of edge semantic encoding feature vectors, k1, k2, k i and k n are the first, second, i-th and n-th edge semantic coding feature vectors in the set of edge semantic coding feature vectors, respectively, and the value of n is the number of edge semantic coding feature vectors, W k 、W q and W v denote the key embedding matrix, query embedding matrix and value embedding matrix respectively, b k 、b q and b v Represent different bias terms, M o represents the key matrix, k1′, k2′, k i ′ and k n ′ are the first, second, i-th and n-th key vectors in the key matrix, i.e., the semantically encoded feature vector after the linear transformation, max(·) represents the maximum value function, v tip represents the prompt template, v1 represents the semantic encoding feature vector of the first node, and v q and v v denote the query vector and value vector respectively, (·) T represents the transpose of the vector, ‖·‖2 represents the two-norm of the vector, and softmax is the normalized exponential function. Represents matrix multiplication operation, ⊙ represents positional multiplication, v pi represents the i-th first node cross-domain optimization query encoding feature vector in the set of the first node cross-domain optimization query encoding feature vectors, v p Indicates that the first node optimizes the semantic encoding feature vector.

[0049] In particular, the user query intent acquisition module 350 and the semantic matching metric module 360 are used to obtain a query intent description input by the user; and based on the semantic matching between the semantic encoding features of the query intent description and the optimized semantic encoding feature vector of the first node, determine whether to return the natural language description of the first node as a query result. In a specific example of the present application, the query intent description is first semantically encoded to obtain a query intent description semantic encoding feature vector; similarly, in order to convert the query intent description into a feature vector representation with semantic information so as to match it with the optimized node semantic features, the query intent description is also semantically encoded using the semantic encoder based on the Bert model to capture the deep semantic information of the query intent and generate a query intent description semantic encoding feature vector. In this way, by using the same semantic encoding model to process the query intent description and node information, it is possible to ensure that the query intent and node information are comparable in the semantic space, thereby improving the accuracy of semantic matching. Next, fuzzy semantic matching is performed on the first node optimized semantic encoding feature vector and the query intent description semantic encoding feature vector to obtain a fuzzy semantic matching coefficient; that is, considering that in actual applications, user queries are often fuzzy. Therefore, in order to improve the flexibility of matching, the present application adopts a fuzzy semantic matching method to calculate the semantic similarity between the first node optimized semantic coding feature vector and the query intention description semantic coding feature vector to obtain a fuzzy semantic matching coefficient. In an embodiment of the present application, the similarity between the two is first measured by the cosine similarity calculation method, and then the similarity result is fuzzified using a membership function (such as a sigmoid function) to obtain the fuzzy semantic matching coefficient. Finally, in response to the fuzzy semantic matching coefficient being greater than or equal to a preset threshold, the natural language description of the first node is returned as the query result. In one example, the fuzzy semantic matching coefficient is compared with a preset threshold. When the fuzzy semantic matching coefficient is greater than or equal to the preset threshold, it indicates that the user's query intention is highly matched with the first node information, thereby returning the natural language description of the first node as the query result.

[0050] In the technical solution of the present application, before fuzzy semantic matching is performed on the first node optimized semantic coding feature vector and the query intention description semantic coding feature vector, it is considered that the first node optimized semantic coding feature vector is obtained through query coding optimization of the node and edge semantic set, so that the first node optimized semantic coding feature vector not only has the semantic information of the first node itself but also includes the implicit semantic association information between it and other nodes to improve the semantic representation richness of the first node, but this will also cause the first node optimized semantic coding feature vector to have a difference in feature manifold density relative to the query intention description semantic coding feature vector, which will cause a structural mismatch in fuzzy matching and thus affect the calculation accuracy of the fuzzy semantic matching coefficient.

[0051] In response to the above technical problem, in a preferred example of the present application, before performing fuzzy semantic matching on the first node optimized semantic coding feature vector and the query intent description semantic coding feature vector, feature manifold modulation is performed on the first node optimized semantic coding feature vector based on the query intent description semantic coding feature vector to obtain a modulated first node optimized semantic coding feature vector, and the process specifically includes the following steps:

[0052] Calculate the statistical moment parameters of the eigenvalues at any two positions in the query intent description semantic encoding feature vector to obtain the query intent description semantic encoding mean field sparse analytic matrix and the query intent description semantic encoding variance field sparse analytic matrix, which are expressed as:

[0053]

[0054] v i ,v j ∈V

[0055] Where V represents the semantic encoding feature vector describing the query intent, v i and v j Represents the feature values of any two positions in the semantic encoding feature vector describing the query intent, M μ (i, j) represents the value of the (i, j) position of the query intent description semantic encoding mean field sparse parsing matrix, M σ (i, j) represents the value at position (i, j) of the sparse parsing matrix of the variance field of the semantic encoding of the query intent description.

[0056] Based on the query intent description semantic coding mean field sparse parsing matrix and the query intent description semantic coding variance field sparse parsing matrix, the first node optimized semantic coding feature vector is subjected to feature manifold sparse parsing to obtain the first node optimized semantic coding mean field sparse parsing coding vector and the first node optimized semantic coding variance field sparse parsing coding vector, which are expressed as:

[0057]

[0058]

[0059] Among them, V1 represents the first node optimized semantic coding feature vector, M μ Represents the query intent description semantic encoding mean field sparse parse matrix, M σ Represents the query intent description semantic encoding variance field sparse parse matrix, Represents matrix multiplication, ⊙ represents positional point multiplication, ReLU represents the rectified linear unit, V2 represents the first node optimized semantic coding mean field sparse analytic coding vector, and V3 represents the first node optimized semantic coding variance field sparse analytic coding vector.

[0060] Based on the first node optimized semantic coding mean field sparse analytic coding vector and the first node optimized semantic coding variance field sparse analytic coding vector, the first node optimized semantic coding autocorrelation manifold field analytic matrix is constructed, which is expressed as:

[0061]

[0062] Among them, T represents the transpose symbol, L represents the length of the vector, softmax represents the normalized exponential function, and M t Represents the first node optimized semantic encoding self-correlation manifold field parsing matrix.

[0063] After the first node optimized semantic coding mean field sparse analytic coding vector and the first node optimized semantic coding variance field sparse analytic coding vector are stacked on the feature manifold local fine-grained manifold, they are mapped to the high-dimensional modulation space of the first node optimized semantic coding self-correlation manifold field analytic matrix to obtain the modulated first node optimized semantic coding feature vector, which is expressed as:

[0064]

[0065] Wherein, V′ represents the first node optimized semantic coding feature vector of the modulation.

[0066] That is, in response to the above technical problems, in this preferred embodiment, a low-dimensional collaborative parsing framework based on statistical moments is constructed by conducting a cross-dimensional interactive analysis based on statistical moments on the semantic coding feature vector of the query intent description, and the low-dimensional collaborative parsing framework based on statistical moments is used to parse the generalized implicit modulation factors of each eigenvalue of the first node optimized semantic coding feature vector. On this basis, a compressed state causal inference network is used to correlate and analyze the sparse local structural features of the feature manifold and the global implicit feature entanglement of the feature vector. Thus, the separability of the feature manifold correlation primitives is enhanced with the help of the self-correlation manifold field parsing matrix to establish a mapping relationship between the correlation of structural elements and the complexity of the system level, thereby enhancing the decoding power of the architectural features activated by the primitives, and realizing multi-order feature disentanglement from local structural features to the overall system behavior, thereby improving the consistency of the feature manifold density of the first node optimized semantic coding feature vector relative to the query intent description semantic coding feature vector, so as to improve the calculation accuracy of the fuzzy semantic matching coefficient.

[0067] As described above, the graph database-based production command knowledge base management system 300 according to the embodiment of the present application can be implemented in various wireless terminals, such as a server with a graph database-based production command knowledge base management algorithm. In one possible implementation, the graph database-based production command knowledge base management system 300 according to the embodiment of the present application can be integrated into the wireless terminal as a software module and / or hardware module. For example, the graph database-based production command knowledge base management system 300 can be a software module in the operating system of the wireless terminal, or it can be an application developed for the wireless terminal; of course, the graph database-based production command knowledge base management system 300 can also be one of the many hardware modules of the wireless terminal.

[0068] Alternatively, in another example, the graph database-based production command knowledge base management system 300 and the wireless terminal may also be separate devices, and the graph database-based production command knowledge base management system 300 may be connected to the wireless terminal through a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0069] Furthermore, a production command knowledge base management method based on a graph database is also provided.

[0070] Figure 4 Flowchart of the production command knowledge base management method based on the graph database according to the embodiment of the present application. Figure 4As shown, according to the production command knowledge base management method based on the graph database of the embodiment of the present application, the steps include: S1, using the Neo4j graph database platform to construct a production command knowledge graph database, wherein the production command knowledge graph database includes a series of nodes and a series of edges between nodes; S2, extracting the natural language description of the first node from the production command knowledge graph database, and extracting the natural language descriptions of all edges connected to the first node to obtain a set of edge natural language descriptions; S3, semantically encoding the natural language description of the first node and each edge natural language description in the set of edge natural language descriptions to obtain a first node semantic encoding feature vector and a set of edge semantic encoding feature vectors; S4, based on the set of edge semantic encoding feature vectors, performing query response optimization on the first node semantic encoding feature vector to obtain a first node optimized semantic encoding feature vector; S5, obtaining the query intent description input by the user; S6, based on the semantic matching between the semantic encoding features of the query intent description and the optimized semantic encoding feature vector of the first node, determining whether to return the natural language description of the first node as the query result.

[0071] In summary, according to the embodiment of the present application, a production command knowledge base management method based on a graph database is explained, which constructs a production command knowledge graph database by using the Neo4j graph database platform, and uses deep learning technology to extract node semantic features and edge semantic features in the graph database. At the same time, the node features are semantically optimized with the significant semantic information of all edges connected to the node as a guide to improve the expressive power of the node semantic features. Furthermore, by fuzzy semantic matching between the user query intent description and the optimized node semantic features, the node information related to the user query intent is intelligently returned as a query result. In this way, efficient management and intelligent query of production command knowledge can be achieved, effectively improving the accuracy and relevance of query results, which helps to better meet the query needs in production command scenarios.

[0072] While various embodiments of the present disclosure have been described above, the above descriptions are illustrative, non-exhaustive, and not intended to be limiting of the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is selected to best explain the principles of the embodiments, their practical applications, or improvements to existing technologies, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A production command knowledge base management system based on a graph database, characterized in that: include: A graph database construction module is used to construct a production command knowledge graph database using the Neo4j graph database platform, wherein the production command knowledge graph database includes a series of nodes and a series of edges between nodes; a node information extraction module, configured to extract a natural language description of a first node from the production command knowledge graph database, and extract natural language descriptions of all edges connected to the first node to obtain a set of edge natural language descriptions; a semantic encoding module, configured to semantically encode the natural language description of the first node and each edge natural language description in the set of edge natural language descriptions to obtain a first node semantic encoding feature vector and a set of edge semantic encoding feature vectors; a node feature optimization module, configured to perform query response optimization on the first node semantic coding feature vector based on the set of edge semantic coding feature vectors to obtain a first node optimized semantic coding feature vector; A user query intention acquisition module is used to obtain the query intention description input by the user; a semantic matching metric module, configured to determine whether to return a natural language description of the first node as a query result based on a semantic match between the semantic coding features of the query intent description and the optimized semantic coding feature vector of the first node; Among them, the node feature optimization module includes: a prompt template construction unit, which is used to extract the significant features of the set of edge semantic coding feature vectors as a prompt template; a cross-domain optimization query unit, which is used to perform cross-domain optimization query encoding on the set of the first node semantic coding feature vector and the edge semantic coding feature vector based on the prompt template to obtain the first node optimized semantic coding feature vector.

2. The production command knowledge base management system based on graph database according to claim 1 is characterized in that: The semantic encoding module is used to: The natural language description of the first node and each edge natural language description in the set of edge natural language descriptions are respectively input into a semantic encoder based on the Bert model to obtain the first node semantic encoding feature vector and the set of edge semantic encoding feature vectors.

3. The production command knowledge base management system based on graph database according to claim 2 is characterized in that: The prompt template construction unit includes: a key matrix construction subunit, configured to construct a key matrix based on the set of edge semantic encoding feature vectors; The significant feature extraction subunit is used to extract the maximum value of each key vector in the key matrix to obtain a significant feature vector of the key matrix as the prompt template.

4. The production command knowledge base management system based on graph database according to claim 3 is characterized in that: The bond matrix construction subunit is used to: Performing linear embedding coding on each edge semantic coding feature vector in the set of edge semantic coding feature vectors using a key embedding matrix to obtain a set of linearly transformed edge semantic coding feature vectors; The semantically coded feature vector after the linear transformation is used as the key vector, and the set of the semantically coded feature vectors after the linear transformation is arranged in a matrix to obtain the key matrix.

5. The production command knowledge base management system based on graph database according to claim 4 is characterized in that: The cross-domain optimization query unit includes: an embedding coding subunit, configured to perform linear embedding coding on the semantic coding feature vector of the first node using a query embedding matrix and a value embedding matrix to obtain a value vector and a query vector; a cross-domain optimized query encoding subunit, configured to input the query vector, the value vector, each key vector in the key matrix, and the prompt template into a heterogeneous converter structure based on template prompt optimization to obtain a set of cross-domain optimized query encoding feature vectors of the first node; The query information aggregation subunit is used to calculate the position mean vector of the set of the first node cross-domain optimized query encoding feature vectors to obtain the first node optimized semantic encoding feature vector.

6. The production command knowledge base management system based on graph database according to claim 5 is characterized in that: The cross-domain optimization query encoding subunit is used to: Multiplying the query vector by the transposed vector of the key vector and dividing by the two-norm of the prompt template to obtain an attention score matrix; Multiplying the attention score matrix by the prompt template after passing it through a softmax function to obtain a template prompt optimized attention weight vector; Calculate the position point multiplication between the template prompt optimized attention weight vector and the value vector to obtain the first node cross-domain optimized query encoding feature vector.

7. The production command knowledge base management system based on graph database according to claim 6 is characterized in that: The semantic matching metric module includes: A query intent semantic encoding unit, configured to semantically encode the query intent description to obtain a query intent description semantic encoding feature vector; a fuzzy semantic matching unit, configured to perform fuzzy semantic matching on the first node optimized semantic encoding feature vector and the query intention description semantic encoding feature vector to obtain a fuzzy semantic matching coefficient; The query result generating unit is configured to return the natural language description of the first node as the query result in response to the fuzzy semantic matching coefficient being greater than or equal to a preset threshold.

8. The production command knowledge base management system based on graph database according to claim 7 is characterized in that: The query intention semantic encoding unit is used to: The query intention description is input into the semantic encoder based on the Bert model for semantic encoding to obtain the query intention description semantic encoding feature vector.

9. A production command knowledge base management method based on a graph database, characterized in that: include: A production command knowledge graph database is constructed using the Neo4j graph database platform, wherein the production command knowledge graph database includes a series of nodes and a series of edges between nodes; Extracting a natural language description of a first node from the production command knowledge graph database, and extracting natural language descriptions of all edges connected to the first node to obtain a set of edge natural language descriptions; Semantically encoding the natural language description of the first node and each edge natural language description in the set of edge natural language descriptions to obtain a set of first node semantic encoding feature vectors and edge semantic encoding feature vectors; Based on the set of edge semantic coding feature vectors, performing query response optimization on the first node semantic coding feature vector to obtain a first node optimized semantic coding feature vector; Obtain the query intent description entered by the user; Based on the semantic matching between the semantic coding features of the query intent description and the optimized semantic coding feature vector of the first node, it is determined whether to return the natural language description of the first node as a query result.

Images