Intelligent counter semantic interaction system based on knowledge graph

Abstract

The invention relates to the technical field of graph analysis, in particular to an intelligent counter semantic interaction system based on a knowledge graph, which comprises a subgraph generation and matrix analysis module, a semantic analysis module and a node mapping module, and the dynamic adjustment factor dynamically adjusts the activation priority of the service node based on the historical jump path frequency, and generates a service flow time sequence priority. According to the method, the dynamic regulation factors and the real-time session features are coupled, so that triple constraint collaborative optimization of association degree scoring, time sequence compliance and resource scheduling is innovatively realized; dynamically cutting the knowledge graph based on the session state vector, and eliminating redundant node interference; the service flow time sequence matrix is combined with historical path attenuation statistics and real-time semantic features, and it is ensured that the node activation priority accurately adapts to service scene changes.

Description

Technical Field

[0001] The present invention relates to the technical field of graph analysis, and specifically to an intelligent counter semantic interaction system based on a knowledge graph. Background Art

[0002] Traditional intelligent counter systems based on knowledge graphs face multiple technical bottlenecks in dynamic business scenarios. First, the static knowledge graph architecture is difficult to capture the context changes of user conversations in real time, resulting in redundant nodes interfering with the semantic parsing efficiency, and the business process jump logic depends on preset rules and cannot dynamically optimize the node activation order according to historical interaction data. Second, the calculation of the correlation score and the compliance verification of the business process are often carried out in isolation, easily resulting in misactivation of nodes that have semantic matching but violate the time sequence (for example, a user triggers a transaction operation without completing identity verification).

[0003] In addition, the resource scheduling strategy lacks collaborative optimization with semantic intentions and time sequence logic, often resulting in resource allocation conflicts or response delays. The traditional system mostly uses the fixed frequency statistics method to represent the jump weights between business nodes, which can neither reflect the timeliness changes of business models nor integrate real-time session features for dynamic priority adjustment, ultimately leading to the recommended path deviating from the actual business needs. Summary of the Invention

[0004] The purpose of the present invention is to provide an intelligent counter semantic interaction system based on a knowledge graph to solve the problems raised in the above background art.

[0005] To achieve the above purpose, the present invention provides the following technical solution: An intelligent counter semantic interaction system based on a knowledge graph, including a sub-graph generation and matrix analysis module, a semantic parsing module, and a node mapping module, where:

[0006] The sub-graph generation and matrix analysis module real-time trims the nodes and relationship edges in the knowledge graph that are irrelevant to the current context by coupling the session state vector and the business process time sequence matrix, and outputs a trimmed sub-graph with time sequence constraint tags;

[0007] The business process time sequence matrix constructs a dynamic adjustment factor according to the historical jump paths of business process nodes, and the dynamic adjustment factor dynamically adjusts the activation priority of business nodes based on the historical jump path frequency to generate a business process time sequence priority;

[0008] The semantic parsing module receives the trimmed sub-graph with time sequence constraint tags, and performs double verification of intention parsing and business node matching based on the dynamic adjustment factor to generate an adjustment parameter with space-time constraint tags;

[0009] The node mapping module extracts a node sequence that conforms to the temporal priority of the business process from the cropped subgraph carrying the temporal constraint tags according to the adjustment parameters marked by the spatio-temporal constraints;

[0010] Through the collaborative optimization of the correlation score and the temporal priority of the business process by the dynamic adjustment factor, it is used to satisfy the triple constraints of semantic accuracy, temporal compliance, and optimal resource scheduling for the business node activation logic.

[0011] As a further improvement of this technical solution, the subgraph generation and matrix analysis module includes a subgraph generation unit, and the subgraph generation unit is used to output a cropped subgraph carrying temporal constraint tags. The specific process includes:

[0012] Generate a session state vector representing the instant interaction context through an encoder;

[0013] Extract relevant submatrices from the business process temporal matrix based on the intent tags in the session state vector;

[0014] Calculate the node correlation score through a bilinear attention mechanism, and determine the cropping threshold in combination with the adaptive quantile method;

[0015] Perform semantic relevance filtering and temporal compliance filtering on the knowledge graph nodes, and output a cropped subgraph carrying temporal constraint tags.

[0016] As a further improvement of this technical solution, the process of the semantic relevance filtering specifically includes:

[0017] Calculate the correlation score between the nodes in the knowledge graph and the session state vector;

[0018] Filter the nodes and relationship edges whose correlation scores are lower than the cropping threshold.

[0019] As a further improvement of this technical solution, the process of the temporal compliance filtering specifically includes:

[0020] Compare the jump weights of the nodes in the knowledge graph in the business process temporal matrix with the preset business rule range;

[0021] Delete the nodes and relationship edges whose jump weights exceed the preset business rule range.

[0022] As a further improvement of this technical solution, the subgraph generation and matrix analysis module includes a matrix analysis unit, and the matrix analysis unit is used to generate a dynamic adjustment factor. The specific process includes:

[0023] Collect node jump sequences based on a sliding time window;

[0024] Enhance the statistical significance of recent paths through an exponential decay model;

[0025] Generate a jump probability matrix with time series sensitivity as a dynamic adjustment factor.

[0026] As a further improvement of this technical solution, the matrix analysis unit is used to generate the time series priority of the business process. The specific process includes:

[0027] Perform a matrix multiplication operation on the dynamic adjustment factor and the session state vector to generate a time series priority parameter reflecting the current business scenario as the time series priority of the business process.

[0028] As a further improvement of this technical solution, the double verification in the semantic parsing module includes:

[0029] Screen a subset of nodes that meet the preset semantic threshold through cosine similarity matching;

[0030] Perform a topological sorting check based on the time series priority of the business process adjusted by the dynamic adjustment factor.

[0031] As a further improvement of this technical solution, the double verification further includes:

[0032] Construct a double verification weight matrix, and non-linearly superimpose the jump weight and the correlation degree score of the nodes in the knowledge graph in the time series matrix of the business process;

[0033] Apply tensor fusion technology to generate a three-dimensional spatio-temporal constraint tag containing node execution priority, time window constraint, and resource allocation weight as an adjustment parameter of the spatio-temporal constraint tag.

[0034] As a further improvement of this technical solution, the node mapping module is used to screen the optimal path, specifically including:

[0035] Parse the adjustment parameter of the spatio-temporal constraint tag;

[0036] Generate a candidate path set through a topological sorting algorithm;

[0037] Apply the dynamic adjustment factor to correct the path weight, and screen the optimal path through a path optimization algorithm.

[0038] As a further improvement of this technical solution, the construction method of the time series matrix of the business process includes:

[0039] Count the frequency of jump paths between nodes in the knowledge graph;

[0040] Calculate the jump weight as the number of historical trigger times of the node path divided by the total trigger times;

[0041] Dynamically update the jump weight through a time decay function.

[0042] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0043] By coupling dynamic adjustment factors with real-time session features, the intelligent counter semantic interaction system based on knowledge graph innovatively realizes the triple-constraint collaborative optimization of correlation score, timing compliance, and resource scheduling; dynamically prunes the knowledge graph based on the session state vector to eliminate the interference of redundant nodes; the business process timing matrix combines historical path decay statistics with real-time semantic features to ensure that the node activation priority accurately adapts to the changes in the business scenario.

[0044] In addition, the dual-verification mechanism synchronously verifies the semantic matching degree and topological logic to eliminate the risk of incorrect node activation; the multi-dimensional spatio-temporal constraint marking coordinates resource allocation and time window through non-linear fusion technology, significantly improving the execution efficiency of the business flow; finally, while ensuring the semantic parsing accuracy, it realizes the dynamic compliance of the entire business process link and the optimization of resource utilization. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 is the overall module schematic diagram of the present invention;

[0046] Figure 2 is the schematic diagram of the sub-graph generation and matrix analysis module unit of the present invention.

[0047] In the figure: 100, sub-graph generation and matrix analysis module; 101, sub-graph generation unit; 102, matrix analysis unit; 200, semantic parsing module; 300, node mapping module. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0048] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0049] Next, please refer to Figure 1 , the present invention provides a technical solution: an intelligent counter semantic interaction system based on knowledge graph, including a sub-graph generation and matrix analysis module 100, a semantic parsing module 200, and a node mapping module 300.

[0050] The sub-graph generation unit 101 in the sub-graph generation and matrix analysis module 100 outputs a cropped sub-graph with timing constraint marks by coupling the session state vector with the business process timing matrix and real-time cropping the nodes and relationship edges in the knowledge graph that are irrelevant to the current context, specifically including:

[0051] Pre-construct a full-scale graph that includes nodes (i.e., business entities such as products and service nodes), relationship edges (such as "dependency"), and node weights as the knowledge graph;

[0052] Based on historical business logs, count the frequency of jump paths between nodes and construct a matrix reflecting the business logic rules as the business process time series matrix, where the jump weight between nodes in the business process time series matrix is equal to the number of historical trigger times of the node path divided by the total trigger times;

[0053] Real-time capture the semantic features of the current session (such as the intent entities in the user input and the keywords in the conversation history) and intent tags, and generate high-dimensional vectors through an encoder (such as Transformer) to represent the immediate interaction context as the session state vector;

[0054] According to the intent tags in the session state vector, extract relevant sub-matrices from the business process time series matrix. For example, when detecting the "refund" intent tag, automatically activate the node cluster related to the after-sales process (such as "order query → logistics tracking → refund application"), and through matrix decomposition technology, screen out the path pattern with the highest matching degree with the current intent;

[0055] Adopt a bilinear attention mechanism to calculate the correlation score between the session state vector and the nodes in the sub-matrix. According to the distribution of the correlation scores, use the adaptive quantile method to determine the pruning threshold to ensure that the sub-graph scale matches the business complexity;

[0056] Perform filtering and verification on the relationship edges corresponding to each node in the knowledge graph, including semantic relevance filtering and temporal compliance filtering. The process of semantic relevance filtering is as follows:

[0057] Calculate the correlation scores of the nodes in the knowledge graph, and filter the nodes and relationship edges whose correlation scores exceed the pruning threshold range; eliminate low-correlation nodes through the correlation score threshold to solve the response delay problem caused by knowledge graph redundancy and improve the semantic focus of the sub-graph;

[0058] The process of temporal compliance filtering is as follows:

[0059] Compare the jump weights of the nodes in the knowledge graph in the business process time series matrix with the preset business rule range, and delete the nodes and relationship edges whose jump weights exceed the preset business rule range; prevent process logic conflicts (such as triggering subsequent nodes without completing the previous steps) through the comparison of jump weights and business rules to ensure the compliance of the business process;

[0060] Through semantic relevance filtering and temporal compliance filtering, real-time prune the nodes and relationship edges in the knowledge graph that are irrelevant to the current context, and output a pruned sub-graph with temporal constraint tags.

[0061] The sub - graph generation unit 101 captures the dynamic coupling of real - time session features and the business process matrix through an encoder, solves the problem that traditional systems cannot dynamically prune the knowledge graph according to the interaction context, and ensures that the generated sub - graph is closely related to the current business scenario.

[0062] The matrix analysis unit 102 in the sub - graph generation and matrix analysis module 100 constructs a dynamic jump weight by statistically analyzing the historical business node jump frequencies and applying a time - decay function as a dynamic adjustment factor; its technical implementation is based on an event - stream analysis framework with a sliding time window; after collecting the node jump sequences, an exponential decay model is used to enhance the statistical significance of recent paths, and a jump probability matrix with time - series sensitivity is generated through normalization; this matrix is input as a core parameter into the priority calculation engine, and combined with the matching degree between the session state vector and the node attribute encoding, the business process time - series priority is dynamically generated.

[0063] The matrix analysis unit 102 proposes a dynamic adjustment factor generation mechanism, uses a sliding window and an exponential decay model to capture the time - series evolution characteristics of business paths, solves the problem that the static weight model cannot adapt to the dynamic changes of business processes; reveals the generation logic of time - series priority parameters, and through the matrix operation of the dynamic adjustment factor and real - time semantics, solves the path recommendation deviation caused by the disconnection between historical data and the current context, and realizes dynamic adaptation of priorities.

[0064] The semantic parsing module 200 receives the pruned sub - graph with time - series constraint tags and performs double - verification of intent parsing and business node matching based on the dynamic adjustment factor to generate adjustment parameters with space - time constraint tags, specifically including:

[0065] Extract the semantic features of nodes and relationship edges in the pruned sub - graph with time - series constraint tags based on the multi - head attention mechanism, and combine the time - series encoding information in the session state vector to generate a set of semantic vectors of candidate nodes;

[0066] Introduce a dynamic adjustment factor to construct a double - verification weight matrix, non - linearly superimpose the node jump weights in the business process time - series matrix and the correlation scores of the nodes in the pruned sub - graph with time - series constraint tags to form a dynamic weighted scoring rule;

[0067] When performing intent parsing verification, match the intent label with the candidate node semantic vectors through cosine similarity calculation, filter out the subset of nodes that meet the preset semantic threshold requirements, and at the same time eliminate the nodes with time - series conflicts (such as business nodes that have not completed the pre - conditions);

[0068] In the service node matching and verification phase, based on the time sequence priority of the service process adjusted by the dynamic adjustment factor, topological sorting verification is performed on the nodes that pass the semantic verification to ensure that the node activation logic conforms to the dependency relationship of the service process (for example, the "payment" node must precede the "delivery" node).

[0069] By using the tensor fusion technology, the dual verification results are combined with the dynamic adjustment factor parameters to generate three-dimensional spatio-temporal constraint marking adjustment parameters including node execution priority, time window constraint, and resource allocation weight. As the adjustment parameters of the spatio-temporal constraint marking, these parameters transmit the dynamic optimization effect of the service process to the node mapping module 300 through the cascading conduction mechanism.

[0070] The node mapping module 300 extracts the node sequence that conforms to the time sequence priority of the service process from the cropped sub-graph carrying the time sequence constraint marking, specifically including:

[0071] Parse the three-dimensional spatio-temporal constraint marking in the adjustment parameters of the spatio-temporal constraint marking, use the node execution priority as the main sorting basis, the time window constraint as the node availability determination condition, and the resource allocation weight as the scheduling optimization index;

[0072] Based on the topological sorting algorithm, perform preliminary path planning on the nodes in the cropped sub-graph carrying the time sequence constraint marking, and at the same time superimpose the jump weight of the service process time sequence matrix as the edge weight to generate a candidate path set;

[0073] Apply the dynamic adjustment factor to perform secondary correction on the edge weight (path weight), enhance the priority of the high-frequency jump path, and screen the optimal path that meets the spatio-temporal constraints through a path optimization algorithm (such as the improved Dijkstra algorithm);

[0074] Perform dynamic pruning operations, eliminate the timeout nodes according to the time window constraint, and adjust the execution order of the parallel nodes according to the resource allocation weight (for example, preferentially schedule the device nodes with a high idle rate);

[0075] Convert the optimal path into a node sequence through path compression technology, and structure it into a chained execution flow (such as "order confirmation → inventory locking → payment processing") or a parallel execution flow (such as simultaneous sorting in multiple warehouses), and inject time sequence dependency markings to ensure that physical devices respond in the order of business rules; the finally generated node sequence simultaneously meets the triple constraints of semantic relevance, time sequence compliance, and optimal resource scheduling, providing a standardized input for subsequent instruction conversion.

[0076] The foregoing has shown and described the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments, and the above embodiments and the descriptions in the specification are only preferred examples of the present invention and are not used to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of the present invention claimed is defined by the appended claims and their equivalents.

Claims

1. The intelligent counter semantic interaction system based on knowledge graph is characterized by: The method comprises a subgraph generation and matrix analysis module (100), a semantic analysis module (200) and a node mapping module (300), wherein: The subgraph generation and matrix analysis module (100) couples the session state vector and the business process timing matrix to clip nodes and relationship edges in the knowledge graph that are irrelevant to the current context in real time, and outputs a clipped subgraph carrying a timing constraint mark; The business process timing matrix constructs a dynamic adjustment factor according to the historical jump path of the business process node, and the dynamic adjustment factor dynamically adjusts the activation priority of the business node based on the frequency of the historical jump path to generate the business process timing priority; The semantic parsing module (200) receives a clipped subgraph carrying a timing constraint mark, and performs a double verification of intent parsing and service node matching based on the dynamic adjustment factor, and generates an adjustment parameter carrying a time-space constraint mark; The node mapping module (300) extracts a node sequence that meets the business process timing priority from the clipped subgraph carrying the timing constraint mark according to the adjustment parameters of the spatiotemporal constraint mark; The system optimizes the intent recognition accuracy and business process node mapping efficiency in the counter semantic interaction scenario by dynamically optimizing the cropped subgraphs of the knowledge graph marked with timing constraints and combining them with the business process timing priority adjustment.

2. The knowledge graph-based intelligent counter semantic interaction system according to claim 1 is characterized in that: The sub-graph generation and matrix analysis module (100) comprises a sub-graph generation unit (101), and the sub-graph generation unit (101) is used to output a cropped sub-graph carrying a timing constraint mark. The specific process comprises: Generate a conversation state vector representing the real-time interaction context through an encoder; Extracting a relevant sub-matrix from a business process timing matrix based on the intent label in the session state vector; The node relevance score is calculated through the bilinear attention mechanism, and the pruning threshold is determined by combining the adaptive quantile method; Perform semantic relevance filtering and temporal compliance filtering on knowledge graph nodes, and output a cropped subgraph with timing constraint tags.

3. The intelligent counter semantic interaction system based on knowledge graph according to claim 2 is characterized in that: The process of semantic relevance filtering specifically includes: Calculating the relevance score between the node in the knowledge graph and the session state vector; Filter nodes and relationship edges whose relevance scores are lower than the trimming threshold.

4. The knowledge graph-based intelligent counter semantic interaction system according to claim 2 is characterized in that: The timing compliance filtering process specifically includes: Compare the jump weights of nodes in the knowledge graph in the business process timing matrix with the preset business rule range; Delete the nodes and relationship edges whose jump weights exceed the range of the preset business rules.

5. The intelligent counter semantic interaction system based on knowledge graph according to claim 1 is characterized in that: The subgraph generation and matrix analysis module (100) comprises a matrix analysis unit (102), and the matrix analysis unit (102) is used to generate a dynamic adjustment factor. The specific process includes: Collect node jump sequences in knowledge graphs based on sliding time windows; Enhance the statistical significance of recent paths through an exponential decay model; A timing-sensitive jump probability matrix is ​​generated as a dynamic adjustment factor.

6. The knowledge graph-based intelligent counter semantic interaction system according to claim 5 is characterized in that: The matrix analysis unit (102) is used to generate the business process time sequence priority, and the specific process includes: A matrix product operation is performed on the dynamic adjustment factor and the session state vector to generate a timing priority parameter reflecting the current business scenario as the business process timing priority.

7. The intelligent counter semantic interaction system based on knowledge graph according to claim 1 is characterized in that: The double verification in the semantic parsing module (200) includes: Filter the node subsets that meet the preset semantic threshold through cosine similarity matching; A topological sorting check is performed based on the business process timing priority adjusted by the dynamic adjustment factor.

8. The knowledge graph-based intelligent counter semantic interaction system according to claim 7 is characterized in that: The dual verification further includes: Construct a double-verification weight matrix to nonlinearly superimpose the jump weights and relevance scores of nodes in the knowledge graph in the business process time series matrix; Tensor fusion technology is used to generate three-dimensional spatiotemporal constraint labels including node execution priority, time window constraints and resource allocation weights as adjustment parameters of spatiotemporal constraint labels.

9. The intelligent counter semantic interaction system based on knowledge graph according to claim 1 is characterized in that: The node mapping module (300) is used to select the optimal path, specifically comprising: Analyze the tuning parameters of spatiotemporal constraint markers; Generate a set of candidate paths through a topological sorting algorithm; Dynamic adjustment factors are applied to correct path weights, and the optimal path is selected through the path optimization algorithm.

10. The intelligent counter semantic interaction system based on knowledge graph according to claim 1 is characterized in that: The method for constructing the business process timing matrix includes: Count the frequency of jump paths between nodes in the knowledge graph; The jump weight is calculated as the number of historical triggers of the node path divided by the total number of triggers; The jump weight is dynamically updated through a time decay function.

Images