A knowledge graph-based intelligent consulting service system

Abstract

This invention relates to the field of intelligent consulting service systems, and discloses an intelligent consulting service system based on a knowledge graph. The system includes a user terminal, a knowledge processing device, and a cloud-based inference center. The user terminal inputs the text of the consulting question and contextual semantic data; the knowledge processing device receives instructions to update the status of knowledge nodes; the knowledge extraction unit in the cloud-based inference center determines the associated regions, generates semantic feature curves and maps them, the relationship matching unit optimizes the inference path and generates a decision scheme, and the decision generation unit performs inference or intervention operations according to the device status. This system can accurately understand semantics, optimize inference paths, and handle anomalies. By dynamically updating the knowledge graph, it provides users with efficient and accurate intelligent consulting services.

Description

Technical Field

[0001] This invention relates to the field of intelligent consulting service system technology, specifically to an intelligent consulting service system based on knowledge graphs. Background Technology

[0002] With the rapid development of information technology, intelligent consultation service systems have been widely applied in many fields such as finance, healthcare, and education. Traditional intelligent consultation service systems are mostly based on rule bases or simple keyword matching technology. This approach often struggles to accurately understand user intent when dealing with complex semantics and ambiguity, resulting in poor accuracy and practicality of consultation results. For example, in medical consultation scenarios, if users describe their symptoms vaguely or unprofessionally, traditional systems may fail to accurately match the corresponding disease knowledge, providing incorrect or incomplete advice.

[0003] While some intelligent consulting service systems employing machine learning technologies have improved semantic understanding to some extent, they struggle to handle complex situations such as semantic shifts and anomalous semantics due to a lack of in-depth analysis of knowledge structures and semantic relationships. When users suddenly change topics during a consultation or input logically contradictory information, the system is prone to reasoning errors and fails to provide effective solutions. Furthermore, the existing knowledge update mechanisms of intelligent consulting service systems are not flexible enough to adapt to dynamic changes in knowledge in real time, resulting in outdated knowledge graphs that fail to reflect the latest industry knowledge and technological advancements.

[0004] Furthermore, traditional systems have shortcomings in reasoning path optimization and anomaly handling. During reasoning, the inability to rationally plan reasoning paths based on the semantic distribution of the knowledge graph can lead to low reasoning efficiency. When system anomalies occur, the lack of effective intervention mechanisms prevents rapid recovery, severely impacting user experience and service quality. With users' increasing demands for accuracy, efficiency, and real-time performance in intelligent consulting services, the development of an intelligent consulting service system capable of deep semantic understanding, flexible anomaly handling, and dynamic knowledge updates is urgently needed. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent consulting service system based on knowledge graphs to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an intelligent consultation service system based on knowledge graph, the system comprising a user terminal, a knowledge processing device, and a cloud-based reasoning center, wherein the user terminal and the knowledge processing device are respectively communicatively connected to the cloud-based reasoning center, and the cloud-based reasoning center comprises a knowledge extraction unit, a relation matching unit, and a decision generation unit;

[0007] The user terminal is used to input consultation questions and contextual semantic data in real time.

[0008] The knowledge processing device is used to receive processing instructions from the cloud-based reasoning center to update the status of knowledge nodes.

[0009] The knowledge extraction unit is used to record the semantic jump period and logical recovery period during the operation of the knowledge processing device, determine the associated region based on the semantic jump period and logical recovery period, control the knowledge processing device to maintain a preset constraint threshold for abnormal semantics and record the changing parameters, generate semantic feature curves based on the changing parameters and perform semantic distribution mapping on the knowledge graph.

[0010] The relation matching unit is used to optimize the reasoning path of the knowledge graph that has completed semantic distribution mapping and generate a decision scheme.

[0011] The decision generation unit includes a regular reasoning unit and an emergency reasoning unit. The regular reasoning unit is used to control the knowledge processing device to execute the baseline reasoning mode when the knowledge processing device is in a normal state. The emergency reasoning unit is used to control the knowledge processing device to perform intervention operations in the associated area according to the decision scheme when the knowledge processing device detects an abnormal state.

[0012] Preferably, the user terminal includes a terminal housing and a multimodal input device, a semantic parsing device, a logic evaluation module, and a communication module disposed in the housing;

[0013] The multimodal input device is used to simultaneously acquire text input data and voice input data;

[0014] The semantic parsing device is used to extract contextual features from the input data;

[0015] The logical evaluation module is used to calculate the similarity between the current context-related features and the historical standard patterns based on the attention mechanism, determine the semantic coherence index, and trigger a correction signal based on the calculation result.

[0016] The communication module is used to upload input data to the cloud-based inference center.

[0017] Preferably, the knowledge processing device includes a processing chassis and a logic operation module, a parameter detection module, an instruction response module, and a protocol conversion module disposed in the chassis.

[0018] Preferably, the semantic transition period and logical recovery period during the operation of the knowledge processing device include:

[0019] When the knowledge processing device detects a semantic jump, it records the timestamp data of the jump occurrence time, continuously monitors the logical recovery critical point through the parameter detection module, and records the sequence number of the recovery time.

[0020] Preferably, determining the associated region based on the semantic jump period and the logical recovery period includes:

[0021] Generate jump coordinates and recovery coordinates based on timestamp data and sequence numbers;

[0022] Spatially correlate the jump coordinates, recovery coordinates, and user terminal location to form a closed region, and mark the closed region as the correlated region.

[0023] Preferably, the control knowledge processing device operates under a preset constraint threshold for abnormal semantics and records changing parameters. Generating semantic feature curves based on these changing parameters and performing semantic distribution mapping on the knowledge graph includes the following steps:

[0024] The knowledge processing device is controlled to adjust the output intensity of the associated region, and semantic parameters are obtained in real time through the parameter detection module.

[0025] If the semantic parameters exceed the preset constraint threshold range, the knowledge processing device will be controlled to maintain the preset constraint threshold operation.

[0026] Continuously record the semantic output values ​​of the knowledge processing device to form a set of changing parameters;

[0027] Based on the set of changing parameters, a hidden Markov model algorithm is used to generate semantic dynamic feature curves;

[0028] Several key feature points are extracted from the semantic dynamic feature curve at equal time intervals, and the number of key feature points is proportional to the duration of the curve.

[0029] The knowledge graph is semantically distributed and labeled based on the extracted key feature points.

[0030] Preferably, the optimization of reasoning paths for the knowledge graph that has completed semantic distribution mapping includes, after semantic distribution mapping and annotation of the knowledge graph, logically isolating the preset reasoning paths in the knowledge graph that overlap with highly related segments.

[0031] Preferably, the step of controlling the knowledge processing device to perform intervention operations within the associated area according to the decision scheme when the device detects an abnormal state includes:

[0032] S1. Select the associated region access location with the shortest operation flow based on the next inference node to be switched.

[0033] S2. Control the knowledge processing device to switch to the access location and synchronize parameters;

[0034] S3. Control the knowledge processing device to switch to the associated area from the access location and adjust the execution status according to the reasoning path in the decision-making scheme;

[0035] S4. Real-time acquisition of the output indicators of the knowledge processing device, and dynamic correction of the output indicators using the stochastic gradient descent algorithm;

[0036] S5. After each continuous reasoning operation is completed in the associated region, control the knowledge processing device to pause output and exit the associated region, and re-execute S1.

[0037] Preferably, step S1 includes the following steps:

[0038] The feasible access locations for each associated region are calculated using the A* algorithm, and a comprehensive evaluation is performed based on the operation process and logical distance parameters.

[0039] Based on the comprehensive evaluation results, the access location of the associated area with the shortest operation process and the smallest logical distance was selected.

[0040] The comprehensive evaluation based on operational procedures and logical distance parameters includes:

[0041] Establish a two-dimensional evaluation matrix that includes the number of operation steps and logical distance values;

[0042] After normalizing the two-dimensional evaluation matrix, factor analysis was used to extract the first principal factor as a comprehensive evaluation index.

[0043] The decision generation unit is used to select the associated region with the highest comprehensive evaluation index value as the access location.

[0044] Preferably, the knowledge extraction unit further includes a dynamic update module, which is used to incrementally update the knowledge graph according to the semantic feature curve after the knowledge processing device completes the semantic distribution mapping.

[0045] Compared with the prior art, the beneficial effects of the present invention are:

[0046] The knowledge graph-based intelligent consultation service system provided by this invention significantly improves the quality and efficiency of intelligent consultation services through the collaborative work of multiple components. The user terminal's multimodal input device supports simultaneous text and voice input, catering to different user habits and broadening information acquisition channels. The semantic parsing device extracts contextual features, and combined with the attention-based similarity calculation by the logical evaluation module, it can accurately determine the semantic coherence index, effectively avoiding semantic misunderstandings and improving the accuracy of understanding user questions.

[0047] The logic operation module and parameter detection module in the knowledge processing device work together to achieve accurate updates of knowledge node status and real-time monitoring of device operation status. The knowledge extraction unit determines the associated region by recording semantic jump periods and logical recovery periods, performs constraint processing on abnormal semantics and generates semantic feature curves, realizing semantic distribution mapping of the knowledge graph, enabling the system to deeply understand complex semantic relationships and mine potential knowledge.

[0048] The relation matching unit optimizes the reasoning path of the knowledge graph that has completed semantic distribution mapping. By logically isolating pre-defined reasoning paths that overlap in highly correlated segments, it improves reasoning efficiency and accuracy. The decision generation unit's regular reasoning unit and emergency reasoning unit function in the normal and abnormal states of the equipment, respectively. In particular, the emergency reasoning unit performs intervention operations in the correlated area based on the decision plan. Through a series of operations such as scientific access location selection, parameter synchronization, state adjustment, and output indicator correction, it ensures that the system can quickly resume normal operation in abnormal situations.

[0049] The dynamic update module of the knowledge extraction unit incrementally updates the knowledge graph based on semantic feature curves, ensuring that the knowledge graph is always synchronized with the latest knowledge. This enables the system to adapt to the ever-changing knowledge environment, providing users with more accurate, efficient, and timely intelligent consulting services, greatly improving user experience and system usability. Attached Figure Description

[0050] Figure 1 This is a schematic diagram illustrating the working principle of an intelligent consulting service system based on knowledge graphs as described in this invention.

[0051] Figure 2 This is a schematic diagram of the user terminal's working principle.

[0052] Figure 3 Define the flowchart for the related regions;

[0053] Figure 4 This is a flowchart of the intervention operation for the associated area under abnormal conditions. Detailed Implementation

[0054] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0055] Please see Figures 1-4This invention relates to an intelligent consultation service system based on a knowledge graph. The system comprises a user terminal, a knowledge processing device, and a cloud-based reasoning center. The user terminal and the knowledge processing device are communicatively connected to the cloud-based reasoning center. The cloud-based reasoning center includes a knowledge extraction unit, a relation matching unit, and a decision generation unit.

[0056] The user terminal receives the user's input of consultation questions and contextual semantic data in real time and uploads them to the cloud-based inference center. The knowledge processing device receives processing instructions from the cloud-based inference center and updates the status of knowledge nodes. During the operation of the knowledge processing device, the knowledge extraction unit records semantic jump periods and logical recovery periods, determines associated regions, controls the knowledge processing device to operate according to preset constraint thresholds for abnormal semantics and records changing parameters, generates semantic feature curves, and completes the semantic distribution mapping of the knowledge graph. The relationship matching unit optimizes the inference path of the knowledge graph with completed semantic distribution mapping and generates a decision scheme. When the knowledge processing device is in normal state, the regular inference unit of the decision generation unit controls it to execute the baseline inference mode; when an abnormal state is detected, the emergency inference unit controls the knowledge processing device to perform intervention operations in the associated region according to the decision scheme.

[0057] Example 1:

[0058] The user terminal serves as the direct interface between the system and the user. Its internal structure includes a terminal housing and multimodal input devices, semantic parsing devices, logic evaluation modules, and communication modules mounted on the housing. These components work together to collect, process, and upload user inquiries.

[0059] The terminal housing, serving as the physical carrier for all functional modules, is manufactured using a unibody engineering plastic molding process, with a frosted surface to enhance grip comfort. The housing features a multi-layered internal structure. The outermost layer is an electromagnetic shielding layer made of copper foil composite material, with a thickness controlled to 0.1 mm, effectively preventing external electromagnetic interference from affecting the internal circuitry. The middle layer is a structural support layer made of high-strength engineering plastic, ensuring the housing will not deform under certain external impacts. The inner layer is a heat dissipation channel layer with a dense microchannel structure designed to quickly conduct heat generated by internal electronic components to the housing surface. A touchscreen display is located on the front of the housing for text input and information display, and a microphone array is located below the display for voice data acquisition.

[0060] The multimodal input device integrates both text and voice input. Text input utilizes a combination of touch keyboard and handwriting recognition. The touch keyboard area features an adaptive layout algorithm that dynamically adjusts the size and position of the keys based on the user's input habits. For example, keys for frequently used characters are enlarged and their positions optimized for easier input. The handwriting recognition function supports various writing styles and has a large internal handwriting font library. It uses a convolutional neural network algorithm to analyze and recognize the user's handwriting trajectory in real time. When the user's writing speed changes, the system automatically adjusts the sampling frequency to ensure accuracy. Voice input employs a four-microphone array design, with four microphones positioned at the four corners of the terminal casing. This layout effectively suppresses ambient noise. Through beamforming algorithms, the system can accurately locate the user's voice position and align the primary sound pickup direction with that location. Simultaneously, the system features echo cancellation, which identifies and filters out sound signals that have been output from the speaker and then re-acquired by the microphones, ensuring the purity of the acquired voice data.

[0061] The semantic parsing unit is responsible for in-depth analysis of the data collected by the multimodal input device. First, the text data undergoes preprocessing, including stop word removal, lexical analysis, and syntactic analysis. The stop word list is dynamically updated according to different application scenarios to ensure the accuracy of preprocessing. Lexical analysis uses a bidirectional LSTM model, which can more accurately identify word boundaries and parts of speech. Syntactic analysis uses a dependency parsing algorithm to construct a grammatical structure tree of sentences, revealing the dependency relationships between words. For speech data, speech recognition is performed first to convert speech into text, and then the same processing flow as for text data is applied. When extracting contextual features, the system comprehensively considers word co-occurrence relationships, semantic similarity, and temporal information. By constructing a semantic network, related words and concepts are connected to form a complex semantic association graph. The system also records users' historical consultation records, analyzes users' interests and question types, and provides a reference for extracting contextual features for the current question.

[0062] The logic evaluation module assesses context-related features based on an attention mechanism. This module maintains a historical standard pattern library, which is continuously updated and improved as the system runs. When a new context-related feature is received, the system calculates its similarity to historical standard patterns. The attention mechanism automatically focuses on important parts of the features, assigning different weights to different features. For example, features closely related to the core of the problem are given higher weights, while less important features are given lower weights. This approach allows for a more accurate assessment of the semantic coherence index. If the semantic coherence index falls below a preset threshold, the system triggers a correction signal. The correction signal may take different actions depending on the specific circumstances, such as prompting the user to re-enter the information or providing supplementary information for the user to choose from.

[0063] The communication module is responsible for uploading the processed input data to the cloud inference center. This module supports multiple communication protocols, including 4G, 5G, and WiFi. When selecting a communication method, the system automatically switches based on the current network environment. For example, when WiFi signal is strong, it prioritizes WiFi for data transmission to improve speed and stability; when WiFi signal is unstable or unavailable, it automatically switches to 4G or 5G networks. The communication module also features data encryption, employing the AES-256 encryption algorithm to encrypt transmitted data, ensuring data security during transmission. Furthermore, to guarantee data transmission reliability, the system uses breakpoint resume technology. When data transmission is interrupted, the system records the position of the transmitted data and resumes transmission from the point of interruption after the network is restored, avoiding duplicate transmission and data loss.

[0064] In actual operation, the various modules of the user terminal collaborate to form an organic whole. The multimodal input device provides users with flexible and diverse input methods to meet the usage habits of different users. The semantic parsing device deeply mines the semantic information of the input data, providing a solid foundation for subsequent processing. The logic evaluation module ensures the semantic coherence and accuracy of the input data, improving the system's processing efficiency and quality. The communication module guarantees the secure and reliable transmission of data, enabling the cloud-based inference center to obtain users' inquiry information in a timely manner. This collaborative approach allows the user terminal to efficiently and accurately collect and process users' inquiries, providing strong support for the normal operation of the entire intelligent consultation service system.

[0065] Example 2:

[0066] The knowledge processing device's internal structure includes a processing chassis and, within the chassis, logic operation modules, parameter detection modules, instruction response modules, and protocol conversion modules. These modules work together to process and analyze the knowledge graph and interact with the cloud-based inference center.

[0067] The chassis is made of aluminum alloy, precision cast and CNC machined. The internal design is modular, with each functional module connecting to the motherboard via standardized interfaces for easy installation, maintenance, and upgrades. The inner walls of the chassis are coated with an electromagnetic shielding layer, effectively preventing electromagnetic interference from internal electronic components from affecting external devices, while also protecting internal modules from external electromagnetic interference. Ventilation holes at the bottom of the chassis, combined with the internal cooling fan, create excellent airflow, ensuring stable temperature control during extended operation.

[0068] The logic processing module, as the core of the knowledge processing device, adopts a multi-core processor architecture and is equipped with a high-performance graphics processing unit and a dedicated artificial intelligence acceleration chip. This module possesses powerful parallel computing capabilities, enabling it to handle multiple complex logical tasks simultaneously. When processing the knowledge graph, the logic processing module first parses the nodes and edges in the graph, extracting the attribute information of the nodes and the relationship information of the edges. Then, according to preset logical rules, it performs reasoning and calculations on this information. For example, when receiving a user's inquiry, the logic processing module searches for relevant nodes and paths in the knowledge graph, and through logical reasoning and semantic matching, finds the most likely answer. The logic processing module also supports deep learning algorithms, enabling it to automatically learn and optimize the knowledge graph. By continuously analyzing and processing large amounts of knowledge data, the module can automatically adjust its reasoning rules and algorithm parameters, improving the accuracy and efficiency of knowledge processing.

[0069] The parameter detection module monitors various operating parameters of the knowledge processing device in real time. This module integrates multiple sensors, including temperature, voltage, current, and memory usage sensors. Temperature sensors, located in key positions inside the chassis, monitor the temperature of heat-generating components such as the processor and chipset. When the detected temperature exceeds a preset threshold, the system automatically activates the cooling fan to increase heat dissipation efficiency. Voltage and current sensors monitor the device's power supply in real time, ensuring stable voltage and current operation. The memory usage sensor monitors system memory usage in real time; when memory usage is too high, the system automatically optimizes memory allocation, releasing unnecessary memory space and improving system efficiency. In addition to hardware parameters, the parameter detection module also monitors semantic parameters. For example, during knowledge graph processing, the module monitors parameters such as the strength of associations between nodes and the credibility of inference paths. Changes in these parameters reflect the operating status and processing effectiveness of the knowledge processing device, providing a basis for system optimization and adjustment.

[0070] The instruction response module is responsible for receiving and processing instructions from the cloud-based inference center. This module employs an efficient message queue mechanism to ensure that instructions are received and processed promptly and accurately. Upon receiving an instruction, the instruction response module first parses it, extracting the instruction type, parameters, and execution requirements. Then, based on the instruction type, it forwards it to the appropriate functional module for processing. For example, if it is a knowledge graph update instruction, the instruction response module will forward it to the logic operation module, which is responsible for updating the content and structure of the knowledge graph. The instruction response module also has instruction priority management capabilities. For urgent instructions, the system will prioritize processing them to ensure that critical tasks can be executed in a timely manner. Simultaneously, the module records the execution status and results of the instructions and feeds them back to the cloud-based inference center. If an anomaly occurs during instruction execution, the instruction response module will automatically take appropriate measures, such as retries or degraded processing, and promptly report the anomaly to the cloud-based inference center.

[0071] The protocol conversion module is responsible for converting between different communication protocols. This module supports various common communication protocols, including HTTP, HTTPS, MQTT, and AMQP. When the knowledge processing device communicates with the cloud inference center, the module automatically selects the appropriate conversion method based on the supported protocol types of both parties. For example, when the cloud inference center uses the HTTP protocol to send instructions, while the internal modules of the knowledge processing device use the MQTT protocol, the module converts the HTTP instructions into MQTT messages, ensuring that the instructions are correctly received and processed. The module also features data format conversion. Different communication protocols may use different data formats, such as JSON, XML, and Protobuf. The module can identify and convert these different data formats, ensuring data consistency and integrity during transmission. Furthermore, the module supports protocol extensions and custom protocols. As technology advances and application scenarios change, new communication protocols and data formats may emerge. The module provides flexible extension interfaces, allowing developers to add support for new protocols as needed.

[0072] In actual operation, the various modules of the knowledge processing device collaborate to form an organic whole. The logic operation module, as the core, is responsible for processing and analyzing the knowledge graph; the parameter detection module monitors the device's operating status in real time, ensuring stable system operation; the instruction response module ensures the device can respond promptly to instructions from the cloud-based inference center, enabling collaborative system operation; and the protocol conversion module solves compatibility issues between different communication protocols, ensuring smooth communication between the device, the cloud-based inference center, and other devices. This collaborative approach enables the knowledge processing device to efficiently and accurately process and analyze the knowledge graph, providing strong support for the intelligent consulting service system.

[0073] Example 3:

[0074] When the knowledge processing device performs knowledge graph reasoning tasks, it monitors changes in the semantic features of the input data in real time. Once a change in semantic features exceeds a preset threshold, the system determines that a semantic shift has occurred. At this point, the system immediately records the timestamp of the shift, accurate to the millisecond level, to ensure the accuracy of the time information.

[0075] Upon detecting a semantic shift, the parameter detection module initiates continuous monitoring of the logical recovery threshold. The logical recovery threshold refers to the critical point at which the knowledge processing device recovers from a semantic shift state to normal logical reasoning. During monitoring, the parameter detection module analyzes the output of the knowledge processing device, evaluating its logical coherence and semantic accuracy. When the logical coherence and semantic accuracy of the output reach a preset recovery standard, the system determines that the logical recovery threshold has been reached and records the sequence number of the recovery time. This sequence number is a unique identifier for the recovery event in chronological order, used for subsequent data processing and analysis.

[0076] Based on the recorded timestamp data and sequence numbers, the system generates transition coordinates and recovery coordinates. Transition coordinates are multi-dimensional coordinate points composed of the timestamp of the transition and the corresponding semantic feature vector. The semantic feature vector contains key semantic information of the input data at the time of the transition, such as keywords, themes, and sentiment tendencies. Recovery coordinates are multi-dimensional coordinate points composed of the sequence number at the recovery time and the corresponding semantic feature vector. These two coordinate points represent the start and end points of the semantic transition, respectively, providing foundational data for subsequent spatial correlation analysis.

[0077] After generating the transition coordinates and recovery coordinates, the system spatially associates these two coordinate points with the user's terminal location. The user's terminal location refers to the geographical location of the user when inputting the inquiry, typically obtained through GPS positioning or network IP address resolution. The spatial association process employs a graph-based association algorithm, treating the transition coordinates, recovery coordinates, and user's terminal location as nodes in a graph, with the connections between nodes representing their semantic relevance. By calculating the semantic similarity and spatial distance between nodes, the system constructs a semantic association network.

[0078] Based on the semantic association network, the system searches for the smallest closed region that includes the jump coordinates, the recovery coordinates, and the user terminal location. This closed region consists of a set of highly semantically related nodes, whose connections form a closed loop. The process of determining the closed region uses a combination of the minimum spanning tree algorithm and the convex hull algorithm. First, the minimum spanning tree algorithm finds the shortest path connecting the jump coordinates, the recovery coordinates, and the user terminal location. Then, the convex hull algorithm calculates the smallest convex polygon containing these paths; this convex polygon is the final determined association region.

[0079] Identifying associated regions is crucial for targeted handling of subsequent anomalous semantics. Within associated regions, knowledge processing devices can more accurately pinpoint the source and scope of impact of anomalous semantics, enabling more effective processing measures. For example, the system can focus on analyzing knowledge nodes within associated regions to identify anomalous knowledge or reasoning rules that may lead to semantic shifts. Simultaneously, identifying associated regions also provides a basis for local optimization of the knowledge graph. The system can adjust and improve the knowledge structure within associated regions, enhancing the accuracy and robustness of the knowledge graph.

[0080] In practical applications, the process of semantic jump detection and related region determination is dynamic and continuous. As the knowledge processing equipment continues to operate and users continuously input their questions, the system constantly detects semantic jumps, records relevant data, determines related regions, and optimizes and updates the knowledge graph based on this information. This dynamic processing mechanism enables the system to adapt to the ever-changing semantic environment, improving the quality and effectiveness of intelligent consultation services.

[0081] Furthermore, to improve the accuracy and efficiency of semantic transition detection, the system employs a multi-dimensional detection strategy. In addition to monitoring the magnitude of semantic feature changes, the system also analyzes factors such as the speed, direction, and frequency of semantic changes. For example, if semantic features undergo drastic changes within a short period, or if semantic changes exhibit periodic fluctuations, the system will pay closer attention to these situations and conduct timely in-depth analysis and processing.

[0082] In representing and storing associated regions, the system employs a hierarchical data structure. An associated region is represented as a subgraph containing multiple knowledge nodes and relationships, with clearly defined boundaries and internal structure. Nodes and relationships within the subgraph are labeled with semantic tags and weight information, reflecting their importance and relevance within the associated region. This hierarchical data structure allows the system to manage and manipulate associated regions more efficiently.

[0083] Semantic jump detection and associated region identification are indispensable and crucial components of intelligent consultation service systems. By accurately detecting semantic jumps, recording relevant data, generating coordinate points, performing spatial associations, and identifying associated regions, the system can more effectively handle anomalous semantics, improving the quality of the knowledge graph and the effectiveness of intelligent consultation services. This processing mechanism not only meets users' consultation needs but also continuously learns and optimizes, adapting to the ever-changing semantic environment and providing users with more accurate and efficient intelligent consultation services.

[0084] Example 4:

[0085] After identifying the relevant regions, the system will perform fine-grained semantic parameter adjustments and mapping on those regions. Taking an intelligent consulting scenario involving financial investment as an example, when a user asks a question about "high-risk investment strategies," the system may detect semantic shifts and identify relevant regions that cover knowledge nodes such as "stock investment," "futures trading," and "risk assessment."

[0086] First, the system controls the knowledge processing device to adjust the output intensity of the associated region. This process is similar to adjusting the brightness of lights, making the knowledge nodes within the associated region more "prominent" in the knowledge graph. By increasing the activation threshold of these nodes, they are more easily invoked during reasoning, thereby improving the retrieval efficiency of knowledge related to the user's question. For example, the node "futures trading risk control," which originally had a low weight in regular retrieval, has its weight increased in the associated region, allowing it to be retrieved by the system more quickly when dealing with questions related to high-risk investments.

[0087] Simultaneously, the parameter detection module begins acquiring semantic parameters in real time. These parameters include, but are not limited to, the strength of association between knowledge nodes, the credibility of reasoning paths, and semantic similarity. In the financial investment example, the system monitors the strength of association between the "stock investment" and "risk assessment" nodes, as well as the credibility parameters of different investment strategies in the current market environment. The real-time acquisition of these parameters provides a data foundation for subsequent semantic analysis.

[0088] Pre-defined constraint thresholds act as a safety barrier for the system. For example, for semantic parameters of investment risk assessment, the system sets a reasonable range. When it detects that the risk assessment parameters of an investment strategy exceed this range, the system immediately controls the knowledge processing equipment to maintain operation within the preset constraint threshold. This is similar to a speed limiter on a car, ensuring that the system does not deviate from a reasonable range in pursuit of an excessive outcome. In financial investment scenarios, if the risk assessment parameters of a high-risk investment strategy exceed the safety threshold, the system automatically adjusts the recommendation strength for that strategy to avoid providing users with overly aggressive investment advice.

[0089] The system continuously records the semantic output values ​​of the knowledge processing device, forming a set of changing parameters. These output values ​​reflect the dynamic changes of the knowledge processing device in processing user questions. In financial investment consulting, these output values ​​may include recommendation indices for different investment strategies, changes in risk assessment results over time, etc. By recording these output values, the system can capture subtle changes in the knowledge processing process, providing rich data for subsequent semantic feature analysis.

[0090] Based on a set of changing parameters, the system uses a Hidden Markov Model (HMM) algorithm to generate semantic dynamic feature curves. This process is similar to meteorologists plotting weather change curves based on meteorological data. HMMs can discover underlying patterns and regularities from seemingly random semantic output values. In financial investment consulting, semantic dynamic feature curves may reflect the fluctuations in the recommendation strength of different investment strategies as the market changes, or the changing trends in users' interest in investment products with different risk levels.

[0091] In the semantic dynamic feature curve, the system extracts several key feature points at equal time intervals, with the number of key feature points proportional to the duration of the curve. This is analogous to marking an important note at regular intervals in a piece of music. These key feature points represent significant turning points in semantic changes or moments when features become prominent. In financial investment consulting, these key feature points may correspond to the timing of major market events or moments when a user's attitude towards consulting changes significantly.

[0092] Finally, based on the extracted key feature points, the system performs semantic distribution mapping annotation on the knowledge graph. This process is similar to marking important locations and routes on a map. By mapping key feature points onto the knowledge graph, the system can more accurately reflect the distribution of semantic information within the knowledge graph. In financial investment consulting, semantic distribution mapping annotation may highlight the connections between knowledge nodes related to high-risk investment strategies, or mark the most valuable investment advice in a specific market environment.

[0093] In this way, knowledge graphs can more accurately reflect semantic information, providing more reliable support for intelligent consulting services. In practical applications, this semantic distribution mapping annotation method can help the system better understand user questions and provide more accurate answers and suggestions. For example, in financial investment consulting, the system can recommend products that best match the user's risk tolerance and investment goals based on the results of semantic distribution mapping annotation, or adjust the recommendation strategy in a timely manner when the market changes, providing users with more timely investment advice.

[0094] Example 5:

[0095] When the knowledge processing device detects an abnormal state, the system immediately activates an emergency response mechanism. This process begins by calculating feasible access locations for each associated region using Algorithm A. Algorithm A is similar to finding the shortest path in a maze; the system comprehensively considers the number of operation steps and the logical distance value, assigning an evaluation value to each possible access location. The number of operation steps represents the number of operations required to reach the target state from the current state, while the logical distance value measures the semantic association between two knowledge nodes.

[0096] The system will establish a two-dimensional evaluation matrix containing the number of operation steps and the logical distance value. This matrix is ​​similar to a coordinate graph, with the horizontal axis representing the number of operation steps and the vertical axis representing the logical distance value. Each feasible access location has a corresponding coordinate point in this matrix. To make these two different dimensions of indicators comparable, the system will normalize the matrix. Normalization is like converting measurements from different units to the same standard unit, allowing the number of operation steps and the logical distance value to be compared on the same scale.

[0097] After normalization, the system uses factor analysis to extract the first principal factor as a comprehensive evaluation index. Factor analysis is similar to finding hidden key factors from a large amount of observational data. The first principal factor represents the main factor that can explain most of the data variation. By extracting this factor, the system can simplify the two-dimensional evaluation matrix into a one-dimensional comprehensive evaluation index, thereby more intuitively comparing the advantages and disadvantages of different access locations.

[0098] Based on the comprehensive evaluation results, the system selects the access location in the associated area with the shortest operation process and the smallest logical distance. This is analogous to choosing the shortest and easiest route among multiple roads leading to a destination. Once the access location is determined, the system controls the knowledge processing device to switch to that location and synchronize its parameters. The parameter synchronization process is similar to data transmission and calibration between two devices, ensuring that the knowledge processing device can accurately acquire and process relevant information at the new access location.

[0099] Subsequently, the knowledge processing device enters the relevant area and performs state adjustments according to the reasoning path in the decision-making scheme. This process is similar to driving along a route on a map; the knowledge processing device operates and adjusts the knowledge nodes within the relevant area along the predetermined reasoning path. During this process, the system acquires the output metrics of the knowledge processing device in real time, which reflect the performance and effectiveness of the knowledge processing device in performing state adjustments.

[0100] To ensure the accuracy and stability of the output metrics, the system employs a stochastic gradient descent algorithm for dynamic correction. This algorithm is analogous to continuously adjusting direction while climbing a mountain to find the highest peak. Based on the real-time acquired output metrics, the system continuously adjusts the operating parameters of the knowledge processing device, gradually bringing the output metrics closer to the ideal value.

[0101] After each continuous reasoning operation is completed in the associated region, the system controls the knowledge processing device to pause output and exit the associated region. This is similar to taking a break and reviewing the results after completing a task. After exiting the associated region, the system will select a new access location and repeat the above operation process. This cyclical process will continue until the abnormal state of the knowledge processing device is resolved.

[0102] Furthermore, after the knowledge processing device completes the semantic distribution mapping, the dynamic update module of the knowledge extraction unit incrementally updates the knowledge graph based on the semantic feature curves. The semantic feature curves record the semantic changes during knowledge processing, and the dynamic update module analyzes these changes to identify the parts of the knowledge graph that need updating. Incremental updates are like adding new colors to a painting; the system adds new knowledge nodes or adjusts the relationships between existing nodes without disrupting the original knowledge structure, ensuring the timeliness and accuracy of the knowledge graph.

[0103] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0104] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A knowledge graph-based intelligent consultation service system, characterized in that, include: The system comprises a user terminal, a knowledge processing device, and a cloud-based reasoning center. The user terminal and the knowledge processing device are respectively communicatively connected to the cloud-based reasoning center, which includes a knowledge extraction unit, a relation matching unit, and a decision generation unit. The user terminal is used to input consultation questions and contextual semantic data in real time. The knowledge processing device is used to receive processing instructions from the cloud-based reasoning center to update the status of knowledge nodes. The knowledge extraction unit is used to record the semantic jump period and logical recovery period during the operation of the knowledge processing device, determine the associated region based on the semantic jump period and logical recovery period, control the knowledge processing device to maintain a preset constraint threshold for abnormal semantics and record the changing parameters, generate semantic feature curves based on the changing parameters and perform semantic distribution mapping on the knowledge graph. The control knowledge processing device maintains operation at a preset constraint threshold for abnormal semantics and records changing parameters, including: the control knowledge processing device adjusts the output intensity of the associated region and obtains semantic parameters in real time through the parameter detection module; a preset constraint threshold range, and if the semantic parameters exceed the preset constraint threshold range, the control knowledge processing device maintains operation at the preset constraint threshold. The relation matching unit is used to optimize the reasoning path of the knowledge graph that has completed semantic distribution mapping and generate a decision scheme. The decision generation unit includes a regular reasoning unit and an emergency reasoning unit. The regular reasoning unit is used to control the knowledge processing device to execute a baseline reasoning mode when the knowledge processing device is in a normal state. The emergency reasoning unit is used to control the knowledge processing device to perform an intervention operation in the associated area according to the decision scheme when the knowledge processing device detects an abnormal state. The knowledge processing device includes a processing chassis and a logic operation module, a parameter detection module, an instruction response module, and a protocol conversion module disposed in the chassis; The semantic transition period and logical recovery period during the operation of the knowledge processing device include: When the knowledge processing device detects a semantic jump, it records the timestamp data of the jump occurrence time, continuously monitors the logical recovery critical point through the parameter detection module, and records the sequence number of the recovery time. The process of determining the associated region based on the semantic jump period and the logical recovery period includes: Generate jump coordinates and recovery coordinates based on timestamp data and sequence numbers; Spatial correlation is performed on the jump coordinates, recovery coordinates, and user terminal location to form a closed region, and the closed region is marked as the correlation region; The step of controlling the knowledge processing device to perform intervention operations within the associated area according to the decision-making scheme when the knowledge processing device detects an abnormal state includes: S1. Select the associated region access location with the shortest operation flow based on the next inference node to be switched. S2. Control the knowledge processing device to switch to the access location and synchronize parameters; S3. Control the knowledge processing device to switch to the associated area from the access location and adjust the execution status according to the reasoning path in the decision-making scheme; S4. Real-time acquisition of the output indicators of the knowledge processing device, and dynamic correction of the output indicators using the stochastic gradient descent algorithm; S5. After each continuous reasoning operation is completed in the associated region, control the knowledge processing device to pause output and exit the associated region, and re-execute S1.

2. The intelligent consultation service system based on knowledge graphs according to claim 1, characterized in that, The user terminal includes a terminal housing and a multimodal input device, a semantic parsing device, a logic evaluation module, and a communication module disposed in the housing; The multimodal input device is used to simultaneously acquire text input data and voice input data; The semantic parsing device is used to extract contextual features from the input data; The logical evaluation module is used to calculate the similarity between the current context-related features and the historical standard patterns based on the attention mechanism, determine the semantic coherence index, and trigger a correction signal based on the calculation result. The communication module is used to upload input data to the cloud-based inference center.

3. The intelligent consultation service system based on knowledge graphs according to claim 1, characterized in that, The process of generating semantic feature curves based on changing parameters and performing semantic distribution mapping on the knowledge graph includes the following steps: Continuously record the semantic output values ​​of the knowledge processing device to form a set of changing parameters; Based on the set of changing parameters, a hidden Markov model algorithm is used to generate semantic dynamic feature curves; Several key feature points are extracted from the semantic dynamic feature curve at equal time intervals, and the number of key feature points is proportional to the duration of the curve. The knowledge graph is semantically distributed and labeled based on the extracted key feature points.

4. The intelligent consultation service system based on knowledge graphs according to claim 3, characterized in that, The optimization of reasoning paths for the knowledge graph that has completed semantic distribution mapping includes, after semantic distribution mapping and annotation of the knowledge graph, logically isolating the preset reasoning paths in the knowledge graph that overlap with highly related segments.

5. The intelligent consultation service system based on knowledge graphs according to claim 1, characterized in that, S1 includes the following steps: Through A The algorithm calculates feasible access locations for each associated region and performs a comprehensive evaluation based on the operation process and logical distance parameters; Based on the comprehensive evaluation results, the access location of the associated area with the shortest operation process and the smallest logical distance was selected. The comprehensive evaluation based on operational procedures and logical distance parameters includes: Establish a two-dimensional evaluation matrix that includes the number of operation steps and logical distance values; After normalizing the two-dimensional evaluation matrix, factor analysis was used to extract the first principal factor as a comprehensive evaluation index. The decision generation unit is used to select the associated region with the highest comprehensive evaluation index value as the access location.

6. The intelligent consultation service system based on knowledge graphs according to claim 1, characterized in that, The knowledge extraction unit also includes a dynamic update module, which is used to incrementally update the knowledge graph according to the semantic feature curve after the knowledge processing device completes the semantic distribution mapping.

Images