A knowledge graph-based customer service knowledge base construction method and system

Abstract

This invention relates to the field of intelligent customer service, specifically disclosing a method and system for constructing a customer service knowledge base based on a knowledge graph. The method includes: identifying implicit needs by analyzing the spatiotemporal correlation between user actions and customer service dialogues, generating enhanced knowledge units that integrate explicit and implicit knowledge; constructing a knowledge graph, achieving adaptive optimization of the knowledge graph through real-time monitoring of relationship strength changes, forming a dynamic knowledge graph; capturing the user's current contextual information, matching it with the dynamic knowledge graph to generate contextualized knowledge subgraphs, and using multi-path reasoning to generate personalized response content; the system includes modules for implementing the above steps. This invention effectively solves the shortcomings of traditional customer service knowledge bases in implicit knowledge mining, dynamic updating, and deep contextual awareness, achieving improvements in knowledge acquisition from partial to comprehensive, knowledge base from static to dynamic, and responses from general to personalized, significantly enhancing the cognitive depth and service quality of intelligent customer service systems.

Description

Technical Field

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

[0002] In the current field of intelligent customer service systems, knowledge graph-based customer service knowledge base construction technology has become a key means to improve service quality and efficiency. However, existing technical solutions still have significant technical bottlenecks in terms of the completeness of knowledge acquisition, the timeliness of knowledge updates, the accuracy of context awareness, and cross-conversation continuity, which restricts the further development of intelligent customer service systems.

[0003] First, traditional knowledge acquisition mechanisms suffer from a significant lack of tacit knowledge: existing customer service knowledge base construction methods primarily rely on explicit dialogue content and structured product documents, failing to effectively integrate tacit knowledge containing crucial needs information, such as user behavior and interaction patterns. For example, behavioral data such as user page browsing time before consultation, changes in search keywords, and function click sequences often contain unexpressed, deeper needs. However, current technologies lack spatiotemporal correlation analysis between this behavioral data and dialogue content, resulting in incomplete knowledge acquisition and an inability to form a true panoramic view of user needs.

[0004] Secondly, the static architecture of knowledge graphs cannot adapt to dynamically changing business environments: current customer service systems based on knowledge graphs mostly adopt a periodic batch update mechanism, lacking real-time monitoring and adaptive optimization capabilities for knowledge decay. In practical applications, factors such as product feature updates, changes in user needs, and adjustments to business rules can all lead to a continuous weakening of the strength of node relationships in the knowledge graph. Existing technologies have failed to establish an effective knowledge decay monitoring model and an automatic triggering update mechanism, causing the knowledge base to gradually deviate from actual needs over time, resulting in a continuous decline in service accuracy.

[0005] Third, the superficial nature of context awareness limits the personalization of services: Traditional customer service systems typically employ a "one-size-fits-all" response strategy when processing user queries, failing to fully consider multi-dimensional contextual factors such as user device type, access time, and historical preferences. Although some advanced systems have attempted to introduce basic context awareness, they lack a mechanism for deep matching of contextual information with knowledge graphs, making it impossible to generate knowledge subgraphs that truly fit the user's current state, resulting in insufficient personalization of response content.

[0006] Fourth, the break in the continuity of knowledge across sessions becomes a hidden obstacle to long-term service optimization: Existing technologies treat each customer service session as an independent event, lacking the tracking and analysis of the evolution of user needs. Multiple inquiries by users at different points in time often reflect gradual changes in their cognitive level and needs, but traditional systems cannot capture this pattern of knowledge evolution, causing each session to start from scratch. This makes it impossible to provide coherent services based on the cognitive progress of past sessions, seriously affecting the consistency of user experience and service efficiency.

[0007] Furthermore, existing technologies suffer from technological silos in key areas such as multi-source data fusion, dynamic knowledge evolution, and context-adaptive responses. Although some research has optimized individual technical aspects, such as improving intent recognition accuracy through attention mechanisms or enhancing response tone through sentiment analysis, these improvements have failed to form a system-wide technological loop and cannot fundamentally solve the comprehensive challenges in building customer service knowledge bases.

[0008] In summary, existing knowledge graph-based customer service knowledge base construction methods have significant shortcomings in areas such as tacit knowledge mining, dynamic knowledge updating, deep context awareness, and cross-conversation continuity. There is an urgent need for an innovative technical solution that can systematically solve these problems in order to achieve a qualitative leap in intelligent customer service systems from "static knowledge bases" to "dynamic intelligent agents". Summary of the Invention

[0009] The purpose of this invention is to address the shortcomings of existing technologies by proposing a method and system for constructing a customer service knowledge base based on knowledge graphs.

[0010] To achieve the above objectives, the present invention adopts the following technical solution:

[0011] A method for constructing a customer service knowledge base based on knowledge graphs, comprising:

[0012] By analyzing the spatiotemporal correlation between user behavior and customer service dialogue, implicit needs are identified, and enhanced knowledge units that integrate explicit and implicit knowledge are generated.

[0013] A knowledge graph is constructed based on the enhanced knowledge units, and adaptive optimization of the knowledge graph is achieved by monitoring changes in relationship strength in real time, thus forming a dynamically evolving knowledge graph.

[0014] Capture the user's current context information, match it with the dynamically evolving knowledge graph to generate a contextualized knowledge subgraph, and use multi-path reasoning to generate personalized response content.

[0015] As a further technical solution of the present invention, the step of analyzing the spatiotemporal correlation between user operation behavior and customer service dialogue, identifying implicit needs, and generating enhanced knowledge units that integrate explicit and implicit knowledge specifically includes: acquiring customer service dialogue records, product documents and user operation behavior data, and establishing a multi-source heterogeneous data pool;

[0016] By aligning user action sequences with corresponding customer service dialogue records through spatiotemporal correlation analysis, a behavior-dialogue correlation mapping table is formed. Based on the behavior-dialogue correlation mapping table, implicit demand nodes of users in the dialogue process are identified, and the implicit demand nodes are fused with explicit dialogue content to generate enhanced knowledge units.

[0017] As a further technical solution of the present invention, the method of identifying implicit demand nodes of users in the dialogue process based on behavior-dialogue association mapping table, and merging implicit demand nodes with explicit dialogue content to generate enhanced knowledge units, specifically includes: analyzing users' high-frequency search behavior before dialogue and identifying deep interest points that are not explicitly expressed.

[0018] Detect users' rapid page-jumping behavior during conversations to identify urgency and preference tendencies;

[0019] Analyze users' repeated actions on specific product functions to identify potential usage obstacles;

[0020] The identified implicit demand nodes are semantically fused with explicit dialogue content to generate enhanced knowledge units containing complete user intent.

[0021] As a further technical solution of the present invention, the construction of a knowledge graph based on enhanced knowledge units, and the adaptive optimization of the knowledge graph by real-time monitoring of relationship strength changes to form a dynamically evolving knowledge graph, specifically includes: constructing an initial knowledge graph based on the enhanced knowledge units; detecting the trend of relationship strength changes between nodes in the knowledge graph by continuously monitoring real-time interaction data between users and the customer service system; and automatically triggering a knowledge update mechanism when the relationship strength decay exceeds a preset threshold, re-evaluating and optimizing the knowledge graph topology to form a dynamically evolving knowledge graph.

[0022] As a further technical solution of the present invention, the construction of the initial knowledge graph based on the enhanced knowledge unit specifically includes: extracting entities, attributes and relations from the enhanced knowledge unit, and constructing entity-relation-entity triples;

[0023] The triples are stored in a graph database to form an initial knowledge graph structure;

[0024] Based on the implicit requirement nodes in the enhanced knowledge units, requirement association edges are established in the knowledge graph.

[0025] As a further technical solution of the present invention, the step of continuously monitoring the real-time interaction data between users and the customer service system to detect the changing trend of the relationship strength between nodes in the knowledge graph specifically includes: real-time collection of user query logs, click feedback and conversation evaluation data;

[0026] Calculate the usage frequency, correlation query accuracy, and user satisfaction score for each relation edge in the knowledge graph;

[0027] Based on time series analysis, a decay model of relationship strength is established to predict the trend of strength change of each relationship edge.

[0028] As a further technical solution of the present invention, the step of capturing the user's current context information, matching it with a dynamically evolving knowledge graph to generate a contextualized knowledge subgraph, and generating personalized response content using multi-path reasoning specifically includes: capturing the user's current context information in real time when the user initiates a customer service inquiry, including device type, access time, and historical behavioral preferences; performing multi-dimensional matching of the context information with a dynamically evolving knowledge graph to generate a contextualized knowledge subgraph; and generating personalized response content based on the contextualized knowledge subgraph and the user's current query intent using a multi-path reasoning mechanism.

[0029] As a further technical solution of the present invention, the step of performing multi-dimensional matching between contextual information and dynamically evolving knowledge graphs to generate contextualized knowledge subgraphs specifically includes: filtering and adapting knowledge nodes based on device type, and distinguishing differentiated knowledge between mobile and PC terminals;

[0030] By combining access time characteristics, activate time-sensitive knowledge nodes to highlight important information in the current time period;

[0031] Based on users' historical behavior preferences, strengthen knowledge paths with high relevance and weaken irrelevant knowledge branches;

[0032] The graph pruning algorithm is used to extract the knowledge subgraph that is most relevant to the current context.

[0033] As a further technical solution of the present invention, the generation of personalized response content based on contextualized knowledge subgraphs and the user's current query intent using a multi-path reasoning mechanism specifically includes: exploring multiple reasoning paths in parallel within the contextualized knowledge subgraph and evaluating the confidence level of each path;

[0034] By combining the keywords of the query intent with contextual features, the matching weight of each reasoning path is calculated;

[0035] Based on a comprehensive score of path confidence and matching weight, the optimal inference path is selected to generate a basic response;

[0036] Based on device type and user preferences, the basic response is formatted and reorganized to generate the final personalized response content. A customer service knowledge base construction system based on knowledge graphs is characterized by implementing a method for constructing a customer service knowledge base based on knowledge graphs, including:

[0037] The multi-source data acquisition and processing module is used to acquire customer service dialogue records, product documents and user operation behavior data to establish a multi-source heterogeneous data pool;

[0038] The Behavior-Dialogue Association Analysis module is used to align user operation behavior sequences with corresponding customer service dialogue records through spatiotemporal association analysis, forming a behavior-dialogue association mapping table, and based on this, identifying implicit demand nodes and generating enhanced knowledge units that integrate explicit and implicit knowledge.

[0039] The knowledge graph construction and storage module is used to extract entities, attributes and relationships based on the enhanced knowledge units to construct an initial knowledge graph, and to store the initial knowledge graph in a graph database;

[0040] The knowledge graph dynamic evolution module is used to monitor the changes in the strength of relationships between nodes in the knowledge graph in real time, and automatically trigger the knowledge update mechanism when the relationship strength decays beyond a preset threshold, so as to realize the adaptive optimization of the knowledge graph and form a dynamically evolving knowledge graph.

[0041] The user context capture and matching module is used to capture the current context information of a user when the user initiates a consultation, and match the context information with the dynamically evolving knowledge graph to generate a contextualized knowledge subgraph;

[0042] The multi-path reasoning and response generation module is used to generate personalized response content based on the contextualized knowledge subgraph and the user's current query intent, using a multi-path reasoning mechanism.

[0043] The beneficial effects of this invention are as follows:

[0044] 1. It has deepened and improved the knowledge structure: By integrating implicit knowledge such as user behavior with explicit dialogue content, it has constructed enhanced knowledge units that better reflect the user's true intentions, fundamentally improving the system's cognitive depth and knowledge completeness.

[0045] 2. The knowledge base has achieved self-evolution and continuous optimization: By monitoring the strength of knowledge graph relationships in real time and automatically triggering the update mechanism, the knowledge base has the ability to evolve dynamically, proactively adapt to business changes, and effectively ensure the long-term accuracy and timeliness of services.

[0046] 3. It achieves precise and personalized response strategies: By matching the user's real-time context with a dynamic knowledge graph and adopting a multi-path reasoning mechanism, it can generate response content that is deeply tailored to the user's current state and preferences, significantly improving the intelligence level of the service and the user experience. Attached Figure Description

[0047] Figure 1 This is a flowchart of a customer service knowledge base construction method based on knowledge graph proposed in this invention;

[0048] Figure 2 This is a module diagram of a customer service knowledge base construction system based on knowledge graphs proposed in this invention.

[0049] Figure 3 This is a comparison diagram of the effects of the method of the present invention and the prior art method in Example 1. Detailed Implementation

[0050] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0051] Please see the appendix Figure 1 A method for constructing a customer service knowledge base based on knowledge graphs, comprising:

[0052] S1. By analyzing the spatiotemporal correlation between user actions and customer service dialogues, implicit needs are identified, and enhanced knowledge units that integrate explicit and implicit knowledge are generated; specifically including:

[0053] S11. Obtain customer service dialogue records in real time via API interface, including text dialogue and voice-to-text data; crawl product documents and manuals using web scraping technology; collect user operation behavior data through front-end tracking and log analysis, including page dwell time, click sequence, search keywords, and operation trajectory; store these three types of data uniformly in a distributed database to form a multi-source heterogeneous data pool; S12. Based on timestamps and session IDs, align user operation behavior sequences with corresponding customer service dialogue records in time and space, and construct a behavior-dialogue association mapping table; specifically including: S121. Establish the correspondence between the operation behavior timeline and the dialogue timeline based on the session start time:

[0054] Let the session start time be The time points on the operation behavior timeline are The time points on the dialogue timeline are For each operation action time point Find the most recent conversation time. ,satisfy: ,in: Represents the time points on the dialogue timeline corresponding to the time of the action. The index of the time point with the smallest time difference establishes the correspondence between the time point of the operation and the time point of the dialogue;

[0055] S122. Eliminate the time scale difference between operational behavior and dialogue content through Dynamic Time Warping (DTW):

[0056] Let the time series of the operation be... The time sequence of the dialogue content is as follows The goal of DTW is to find an optimal alignment path. , so that: ,in: It is an operational behavior and dialogue content The distance between them is measured using Euclidean distance;

[0057] S123. Calculate the correlation between behavior sequences and dialogue content based on three dimensions: operation frequency, operation type, and operation object:

[0058] Let the sequence of operations be... and dialogue content sequence The features in the three dimensions are as follows and correlation Calculated using a weighted sum: ,in: These are weighting coefficients, satisfying... ; It is a similarity measurement function, using cosine similarity.

[0059] S13. By analyzing user behavior characteristics, implicit demand nodes are identified and integrated with explicit dialogue content to generate enhanced knowledge units containing complete user intent, providing a foundation for subsequent knowledge graph construction; specifically including:

[0060] S131. Suppose the user's search keyword sequence before the conversation is... Statistical analysis of the search frequency for each keyword ,like ( If the threshold is used, then the keyword is considered... The corresponding deep interest points are implicit demand nodes;

[0061] S132. Suppose the page navigation sequence during the user's conversation is as follows: Calculate the time interval between adjacent page jumps. ,like ( If the threshold is used, then the user is considered to have a positive impression of the page. arrive The redirection behavior shows a high degree of urgency and preference, and the corresponding page content is a node of implicit demand;

[0062] S133. Suppose the user's sequence of operations on a specific product function is as follows: Count the number of operations for each function. ,like ( If the threshold is set to , then the user is considered to have a potential obstacle to using the product function, and the corresponding implicit demand node is the function itself; S134. Let the set of implicit demand nodes be . The explicit dialogue content set is Enhanced knowledge units are generated through semantic fusion. Using word embedding technology to... and Convert to vector representation, and then fuse them using a weighted vector sum: ,in: It is a fusion weight. It is a vector representation function.

[0063] S2. Construct a knowledge graph based on the enhanced knowledge units, and achieve adaptive optimization of the knowledge graph by real-time monitoring of changes in relationship strength, forming a dynamically evolving knowledge graph; specifically including:

[0064] S21. Construct an initial knowledge graph based on enhanced knowledge units, specifically including:

[0065] S211. Let the set of enhanced knowledge units be... For each enhanced knowledge unit Extract the entities from it ,property and relationships Construct triples ,in: ;

[0066] S212. All triples Stored in a graph database to form an initial knowledge graph structure;

[0067] S213. Let the set of implicit demand nodes be... For each implicit requirement node Find related entities in the knowledge graph Establish demand-related edges And store it in the graph database.

[0068] S22. By continuously monitoring real-time interaction data between users and the customer service system, detect the changing trends in the strength of relationships between nodes in the knowledge graph, specifically including:

[0069] S221. Real-time collection of user query logs Click to provide feedback and conversation evaluation data ;

[0070] S222. For each relation edge Count the number of times it appears in user query logs Frequency of use Defined as: ,in: This represents the total number of user query logs;

[0071] For each relation edge Count the number of times it was correctly clicked in the click feedback. Accuracy of related queries Defined as: ;

[0072] For each relation edge The average satisfaction score of the participants in the conversation evaluation data was statistically analyzed. User satisfaction rating Defined as: ,in: It is the total number of session evaluation data;

[0073] S223. Assume the strength of the relationship. It is time Based on time series analysis, a decay model is established using the function: ,in: It is the initial relationship strength. It is the attenuation coefficient.

[0074] S23. When the detected relationship strength decay exceeds a preset threshold, the knowledge update mechanism is automatically triggered to re-evaluate and optimize the knowledge graph topology, forming a dynamically evolving knowledge graph, specifically including:

[0075] S231. For each relation edge Calculate its current relationship strength ,like ( If the threshold is used, then the edge of the relationship is considered to decay.

[0076] S232. Knowledge units corresponding to the decay relation edges The accuracy and timeliness of the data were reassessed. The accuracy assessment was conducted through user feedback and expert review, while the timeliness assessment was conducted through data update frequency and user query time distribution.

[0077] S233. Based on the latest user interaction data, perform node merging, relationship reconstruction, and weight adjustment: If two nodes and If multiple relationships exhibit high similarity, they are merged into a single node. Adjust the direction and type of relationships based on user query patterns and feedback; adjust the weights of relationship edges based on changes in relationship strength. : ,in: and These are weighting coefficients, satisfying... ;

[0078] S234. Use incremental learning algorithms, such as online gradient descent or incremental neural networks, to update the knowledge graph, ensuring the continuity and stability of the evolution process and avoiding drastic changes in the global structure due to local updates.

[0079] S3. Capture the user's current contextual information, match it with the dynamically evolving knowledge graph to generate a contextualized knowledge subgraph, and use multi-path reasoning to generate personalized response content; specifically including:

[0080] S31. Parse HTTP request headers to obtain device type ,operating system and browser information Extract access time information, including time periods. Weekly information and holiday markers Obtain historical behavioral preferences from the user profile database, including frequently used functions. Consultation frequency and problem type preference ;

[0081] S32. Perform multi-dimensional matching of contextual information with dynamically evolving knowledge graphs to generate contextualized knowledge subgraphs, specifically including:

[0082] S321. Select suitable knowledge nodes based on device type, assuming the knowledge graph is... ,in: It is a set of nodes. It is an edge set; it differentiates between mobile and PC knowledge and selects appropriate knowledge nodes accordingly:

[0083]

[0084]

[0085] in: and This is a function that determines whether a node is compatible with both mobile and PC platforms.

[0086] S322. Based on the characteristics of access time, activate time-sensitive knowledge nodes to highlight important information in the current time period: Let the set of time-sensitive knowledge nodes be... : ,in: It is a function to determine whether a node is related to the current time period, weekday, or holiday.

[0087] S323. Based on user historical behavior preferences, strengthen knowledge paths with high relevance and weaken irrelevant knowledge branches: Let the user historical behavior preference relevance function be... For each node Calculate its correlation with users' historical behavioral preferences: ,in: These are weighting coefficients, satisfying... ;

[0088] S324. Consider a contextualized knowledge subgraph. Using a graph pruning algorithm, extract the knowledge subgraph most relevant to the current context:

[0089]

[0090]

[0091] in: It is the relevance threshold.

[0092] S33. Based on contextualized knowledge subgraphs and the user's current query intent, a multi-path reasoning mechanism is used to generate personalized response content, specifically including:

[0093] S331. In contextualized knowledge subgraphs Parallel exploration of multiple inference paths Let the path confidence function be... For each path Calculate its confidence level: ,in: It is the edge Confidence level;

[0094] S332. Let the keywords of the query intent be... For each path Calculate its matching weight with query intent and contextual features: ,in: It is a node With keywords Similarity;

[0095] S333. Let the comprehensive scoring function be... : ,in: These are weighting coefficients, satisfying... Based on comprehensive score Select the path with the highest overall score to generate the basic response: ,in: The path with the highest overall score;

[0096] S334. According to equipment type and user preferences The basic response is formatted and reorganized to generate the final personalized response content: ,in: It is an optimization function. For the final personalized response content, Based on the response.

[0097] Please see the appendix Figure 2 A knowledge graph-based customer service knowledge base construction system is provided to implement a knowledge graph-based customer service knowledge base construction method, including:

[0098] The multi-source data acquisition and processing module is used to acquire customer service dialogue records, product documents and user operation behavior data to establish a multi-source heterogeneous data pool;

[0099] The Behavior-Dialogue Association Analysis module is used to align user operation behavior sequences with corresponding customer service dialogue records through spatiotemporal association analysis, forming a behavior-dialogue association mapping table, and based on this, identifying implicit demand nodes and generating enhanced knowledge units that integrate explicit and implicit knowledge.

[0100] The knowledge graph construction and storage module is used to extract entities, attributes and relationships based on the enhanced knowledge units to construct an initial knowledge graph, and to store the initial knowledge graph in a graph database;

[0101] The knowledge graph dynamic evolution module is used to monitor the changes in the strength of relationships between nodes in the knowledge graph in real time, and automatically trigger the knowledge update mechanism when the relationship strength decays beyond a preset threshold, so as to realize the adaptive optimization of the knowledge graph and form a dynamically evolving knowledge graph.

[0102] The user context capture and matching module is used to capture the current context information of a user when the user initiates a consultation, and match the context information with the dynamically evolving knowledge graph to generate a contextualized knowledge subgraph;

[0103] The multi-path reasoning and response generation module is used to generate personalized response content based on the contextualized knowledge subgraph and the user's current query intent, using a multi-path reasoning mechanism.

[0104] Example 1

[0105] To verify the effectiveness of the method proposed in this invention, a comparative test was designed and implemented.

[0106] 1. Test Setup

[0107] Test group: A customer service knowledge base system constructed using the method proposed in this invention.

[0108] Control group: A traditional customer service system based on keyword matching and static knowledge base.

[0109] Test environment: Two systems were deployed on the same cloud server cluster, using the same hardware configuration and network environment.

[0110] Data source: Anonymized customer service dialogue records, user operation logs, and product documents from a certain e-commerce platform for three consecutive months were used as training and testing data.

[0111] Evaluation metrics: Response accuracy, the percentage of responses provided by the system that are judged as "correct and complete" by domain experts; Implicit needs coverage, the percentage of implicit needs that the system is able to identify and respond to that are not explicitly expressed by the user; User satisfaction, the average satisfaction rating (out of 1-5) given by test users after the test session ends; Knowledge update timeliness, the average time required from the appearance of new user feedback to the knowledge base completing adaptive updates.

[0112] 2. Testing Process

[0113] System initialization: The test group and the control group use the same historical data to build the initial knowledge base.

[0114] Static testing: The two systems are tested using a test set containing 500 standard questions to evaluate their basic responsiveness.

[0115] Dynamic Interaction Test: Recruit 500 real users to interact with two systems in a simulated e-commerce customer service scenario through multiple rounds of interaction. This test scenario is specifically designed with complex questions that require users to combine their operation history (such as repeatedly viewing a product comparison page) to answer accurately.

[0116] Long-term evolution observation: During the 30-day experimental period, new user interaction data and product update information were continuously injected into the two systems to observe the stability and evolution capabilities of the system performance.

[0117] 3. Test Results and Analysis

[0118] The effects of the method of the present invention and the existing technology are compared in Table 1 and below. Figure 3 As shown:

[0119] Table 1: Comparison of Results

[0120] Evaluation indicators Prior art (control group) Method of the present invention (test group) Effect Improvement and Analysis Response accuracy 68.5% 89.2% An improvement of 20.7%. This invention significantly improves the accuracy of understanding and responding to complex and fuzzy problems by generating enhanced knowledge units through the fusion of tacit knowledge and using contextualized knowledge subgraphs for multi-path reasoning. Implicit demand coverage 15.0% 73.5% An improvement of 58.5%. This invention accurately identifies the underlying intentions behind users' high-frequency search and rapid navigation behaviors through spatiotemporal correlation analysis, and integrates them into a knowledge graph, thereby achieving effective capture and response to implicit needs. User satisfaction 3.2 points 4.5 points An improvement of 1.3 points. The user experience was significantly improved, and satisfaction was greatly increased, thanks to more accurate and personalized responses that anticipated user needs. Timeliness of knowledge updates Approximately 7 days (depending on manual batch updates) Near real-time (automatically completed within minutes) The accuracy rate has been improved from the "day" level to the "minute" level. The dynamic evolution mechanism of this invention monitors the strength of relationships in real time and automatically triggers updates, enabling the knowledge base to keep up with business changes and avoiding a decline in accuracy due to outdated knowledge. Cross-session continuity Weak (independent for each session) powerful Fundamental improvement. This invention, through a continuously evolving knowledge graph and user profile, can remember a user's historical preferences and cognitive progress, providing coherent and progressive services across multiple consultations, whereas traditional systems start almost from scratch for each session.

[0121] 4. Conclusion

[0122] The comparative tests described above demonstrate that, compared to existing technologies, the method proposed in this invention achieves significant performance improvements across multiple key dimensions, including response accuracy, understanding of implicit needs, user satisfaction, knowledge base self-evolution capabilities, and cross-session service continuity. This method effectively addresses the technical bottlenecks of traditional systems in terms of knowledge completeness, timeliness, and context awareness, realizing a qualitative leap in customer service systems from a "static knowledge base" to a "dynamic intelligent agent," thus possessing extremely high application value and market potential.

[0123] As can be seen from the above description, the embodiments of the present invention achieve the following technical effects:

[0124] This invention achieves a leap from "partial explicitness" to "comprehensive enhancement" in knowledge acquisition, significantly improving the cognitive depth and response accuracy of customer service systems. Breaking through the limitations of traditional methods that rely solely on explicit dialogue content, this invention aligns and integrates user actions (such as search sequences, page jumps, and repeated function operations) with customer service dialogues through spatiotemporal correlation analysis, identifying and constructing enhanced knowledge units containing implicit user needs. This mechanism enables the customer service system to understand the user's "implied meaning," achieving a deep insight into user intent. As shown in the embodiments, this method increases the implicit need coverage rate from 15% to 73.5%, thereby significantly improving the overall response accuracy from 68.5% to 89.2%, fundamentally solving problems such as superficial responses and irrelevant answers caused by incomplete knowledge acquisition.

[0125] This invention achieves the evolution of the knowledge base architecture from "static lag" to "dynamic intelligence," ensuring the timeliness and long-term accuracy of services. It constructs a dynamically evolving knowledge graph with adaptive optimization capabilities. Its core lies in the introduction of a relationship strength decay model and a real-time monitoring mechanism, which automatically detects the "aging" phenomenon of node relationships in the knowledge graph and automatically triggers a knowledge update mechanism when the strength falls below a threshold, using incremental learning and other technologies to optimize the topology. This allows the knowledge base to autonomously evolve in line with product updates and changes in user needs, improving the timeliness of knowledge updates from "days" (relying on manual intervention) to "minutes" (automatic completion), effectively curbing the decline in knowledge base accuracy over time. Simultaneously, the dynamically evolving knowledge graph naturally records the evolutionary trajectory of user needs, achieving cross-session knowledge continuity and service coherence, providing users with a continuously optimized long-term service experience.

[0126] This invention represents a significant leap from generic, one-size-fits-all responses to highly contextualized and precise ones, greatly optimizing user experience and satisfaction. The invention proposes a contextualized knowledge subgraph generation and multi-path reasoning mechanism. During response generation, the system captures multi-dimensional contextual information such as user device, time, and historical preferences in real time. It then rapidly extracts the most relevant knowledge subgraphs from a dynamic knowledge graph using a graph pruning algorithm, and explores multiple reasoning paths in parallel. The optimal response is generated by combining confidence level and contextual matching weights. This mechanism ensures that every response is deeply personalized. The system not only provides the correct answer but also intelligently restructures the format, detail, and focus of the response based on contextual factors such as whether the user is using a mobile phone or computer, whether they are accessing the system on a holiday, and whether they have specific functional preferences. As the test results show, this series of innovations significantly improved user satisfaction from 3.2 to 4.5, marking a successful transformation of the customer service system from a rigid question-and-answer machine to a user-aware intelligent assistant.

[0127] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of the invention is limited to these examples; within the framework of the invention, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of the different aspects of the invention as described above, which are not provided in detail for the sake of brevity.

[0128] This invention is intended to cover all such substitutions, modifications, and variations that fall within the broad scope of this specification. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for constructing a customer service knowledge base based on knowledge graphs, characterized in that, include: By analyzing the spatiotemporal correlation between user behavior and customer service dialogue, implicit needs are identified, and enhanced knowledge units that integrate explicit and implicit knowledge are generated. A knowledge graph is constructed based on the enhanced knowledge units, and adaptive optimization of the knowledge graph is achieved by monitoring changes in relationship strength in real time, thus forming a dynamically evolving knowledge graph. Capture the user's current context information, match it with the dynamically evolving knowledge graph to generate a contextualized knowledge subgraph, and use multi-path reasoning to generate personalized response content.

2. The method for constructing a customer service knowledge base based on a knowledge graph according to claim 1, characterized in that, The method involves analyzing the spatiotemporal correlation between user actions and customer service dialogues to identify implicit needs and generate enhanced knowledge units that integrate explicit and implicit knowledge. Specifically, this includes: acquiring customer service dialogue records, product documents, and user action data to establish a multi-source heterogeneous data pool. By aligning user action sequences with corresponding customer service dialogue records through spatiotemporal correlation analysis, a behavior-dialogue correlation mapping table is formed. Based on the behavior-dialogue correlation mapping table, implicit demand nodes of users in the dialogue process are identified, and the implicit demand nodes are fused with explicit dialogue content to generate enhanced knowledge units.

3. The method for constructing a customer service knowledge base based on a knowledge graph according to claim 2, characterized in that, The behavior-dialogue association mapping table identifies implicit demand nodes of users during the dialogue process and integrates implicit demand nodes with explicit dialogue content to generate enhanced knowledge units. Specifically, it includes: analyzing users' high-frequency search behavior before the dialogue and identifying deep interest points that are not explicitly expressed. Detect users' rapid page-jumping behavior during conversations to identify urgency and preference tendencies; Analyze users' repeated actions on specific product functions to identify potential usage obstacles; The identified implicit demand nodes are semantically fused with explicit dialogue content to generate enhanced knowledge units containing complete user intent.

4. The method for constructing a customer service knowledge base based on a knowledge graph according to claim 1, characterized in that, The method of constructing a knowledge graph based on enhanced knowledge units, and achieving adaptive optimization of the knowledge graph by real-time monitoring of changes in relationship strength to form a dynamically evolving knowledge graph, specifically includes: constructing an initial knowledge graph based on the enhanced knowledge units; detecting the trend of relationship strength changes between nodes in the knowledge graph by continuously monitoring real-time interaction data between users and the customer service system; and automatically triggering a knowledge update mechanism when the relationship strength decay exceeds a preset threshold, re-evaluating and optimizing the knowledge graph topology to form a dynamically evolving knowledge graph.

5. The method for constructing a customer service knowledge base based on a knowledge graph according to claim 4, characterized in that, The construction of the initial knowledge graph based on enhanced knowledge units specifically includes: extracting entities, attributes, and relations from the enhanced knowledge units, and constructing entity-relation-entity triples; The triples are stored in a graph database to form an initial knowledge graph structure; Based on the implicit requirement nodes in the enhanced knowledge units, requirement association edges are established in the knowledge graph.

6. The method for constructing a customer service knowledge base based on a knowledge graph according to claim 5, characterized in that, The method of continuously monitoring real-time interaction data between users and the customer service system to detect the changing trend of relationship strength between nodes in the knowledge graph specifically includes: real-time collection of user query logs, click feedback, and conversation evaluation data; Calculate the usage frequency, correlation query accuracy, and user satisfaction score for each relation edge in the knowledge graph; Based on time series analysis, a decay model of relationship strength is established to predict the trend of strength change of each relationship edge.

7. The method for constructing a customer service knowledge base based on a knowledge graph according to claim 1, characterized in that, The process of capturing the user's current contextual information, matching it with a dynamically evolving knowledge graph to generate a contextualized knowledge subgraph, and using multi-path reasoning to generate personalized response content specifically includes: capturing the user's current contextual information in real time when the user initiates a customer service inquiry, including device type, access time, and historical behavioral preferences; performing multi-dimensional matching of the contextual information with a dynamically evolving knowledge graph to generate a contextualized knowledge subgraph; and generating personalized response content based on the contextualized knowledge subgraph and the user's current query intent using a multi-path reasoning mechanism.

8. The method for constructing a customer service knowledge base based on a knowledge graph according to claim 7, characterized in that, The process of performing multi-dimensional matching of contextual information with dynamically evolving knowledge graphs to generate contextualized knowledge subgraphs specifically includes: selecting appropriate knowledge nodes based on device type and distinguishing differentiated knowledge between mobile and PC terminals; By combining access time characteristics, activate time-sensitive knowledge nodes to highlight important information in the current time period; Based on users' historical behavior preferences, strengthen knowledge paths with high relevance and weaken irrelevant knowledge branches; The graph pruning algorithm is used to extract the knowledge subgraph that is most relevant to the current context.

9. The method for constructing a customer service knowledge base based on a knowledge graph according to claim 8, characterized in that, The contextualized knowledge subgraph and the user's current query intent are used to generate personalized response content using a multi-path reasoning mechanism, specifically including: exploring multiple reasoning paths in parallel in the contextualized knowledge subgraph and evaluating the confidence of each path; By combining the keywords of the query intent with contextual features, the matching weight of each reasoning path is calculated; Based on a comprehensive score of path confidence and matching weight, the optimal inference path is selected to generate a basic response; Based on device type and user preferences, the basic response is formatted and reorganized to generate the final personalized response content.

10. A customer service knowledge base construction system based on knowledge graphs, characterized in that, A method for constructing a customer service knowledge base based on a knowledge graph, as described in any one of claims 1-9, includes: The multi-source data acquisition and processing module is used to acquire customer service dialogue records, product documents and user operation behavior data to establish a multi-source heterogeneous data pool; The Behavior-Dialogue Association Analysis module is used to align user operation behavior sequences with corresponding customer service dialogue records through spatiotemporal association analysis, forming a behavior-dialogue association mapping table, and based on this, identifying implicit demand nodes and generating enhanced knowledge units that integrate explicit and implicit knowledge. The knowledge graph construction and storage module is used to extract entities, attributes and relationships based on the enhanced knowledge units to construct an initial knowledge graph, and to store the initial knowledge graph in a graph database; The knowledge graph dynamic evolution module is used to monitor the changes in the strength of relationships between nodes in the knowledge graph in real time, and automatically trigger the knowledge update mechanism when the relationship strength decays beyond a preset threshold, so as to realize the adaptive optimization of the knowledge graph and form a dynamically evolving knowledge graph. The user context capture and matching module is used to capture the current context information of a user when the user initiates a consultation, and match the context information with the dynamically evolving knowledge graph to generate a contextualized knowledge subgraph; The multi-path reasoning and response generation module is used to generate personalized response content based on the contextualized knowledge subgraph and the user's current query intent, using a multi-path reasoning mechanism.

Images