Interactive intelligent customer service knowledge base construction method and system based on knowledge graph

Abstract

The application relates to the technical field of human-computer interaction, and particularly provides a knowledge graph-based interactive intelligent customer service knowledge base construction method and system. The method comprises the following steps: processing through semantic alignment and timestamp fusion technology to obtain a multi-source heterogeneous power service data fusion body; inputting the obtained multi-source heterogeneous power service data fusion body into a deep relationship mining module for processing, identifying and establishing three types of core associations through a bidirectional association discovery mechanism, and forming an initial power service knowledge graph with dynamic weights; connecting the initial power service knowledge graph with dynamic weights with real-time customer interaction conversation flows, processing through an interactive graph enhancement engine, and forming a closed-loop evolving interactive power service knowledge base. The system comprises a data fusion module, a core association module and an enhancement processing module. The application forms a high-precision intelligent decision core which evolves synchronously with business flows.

Description

Technical Field

[0001] This invention relates to the field of human-computer interaction technology, and in particular to a method and system for constructing an interactive intelligent customer service knowledge base based on knowledge graphs. Background Technology

[0002] Driven by both the digital transformation of the energy industry and the reform of the power system, the State Grid Corporation of China, as a large-scale state-owned backbone enterprise vital to the national economy and energy security, is facing multiple challenges in its customer service centers, including a rapid expansion of service scale, increasingly complex user needs, and continuously rising service quality requirements. Currently, the State Grid customer service centers handle millions of service requests daily, including electricity inquiries, fault reports, business processing, and electricity bill inquiries. Traditional manual customer service models are no longer sufficient to meet the demands for efficient, accurate, and 24 / 7 service. With the deepening of smart grid construction and the rapid development of the "Internet and power services" model, users are placing higher demands on the personalization, intelligence, and interactivity of power services. Especially in emerging business areas such as electricity price policy interpretation, distributed photovoltaic grid connection, electric vehicle charging services, and energy consumption analysis guidance, user questions often involve high levels of professionalism and complex scenarios. However, the application of existing intelligent customer service systems in the power sector still has significant limitations: knowledge base updates lag behind policy and regulatory changes, there is insufficient understanding of professional terminology, weak cross-business scenario collaboration capabilities, and a lack of deep integration with power system operation knowledge.

[0003] Prior art 1, Chinese Patent Application No. 202411989230.5, discloses a composite content generation method. This technology includes: a knowledge base-driven content generation and human-computer interaction system framework, a knowledge base engine construction method, a language model integration module, and a human-computer interaction implementation module. It proposes a content generation system combining a language model and a knowledge base. This system can effectively generate content based on human-computer interaction. Users can interact with the content generation system through text-based questions or by operating auxiliary design tools, jointly operating application software for interactive design and development. While intelligent technologies greatly reduce reliance on manual technical capabilities, in complex business scenarios like the power industry, which involve strong time-series interactions and multiple system linkages, data sources are scattered, formats are inconsistent, and it is difficult to directly use them for accurate knowledge construction.

[0004] Prior art two, Chinese patent application number 202010029849.7, discloses an interactive intelligent customer service engine based on a relevant feedback algorithm, belonging to the field of customer service engine technology. It provides users with concise and accurate answers through the design of data sources, acquisition, mining, and storage, and by combining natural language processing (NLP) with search technology. The design of data sources, acquisition, mining, and storage includes: the intelligent response system first needs data, which can come from internet crawling, existing knowledge bases, or specific corpora. While avoiding the inaccuracy and redundancy of knowledge provided by website search functions, and providing users with simple and accurate answers, it avoids scenarios where interactive intelligent customer service engines respond impolitely or unseriously. It handles the special characteristics of government services well, as its corpus or thesaurus is exclusive to government services, preventing responses from other fields and allowing government service websites to quickly and accurately display the results users want. However, when dealing with complex business logic, rule version changes, and cross-business work order associations, it suffers from weak knowledge correlation, static nature, and a lack of business context depth.

[0005] Prior art three, Chinese patent application number 202111665535.7, discloses an interactive intelligent question-answering method based on a power grid training question-answering knowledge base, including the following steps: constructing a power grid training knowledge base containing question descriptions and answers from power grid training; pre-training a neural network-based semantic model library based on the question descriptions and answers in the power grid training knowledge base; having a user raise a question, obtaining the original text raised by the questioner and preprocessing it to convert the original text into standard text; obtaining the user's historical questions, combining the historical questions to perform context recognition on the standard text, and obtaining the corresponding context recognition results; performing entity extraction on the standard text based on the context recognition results, and obtaining entity extraction results; inputting the context recognition results and entity extraction results into the semantic recognition model, and outputting the corresponding answer; converting the answer into natural language descriptive text and replying it to the user. Although the method achieves the retrieval of the power grid training knowledge base by combining the semantics of the question, the background knowledge of the questioner, and the coherent intent of the user, and converts the retrieved relevant answers into natural language description text, thereby realizing a question-and-answer and interactive response method from user question understanding, question retrieval, and answer generation; however, the knowledge base is lagging in updates, cannot continuously learn and optimize from real-time customer service interactions, and has a passive knowledge lifecycle management.

[0006] Current technologies 1, 2, and 3 suffer from difficulties in effectively integrating diverse and time-series-dependent data from different business systems, resulting in fragmented and inconsistent knowledge sources. They also lack the ability to automatically mine and structurally represent deep, dynamic relationships between business entities, leading to shallow and static knowledge graphs that fail to reflect complex business logic and state transitions. Furthermore, knowledge base updates and maintenance are often offline and passive, unable to dynamically and in a closed-loop manner self-correct and enhance based on dialogue context and feedback during real-time customer service interactions, resulting in insufficient timeliness, accuracy, and adaptability of the knowledge base. Therefore, this invention provides a method and system for constructing an interactive intelligent customer service knowledge base based on knowledge graphs. Summary of the Invention

[0007] To achieve the above objectives, the present invention adopts the following technical solution: One aspect of the present invention provides a method for constructing an interactive intelligent customer service knowledge base based on knowledge graphs, comprising the following steps: Structured business data, unstructured dialogue records, and real-time power grid status data are extracted from multiple independent business systems of the State Grid Customer Service Center. These data are then processed using semantic alignment and timestamp fusion technologies to obtain a multi-source heterogeneous power service data fusion body. The resulting multi-source heterogeneous power service data is fused together and input into the deep relationship mining module for processing. Through a bidirectional association discovery mechanism, three types of core associations are identified and established: implicit associations between user issues and power grid equipment status, version associations between business rules and real-time policy documents, and cross-work order time-series associations between fault phenomena and handling measures; thus forming an initial power service knowledge graph with dynamic weights. An initial power service knowledge graph with dynamic weights is integrated with the real-time inflow of customer interaction conversations and processed through an interactive graph enhancement engine. In each round of customer dialogue, answers are retrieved from the graph, and context-sensitive subgraph extraction and correction operations are performed. Based on the knowledge nodes involved in the current conversation and their dynamic weights, a subgraph for the current question is generated in real time. Information confirmed, supplemented, or corrected by customer service representatives or users in the dialogue is fed back to the initial knowledge graph according to preset confidence assessment rules, and the node attributes and associated edge weights are adjusted and enhanced online to form a closed-loop evolving interactive power service knowledge base.

[0008] In another aspect, the present invention provides a knowledge graph-based interactive intelligent customer service knowledge base construction system for implementing the knowledge graph-based interactive intelligent customer service knowledge base construction method, wherein the knowledge graph-based interactive intelligent customer service knowledge base construction system comprises: The data fusion module is responsible for extracting structured business data, unstructured dialogue records, and real-time power grid status data from multiple independent business systems of the State Grid Customer Service Center. It processes these data through semantic alignment and timestamp fusion technology to obtain a multi-source heterogeneous power service data fusion body. The core association module is responsible for inputting the obtained multi-source heterogeneous power service data fusion into the deep relationship mining module for processing. Through a two-way association discovery mechanism, it identifies and establishes three types of core associations: implicit association between user issues and power grid equipment status, version association between business rules and real-time policy documents, and cross-work order time-series association between fault phenomena and handling measures; forming an initial power service knowledge graph with dynamic weights. The enhancement processing module is responsible for connecting the initial power service knowledge graph with dynamic weights with the real-time inflow of customer interaction conversations, and processing it through an interactive graph enhancement engine. In each round of customer dialogue, it retrieves answers based on the graph and performs context-sensitive subgraph extraction and correction operations. Based on the knowledge nodes involved in the current conversation and their dynamic weights, it generates a subgraph for the current question in real time. Information confirmed, supplemented, or corrected by customer service representatives or users in the dialogue is fed back to the initial knowledge graph according to preset confidence assessment rules, and the node attributes and associated edge weights are adjusted and enhanced online to form a closed-loop evolving interactive power service knowledge base.

[0009] In the data processing stage, this invention eliminates semantic ambiguity and spatiotemporal misalignment inherent in traditional data integration through cross-system semantic alignment and timestamp fusion. The result is a knowledge base with precise spatiotemporal consistency and machine-understandable consistency. This fused entity ensures the integrity and accuracy of the data in both logical and temporal dimensions for subsequent in-depth analysis. Secondly, in the knowledge construction stage, deep bidirectional relationship mining of the fused data systematically extracts complex, cross-domain, implicit, and time-dependent logical relationships from the mixed data. This results in the construction of a dynamic knowledge network that not only contains explicit facts but also embeds business causal relationships and policy evolution contexts, giving the knowledge base a preliminary reasoning foundation and weight adjustment capabilities. Finally, in the application interaction stage, closed-loop integration of the dynamic knowledge graph and real-time session stream enables online verification and immediate enhancement of the knowledge base in real-world business scenarios. This transforms the knowledge base update mode from passive, lagging manual maintenance to proactive, real-time automatic evolution. Every customer interaction serves as a process of knowledge retrieval and calibration, ensuring that the accuracy and applicability of the knowledge base continuously optimize themselves as the service process continues, ultimately forming a high-precision intelligent decision-making core that evolves in sync with the business flow. Attached Figure Description

[0010] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of the interactive intelligent customer service knowledge base construction method based on knowledge graph provided in Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the interactive intelligent customer service knowledge base construction method based on knowledge graph provided in Embodiment 1 of the present invention; Figure 3 This is a process diagram of obtaining a multi-source heterogeneous power service data fusion body provided in Embodiment 2 of the present invention; Figure 4 This is a process diagram of forming an initial power service knowledge graph with dynamic weights, as provided in Embodiment 4 of the present invention. Figure 5 This is a process diagram of online adjustment and enhancement of node attributes and associated edge weights provided in Embodiment 10 of the present invention; Figure 6 This is a block diagram of the interactive intelligent customer service knowledge base construction system based on knowledge graphs provided in Embodiment 11 of the present invention.

[0011] Figure 7 A block diagram of the electronic device provided by the present invention; Figure 8 A block diagram of a computer-readable storage medium provided for this invention.

[0012] Reference numerals: 1. Data fusion module; 2. Core association module; 3. Enhanced processing module; 4. Central processing unit / microprocessor / main control chip; 5. Storage medium; 6. Data bus; 7. Input / output bus / external bus / device bus; 8. Display; 9. Input / output device; 10. Computer-readable instructions; 11. Non-transitory computer-readable storage medium. Detailed Implementation

[0013] The technical solutions of the present invention will now be described with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0014] Hereinafter, the terms "first," "second," etc., are used for descriptive convenience only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0015] In this invention, unless otherwise explicitly specified and limited, the term "connection" should be interpreted broadly. For example, "connection" can be a fixed mechanical connection, a detachable mechanical connection, or an integral part; or, "connection" can be a direct connection or an indirect connection through an intermediate medium. Furthermore, unless otherwise explicitly specified and limited, the term "coupling" should be interpreted broadly. For example, "coupling" can be a direct electrical connection, such as physical contact and electrical conduction between two components; it can also be understood as an electrical connection between different components in a circuit structure through physical lines capable of transmitting electrical signals, such as copper foil or wires on a printed circuit board (PCB), to transmit electrical signals; or, "coupling" can be an indirect electrical connection between two components through an intermediate medium; or, "coupling" can be an electrical connection between two components in a non-contact manner, such as an electrical connection between two components using capacitive coupling to transmit electrical signals.

[0016] In this embodiment of the invention, directional terms such as "up," "down," "left," and "right" may be defined relative to the orientation of the components shown in the accompanying drawings. It should be understood that these directional terms can be relative concepts, used for relative description and clarification, and can change accordingly depending on the orientation of the components in the accompanying drawings.

[0017] Example 1: As Figure 1 As shown, this embodiment of the invention provides a method for constructing an interactive intelligent customer service knowledge base based on a knowledge graph, comprising the following steps: Step S100: Extract structured business data, unstructured dialogue records, and real-time power grid status data from multiple independent business systems of the State Grid Customer Service Center, and process them through semantic alignment and timestamp fusion technology to obtain a multi-source heterogeneous power service data fusion body; Step S200: The obtained multi-source heterogeneous power service data fusion body is input into the deep relationship mining module for processing. Through the bidirectional association discovery mechanism, three types of core associations are identified and established: implicit association between user problems and power grid equipment status, version association between business rules and real-time policy documents, and cross-work order time sequence association between fault phenomena and handling measures; forming an initial power service knowledge graph with dynamic weights. Step S300: The initial power service knowledge graph with dynamic weights is connected to the real-time customer interaction conversation flow and processed through an interactive graph enhancement engine. In each round of customer dialogue, answers are retrieved based on the graph, and context-sensitive subgraph extraction and correction operations are performed. Based on the knowledge nodes involved in the current conversation and their dynamic weights, a subgraph for the current question is generated in real time. Information confirmed, supplemented, or corrected by customer service representatives or users in the dialogue is fed back to the initial knowledge graph according to preset confidence evaluation rules. The node attributes and associated edge weights are adjusted and enhanced online to form a closed-loop evolving interactive power service knowledge base.

[0018] For details on the principles described in the above embodiments, please refer to the appendix. Figure 2 In the data processing stage, this embodiment eliminates semantic ambiguity and spatiotemporal misalignment in traditional data integration through cross-system semantic alignment and timestamp fusion. The result is a knowledge base with precise spatiotemporal consistency and machine-understandable consistency. This fused entity ensures the integrity and accuracy of the data in both logical and temporal dimensions for subsequent in-depth analysis. Secondly, in the knowledge construction stage, deep bidirectional relationship mining of the fused data systematically extracts complex, cross-domain, implicit, and time-dependent logical relationships from the mixed data. This results in the construction of a dynamic knowledge network that not only contains explicit facts but also embeds business causal relationships and policy evolution, giving the knowledge base a preliminary reasoning foundation and weight adjustment capability. Finally, in the application interaction stage, closed-loop integration of the dynamic knowledge graph and real-time session stream enables online verification and immediate enhancement of the knowledge base in real-world business scenarios. This transforms the knowledge base update mode from passive, lagging manual maintenance to proactive, real-time automatic evolution. Every customer interaction serves as a process of knowledge retrieval and calibration, ensuring that the accuracy and applicability of the knowledge base continuously optimize themselves as the service process continues, ultimately forming a high-precision intelligent decision-making core that evolves in sync with the business flow.

[0019] Example 2: Figure 3 As shown, based on Embodiment 1, the process of obtaining a multi-source heterogeneous power service data fusion body in step S100 of this embodiment of the invention specifically includes the following steps: Step S101: Structured business data, unstructured dialogue records, and real-time power grid status data are input into a pre-built power business entity dictionary. The power business entity dictionary contains three core anchor entities and their synonym mappings: equipment codes, work order type identifiers, and policy document numbers. Entity mentions matching the dictionary are identified and annotated in the unstructured dialogue records through pattern matching. Key fields are located in the structured business data, and equipment identification codes are associated with them in the real-time power grid status data. A set of semantic anchors with unified entity identifiers is output. Step S102: For the same entity identifier, establish a timeline based on business events, with work order creation, status transition, dialogue rounds, and power grid event alarms as key event nodes; align data records from different data sources that describe the same business event but have timestamp discrepancies or accuracy differences to a unified event timeline through timeline alignment; for data with content conflicts at the same event node, use time freshness conflict resolution rules for adjudication; output a multi-source data sequence organized along the unified business event timeline; Step S103: Encapsulate the multi-source data sequence, add two types of tags to each data record: time-series consistency tag and source tracing tag, and output the multi-source heterogeneous power service data fusion body; Among them, the time sequence consistency label is used to identify the position of the record on the business event timeline and its logical relationship with the preceding and following events; the source tracing label is used to record the original data source of the record, the original value before conflict resolution, and the identifier of the conflict resolution rule applied.

[0020] In the above embodiments, this embodiment generates a standardized data entity with precise entity consistency, complete business temporality, and full operational traceability. Through pattern matching and unified identifier mapping based on a predefined power business entity dictionary, it achieves accurate identification and semantic unification of business entities in data from different sources and formats. It maps descriptions in unstructured text, fields in structured databases, and device codes in real-time data streams to the same entity identification system, fundamentally solving the problems of entity ambiguity and association breakage caused by naming and format differences in cross-system data fusion, ensuring the uniqueness of core business objects across different data sources. The established unified timeline based on business events and its alignment and conflict resolution mechanisms enable the temporal logic reconstruction and conflict adjudication of scattered data fragments. It reorganizes and calibrates discrete data originally recorded according to their respective system clocks, which may have deviations or contradictions, based on the actual occurrence logic of business events, generating a data sequence that is temporally coherent and internally consistent at the business level; providing a reliable temporal dimension foundation for analyzing event causal relationships and state evolution. The encapsulation and dual-labeling process achieves complete encapsulation of aligned data metadata and evolution process; the temporal consistency label retains the logical position of the data in the business event chain, while the source tracing label records the original state of the data and the basis for processing; the output is a data fusion body that not only contains the fused data value, but also embeds its temporal context and complete tracing path; the fusion body ensures that any downstream reasoning or decision can be verified and traced, greatly improving the reliability and transparency of the entire system processing process.

[0021] Example 3: Based on Example 2, the process of attaching two types of tags, namely a time-series consistency tag and a source attribution tag, to each data record in step S103 of this embodiment of the invention specifically includes the following steps: Step S1031: Based on the multi-source data sequence organized along a unified business event timeline, extract the positional features of each data record on the timeline; the positional features include the sequence number of the event node corresponding to the record in the overall timeline, and the dependency relationship type between the record and the previous and next event nodes in business logic; the dependency relationship type is determined according to the association rules between entities in the established semantic anchor set to form an initial temporal relationship description; combine the sequence number and the dependency relationship type to generate a preliminary temporal consistency label for the record; Step S1032: For multiple data records of the same event node, based on the results of the conflict resolution rule execution, record the original data source type and its original value for each record; the original data source type is categorized and identified according to the three major categories of data input; the original value is taken from the initial value of the record in the corresponding data source before conflict resolution; at the same time, the conflict resolution rule identifier finally adopted by the record is bound to the data source type and the original value to form the initial content of the source tracing tag; Step S1033: Merge and encapsulate the generated prototype of the time-series consistency label with the initial content of the source tracing label; use the event node sequence number in the time-series consistency label as the association key to merge the source tracing information corresponding to the same event node; during merging, verify the conflict resolution rule identifier in the source tracing label according to the time freshness adjudication result; after the verification is completed, formally attach the two types of labels to the data record and output a multi-source heterogeneous power service data fusion body with a complete label system.

[0022] In the above embodiments, this embodiment generates a structured data encapsulation with temporal logical associations and complete data genealogy records. By extracting the event sequence and logical dependencies and encoding them as temporal consistency tags, discrete data records are strictly anchored in a unified business event sequence, clarifying the sequential position and causal relationship of each record in the business process, thereby providing a directly computable structured foundation for time-based logical reasoning and state evolution analysis. By recording the original data source, original values, and applied conflict resolution rules and encapsulating them as source tracing tags, a complete evidence chain of generation, transformation, and adjudication is preserved for each piece of data. This not only makes all operations in the data fusion process auditable but also ensures that in the event of analysis or disputes, the original information can be accurately traced back for verification or reassessment, greatly enhancing the reliability and credibility of the entire data system. By using event nodes as keys to fuse and merge the two types of tags and verify the rules, seamless integration and inherent consistency assurance of temporal logic and data genealogy information are achieved. The output data fusion is a high-fidelity, traceable information entity in which each record comes with its own temporal context and complete data genes, laying a solid and transparent data foundation for building a high-quality knowledge graph.

[0023] Example 4: Figure 4 As shown, based on Example 1, the process of forming an initial power service knowledge graph with dynamic weights in step S200 of this embodiment of the invention specifically includes the following steps: Step S201: Utilize the unified entity identifier and time series of the multi-source heterogeneous power service data fusion body to analyze the co-occurrence pattern of user problems and power grid equipment status on the timeline, and verify the causal relationship through time sequence logic, filter out accidental associations, and output a problem-state probability association network with confidence weights. Step S202: Based on the generated problem-state probability association network, extract and arrange policy document entities in timeline from the multi-source heterogeneous power service data fusion body; identify clause changes by comparing the deep semantic differences between adjacent version documents, and bind and adjudicate the clause changes with the affected business rules in the problem-state probability association network according to the pre-set conflict resolution rules in the data, and output a version association evolution chain indicating how the evolution of policy documents affects specific businesses; Step S203: Use the problem-state probability association network as the state logic background and the version association evolution chain as the rule constraint; analyze the historical event sequence of all work orders to mine the repetitive patterns from fault phenomena to handling measures; the repetitive patterns are integrated into a weighted temporal reasoning path through the validity screening of composite constraints, forming an initial knowledge graph with dynamic weights.

[0024] In the above embodiments, this embodiment realizes the deep integration and dynamic knowledge extraction of multi-dimensional power service data. The constructed knowledge graph can reflect the temporal evolution relationship of state, policy and business behavior in power services, and dynamically adjust the correlation strength through a weight mechanism, providing a structured and interpretable knowledge foundation for power service analysis, fault reasoning and decision support.

[0025] Example 5: Based on Example 4, the process of filtering the repeatability pattern through the validity of composite constraints in step S203 of this embodiment of the invention specifically includes the following steps: Step S2031: Taking the full sequence of work order events with a unified time-series label as input, identify the fault-handling event chain that recurs in different work orders through cross-instance sequence alignment and mining, and output a set of candidate patterns with initial frequencies. Step S2032: Taking the candidate pattern set as input, the established problem-state probability correlation network and version correlation evolution chain are introduced as verification criteria; each candidate pattern is checked one by one to see if its logic is consistent with the state causal network and whether its operation basis conforms to the latest policy provisions; the patterns that pass the double verification are retained, their weights are recalibrated according to the degree of conformity, and the verified enhanced pattern set is output. Step S2033: The verified set of enhanced patterns is added to the knowledge graph as a whole path; when multiple paths point to the same disposal target but have different processes, arbitration is carried out based on their weights and the depth of their alignment with the state network and policy evolution chain, and the path with the highest comprehensive confidence is selected as the final temporal reasoning path; the temporal reasoning path is integrated with the knowledge graph to jointly form the initial knowledge graph.

[0026] In the above embodiments, this embodiment realizes a complete pattern processing flow from extraction, verification to fusion of work order data. The constructed knowledge graph not only includes static associations, but also incorporates temporal reasoning paths that have been verified by multiple elements, providing a dynamic knowledge foundation for power service analysis that has both statistical significance and domain reliability.

[0027] Example 6: Based on Example 5, the process of fusing the temporal reasoning path and knowledge graph in step S2033 of this embodiment of the invention specifically includes the following steps: Step S20331: Each enhancement pattern in the verified enhancement pattern set is a complete event chain from the fault phenomenon to the handling measure; perform structured decomposition of the path, breaking down each event chain into sequentially connected event nodes and transition edges; initiate primitive mapping, based on the existing unified entity identification system of the knowledge graph, mapping each event node to a specific entity node or concept node in the knowledge graph, and mapping each transition edge to a relation edge with type and direction in the knowledge graph; after mapping, each enhancement pattern is converted into a path segment composed of standard graph elements and with calibrated weights; all path segments form a set of path segments to be merged. Step S20332: Perform multi-evidence conflict detection: When multiple path segments provide different intermediate nodes or relation sequences between the same start and end nodes, it is determined that there is a path conflict; for each group of conflicting paths, a dynamic weight arbitration mechanism is initiated, using the calibrated weight of each path segment as the baseline, calling the existing problem-state network in the knowledge graph to verify the causal logic strength of the path, and calling the version evolution chain to verify the timeliness strength of its policy provisions; the dynamic weight arbitration mechanism integrates the three pieces of evidence, calculates the comprehensive confidence of each conflicting path, selects the only path with the highest comprehensive confidence for each group of conflicting paths, and forms a set of optimal path decisions without conflict; Step S20333: Decompose the path into the most basic nodes and relation edges, and insert them one by one into the storage structure of the knowledge graph in the form of atomic operations; when inserting, if the node or edge already exists, perform attribute and weight superposition update; after all atomic operations are completed, perform global consistency recalibration; after recalibration, the attributes of all nodes and the dynamic weights of all relation edges in the knowledge graph reach a new round of balance, which serves as the initial power service knowledge graph output for completing the temporal reasoning path fusion.

[0028] In the above embodiments, this embodiment completes the standardized transformation from pattern to knowledge, conflict resolution and system integration, so that the initial power service knowledge graph not only contains static associations and policy evolution information, but also incorporates optimized and verified time-series handling paths, forming a unified knowledge foundation that can support fault reasoning, handling plan generation and decision optimization.

[0029] Example 7: Based on Example 6, the process of calculating the comprehensive confidence of each conflicting path in step S20332 of this embodiment of the invention specifically includes the following steps: Step S203321: Taking the conflicting path segments as input, extract three original pieces of evidence: their calibrated weights, the causal logic strength values ​​retrieved from the knowledge graph, and the policy timeliness strength values; using ranking-based normalization, convert the original value sequence of each piece of evidence in all conflicting paths into scalar scores ranked according to their relative positions; output a standard comparable evidence vector for each path consisting of the three standardized scores. Step S203322: Using all the output standard comparable evidence vectors as input, calculate the dispersion coefficient of the score within the evidence vector of each path as an adjustment factor to measure its internal consistency; use the adjustment factor to weight the original evidence vector of the path, enhance the scores of evidence with high consistency and suppress the scores of evidence with low consistency; output the new evidence set after weight adjustment, i.e. the weighted verification evidence matrix. Step S203323: Use the weighted verification evidence matrix as the basis for decision-making. Based on the preset priority sequence, policy timeliness is the highest, followed by causal logic strength, and path self-weight is the last to be sorted. First, sort and group all paths according to the highest priority evidence score; then sort them according to the second priority evidence score within the same group. If they still cannot be distinguished, use the third piece of evidence to make a decision. Output the final comprehensive confidence ranking and unique path selection decision obtained based on this sequence comparison.

[0030] In the above embodiments, this embodiment realizes a complete confidence calculation process from standardized processing of multi-source heterogeneous evidence, self-consistency weighted adjustment to priority hierarchical decision-making; it not only integrates information from three dimensions of statistics, logic and policy, but also achieves an optimized balance in evidence coordination, business compliance and causal rationality through consistency adjustment and priority adjudication, providing a reliable, interpretable and conflict resolution basis that meets the requirements of power service business for the temporal reasoning path in the knowledge graph.

[0031] Example 8: Based on Example 7, the process of calculating the discrete coefficients of the score within the evidence vector of each path in step S203322 provided in this embodiment of the invention specifically includes the following steps: Step S2033221: Take a standard comparable evidence vector of a path as input, which includes three standardized scores: policy timeliness, causal logic strength, and path self-weight; in a fixed order of policy timeliness, causal logic strength, and path self-weight, obtain the numerical difference between the next score and the previous score in turn to obtain two directional original difference values; the two original difference values ​​constitute the original difference sequence of the evidence vector. Step S2033222: Using the obtained original difference sequence as input; first, take the absolute value of the two original difference values ​​in the sequence to eliminate directionality and obtain the absolute difference sequence; then, identify the maximum value in the absolute difference sequence; using the maximum absolute difference value as a benchmark, convert the two absolute difference values ​​in the difference sequence into proportions relative to the maximum value to obtain two proportion values; the two proportion values ​​constitute the normalized difference measure. Step S2033223: Take the generated normalized difference measure as input, which contains two ratio values; calculate the arithmetic mean of the two ratio values ​​to obtain an intermediate value that represents the average relative difference; then perform a subtraction transformation, subtracting the intermediate value from a preset constant, and the final result is the discrete coefficient of the path evidence vector; the smaller the discrete coefficient value, the lower the consistency of the scores within the original evidence vector.

[0032] In the above embodiments, this embodiment realizes the conversion from multiple evidence scores to internal consistency quantitative indicators; through differential sequence construction, normalization processing and comprehensive calculation, a discrete coefficient that can sensitively reflect the degree of coordination between evidence is generated, providing a basis for evidence weighting adjustment, thereby giving priority to supporting the path with higher internal consistency in conflict path arbitration, and improving the reliability and stability of decision results.

[0033] Example 9: Based on Example 8, the process of converting the two absolute difference values ​​in the difference sequence into proportions relative to the maximum value in step S2033222 of this embodiment of the invention specifically includes the following steps: Step S20332221: Using the identified maximum absolute difference value as input, set the maximum value as the reference benchmark for conversion; treat the benchmark value as a complete unit, and the conversion of all other difference values ​​will be measured based on the unit; establish a standard dimensional benchmark; Step S20332222: Using the first absolute difference value in the generated standard dimension benchmark and absolute difference sequence as input, the absolute difference value is proportionally calculated with the standard dimension benchmark value to obtain an intermediate value between zero and one, which serves as the first intermediate scaling factor; subsequently, the same proportional calculation is performed using the second absolute difference value in the same standard dimension benchmark and absolute difference sequence to obtain the second intermediate scaling factor. Step S20332223: Output the two intermediate scaling factors directly to form a scaling value set containing two elements; the scaling value set consists of two scaling values.

[0034] In the above embodiments, this embodiment realizes the conversion of the absolute difference sequence into a set of proportional values ​​relative to the maximum difference; by establishing a unified benchmark and proportional calculation, the difference values ​​of different orders of magnitude are normalized to a comparable proportional scale, so that the consistency assessment can focus on the relative distribution pattern of differences between evidence, without being disturbed by the numerical range of individual dimensions, thereby improving the robustness and comparability of the coefficient of variation as an internal consistency indicator.

[0035] Example 10: As Figure 5 As shown, based on Example 1, the process of online adjustment and enhancement of node attributes and associated edge weights in step S300 of this embodiment of the invention specifically includes the following steps: Step S301: Take the information confirmed, supplemented, or corrected by the customer service representative or user in the real-time dialogue as input; combine the sub-graph generated in real time in the current conversation, map the information fragments to specific operation types for knowledge nodes or related edges, including attribute value update, relationship edge addition / deletion, and weight modification; each mapping generates a pending instruction containing the operation target, operation content, and original conversation fragment; all instructions constitute a graph operation instruction set to be verified. Step S302: Using the generated set of operation instructions for the graph to be verified as input, calculate a comprehensive confidence score for each operation instruction for the graph to be verified, and merge the three scores to obtain a comprehensive confidence score; only when the comprehensive confidence score exceeds a preset threshold will the corresponding operation instruction for the graph to be verified be retained and a comprehensive confidence score be attached, generating an enhanced set of operations with confidence rating; The calculation is based on three dimensions: the conversation context dimension, which derives the context strength score based on the co-occurrence frequency and consistency of the entities mentioned in the current dialogue history; the information source dimension, which derives the source authority score based on whether the party issuing the confirmation or correction instruction is a verified expert or a regular user; and the graph consistency dimension, which derives the path consistency score based on the compatibility between the new value proposed in the operation and the existing related paths in the knowledge graph. Step S303: The set of enhanced operations with credibility ratings is used as input and applied to the initial knowledge graph; after a series of real-time and orderly adjustments and dissemination, the initial knowledge graph is updated into a new online version with optimized content and weights. The new version serves as the output of the interactive power service knowledge base after the current closed-loop iteration and is ready for the next dialogue service. The operation includes two sub-processes: first, targeted enhancement, which directly adjusts the attributes of the target node or the weights of the target relation edges based on the operation type and the accompanying confidence score, with the adjustment range controlled by the confidence score; second, subgraph backtracking correction, which, after completing the local enhancement, initiates a graph structure-based propagation mechanism to check and fine-tune the attributes and weights of other adjacent nodes that the node or edge was once associated with in the subgraph, in order to maintain the consistency of the local subgraph.

[0036] In the above embodiments, this embodiment realizes online dynamic enhancement of the knowledge graph in a real-time interactive environment. This process, through instruction mapping, credibility filtering, and controllable adjustment propagation, transforms real-time dialogue information into orderly updates to the knowledge graph structure and weights. This allows the knowledge base to continuously evolve and maintain a high-quality state as the interaction process progresses, thereby providing accurate, consistent, and timely knowledge support for dialogue services.

[0037] Example 11: As Figure 6 As shown, based on Examples 1-10, the interactive intelligent customer service knowledge base construction system based on knowledge graph provided in this embodiment of the invention includes: Data fusion module 1 is responsible for extracting structured business data, unstructured dialogue records, and real-time power grid status data from multiple independent business systems of the State Grid Customer Service Center, and processing them through semantic alignment and timestamp fusion technology to obtain a multi-source heterogeneous power service data fusion body. Core Association Module 2 is responsible for inputting the obtained multi-source heterogeneous power service data fusion into the deep relationship mining module for processing. Through a two-way association discovery mechanism, it identifies and establishes three types of core associations: implicit association between user problems and power grid equipment status, version association between business rules and real-time policy documents, and cross-work order time-series association between fault phenomena and handling measures; forming an initial power service knowledge graph with dynamic weights. The enhanced processing module 3 is responsible for connecting the initial power service knowledge graph with dynamic weights with the real-time customer interaction conversation flow, and processing it through an interactive graph enhancement engine. In each round of customer dialogue, it retrieves answers based on the graph and performs context-sensitive subgraph extraction and correction operations. Based on the knowledge nodes involved in the current conversation and their dynamic weights, it generates a subgraph for the current question in real time. Information confirmed, supplemented, or corrected by customer service representatives or users in the dialogue is fed back to the initial knowledge graph according to preset confidence evaluation rules, and the node attributes and associated edge weights are adjusted and enhanced online to form a closed-loop evolving interactive power service knowledge base.

[0038] In the above embodiments, this embodiment achieves contextualized and precise alignment of data from different business domains and forms through the unified processing of data fusion module 1. Utilizing semantic alignment and timestamp fusion technology, it generates a standardized data entity that is semantically interoperable and time-traceable. This provides a high-quality data foundation for analysis, eliminating source ambiguity and possessing a complete business context, overcoming the logical confusion and information fragmentation caused by the simple aggregation of traditional multi-source data. Core association module 2 operates on the aforementioned data fusion body, achieving systematic in-depth mining and networked representation of the complex logic inherent in the business. Through a bidirectional association discovery mechanism, it constructs a knowledge network that not only includes entities and facts but also embeds multi-dimensional dynamic relationships such as the impact of equipment status, policy version changes, and the inheritance of handling experience. This enables the knowledge graph to possess preliminary quantitative reasoning capabilities based on weight changes, surpassing the simple query function of a static knowledge base. Enhancement processing module 3, through a closed-loop design, achieves real-time interaction and adaptive enhancement between the knowledge system and the business front end. By connecting the dynamic knowledge graph with the real-time conversation stream, it ensures that every customer service interaction simultaneously triggers two processes: knowledge retrieval and knowledge base calibration. The knowledge base can adjust node attributes and association strength online based on real-time feedback and guided by the conversation context; this frees the system from the traditional update mode that relies on periodic offline maintenance, transforming it into a high-precision decision-making core that continuously and autonomously evolves in business operations, with its output accuracy and scenario adaptability constantly improving over time.

[0039] Figure 7 A block diagram of an exemplary electronic device suitable for implementing embodiments of the present invention is shown.

[0040] The electronic device may include a central processing unit / microprocessor / main control chip 4; and a storage medium 5 coupled to the central processing unit / microprocessor / main control chip 4 and storing computer-executable instructions therein for performing the steps of various methods of embodiments of the present invention when executed by the processor.

[0041] The central processing unit / microprocessor / main control chip 4 may include, but is not limited to, one or more processors or microprocessors.

[0042] Storage medium 5 may include, but is not limited to, random access memory (RAM), read-only memory (ROM), flash memory, EPROM memory, EEPROM memory, registers, computer storage media (e.g., hard disk, floppy disk, solid-state drive, removable disk, CD-ROM, DVD-ROM, Blu-ray disc, etc.).

[0043] In addition, the electronic device may include (but is not limited to) a data bus 6, an input / output bus / external bus / device bus 7, a display 8, and input / output devices 9 (e.g., keyboard, mouse, speaker, etc.).

[0044] The central processing unit / microprocessor / main control chip 4 can communicate with external devices (8, 9, etc.) via wired or wireless networks (not shown) through the input / output bus / external bus / device bus 7.

[0045] The storage medium 5 may also store at least one computer-executable instruction for performing the steps of various functions and / or methods in the embodiments described herein when the central processing unit / microprocessor / main control chip 4 is running.

[0046] In one embodiment, the at least one computer-executable instruction may also be compiled into or comprise a software product, wherein one or more computer-executable instructions are executed by a processor to perform the steps of the various functions and / or methods in the embodiments described herein.

[0047] Figure 8 A schematic diagram of a computer-readable storage medium according to an embodiment of the present invention is shown.

[0048] like Figure 8 As shown, the non-transitory computer-readable storage medium 11 stores instructions, such as computer-readable instructions 10. When the computer-readable instructions 10 are executed by a processor, the various methods described above can be performed. The non-transitory computer-readable storage medium includes, but is not limited to, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory. Non-transitory non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. For example, the non-transitory computer-readable storage medium 11 can be connected to a computing device such as a computer, and then, when the computing device executes the computer-readable instructions 10 stored on the non-transitory computer-readable storage medium 11, the various methods described above can be performed.

[0049] In the embodiments provided by this invention, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between systems or units may be electrical, mechanical, or other forms.

[0050] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0051] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0052] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for executing all or part of the steps of the methods of the various embodiments of this invention through a computer device (which may be a personal computer, server, or network device, etc.). The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0053] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A knowledge graph-based interactive intelligent customer service knowledge base construction method, characterized in that, Includes the following steps: The resulting multi-source heterogeneous power service data fusion is input into the deep relationship mining module for processing. Through a two-way association discovery mechanism, three types of core associations are identified and established: implicit associations between user problems and power grid equipment status, version associations between business rules and real-time policy documents, and cross-work order time-series associations between fault phenomena and handling measures. An initial knowledge graph of electricity services with dynamic weights is formed; An initial power service knowledge graph with dynamic weights is connected to the real-time inflow of customer interaction conversations and processed through an interactive graph enhancement engine. In each round of customer dialogue, answers are retrieved based on the knowledge graph, and context-sensitive subgraph extraction and correction operations are performed. Based on the knowledge nodes involved in the current dialogue and their dynamic weights, a subgraph for the current question is generated in real time. Information confirmed, supplemented, or corrected by customer service representatives or users in the dialogue is fed back to the initial knowledge graph according to preset confidence assessment rules. The node attributes and associated edge weights are adjusted and enhanced online to form a closed-loop evolving interactive power service knowledge base. 2.The knowledge graph-based interactive intelligent customer service knowledge base construction method of claim 1, wherein, The process of forming an initial knowledge graph of electricity services with dynamic weights includes the following steps: By utilizing the unified entity identifiers and time series of multi-source heterogeneous power service data fusion, we analyze the co-occurrence patterns of user problems and power grid equipment status on the timeline, verify the causal relationship through time-series logic, filter out accidental correlations, and output a problem-state probabilistic correlation network with confidence weights. Against the backdrop of the generated problem-state probabilistic association network, policy document entities are extracted from the multi-source heterogeneous power service data fusion body and arranged chronologically. By comparing the deep semantic differences between adjacent versions of the document, clause changes are identified. Based on the pre-set conflict resolution rules in the data, clause changes are bound and adjudicated with the affected business rules in the problem-state probabilistic association network, and the version association evolution chain indicating how the evolution of policy documents affects specific businesses is output. Using the problem-state probabilistic association network as the state logic background and the version association evolution chain as the rule constraint, the historical event sequence of all work orders is analyzed to uncover repetitive patterns from fault phenomena to handling measures. These repetitive patterns are then integrated into a weighted temporal reasoning path through the effectiveness screening of composite constraints, forming an initial knowledge graph with dynamic weights. 3.The knowledge graph-based interactive intelligent customer service knowledge base construction method of claim 2, characterized in that, The process of filtering repetitive patterns through the validity of composite constraints includes the following steps: Taking the full sequence of work order events with a unified time label as input, the system identifies the fault-handling event chain that recurs in different work orders through cross-instance sequence alignment and mining, and outputs a set of candidate patterns with initial frequencies. Taking the candidate pattern set as input, the established problem-state probability correlation network and version correlation evolution chain are introduced as verification criteria; each candidate pattern is checked one by one to see if its logic is consistent with the state causal network and whether its operation is in line with the latest policy provisions; the patterns that pass the double verification are retained, their weights are recalibrated according to the degree of compliance, and the verified enhanced pattern set is output. The verified set of enhanced patterns is added to the knowledge graph as a whole path. When multiple paths point to the same disposal goal but have different processes, arbitration is carried out based on their weights and the depth of their alignment with the state network and policy evolution chain. The path with the highest comprehensive confidence is selected as the final temporal reasoning path. The temporal reasoning path is integrated with the knowledge graph to form the initial knowledge graph. 4.The knowledge graph-based interactive intelligent customer service knowledge base construction method of claim 3, characterized in that, The process of integrating temporal reasoning paths with knowledge graphs includes the following steps: Each enhancement pattern in the verified enhancement pattern set is a complete event chain from the fault phenomenon to the handling measure; the structured decomposition of the execution path breaks down each event chain into sequentially connected event nodes and transition edges; Initiate primitive mapping: based on the existing unified entity identification system of the knowledge graph, map each event node to a specific entity node or concept node in the knowledge graph, and map each transition edge to a relation edge with type and direction in the knowledge graph; after mapping, each enhancement pattern is converted into a path segment composed of standard graph elements and with calibrated weights; all path segments form a set of path segments to be merged. Perform multi-evidence conflict detection: When multiple path segments provide different intermediate nodes or relation sequences between the same start and end nodes, a path conflict is determined to exist. For each set of conflicting paths, a dynamic weight arbitration mechanism is initiated. Using the calibrated weights of each path segment as the baseline, the existing problem-state network in the knowledge graph is invoked to verify the causal logic strength of the path, and the version evolution chain is invoked to verify the timeliness of its policy provisions. The dynamic weight arbitration mechanism integrates the three pieces of evidence, calculates the comprehensive confidence of each conflicting path, and selects the unique path with the highest comprehensive confidence for each set of conflicting paths, forming a set of conflict-free optimal path decisions. The path is broken down into the most basic nodes and relation edges, and inserted into the knowledge graph's storage structure one by one in the form of atomic operations. During insertion, if the node or edge already exists, the attributes and weights are updated by superposition. After all atomic operations are completed, a global consistency recalibration is performed. After the recalibration, the attributes of all nodes and the dynamic weights of all relation edges in the knowledge graph reach a new round of balance, which serves as the initial power service knowledge graph output for completing the temporal reasoning path fusion.

5. The method for constructing an interactive intelligent customer service knowledge base based on knowledge graphs as described in claim 4, characterized in that, The process of calculating the overall confidence score for each conflicting path includes the following steps: Taking conflicting path segments as input, we extract three original pieces of evidence: their calibrated weights, causal logic strength values ​​retrieved from the knowledge graph, and policy timeliness strength values. We then use ranking-based normalization to convert the original value sequence of each piece of evidence across all conflicting paths into scalar scores ranked by relative position. Finally, we output a standard comparable evidence vector for each path, consisting of the three standardized scores. Using all the standard comparable evidence vectors output as input, calculate the coefficient of variation of the score within the evidence vector of each path as an adjustment factor to measure its internal consistency; use the adjustment factor to weight the original evidence vector of the path, enhance the scores of evidence with high consistency and suppress the scores of evidence with low consistency; output the new set of evidence after weight adjustment, i.e. the weighted verification evidence matrix. The weighted verification evidence matrix is ​​used as the basis for decision-making. According to the preset priority sequence, the policy timeliness is the highest, followed by the strength of causal logic, and the path itself is the last to be sorted. First, all paths are sorted and grouped according to the highest priority evidence score. Then, within the same group, they are sorted according to the second priority evidence score. If they still cannot be distinguished, the third piece of evidence is used to make a decision. The final comprehensive confidence ranking and the unique path selection decision are output based on this sequence comparison.

6. The method for constructing an interactive intelligent customer service knowledge base based on knowledge graphs as described in claim 5, characterized in that, The process of calculating the discrete coefficients of the score within the evidence vector of each path includes the following steps: Using a standard comparable evidence vector of a path as input, it includes three standardized scores: policy timeliness, causal logic strength, and path self-weight. Following a fixed order of policy timeliness, causal logic strength, and path self-weight, the numerical difference between each subsequent score and the previous score is obtained, resulting in two directional raw difference values. The two raw difference values ​​constitute the raw difference sequence of the evidence vector. Using the obtained original difference sequence as input, firstly, take the absolute value of each of the two original difference values ​​in the sequence to eliminate directionality and obtain the absolute difference sequence; then, identify the maximum value in the absolute difference sequence. Based on the maximum absolute difference, the two absolute difference values ​​in the difference sequence are converted into proportions relative to the maximum value, resulting in two proportion values; the two proportion values ​​constitute the normalized difference measure. The generated normalized difference measure is used as input, which contains two ratio values; the arithmetic mean of the two ratio values ​​is calculated to obtain an intermediate value that represents the average relative difference. Then perform a subtraction transformation, subtracting the intermediate value from the preset constant, and the final result is the discrete coefficient of the path evidence vector; The smaller the coefficient of variation, the lower the consistency of the scores within the original evidence vector.

7. The method for constructing an interactive intelligent customer service knowledge base based on knowledge graphs as described in claim 6, characterized in that, The process of converting two absolute difference values ​​in a difference sequence into proportions relative to the maximum value includes the following steps: The maximum absolute difference value identified is used as input, and this maximum value is set as the reference benchmark for the transformation. The benchmark value is treated as a complete unit, and the transformation of all other difference values ​​will be measured based on this unit. Establish a standard dimensional benchmark; Using the first absolute difference value in the generated standard dimension benchmark and absolute difference sequence as input, the absolute difference value is proportionally calculated to the standard dimension benchmark value to obtain an intermediate value between zero and one, which serves as the first intermediate scaling factor; subsequently, the same proportional calculation is performed using the second absolute difference value in the same standard dimension benchmark and absolute difference sequence to obtain the second intermediate scaling factor. The two intermediate scaling factors are output directly to form a set of scaling values ​​containing two elements; the set of scaling values ​​consists of two scaling values.

8. The method for constructing an interactive intelligent customer service knowledge base based on knowledge graphs as described in claim 1, characterized in that, The process of online adjustment and enhancement of node attributes and associated edge weights includes the following steps: The system takes information confirmed, supplemented, or corrected by customer service representatives or users in real-time conversations as input; combines the sub-graph generated in real time for the current conversation, and maps information fragments to specific operation types for knowledge nodes or related edges; each mapping generates a pending instruction containing the operation target, operation content, and original conversation fragment; all instructions constitute a set of graph operation instructions to be verified. Using the generated set of graph operation instructions to be verified as input, a comprehensive confidence score is calculated for each graph operation instruction to be verified; the three scores are merged to obtain a comprehensive confidence score; only when the comprehensive confidence score exceeds a preset threshold will the corresponding graph operation instruction to be verified be retained and a comprehensive confidence score be attached, generating an enhanced set of operations with confidence rating; The set of enhancement operations with credibility ratings is used as input and applied to the initial knowledge graph; After a series of timely and orderly adjustments and dissemination, the initial knowledge graph was updated into a new online version with optimized content and weights. This new version serves as the output of the interactive power service knowledge base after the current closed-loop iteration and is ready for the next dialogue service.

9. The method for constructing an interactive intelligent customer service knowledge base based on knowledge graphs as described in claim 1, characterized in that, Structured business data, unstructured dialogue records, and real-time power grid status data are extracted from multiple independent business systems of the State Grid Customer Service Center. These data are then processed using semantic alignment and timestamp fusion technologies to obtain a multi-source heterogeneous power service data fusion body.

10. A knowledge graph-based interactive intelligent customer service knowledge base construction system, used to implement the knowledge graph-based interactive intelligent customer service knowledge base construction method as described in any one of claims 1 to 9, characterized in that, The knowledge graph-based interactive intelligent customer service knowledge base construction system includes: The data fusion module is responsible for extracting structured business data, unstructured dialogue records, and real-time power grid status data from multiple independent business systems of the State Grid Customer Service Center. It processes these data through semantic alignment and timestamp fusion technology to obtain a multi-source heterogeneous power service data fusion body. The core association module is responsible for inputting the obtained multi-source heterogeneous power service data fusion into the deep relationship mining module for processing. Through a two-way association discovery mechanism, it identifies and establishes three types of core associations: implicit association between user issues and power grid equipment status, version association between business rules and real-time policy documents, and cross-work order time-series association between fault phenomena and handling measures; forming an initial power service knowledge graph with dynamic weights. The enhancement processing module is responsible for connecting the initial power service knowledge graph with dynamic weights with the real-time inflow of customer interaction conversations, and processing it through an interactive graph enhancement engine. In each round of customer dialogue, it retrieves answers based on the graph and performs context-sensitive subgraph extraction and correction operations. Based on the knowledge nodes involved in the current conversation and their dynamic weights, it generates a subgraph for the current question in real time. Information confirmed, supplemented, or corrected by customer service representatives or users in the dialogue is fed back to the initial knowledge graph according to preset confidence assessment rules, and the node attributes and associated edge weights are adjusted and enhanced online to form a closed-loop evolving interactive power service knowledge base.

Images