Knowledge graph construction and retrieval method and system of diving electronic manual

Abstract

The invention relates to the field of diving, in particular to a knowledge graph construction and retrieval method and system for a diving electronic manual, and the method comprises the steps: obtaining the residence time and interaction frequency data generated in the process of browsing the electronic manual by a user, and obtaining a user behavior feature sequence; performing classification processing on the user behavior feature sequence by adopting a support vector machine model, and determining a user skill level category; adjusting the presentation sequence of the electronic manual according to the skill level category of the user, and generating an interface layout structure; collecting a feedback signal of a user to the interface layout structure, and generating an interface layout adjustment suggestion; and updating an interface layout structure according to the interface layout adjustment suggestion, generating an adjusted interface version, and pushing the adjusted interface version to a corresponding electronic manual device.

Description

Technical Field

[0001] This application relates to the field of diving, and in particular to a method and system for constructing and retrieving a knowledge graph for an electronic diving manual. Background Technology

[0002] Electronic manuals in the diving field serve as crucial tools supporting divers' learning, training, and practical application. Their interface design and content adaptability directly impact users' learning outcomes and operational safety. With the increasing popularity of diving activities and the expansion of the user base, users with different skill levels have raised differentiated demands for knowledge presentation methods. For example, beginners need clear and concise guidance, while professional divers expect to access in-depth content that includes professional parameters and complex procedures.

[0003] However, most existing electronic diving manual systems have significant design limitations. They generally adopt a static content organization method and fail to fully consider the diversity of user roles, resulting in a lack of targeted interface presentation. Beginners often find the interface content too technical and difficult to understand, while professional divers often find the existing content insufficient in depth and unable to meet the reference needs of complex diving missions. This mismatch between content and users' cognitive level not only reduces users' learning efficiency but also creates safety hazards in actual operation. Summary of the Invention

[0004] To effectively address the aforementioned issues, this application provides a method and system for constructing and retrieving a knowledge graph for diving electronic manuals.

[0005] Firstly, this application provides a knowledge graph construction and retrieval method for an electronic diving manual, which adopts the following technical solution: A method for constructing and retrieving a knowledge graph for an electronic diving manual includes the following steps: Data on user dwell time and interaction frequency during browsing of electronic manuals are obtained to generate user behavior feature sequences; Obtain a user behavior dataset; classify the user behavior dataset and combine it with user behavior feature sequences to determine the user skill level category; wherein, the user skill level category includes novice diver and technical diver; Adjust the presentation order of the electronic manual according to the user's skill level category to generate the interface layout structure; Collect user feedback signals on the interface layout structure and generate interface layout adjustment suggestions; The interface layout structure is updated according to the interface layout adjustment suggestions, an adjusted interface version is generated, and the adjusted interface version is pushed to the corresponding electronic manual device.

[0006] Optionally, the step of adjusting the presentation order of the electronic manual according to the user's skill level category includes: If the user's skill level is identified as a beginner diver, simplified chapter information is extracted from the preset content library to obtain a beginner-adapted content set. If the user's skill level category is identified as "technical diver", then content containing in-depth technical details is extracted from the preset content library to obtain an advanced adapted content set; For the primary or advanced adapted content set, the chapter information of the electronic manual is adjusted to determine the final presentation order.

[0007] Optionally, the step of obtaining user behavior feature sequence data by acquiring dwell time and interaction frequency data generated by the user during browsing the electronic manual includes: The data acquisition module obtains data on the user's dwell time and interaction frequency during the browsing of the electronic manual, thus obtaining a preliminary behavioral dataset. The preliminary behavior dataset is cleaned and standardized to obtain a standardized behavior dataset. Based on standard behavioral datasets, determine the distribution of user dwell time on different pages or modules, and identify user browsing behavior preferences; When a user stays on a specific page or module for more than a preset threshold, the interaction frequency data corresponding to that page or module is extracted to determine the user attention index for that specific content. Based on the user attention index and interaction pattern, a user behavior sequence is constructed, and the user behavior sequence is classified to determine the habit category division; Based on habit categories, browsing behavior preferences, and device information from electronic manuals, a final sequence of user behavior characteristics is generated.

[0008] Optionally, the step of extracting simplified chapter information from a preset content library to obtain a basic adapted content set if the user's skill level category is identified as a beginner diver includes: When a user's skill level is identified as a beginner diver, simplified chapter information corresponding to the beginner diver is selected from the preset content library to form a preliminary content set; Based on the user's skill level and the relevance of the content, the chapter information in the initial content set is prioritized to determine the core content units that should be displayed first. The core content units are standardized to obtain a set of standard content. The integrity of the standard content set is verified to generate the final primary adapted content set.

[0009] Optionally, if the user's skill level is identified as "technical diver," then content containing in-depth technical details is extracted from a preset content library to obtain an advanced adapted content set, including: When a user's skill level is identified as "technical diver," a set of original content related to in-depth technical details is retrieved from a pre-set content library. The original content set is layered and divided into data groups of different depths based on technical dimensions. The stratified data groups are cleaned and filtered to obtain cleaned data groups; Extract key information fields related to technical diving missions from the cleaned data groups and construct a content framework that conforms to the cognitive structure of technical divers. The content framework is deeply optimized and restructured according to specific scenarios to generate a content set; The content set is then subjected to consistency calibration to generate the final advanced adapted content set.

[0010] Optionally, the step of collecting user feedback signals on the interface layout structure and generating interface layout adjustment suggestions includes: Collect user feedback signals regarding the interface layout structure; these feedback signals include click frequency and scrolling speed. Key indicators are extracted from the feedback signals to identify user interaction patterns in different interface areas; The interaction pattern is analyzed based on a preset classification model. When the click frequency of a specific area is lower than a preset threshold within a specific time period, the area is determined to be a low-participation area. When the scrolling rate fluctuation of any region exceeds the preset reasonable range, that region is marked as a potential problem region. Based on low-participation areas and potential problem areas, generate targeted interface layout adjustment suggestions.

[0011] Optionally, updating the interface layout structure according to the interface layout adjustment suggestions includes: Based on low-participation areas and potential problem areas, identify the key interface areas that need adjustment and their location coordinates; Compare the current interface layout with the preset standard layout structure to determine the layout deviation; Calculate the position adjustment amount based on the layout deviation, and generate a layout structure adjustment scheme; The execution order of each adjustment parameter in the layout structure adjustment scheme is set, and the interface layout structure is updated through an event listener mechanism. The interface layout structure is iteratively processed during the adjustment process to generate the final interface version.

[0012] Optionally, before the step of adjusting the presentation order of the electronic manual according to the user's skill level category and generating the interface layout structure, the method further includes: A knowledge graph covering the diving domain is pre-constructed, where nodes represent conceptual entities and edges represent semantic relationships between entities; Based on the user's current browsing topic and their skill level category, related subgraphs are extracted from the knowledge graph; specifically, subgraphs containing basic concepts and their directly related entities are extracted for novice users, and subgraphs containing multiple reasoning paths and advanced technical parameter nodes are extracted for technical users. The subgraph is used as a semantic index source for content retrieval.

[0013] Optional, also includes: Identify the user's current task objective when using the electronic manual; Based on a pre-built knowledge graph in the diving field, a personalized knowledge path is generated, with the task objective as the endpoint and the user's existing knowledge as the starting point. The knowledge path is divided into several stages, and adapted content matching the user's skill level category is generated for each stage. Embed a path navigation control in the interface layout structure to display the current stage, completed nodes, and subsequent steps.

[0014] Secondly, the knowledge graph construction and retrieval system for the diving electronic manual provided in this application adopts the following technical solution: A knowledge graph construction and retrieval system for an electronic diving manual includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the method described in the first aspect.

[0015] In summary, this application has at least the following beneficial effects: This solution categorizes users based on their behavioral characteristics, classifying them into different skill levels: beginner and advanced. For beginners, clear and concise guidance, basic concepts, and visual charts are presented, filtering out complex technical parameters to reduce cognitive load and accelerate learning. For advanced users, detailed technical parameters, complex processes, and troubleshooting solutions are directly presented, meeting their requirements for information completeness in high-risk, complex environments. This provides a customized experience for users with different cognitive levels and needs, reducing the time users spend searching and filtering through redundant information, making the learning process more efficient and focused. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the knowledge graph construction and retrieval method in an embodiment of this application. Detailed Implementation

[0017] The present application will be further described below with reference to the accompanying drawings and specific embodiments. It should be noted that the embodiments are merely specific illustrations of the present application, and their purpose is to enable those skilled in the art to better understand the technical solutions of the present application, and should not be regarded as limitations on the present application.

[0018] This application discloses a method for constructing and retrieving a knowledge graph for a diving electronic manual, such as... Figure 1 As shown, it includes the following steps: S101, Obtain data on the user's dwell time and interaction frequency during the browsing of the electronic manual to obtain a user behavior feature sequence.

[0019] Step 101 includes: S1011 acquires data on user dwell time and interaction frequency during browsing of electronic manuals through a data acquisition module to obtain a preliminary behavioral dataset; its data acquisition module can be configured as a touch screen sensor, motion sensor, and system-level event monitoring service built into the mobile terminal device.

[0020] Specifically, when users browse electronic manuals through mobile devices such as tablets and smartphones, the built-in sensors and event log systems of these devices can capture the user's dwell time on each page and interaction frequency data such as clicks, swipes, and zooms. This timestamped information is stored as a log file, forming a preliminary behavioral dataset. Dwell time records the duration a user spends on each individual content page or specific information module; longer dwell times may indicate in-depth reading, difficulty in understanding, or high interest in that section of content. Interaction frequency data quantifies the user's activity level during page interaction. This data includes not only basic clicks and swipes for page turning, but can also be further refined into content zooming operations, navigation jumps, bookmarks, and annotations.

[0021] S1012, perform data cleaning and standardization on the preliminary behavior dataset to obtain a standardized behavior dataset.

[0022] First, the initial behavioral dataset is cleaned to remove invalid or anomalous records. For example, records with a single page dwell time exceeding 10 minutes are considered invalid sessions, and high-frequency click sequences with extremely short durations (less than 0.1 seconds) are removed. Such anomalous data may originate from abnormal user behavior, such as leaving the device without locking the screen, or system recording errors, and is therefore not valuable for analysis. After data cleaning, to eliminate differences in the units and numerical ranges of different behavioral features, Z-score standardization can be used to process the cleaned data. This transforms each feature value into a normalized distribution with a mean of 0 and a standard deviation of 1, placing all features on the same comparable scale. This eliminates noise and unit differences in the initial behavioral dataset, thus avoiding negative impacts on the accuracy of subsequent analytical models.

[0023] S1013, Determine the distribution of user dwell time on different pages or modules based on the standard behavioral dataset, and determine the user's browsing behavior preferences.

[0024] Specifically, based on standard behavioral datasets, the distribution of user dwell time on different pages or modules (such as catalog pages, detailed description pages, safety warning modules, and accessory list modules) is analyzed and compared to determine user browsing behavior preferences. For example, if a user's average dwell time on the safety warning module is 300 seconds, while their average dwell time on the accessory list module is 60 seconds, it can be inferred that the user has a higher preference for safety-related content. This judgment can be visualized using statistical methods such as histograms to aid in understanding the overall preference distribution and provide a core basis for the subsequent personalized presentation of the electronic manual.

[0025] S1014 When the time a user spends on a specific page or module exceeds a preset threshold, extract the interaction frequency data corresponding to that page or module to determine the user attention index for that specific content.

[0026] If a user spends more than a preset threshold time on a certain page or module, the analysis process is initiated. The average value of interaction data such as the page's click-through rate and number of swipes is calculated by weighting the data to generate an indicator between 0 and 1.

[0027] Preset maximum stay time For example, 600 seconds; exceeding this time is generally considered abnormal or non-active learning, and a high interaction frequency is defined, such as 5 valid operations per minute. This is achieved through the formula... Calculate the time spent in the residence score ,in This refers to the actual time spent in the hotel. Set a preset threshold; and use the formula Calculate the interaction frequency score ,in The actual frequency of user interaction. For high interaction frequency.

[0028] The user engagement metric is generated by weighting and combining the dwell time score and interaction frequency score. ,in As a weight for the time spent, As the interaction frequency weight, and .

[0029] For example, when a user spends 450 seconds on the "User Instructions" page, and the user's actual interaction frequency is... At 4.5 times per minute, the interaction frequency is high. 5 times / minute, dwell time weighting =0.4, interaction frequency weight =0.6, its stay time score = (450 - 180) / (600 - 180) = 270 / 420 ≈ 0.64, Interaction Frequency Score = 4.5 / 5 = 0.9, so the user attention index = (0.4 × 0.64) + (0.6 × 0.9) = 0.256 + 0.54 = 0.796 ≈ 0.8. This high score reflects the user's strong interest in the content or need for in-depth reading.

[0030] S1015, construct a user behavior sequence based on the user attention index and interaction pattern, classify the user behavior sequence, and determine the habit category division.

[0031] Specifically, before model training, domain experts define several typical habit categories based on teaching experience and user observations. These habit categories can include deep learning, fast browsing, and gap-filling. Key characteristics of deep learning might include long average time spent on a single page, high interaction frequency, few page jumps, and frequent re-reading. Key characteristics of fast browsing might include short average time spent on a single page, low interaction frequency, frequent page jumps, and high usage of the search function. Key characteristics of gap-filling might include short total session duration, extremely long time spent on specific pages, extremely short time spent on other pages, and extensive use of directories or direct search navigation.

[0032] The system constructs a user behavior sequence based on key user behaviors, such as dwell time, interaction frequency, and page navigation, in chronological order; for example: high dwell time - high interaction - navigation - return. It then extracts feature vectors representing behavioral patterns from these sequences. Average dwell time, average interaction frequency, page bounce rate The input is fed into a pre-trained logistic regression model, which maps the linear combination of features to the probability interval [0,1] based on the sigmoid function, and takes the category with the highest probability as the user's final habit category, thereby achieving the classification.

[0033] S1016 generates the final user behavior feature sequence based on habit categories, browsing behavior preferences, and electronic manual device information.

[0034] The system integrates habit categories, browsing behavior preferences, and device information from electronic manuals to generate a multi-dimensional feature vector. This vector constitutes a complete quantitative description of user behavior patterns, such as the user behavior feature sequence = {Habit Category: 'Deep Learning', Behavior Preference: 'Safety Procedures', Device Information: 'High-Resolution Tablet'}. This is the final generated user behavior feature sequence.

[0035] S102, Obtain the user behavior dataset; classify the user behavior dataset and combine it with the user behavior feature sequence to determine the user skill level category; among which, the user skill level category includes junior diver and technical diver.

[0036] Specifically, user behavior datasets related to diving activities can be obtained from user behavior records. Real-time data on underwater navigation, oxygen consumption monitoring, and buoyancy control can be collected through smart wearable devices. This data includes, but is not limited to, depth change records, breathing rate, oxygen consumption rate, buoyancy control operation frequency, underwater navigation path, and device menu interaction logs. After preliminary processing, these raw data from different sources are uniformly converted into a time-aligned format, such as a tabular dataset containing fields such as timestamps, action types, parameter values, and duration.

[0037] The dataset is classified using a Support Vector Machine (SVM) model. As a supervised learning algorithm, the core principle of SVM is to find an optimal hyperplane that maximizes the margin between different data categories, thereby achieving high-precision classification. In this approach, key features, such as average diving depth and the frequency of emergency ascents or rapid descents, are extracted from the user behavior dataset to construct input feature vectors. The SVM model then calculates and classifies users into novice or technical divers based on these input feature vectors. For example, if a user's feature vector shows "stable depth changes but frequent and inefficient buoyancy adjustments," the model will classify them as a novice diver with limited skills.

[0038] To ensure the reliability of the classification structure, the system deeply integrates and cross-validates the classification results of the Support Vector Machine model based on the user behavior dataset with the user feature sequence. The user behavior dataset reflects practical skills, while the user feature sequence reflects theoretical understanding and learning tendencies. The combination of the two forms a comprehensive insight into user capabilities. When the two types of data point in the same direction (for example, practical data suggests a novice, while browsing data shows a focus on basic content), the system's determination of the user's skill level category is more accurate.

[0039] S103, Adjust the presentation order of the electronic manual according to the user's skill level category and generate the interface layout structure; Step 103 specifically includes the following steps: S1031, if the user's skill level category is identified as a beginner diver, simplified chapter information is extracted from the preset content library to obtain a beginner-adapted content set; Specifically, when a user's skill level is identified as a beginner diver, simplified chapter information corresponding to the beginner diver is selected from a preset content library to form a preliminary content set; based on the correlation between the user's skill level and the content, the chapter information in the preliminary content set is prioritized to determine the core content units to be displayed first; the core content units are standardized to obtain a standard content set; the standard content set is then validated for completeness to generate the final primary adapted content set.

[0040] Based on the user's skill level tags, the system automatically queries and filters simplified chapter information designed specifically for novice divers from a preset content library. These chapters typically cover core modules such as basic breathing techniques, simple underwater navigation, underwater safety basics, and equipment usage. All the selected chapters are integrated into a preliminary adapted content set, which serves as the basic data pool for subsequent processing.

[0041] The system introduces a content matching algorithm to intelligently sort the initially adapted content set. This algorithm analyzes user behavior datasets, such as error types and operational weaknesses in dive logs, to calculate the relevance score of each chapter to the user's current needs. Specifically, user behavior data is first converted into a numerical user feature vector, such as User Feature Vector = [Breathing Control Stability: 0.2, Ear Pressure Equalization Proficiency: 0.5, Equipment Familiarity: 0.8]. The lower the value, the weaker the ability in that dimension. At the same time, the algorithm pre-assigns a content feature vector for each chapter, labeling the skill point targeted by that chapter, such as the "Breathing Techniques" chapter vector = [Target Skill: Breathing Control, Weight: 0.9].

[0042] The algorithm evaluates the relevance score between user feature vectors and content feature vectors by calculating the cosine similarity or dot product between them. The basic formula can be simplified to: Relevance score = User feature vector × Content feature vector.

[0043] For example, if data analysis shows that a user has difficulty with underwater breathing control, the algorithm will prioritize relevant chapters such as "calmness skills training" and "breathing control basics" to ensure that the order of content presentation is tailored to the user's skill gaps and identifies them as core content units.

[0044] The sorted core content units are standardized, such as converting complex professional charts into intuitive cartoon animations or diagrams, or reconstructing text content using large fonts, step-by-step lists, and bold emphasis, making the presentation format suitable for the cognitive characteristics of novice users. A logic verification tool is used to validate the completeness of the standard content set, comparing it with a preset beginner teaching content template to check the coverage of key knowledge points and verify the logical coherence of the content; for example, ensuring that the explanation of "ear pressure equalization techniques" is followed by corresponding practice methods. After verification, the system generates the final beginner-adapted content set, a tailored, structurally complete, and user-friendly learning solution that effectively helps beginners gradually build diving skills in a safe and controlled knowledge environment.

[0045] S1032, if the user's skill level category is identified as a technical diver, then content containing in-depth technical details is extracted from the preset content library to obtain an advanced adapted content set; It should be noted that when a user's skill level is identified as a technical diver, the system retrieves a set of raw content related to in-depth technical details from a pre-defined content library. This raw content set is then layered, divided into data groups of different depths based on technical dimensions. The layered data groups are then cleaned and filtered to obtain cleaned data groups. Key information fields related to the technical diver's task are extracted from these cleaned data groups to construct a content framework that conforms to the technical diver's cognitive structure. This content framework is then deeply optimized and restructured in a contextualized manner to generate a content set. Finally, the content set undergoes consistency calibration to generate the final advanced adapted content set.

[0046] The system retrieves raw content sets containing in-depth technical details from a pre-defined content library and employs a technology-based classification method to stratify the raw content, forming data groups with different levels of professional depth. These stratified data groups undergo rigorous cleaning, and content is filtered based on pre-defined technical validity thresholds to ensure the accuracy and timeliness of the technical content. Using pattern recognition algorithms, key information fields highly relevant to technical diving missions are extracted from the cleaned data to construct a content framework that aligns with the professional cognitive structure of technical divers.

[0047] Based on the constructed content framework, the system applies in-depth analysis methods, such as machine learning models, to reorganize scenarios. For example, for deep-sea diving scenarios, the system will integrate gas management, multi-stage decompression calculation, and emergency response plans to generate a content set that can guide practical application.

[0048] The reorganized content set undergoes consistency calibration, which compares it with the latest industry standards and security specifications to correct content deviations and generate an advanced adapted content set.

[0049] S1033, For the primary or advanced adaptation content set, adjust the chapter information of the electronic manual to determine the final presentation order.

[0050] Specifically, the content reorganization algorithm adjusts the logical order of text, images, and multimedia elements based on user skill levels, behavioral preferences, and real-time feedback data. For novice users, a linear guided path is constructed, placing basic concepts at the beginning, complex charts at the end, and inserting periodic summaries. For technical users, frequently accessed specialized technical parameters are placed at the top, and highly relevant troubleshooting solutions are presented.

[0051] By calculating the semantic relevance between content units and the matching value of user needs, the algorithm generates an optimal content sequence and simultaneously optimizes the text and image layout strategy in the electronic manual, outputting an interface layout structure that conforms to user cognitive habits and operational needs.

[0052] S104, Collect user feedback signals on the interface layout structure and generate interface layout adjustment suggestions; Specifically, the system collects user feedback signals regarding the interface layout structure, including click frequency and scrolling speed. Key indicators are extracted from these signals to identify user interaction patterns in different interface areas. These interaction patterns are analyzed based on a pre-defined classification model. When the click frequency of a specific area falls below a pre-defined threshold within a specific time period, that area is identified as a low-engagement area. When the scrolling speed fluctuation in any area exceeds a pre-defined reasonable range, that area is marked as a potential problem area. Based on the low-engagement areas and potential problem areas, targeted interface layout adjustment suggestions are generated.

[0053] The system continuously monitors real-time behavioral data generated by users during their interaction with the electronic manual interface. Key feedback signals include click frequency and scrolling rate. Click frequency records the number of clicks made per unit of time on specific buttons, links, or content blocks. Scrolling rate captures the speed changes of users scrolling the page, including fast scrolling, slow browsing, or repeated up-and-down scrolling. Next, the system extracts key indicators from these feedback signals and uses a preset classification model to diagnose and analyze user interaction patterns. When the system detects that the click frequency of a specific content area or functional module is consistently below a preset threshold in consecutive sessions, it classifies that area as a low-engagement area, indicating that the current content or its presentation has failed to effectively attract user interest. When the system detects abnormal fluctuations in the user's scrolling rate in a certain area—for example, continuously scrolling rapidly in theoretical explanations or frequently scrolling back and forth with small amplitudes at key charts—and this fluctuation exceeds a preset reasonable range, it marks that area as a potential problem area. This may indicate difficulty in understanding the content, a chaotic layout, or inappropriate information density.

[0054] Based on the above diagnostic results, the system generates specific and actionable suggestions for adjusting the interface layout, such as adjusting the visual priority of low-engagement areas, reconstructing their content summaries to enhance their appeal, or associating them with high-engagement content for push notifications; or simplifying and reconstructing the content of potentially problematic areas, such as simplifying or displaying complex text paragraphs in steps, adding interactive annotations or decomposition animations to complex charts, or replanning the information layout of this section to reduce cognitive load.

[0055] S105, update the interface layout structure according to the interface layout adjustment suggestion, generate the adjusted interface version, and push the adjusted interface version to the corresponding electronic manual device.

[0056] Specifically, based on low-participation areas and potential problem areas, the key interface areas that need adjustment and their location coordinates are determined; the current interface layout is compared with the preset standard layout structure to determine the layout deviation; the position adjustment amount is calculated based on the layout deviation to generate a layout structure adjustment plan; the execution order of each adjustment parameter in the layout structure adjustment plan is set, and the interface layout structure is driven to update through an event listening mechanism; the interface layout structure is iteratively processed during the adjustment process to generate the final interface version.

[0057] The system acquires the initial state data of the current interface layout. This initial state data includes element coordinates, dimensions, and hierarchical relationships. By comparing this data with the preset standard layout structure, layout deviations are identified. An automated comparison is performed using a rule base based on JSON format. When an element attribute is detected to be inconsistent with the standard, such as the navigation bar width being a fixed value of 800px instead of the expected 100% viewport width, the system generates a log file containing the specific deviation coordinates and types. Based on the deviation analysis results, an iterative update mechanism generates a layout structure adjustment plan. This plan includes a complete set of CSS rules and DOM operation sequences to ensure that the interface has cross-device adaptive capabilities.

[0058] The system sets execution priorities for parameters in the adjustment plan, drives UI updates through an event listener mechanism (such as MutationObserver), generates multiple intermediate versions and verifies them in real time during the iterative process, and finally converts the adjustment parameters into final UI attributes through key-value mapping rules to generate UI version data that conforms to the display standards. All adjustment processes are fully recorded in the SQL database, including timestamps and change history, which facilitates subsequent auditing and version rollback.

[0059] After obtaining the final interface version, the system will collect the system environment information of the user's device, adapt the updated content through a compatibility conversion tool, and convert it into a static image or simplify the effect when it detects that the device does not support certain features. The system will generate an adaptation data package and distribute it using a push notification service (such as Firebase Cloud Messaging). The system will monitor the push status in real time to ensure that the content is delivered.

[0060] Based on the device's browsing mechanism configuration, the system will adjust rendering parameters, such as enabling WebGL acceleration or compatibility mode, and perform secondary processing through rendering tools such as the Canvas API to ensure that the display effect meets the preset standards.

[0061] Furthermore, step 103 includes the following preceding steps: A knowledge graph covering diving terminology, operating procedures, safety standards, and equipment parameters is pre-constructed. Nodes in the knowledge graph represent conceptual entities, and edges represent semantic relationships between entities, forming a machine-readable domain knowledge network.

[0062] Based on the user's current browsing topic and skill level, relevant subgraphs are extracted from the knowledge graph. For beginner users, subgraphs containing basic concepts and their directly related entities are extracted. For example, when browsing "mask drainage," the subgraph only includes directly related entities such as the mask, water, and drainage techniques, ensuring a focused knowledge base and avoiding information overload. For advanced users, subgraphs containing multiple reasoning paths and advanced technical parameter nodes are extracted. Similarly, for "mask drainage," the subgraph extends to advanced content such as special drainage techniques in dark environments or at great depths, and the impact of different mask materials on drainage effectiveness, linking them to specific operational parameters to meet their needs for in-depth exploration. It should be noted that the subgraph radius adjustment rule is based on a preset skill-depth mapping function, which is optimized online based on historical user behavior feedback data, ensuring that the subgraph coverage converges as the user's actual comprehension ability increases.

[0063] The subgraph is used as a semantic index source for content retrieval, guiding the system to extract the corresponding simplified or in-depth chapter information from the content library, and generating intelligent related content recommendations based on the semantic relationships in the graph.

[0064] Furthermore, during the content reorganization process in step 1033, the presentation order of text and images is sorted according to the semantic path priority in the subgraph, prioritizing the display of content units corresponding to nodes with high centrality or closely related to the user's historical points of confusion.

[0065] Through the above approach, the electronic manual achieves semantic understanding from keyword matching, providing a matching learning path for novice divers and a knowledge network that can be freely explored and deeply connected for technical divers, thereby improving the efficiency and accuracy of knowledge transfer.

[0066] In addition, the method includes the following steps: The system identifies the user's current task objective when using the electronic manual through explicit selection or implicit behavior analysis, such as search keywords and menu access patterns. These task objectives include, but are not limited to: preparing for the PADI Advanced Open Water Exam, executing a decompression dive plan, and troubleshooting regulator freeze malfunctions.

[0067] Based on a pre-built knowledge graph in the diving field, starting with the task objective and the user's existing knowledge, a personalized knowledge path is generated by using the shortest semantic path algorithm or reinforcement learning strategy, covering necessary prerequisite knowledge, key operation steps and safety verification points.

[0068] The knowledge path is divided into several stages, such as dividing PADI AOW exam preparation into modules like deep-diving, navigation, and top-level neutral buoyancy. For each stage, a set of content tailored to the user's skill level is generated, ensuring that the difficulty of the learning materials always matches the user's cognitive abilities. Embedding path navigation controls within the interface layout structure visually displays the current stage, completed nodes, and subsequent steps, providing users with clear learning guidance and progress management.

[0069] This invention also discloses a knowledge graph construction and retrieval system for diving electronic manuals, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements a knowledge graph construction and retrieval method for diving electronic manuals.

[0070] This invention also discloses a computer-readable storage medium storing instructions that, when executed by a processor, configure the processor to perform the knowledge graph construction and retrieval method for the diving electronic manual described in any of the above embodiments. The computer program can be stored in a machine-readable medium. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain middleware, etc. The machine-readable medium includes any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the machine-readable medium includes, but is not limited to, the aforementioned components.

[0071] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A method for constructing and retrieving a knowledge graph for a diving electronic manual, characterized in that, Includes the following steps: Data on user dwell time and interaction frequency during browsing of electronic manuals are obtained to generate user behavior feature sequences; Obtain a user behavior dataset; classify the user behavior dataset and combine it with user behavior feature sequences to determine the user skill level category; wherein, the user skill level category includes novice diver and technical diver; Adjust the presentation order of the electronic manual according to the user's skill level category to generate the interface layout structure; Collect user feedback signals on the interface layout structure and generate interface layout adjustment suggestions; The interface layout structure is updated according to the interface layout adjustment suggestions, an adjusted interface version is generated, and the adjusted interface version is pushed to the corresponding electronic manual device.

2. The knowledge graph construction and retrieval method for a diving electronic manual according to claim 1, characterized in that, The steps for adjusting the presentation order of the electronic manual according to the user's skill level category include: If the user's skill level is identified as a beginner diver, simplified chapter information is extracted from the preset content library to obtain a beginner-adapted content set. If the user's skill level category is identified as "technical diver", then content containing in-depth technical details is extracted from the preset content library to obtain an advanced adapted content set; For the primary or advanced adapted content set, the chapter information of the electronic manual is adjusted to determine the final presentation order.

3. The knowledge graph construction and retrieval method for a diving electronic manual according to claim 1, characterized in that, The step of obtaining user behavior feature sequences by acquiring data on user dwell time and interaction frequency during browsing of electronic manuals includes: The data acquisition module obtains data on the user's dwell time and interaction frequency during the browsing of the electronic manual, thus obtaining a preliminary behavioral dataset. The preliminary behavior dataset is cleaned and standardized to obtain a standardized behavior dataset. Based on standard behavioral datasets, determine the distribution of user dwell time on different pages or modules, and identify user browsing behavior preferences; When a user stays on a specific page or module for more than a preset threshold, the interaction frequency data corresponding to that page or module is extracted to determine the user attention index for that specific content. Based on the user attention index and interaction pattern, a user behavior sequence is constructed, and the user behavior sequence is classified to determine the habit category division; Based on habit categories, browsing behavior preferences, and device information from electronic manuals, a final sequence of user behavior characteristics is generated.

4. The knowledge graph construction and retrieval method for a diving electronic manual according to claim 2, characterized in that: The step of extracting simplified chapter information from the preset content library to obtain the initial adapted content set includes: When a user's skill level is identified as a beginner diver, simplified chapter information corresponding to the beginner diver is selected from the preset content library to form a preliminary content set; Based on the user's skill level and the relevance of the content, the chapter information in the initial content set is prioritized to determine the core content units that should be displayed first. The core content units are standardized to obtain a set of standard content. The integrity of the standard content set is verified to generate the final primary adapted content set.

5. The knowledge graph construction and retrieval method for a diving electronic manual according to claim 1, characterized in that: The step of extracting content containing in-depth technical details from a preset content library to obtain an advanced adapted content set includes: When a user's skill level is identified as "technical diver," a set of original content related to in-depth technical details is retrieved from a pre-set content library. The original content set is layered and divided into data groups of different depths based on technical dimensions. The stratified data groups are cleaned and filtered to obtain cleaned data groups; Extract key information fields related to technical diving missions from the cleaned data groups and construct a content framework that conforms to the cognitive structure of technical divers. The content framework is deeply optimized and restructured according to specific scenarios to generate a content set; The content set is then subjected to consistency calibration to generate the final advanced adapted content set.

6. The knowledge graph construction and retrieval method for a diving electronic manual according to claim 1, characterized in that, The step of collecting user feedback signals on the interface layout structure and generating interface layout adjustment suggestions includes: Collect user feedback signals regarding the interface layout structure; these feedback signals include click frequency and scrolling speed. Key indicators are extracted from the feedback signals to identify user interaction patterns in different interface areas; The interaction pattern is analyzed based on a preset classification model. When the click frequency of a specific area is lower than a preset threshold within a specific time period, the area is determined to be a low-participation area. When the scrolling rate fluctuation of any region exceeds the preset reasonable range, that region is marked as a potential problem region. Based on low-participation areas and potential problem areas, generate targeted interface layout adjustment suggestions.

7. The knowledge graph construction and retrieval method for a diving electronic manual according to claim 6, characterized in that, The interface layout structure will be updated according to the aforementioned interface layout adjustment suggestions, including: Based on low-participation areas and potential problem areas, identify the key interface areas that need adjustment and their location coordinates; Compare the current interface layout with the preset standard layout structure to determine the layout deviation; Calculate the position adjustment amount based on the layout deviation, and generate a layout structure adjustment scheme; The execution order of each adjustment parameter in the layout structure adjustment scheme is set, and the interface layout structure is updated through an event listener mechanism. The interface layout structure is iteratively processed during the adjustment process to generate the final interface version.

8. The knowledge graph construction and retrieval method for a diving electronic manual according to claim 2, characterized in that, Before the step of adjusting the presentation order of the electronic manual according to the user's skill level category and generating the interface layout structure, the following steps are also included: A knowledge graph covering the diving domain is pre-constructed, where nodes represent conceptual entities and edges represent semantic relationships between entities; Based on the user's current browsing topic and their skill level category, related subgraphs are extracted from the knowledge graph; specifically, subgraphs containing basic concepts and their directly related entities are extracted for novice users, and subgraphs containing multiple reasoning paths and advanced technical parameter nodes are extracted for technical users. The subgraph is used as a semantic index source for content retrieval.

9. The knowledge graph construction and retrieval method for a diving electronic manual according to claim 8, characterized in that, Also includes: Identify the user's current task objective when using the electronic manual; Based on a pre-built knowledge graph in the diving field, a personalized knowledge path is generated, with the task objective as the endpoint and the user's existing knowledge as the starting point. The knowledge path is divided into several stages, and adapted content matching the user's skill level category is generated for each stage. Embed a path navigation control in the interface layout structure to display the current stage, completed nodes, and subsequent steps.

10. A knowledge graph construction and retrieval system for an electronic diving manual, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, it implements the method as described in any one of claims 1 to 9.

Images