Data information storage method and system for online Chinese language learning platform

Abstract

The invention relates to the technical field of data processing, in particular to a data information storage method and system for an online Chinese language learning platform. The method comprises the following steps: obtaining platform user interaction data, and carrying out pinyin input track analysis and Chinese character component structure coding to obtain a pinyin alignment input frame; extracting a pinyin-Chinese character mapping relation in the pinyin alignment input frame, and performing Chinese character structure path reconstruction to obtain a user input word structure path set; learning content data are called, multi-modal content node embedding is carried out, and a content linkage fragment cluster is obtained; constructing a single session learning topology track according to the content linkage fragment cluster, and performing bidirectional topology sorting to obtain a learning track topological graph; and establishing a multi-dimensional primary key index and behavior triggering mechanism for the learning track topological graph to obtain a dynamic index distribution block, and transmitting the dynamic index distribution block to an online Chinese language learning platform. According to the invention, the stability and expansibility of the learning platform are obviously improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a data information storage method and system for an online Chinese language learning platform. Background Technology

[0002] Online Chinese language learning platforms provide course resources, interactive teaching, assignment submission, and intelligent assessment via the internet, greatly improving the accessibility and flexibility of Chinese language learning and promoting the transformation and upgrading of Chinese teaching models. During the actual operation of these platforms, a large amount of structured and unstructured data is continuously generated, including learners' actions, learning progress, course content, teacher interaction records, answer data, and assessment feedback, forming a vast data information system. To support the platform's intelligent recommendation, learning path optimization, personalized learning assessment, and subsequent content updates, efficient, controllable, and scalable storage and management of this learning data has become a core requirement for the platform's functional evolution and service quality improvement.

[0003] Traditional online learning platforms primarily rely on relational database systems for content management and user data archiving. While this approach offers advantages in managing structured data, it faces significant limitations when dealing with multi-dimensional and diverse learning data. For instance, traditional methods struggle to uniformly model and manage unstructured data such as voice interaction data, classroom interaction records, mixed text and image notes, and video learning progress, leading to severe data silos. Furthermore, when dealing with high-frequency read and write operations from a large user base, traditional data storage structures often experience access conflicts, data redundancy, and performance bottlenecks, severely impacting the stability and responsiveness of the learning platform. Summary of the Invention

[0004] Therefore, it is necessary for the present invention to provide a data information storage method and system for an online Chinese language learning platform to solve at least one of the above-mentioned technical problems.

[0005] To achieve the above objectives, a data storage method for an online Chinese language learning platform includes the following steps:

[0006] Step S1: Obtain platform user interaction data, and perform pinyin input trajectory analysis and Chinese character component structure encoding based on the platform user interaction data to obtain pinyin aligned input frames;

[0007] Step S2: Extract the pinyin-Chinese character mapping relationship in the pinyin-aligned input frame, and use the preset word structure index table to reconstruct the Chinese character structure path of the pinyin-Chinese character mapping relationship to obtain the user input word structure path set;

[0008] Step S3: Retrieve learning content data associated with user interaction data from the online Chinese language learning platform, and decompose content fragment structure units based on the learning content data; embed the user input word structure path set and content fragment structure units into multimodal content nodes to obtain content linkage fragment clusters;

[0009] Step S4: Construct a single-session learning topology trajectory based on the content linkage fragment clusters, and perform bidirectional topology sorting on the single-session learning topology trajectory to obtain the learning trajectory topology map;

[0010] Step S5: Perform learning state segmentation and aggregation on the learning trajectory topology map to generate Chinese language state compressed frames; establish a multi-dimensional primary key index and behavior triggering mechanism based on the Chinese language state compressed frames to obtain dynamic index distribution blocks, and transmit the dynamic index distribution blocks to the online Chinese language learning platform to deploy the mechanism to orchestrate tasks.

[0011] This invention achieves unified processing and in-depth mining of multimodal and multi-level data in the Chinese language learning process through meticulous analysis and structural encoding of platform user interaction data, greatly improving the scientific nature and precision of data management. The analysis of Pinyin input trajectories combined with the structural encoding of Chinese character components not only accurately captures detailed information from user input but also enhances the semantic expressiveness of the data through structured representation, laying a solid foundation for subsequent mapping relationships and path reconstruction. Using a pre-set word structure index table to reconstruct the structural path of the Pinyin-Chinese character mapping effectively achieves a precise structured representation of user-input words and supports complex semantic association analysis. The decomposition of content fragment structural units and multimodal node embedding technology enable deep integration of user input information and learning content across multiple dimensions such as text, graphics, and audio, greatly enriching the content's presentation and matching accuracy. By constructing conversational learning topology trajectories and bidirectional topological sorting, it not only comprehensively reveals the knowledge structure and logical order in the learning process but also enhances the temporal and semantic coherence description of the learning behavior trajectory. By segmenting and aggregating the state segments of the learning trajectory topology graph and constructing a primary key index, combined with a behavior triggering mechanism, dynamic compression and rapid indexing of learning states are effectively achieved, providing strong data support for intelligent recommendation and personalized learning path planning. Parameters such as the path overlap ratio threshold in topology sorting are set to 0.6 to ensure the rationality of path merging, preventing information loss due to excessive merging and avoiding redundancy in fragmented paths, thus improving the system's robustness and response speed. Overall, this solution effectively solves the challenges of traditional storage in managing multimodal and unstructured data, enhances the platform's ability to process large-scale, high-frequency interactive data, and significantly improves the stability, scalability, and personalized service level of the learning platform.

[0012] Optionally, step S1 includes:

[0013] Step S11: Obtain platform user interaction data and encode it with a unified timestamp to obtain the original input behavior record set;

[0014] Step S12: Perform dynamic pinyin segmentation on the original input behavior record set and mark the pinyin segment boundaries to obtain the pinyin action sequence matrix;

[0015] Step S13: Call the preset local dictionary index to perform pinyin-Chinese character mapping alignment on the pinyin action sequence to obtain the pinyin-Chinese character alignment table;

[0016] Step S14: Encode the Chinese character component structure in the Pinyin-Chinese character alignment table to generate a component structure vector set;

[0017] Step S15: Associate and fuse the Pinyin action sequence matrix and the component structure vector set according to the alignment relationship in the Pinyin-Chinese character alignment table to generate a Pinyin aligned input frame.

[0018] This invention achieves precise time synchronization and sequence management of input behavior by uniformly timestamping user interaction data on the platform. This effectively solves the time inconsistency problem in multi-source, multi-time-series data fusion, ensuring the accuracy and reliability of the data foundation for subsequent processing. Dynamic pinyin segmentation and annotation of pinyin segment boundaries help capture fine-grained pinyin information from user input, improving the coherence and accuracy of pinyin recognition, thus more accurately reflecting real input behavior. Reasonable setting of boundary annotation parameters ensures the continuity and integrity of segmentation, avoiding over-segmentation or omissions. Utilizing a local dictionary index for pinyin-Chinese character mapping and alignment enables pinyin input to be quickly and accurately mapped to corresponding Chinese characters, balancing dictionary coverage and response speed. Parameter design considers both mapping accuracy and computational efficiency. Encoding the pinyin-Chinese character alignment results with Chinese character component structures, by decomposing Chinese characters into basic components and expressing them quantitatively, greatly enhances the computability of Chinese character semantic and structural information, facilitating subsequent deep semantic analysis and content matching. By merging the Pinyin action sequence matrix and the component structure vector set according to the alignment relationship, a Pinyin-aligned input frame is formed, realizing the organic integration of multi-dimensional data and improving the expressive richness and semantic accuracy of the input data.

[0019] Optionally, step S12 includes:

[0020] Step S121: Extract the character input sequence from the original input behavior record set and construct the pinyin fragment stream;

[0021] Step S122: According to the preset initial consonant-final vowel-tone triplet matching logic, the pinyin segment stream is segmented into pinyin substrings to obtain a set of candidate pinyin segments;

[0022] Step S123: Perform pinyin distribution statistics on the candidate pinyin segment set, and calculate the syllable confidence value based on the pinyin distribution statistics results;

[0023] Step S124: Based on the pinyin length, input continuity and corresponding syllable confidence value in the candidate pinyin segment set, a weighted score is performed, the substring segmentation combination with the highest score is selected as the segmentation boundary rule, and the candidate pinyin segment set is segmented using the segmentation boundary rule to obtain the pinyin segmentation boundary sequence;

[0024] Step S125: Perform pinyin unit error correction expansion on the pinyin segmentation boundary sequence to construct a set of pinyin action items;

[0025] Step S126: Align the set of Pinyin action items with the corresponding start and end positions, error correction status and Pinyin type in the Pinyin segment stream to generate a Pinyin action sequence matrix.

[0026] This invention constructs a pinyin segment stream from the character sequence of the original input behavior, realizing a structured representation of the input information, which facilitates subsequent refined processing and recognition. Combining initial consonant-final vowel-tone triplet matching logic to segment the pinyin segment stream can accurately divide pinyin substrings, avoiding recognition errors caused by pinyin breaks or overlaps, and ensuring the integrity and accuracy of the segmentation results. Distribution statistics of candidate pinyin segments and calculation of syllable confidence values ​​effectively quantify the reliability of pinyin substrings, filtering out more reliable pinyin inputs and improving the overall stability of pinyin recognition. By comprehensively considering pinyin length, input continuity, and confidence through weighted scoring, the optimal substring segmentation combination is selected, ensuring the scientific rationality of the segmentation boundary rules, effectively balancing the fineness and robustness of segmentation. The parameter settings reflect a balance between input stream continuity and pinyin accuracy, avoiding sentence segmentation problems caused by over-segmentation and ambiguity caused by under-segmentation. Error correction and expansion of the pinyin segmentation boundary sequence helps to correct potential pinyin errors in user input, improving the error tolerance and recognition accuracy of the input data. Finally, the corrected pinyin action items are aligned with the start and end positions, error correction status, and pinyin type in the pinyin segment stream to construct a pinyin action sequence matrix, forming a unified and accurate input expression structure, which facilitates subsequent mapping of pinyin to Chinese characters and semantic analysis.

[0027] Optionally, step S125 includes:

[0028] Extract pinyin substrings from the pinyin segmentation boundary sequence and construct a list of basic pinyin units;

[0029] Load the platform's built-in Pinyin error correction mapping table, including initial consonant replacement rules, final vowel merging rules, and tone tolerance table;

[0030] Based on the pinyin error correction mapping table, each pinyin substring in the basic pinyin unit list is expanded by initial consonant-final vowel error matching to obtain a pinyin error correction candidate set;

[0031] Extract the word-initial and word-final pinyin collocation probabilities of the left and right context intervals ±2 pinyin units in the pinyin fragment stream, calculate the input context misjudgment probability by combining the self-structural error of each candidate option in the pinyin error correction candidate set, and calculate the confidence reconstruction value based on the input morning and afternoon misjudgment probabilities, and retain the candidate options with confidence reconstruction values ​​≥0.6;

[0032] Each pinyin substring is matched with the candidate with the highest confidence reconstruction value in the corresponding pinyin error correction candidate set to obtain a standardized pinyin action item set.

[0033] This invention extracts pinyin substrings from the pinyin segmentation boundary sequence to construct a basic pinyin unit list, achieving a structured representation of user-input pinyin information and laying a solid foundation for subsequent error correction and standardization processing. The platform's built-in pinyin error correction mapping table, covering initial consonant replacement, final vowel merging, and tone tolerance, effectively covers common pronunciation and writing errors in pinyin input, enhancing the comprehensiveness and flexibility of error correction and enabling the system to adapt to diverse user input. Based on this error correction mapping table, initial consonant-final error matching is extended to the basic pinyin units to construct a pinyin error correction candidate set, effectively capturing the user's possible pinyin error range and improving the accuracy and recall of pinyin error correction. By extracting the probability of word beginning and ending collocations within ±2 pinyin units in the left and right context intervals of the pinyin segment stream, and combining this with the structural errors of the candidate options themselves to calculate the probability of misjudgment in the input context, and further calculating the confidence reconstruction value, a scientific and reasonable confidence evaluation mechanism is formed. This mechanism can accurately identify high-confidence pinyin candidates, effectively reducing the risk of misjudgment. The confidence reconstruction value parameter was set to no less than 0.6, balancing error correction sensitivity and robustness, avoiding over-correction that could introduce errors, and ensuring the reliability of Pinyin standardization. Finally, based on the highest confidence reconstruction value matching standard, standardized processing of Pinyin action items was achieved, ensuring the uniformity and standardization of input data, greatly improving the accuracy of subsequent Chinese character mapping and semantic analysis, and enhancing the overall system interaction experience.

[0034] Optionally, the Chinese character structure path reconstruction in step S2 includes:

[0035] The structural path query and location of each Chinese character in the pinyin-Chinese character mapping relationship are performed using the word structure index table to obtain the structural record sequence;

[0036] The structural hierarchy is expanded based on the structural record sequence, thereby constructing a sequence of Chinese character structural topology graphs;

[0037] Linearly encode the traversal path from the main component to the secondary component in each graph of the Chinese character structure topology graph sequence to obtain the structure path encoding sequence;

[0038] Semantic compression is performed on the structure path encoded sequence to form a set of structure path segments;

[0039] The set of structural path fragments is bound to the user input behavior ID in the pinyin-aligned input frame to generate a set of user input word structural paths.

[0040] This invention utilizes a word structure index table to perform structural path lookup and location for each Chinese character in the Pinyin-Chinese character mapping relationship, achieving a systematic extraction of complex structural information of Chinese characters. This ensures that each component of a Chinese character and its hierarchical relationships are accurately identified and represented, solving the problem of incomplete or confusing Chinese character structural representation in traditional input processing. Based on the structural record sequence, the invention expands the hierarchical structure of Chinese characters and constructs a topological graph sequence, which helps to comprehensively reflect the internal composition and spatial hierarchy of Chinese characters, supports multi-dimensional analysis of Chinese character structures, and improves the accuracy of subsequent encoding and semantic processing. By linearly encoding the traversal paths from main components to secondary components in the topological graph, it transforms them into structural path encoding sequences, enabling the concise expression of complex graphical structures, facilitating computer processing and storage, and improving encoding efficiency and the standardization of expression. Semantic compression is used to process the structural path encoding sequences, effectively removing redundant information, extracting core structural features, reducing the burden of data storage and transmission, while maintaining the integrity of structural expression and semantic association, enhancing the system's response speed and processing performance. By binding the set of structural path fragments with user input behavior IDs to form a set of user input word structural paths, the organic integration of input behavior and Chinese character structure expression is achieved, ensuring the temporal consistency and traceability of input parsing. The overall parameter design takes into account both encoding granularity and compression ratio. For example, the reasonable values ​​of linear encoding length and compression threshold ensure that the details of Chinese character structure can be accurately expressed while effectively controlling the consumption of computing resources and storage space. This significantly improves the performance and user experience of the online Chinese language learning platform in functions such as Chinese character input parsing, learning content association, and intelligent feedback.

[0041] Optionally, the splitting of content fragment structural units in step S3 includes:

[0042] Based on the user input behavior ID and course number in the platform's user interaction data, the corresponding course content set is obtained through the online Chinese language learning platform, and the course text units in the corresponding course content set are loaded to obtain the learning content data;

[0043] Semantic segmentation is performed on natural paragraphs in the learning content data to construct semantic fragment units;

[0044] Based on semantic fragment units, word and structural feature annotations are performed to obtain structured annotation results;

[0045] Semantic tightness analysis is performed on the structured annotation results, and semantic segments with semantic tightness ≥ 0.75 are marked as knowledge focuses to obtain a set of knowledge focuses;

[0046] The semantic fragment units are divided into content fragment units based on the knowledge focus set, and the timestamps of each content fragment unit are relocated to obtain the content fragment structure unit.

[0047] This invention achieves efficient association between learning content and user interaction data by accurately retrieving corresponding course content based on user input behavior ID and course number, ensuring the targeted and real-time nature of content retrieval and improving the efficiency of platform resource access. Semantic segmentation of natural paragraphs within course text units constructs semantic fragment units, enabling fine-grained division of text content for subsequent refined analysis and processing, thereby enhancing the structured expression of the content. Word and phrase annotation and structural feature marking effectively reveal the semantic hierarchy and word relationships of the text, enriching the structured information expression of the content and supporting subsequent intelligent matching and knowledge point mining. Through semantic density analysis, using 0.75 as a threshold, important knowledge focuses are accurately distinguished, ensuring that the marked knowledge points have high semantic relevance and pedagogical value, improving the accuracy and practicality of knowledge focuses, and facilitating the key grasp and effective review of learning content. Based on the knowledge focus set, content fragment units are divided, and combined with timestamp relocation, dynamic adjustment of the content structure and accurate mapping of the time dimension are achieved, supporting fine-grained tracking and management of learning progress. The overall parameter settings balance the precision of segmentation with processing efficiency, enabling the system to fully capture the core knowledge structure of the course content while efficiently responding to user needs, significantly improving the content service quality and learning experience of the online Chinese language learning platform.

[0048] Optionally, the multimodal content node embedding in step S3 includes:

[0049] The user-input word structure path set is encoded using structural topology vectorization to obtain a structure vector encoding set;

[0050] Based on the structural units of the content fragments, extract the text semantic embeddings, graphic texture features of Chinese character components in the accompanying graphics and text, phoneme sequence features in the audio dubbing of the fragments, and text grammatical tag distribution features from the corresponding course content set to construct a multimodal fragment feature set;

[0051] Bidirectional matching is performed on the structure vector encoding set and the multimodal fragment feature set to obtain the association matching matrix;

[0052] Filter matching pairs with a matching degree ≥ 0.7 from the association matching matrix to construct a content node matching set;

[0053] Based on the content node matching set, content fragments are grouped and aggregated to generate content-linked fragment clusters.

[0054] This invention achieves accurate digital representation of Chinese character structure information by converting user-input word structure path sets into structural topology vector codes, enhancing the compatibility and efficiency of input data in computer processing and effectively supporting subsequent multimodal content fusion. Combining text semantic embeddings, graphic texture features of Chinese character components, phoneme sequence features of dubbing audio, and text grammatical tag distribution features in the course content, a rich multimodal fragment feature set is constructed, improving the semantic, visual, and auditory multi-dimensional representation capabilities of the content, thereby enhancing the expressiveness and matching accuracy of the content. Performing a bidirectional matching process enables a comprehensive comparison between the input structure code and multimodal content features, ensuring bidirectional verification and high consistency of information, avoiding bias caused by unidirectional matching, and improving the accuracy and robustness of the matching results. Selecting a matching degree threshold of 0.7 balances matching accuracy and coverage, ensuring high-quality content association while avoiding content omissions due to over-screening, thus improving the overall matching performance of the system and the utilization efficiency of learning resources. Based on the matching results, a content node matching set is constructed and grouped and aggregated to form a content linkage fragment cluster. This effectively integrates the association structure between user input and learning content, promotes hierarchical management and dynamic linkage of content, and improves the intelligence level and user experience of the online Chinese language learning platform in terms of personalized recommendation, learning path planning and knowledge point tracking.

[0055] Optionally, step S4 includes:

[0056] Step S41: Extract the list of segments within each cluster of content-linked segments, and sort the segment list by time according to the trigger time of each content segment unit to obtain the initial segment trajectory sequence;

[0057] Step S42: Perform semantic adjacency analysis on the initial segment trajectory sequence to generate a segment transition relationship graph;

[0058] Step S43: Calculate the transition path weights of the edge set in the fragment transition graph, and prune the edges with transition path weights < 0.45 to obtain the optimized learning sub-trajectory set;

[0059] Step S44: Aggregate the optimized learning sub-trajectory set into a directed graph structure to obtain a single-session learning topology trajectory graph;

[0060] Step S45: Perform bidirectional topological sorting on the single-session learning topology trajectory graph to obtain the learning trajectory topology graph.

[0061] This invention accurately reconstructs the temporal sequence of a user's learning process by sorting the list of segments within a content-linked segment cluster according to their trigger times, ensuring the temporal continuity and integrity of the learning trajectory and improving the temporal accuracy of subsequent analysis. Semantic adjacency analysis of the initial segment trajectory sequence effectively reveals the semantic connections and logical transitions between different content segments, promoting a deeper understanding of the learning content structure and the mining of its relevance, thus enhancing the semantic expressiveness of the learning trajectory. An edge-weighted calculation mechanism is employed, which quantitatively reflects the intensity of learning transfer between segments by comprehensively weighting semantic similarity and knowledge point relevance, enabling effective identification of key nodes and important transitions in the learning path. A weight threshold of 0.45 is set for pruning, reasonably filtering weakly related paths, avoiding redundancy and noise interference, improving the clarity and accuracy of the learning trajectory, and facilitating subsequent trajectory aggregation and analysis. The optimized sub-trajectory set is aggregated through a directed graph structure, which helps to construct a multi-path topology structure that conforms to actual learning logic, comprehensively reflecting the diverse learning behavior trajectories of users, and enhancing the expressive richness and dynamic adaptability of the trajectory. Bidirectional topology sorting ensures that the learning trajectory topology map unfolds in an orderly manner in both directions, supporting multi-angle learning path analysis and evaluation, and improving the completeness and flexibility of trajectory analysis. The overall parameter configuration balances data accuracy and processing efficiency, guaranteeing the stability and accuracy of the learning trajectory construction process, and significantly enhancing the personalized learning path optimization capabilities and intelligent teaching support level of the online Chinese language learning platform.

[0062] Optionally, step S5 includes:

[0063] Step S51: Divide the learning trajectory topology graph into semantic stages and generate semantic stage segment groups;

[0064] Step S52: Aggregate stages based on the start and end time intervals and structural jumps in the semantic stage fragment group to construct a learning state evolution chain;

[0065] Step S53: Based on the learned state evolution chain, extract the principal component feature fields to construct a compressed frame of Chinese language state;

[0066] Step S54: Extract the topic density, structural complexity, and learning duration segments from the Chinese language state compression frame, and convert the topic density, structural complexity, and learning duration segments into multiple primary key combinations to generate a primary key index set;

[0067] Step S55: Aggregate primary keys with similar topic density and structural complexity in the primary key index set into behavior patterns and establish a trigger mapping table;

[0068] Step S56: Merge and package the primary key index set and the trigger mapping table to obtain a dynamic index distribution block, and transmit the dynamic index distribution block to the online Chinese language learning platform to deploy the orchestration task.

[0069] This invention achieves a detailed decomposition of complex learning processes by semantically segmenting the learning trajectory topology graph. This allows learning behaviors to be divided into stages with clear semantic features, effectively improving the understanding of the internal logic and knowledge hierarchy of the learning content. Aggregating based on the start and end time intervals and structural jumps of these stage segments allows for the reasonable integration of temporally continuous and structurally relevant learning content, forming a coherent learning state evolution chain. This accurately reflects the dynamic evolution of the learning process and the progress of knowledge acquisition. By extracting principal component feature fields from the learning state evolution chain, a Chinese language state compression frame is constructed, significantly reducing data dimensionality and redundant information. This achieves efficient expression and storage of learning states, improving the computational efficiency of subsequent processing. The topic density, structural complexity, and learning duration segments are extracted from the state compression frame and transformed into multi-primary key combinations to form a primary key index set. This enhances the multi-dimensional expressive power of the index, supporting fine differentiation and rapid retrieval of different learning states. Aggregating primary keys with similar topic density and structural complexity into behavioral patterns and establishing a trigger mapping table enables patterned management and automatic triggering mechanisms for learning behaviors, improving the response speed and accuracy to changes in learning states. Finally, a dynamic index distribution block is generated by merging and packaging the primary key index set and the trigger mapping table, and then transmitted to the learning platform. This ensures efficient data transfer and dynamic task scheduling, effectively supporting the platform's intelligent management and personalized teaching services. Parameters in each step, such as the structural jump threshold, principal component selection ratio, and multiple primary key combination methods, have been optimized to balance data compression and information retention. This ensures both a detailed portrayal of the learning state and improved system processing performance and response speed, guaranteeing the stable operation and efficient service of the online Chinese language learning platform in a large-scale user environment.

[0070] Optionally, this specification also provides a data storage system for an online Chinese language learning platform, used to execute the data storage method for the online Chinese language learning platform as described above. The data storage system for the online Chinese language learning platform includes:

[0071] The Chinese character structure alignment module is used to acquire platform user interaction data, and to perform pinyin input trajectory analysis and Chinese character component structure encoding based on the platform user interaction data to obtain pinyin aligned input frames;

[0072] The mapping relationship analysis module is used to extract the pinyin-Chinese character mapping relationship in the pinyin-aligned input frame, and to reconstruct the Chinese character structure path of the pinyin-Chinese character mapping relationship using a preset word structure index table to obtain the user input word structure path set;

[0073] The content fragment embedding module is used to retrieve learning content data associated with user interaction data from the online Chinese language learning platform, and to split the content fragment structure units according to the learning content data; the user input word structure path set and the content fragment structure units are embedded with multimodal content nodes to obtain content linkage fragment clusters;

[0074] The learning trajectory analysis module is used to construct a single-session learning topology trajectory based on content-linked fragment clusters, and to perform bidirectional topological sorting on the single-session learning topology trajectory to obtain a learning trajectory topology map.

[0075] The learning state aggregation module is used to aggregate the learning state segments of the learning trajectory topology map and generate Chinese language state compressed frames. Based on the Chinese language state compressed frames, a multi-dimensional primary key index and behavior triggering mechanism are established to obtain dynamic index distribution blocks. The dynamic index distribution blocks are then transmitted to the online Chinese language learning platform to deploy the mechanism and orchestrate tasks. Attached Figure Description

[0076] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0077] Figure 1 This is a flowchart illustrating the steps of the data storage method for the online Chinese language learning platform of the present invention.

[0078] Figure 2 This is a detailed flowchart of step S1 in the present invention;

[0079] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0080] The technical method of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0081] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0082] It should be understood that although terms such as "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0083] To achieve the above object, please refer to Figures 1 to 2 , the present invention provides a method for storing data information of an online Chinese language learning platform, and the method includes the following steps:

[0084] Step S1: Obtain platform user interaction data, and perform pinyin input trajectory parsing and Chinese character component structure encoding according to the platform user interaction data to obtain a pinyin-aligned input frame;

[0085] In this embodiment, during the process of the user practicing real-time pinyin input through the online Chinese language learning platform, interaction data such as the user input sequence, key timestamp, key position start and end interval, and input device type are automatically collected and uniformly converted into a time-aligned format with an accuracy of 0.01 seconds. The original input behavior records extracted from this data set contain fields {user_id, key_seq, timestamp, device_type}. Then, the platform uses the built-in pinyin segmentation module to perform a three-segment segmentation of the character sequence into initials, finals, and tones. For example, "zhong1guo2" is segmented into {zh-ong-1, guo-2}, forming a pinyin action matrix where n is the number of pinyin units. Subsequently, the locally cached lexicon index (constructed based on GB2312) is called to map the pinyin to standard Chinese characters. For example, "zhong1" is mapped to "中", forming a pinyin-Chinese character alignment table T ph = [(zhong1, 中), (guo2, 国)]. Finally, according to the Chinese character decomposition rules, "中" is parsed into "丨 + 口", and "国" is parsed into "囗 + 玉". Each component structure vector uses a 5-dimensional one-hot encoding, and the structure encoding matrix Finally, combine P and S and align them with each pinyin in T ph to obtain a pinyin-aligned input frame I 拼 for structure path reconstruction.

[0086] Step S2: Extract the pinyin-Chinese character mapping relationship in the input frame by pinyin alignment, and use the preset word structure index table to reconstruct the Chinese character structure path for the pinyin-Chinese character mapping relationship, obtaining the user input word structure path set;

[0087] In this embodiment, for each record in the pinyin alignment input frame I 拼 in, call the preset word structure index table T struct (in the format of JSON, recording the component path and hierarchical information of each Chinese character) for path reconstruction. For example, query the structure index corresponding to "国" as {"国": ["囗", ["玉", ["丶", "丶"]]]}, which is expanded into a linear structure path [囗 → 玉 → 丶 → 丶] by the path parser. Perform a topological expansion process with the main component prioritized on the structure path to generate a tree-like graph structure G, where the nodes are component codes and the edges are structure position offsets (such as up, down, left, right). This graph structure is traversed and converted into a path vector sequence V = [(囗, 1), (玉, 2), (丶, 3), (丶, 4)], and further, the path segment compression rule is adopted to only retain the topological backbone path and branch numbers, obtaining the path compression code C path = ["囗 → 玉", "玉 → 丶 × 2"]. Finally, the system binds the encoded path set with the input ID of the original pinyin input frame to generate the user input word structure path set P 字 = {user_id, [path_1, path_2,...]}, for subsequent multi-modal content node matching use.

[0088] Step S3: Retrieve the learning content data associated with the interaction data of the platform users through an online Chinese language learning platform, and split the content fragment structure units according to the learning content data; perform multi-modal content node embedding on the user input word structure path set and the content fragment structure units to obtain the content linkage fragment cluster;

[0089] In this embodiment, during the user's learning process, the platform calls the course number associated with the current input behavior ID and retrieves the corresponding course resource package. Each text in the resource package is segmented by paragraph, and词性标注与语义句法分析 is performed on each paragraph. For example, the paragraph "我们学习汉字的结构" will be decomposed into structure units ["我们", "学习", "汉字", "结构"], and the dependency relationship of each unit is extracted. Subsequently, the platform performs vector matching based on the structure vectors in the user input word structure path set P 字 and extracts the following feature vector groups for each content fragment: (1) text semantic vector (generated by the embedding model), (2) graphic texture vector (derived from the convolution features of Chinese character stroke images), (3) voice phoneme spectrum vector The above three types of feature vectors are matched with user structure vectors using cosine similarity. Those with a matching score >0.7 are selected to form a set of content-linked segments. Each cluster of structures is denoted as . Ultimately, a cluster of interconnected content segments is formed for learning path modeling.

[0090] Step S4: Construct a single-session learning topology trajectory based on the content linkage fragment clusters, and perform bidirectional topology sorting on the single-session learning topology trajectory to obtain the learning trajectory topology map;

[0091] In this embodiment, each cluster in the content linkage segment cluster is sorted along the time dimension to generate an initial learning trajectory sequence S. 轨 = [f1, f2, ..., fn], where each fi is a fragment node. Semantic adjacency scoring is performed between adjacent fragments in the sequence to form a fragment transition graph G. 转 = (V, E), where the weight of E is calculated from the intersection ratio of keywords (weight range [0,1]). Edges with weights < 0.45 are pruned to preserve highly coherent paths. The connected subgraphs are then aggregated to form a directed graph T. 轨 The system records the learning flow. It then performs a forward topology scan and a backward backtracking, cross-referencing bidirectional paths, calculating the proportion of overlapping nodes, and selecting path pairs with an intersection-union ratio (IU) ≥ 0.6, retaining them as the main path chain M. 链 Redundant paths in non-main path chains undergo semantic merging and are mapped to nodes on the main path. Finally, the main path and semantic projections form a structured "learning trajectory topology graph" T. 拓 This includes node hierarchy, transition probabilities, and semantic annotations, which are used for subsequent state evolution.

[0092] Step S5: Perform learning state segmentation and aggregation on the learning trajectory topology map to generate Chinese language state compressed frames; establish a multi-dimensional primary key index and behavior triggering mechanism based on the Chinese language state compressed frames to obtain dynamic index distribution blocks, and transmit the dynamic index distribution blocks to the online Chinese language learning platform to deploy the mechanism to orchestrate tasks.

[0093] In this embodiment, based on T 拓 The learning process is divided into multiple semantic stages based on the semantic topics, duration, and structural jumps between segments. Each stage has a minimum time interval of 60 seconds and a structural jump of no more than two layers. Principal component features are extracted from each stage, including Structural Expansion Index (SEI), Semantic Density Index (SDI), and Mean Node Stay (NTS), to construct a state vector matrix. For M 压 Perform K-means compression (k=3) to generate Chinese language state compressed frames S. 帧 The system then retrieved data from S. 帧Extracting topic density (e.g., "culture" topic density = 0.82), structural complexity (e.g., average number of syntactic branches = 4.3), and learning duration (30-180 seconds per segment) to construct the primary key combination K. 主 ={0.82,4.3,120}. Primary key combinations with the same topic density ±0.05 and structural complexity ±1 are grouped into behavior groups, aggregated into a behavior pattern set. Finally, a trigger mapping table is constructed, including primary key mapping behavior decision templates, and the dynamic index distribution block D is output. 索 ={K 主 The template ID is sent to the platform controller in real time for deployment strategies.

[0094] Optionally, step S1 includes:

[0095] Step S11: Obtain platform user interaction data and encode it with a unified timestamp to obtain the original input behavior record set;

[0096] In this embodiment, when a user performs input operations through the platform, the platform backend records the complete input behavior sequence, including the user ID (user_id), key sequence (key_seq), key timestamp (timestamp), device identifier (device_type), and other raw input behavior information. The platform sets the timestamp precision to 10 milliseconds and maps all input records to a standard timeline to avoid cross-device offset issues. For example, the user's input of "zhongguo" is recorded in the following structured format: {"user_id":"U20240601","key_seq":["z","h","o","n","g","g","u","o"],"timestamp":[100,150,180,220,260,300,340,380],"device_type":"mobile"}; linear time series interpolation is used to fill in some missing key intervals to obtain an input sequence with a uniform time step. The completed raw input behavior record set will serve as the base dataset for subsequent pinyin segmentation and input parsing.

[0097] Step S12: Perform dynamic pinyin segmentation on the original input behavior record set and mark the pinyin segment boundaries to obtain the pinyin action sequence matrix;

[0098] In this embodiment, the key sequence is extracted from the original input behavior record and assembled into a Pinyin segment stream, such as ["zh","o","n","g","g","u","o"]. Based on the built-in initial consonant-final vowel-tone triplet matching table (such as {"zh":S,"ong":V,"1":T}), the system combines this segment stream into a candidate Pinyin segment set, such as ["zhong","guo"]. The structure of each segment is calibrated based on the initial consonant position, syllable boundary, and tone mapping. During this process, the Pinyin length threshold is set to 1-6 characters, and the syllable confidence value is calculated using word frequency inverse mapping, with a minimum threshold set to 0.6. Finally, a scoring function S = 0.4·L + 0.3·C + 0.3·P (where L is the length weight, C is the continuity score, and P is the confidence level) is used to generate the pinyin segmentation boundary sequence, for example: [{"start":0,"end":4,"pinyin":"zhong","sc ore":0.83},{"start":5,"end":7,"pinyin":"guo","score":0.79}]; this sequence is then arranged into a matrix form M. 拼每行分别表示起始、结束、拼音内容及得分，用于拼音动作生成。

[0099] 步骤S13：调用预设的本地词库索引，对拼音动作序列进行拼音-汉字映射对齐，得到拼音-汉字对齐表；

[0100] 本实施例中，调用本地缓存词库索引(例如基于《通用规范汉字表》构建的索引结构，包含约8105条拼音-汉字映射关系)，对步骤S12中拼音动作序列矩阵M拼进行一对一对齐。例如，"zhong"→"中"，"guo"→"国"。系统内部设定候选汉字置信度列表，例如对"guo”可对应"国(0.85)”,"过(0.70)”,"果(0.58)”等，通过输入上下文进行置信度筛选，保留置信度最高项。对齐结果整理为表格结构：[{"pinyin":"zhong","hanzi":"中","confidence":0.92},{"pi nyin":"guo","hanzi":"国","confidence":0.85}]；该拼音-汉字对齐表Tph用于指导下一步的结构编码绑定操作。

[0101] 步骤S14：对拼音-汉字对齐表进行汉字部件结构编码，生成部件结构向量集；

[0102] 本实施例中，依据Tph中的汉字项，调用结构分解器对每个汉字执行部件级拆解。例如，"中”拆解为"丨”和"口”，"国”拆解为"囗”和"玉”。每个部件采用五维属性向量编码，包括笔画数(strokes)、笔顺类型(stroke_order)、相对位置(position_index)、部件类型(component_type)、部件层级(hierarchy_level)。示例如下：{"中":[{"component":"丨","vector":[1,2,1,1,1]},{"component":"口","vector":[3,1,2,0,1]}],"国":[{"component":"囗","vector":[3,1,1,0,1]},{"component":"玉","vector":[5,2,2,1,2]}]}；将这些部件向量按汉字顺序构建结构向量集其中n为部件总数。该向量集用于与拼音动作序列的融合。

[0103] 步骤S15：将拼音动作序列矩阵与部件结构向量集按照拼音-汉字对齐表中的对齐关系进行关联融合，生成拼音对齐输入帧。

[0104] 本实施例中，将拼音动作序列矩阵M拼与结构向量集Sstruct对应拼音项进行对齐融合。融合方式依据拼音-汉字对齐表中的位置信息，将拼音动作的起止时间及内容分别映射至结构向量集中的对应汉字结构项。融合结果组织为拼音对齐输入帧I拼，结构如下：[{"pinyin":"zhong","time_range":[100,260],"hanzi":"中","components":[{"co mponent":"丨","vector":[1,2,1,1,1]},{"component":"口","vector":[3,1,2,0,1]}]},{"pinyin":"guo","ti me_range":[300,380],"hanzi":"国","components":[{"component":"囗","vector":[3,1,1,0,1]},{"comp onent":"玉","vector":[5,2,2,1,2]}]}]，该结构将拼音行为轨迹与汉字结构表达进行语义级融合，是后续进行结构路径重建及多模态匹配的关键基础帧。

[0105] 可选地，步骤S12包括：

[0106] 步骤S121：提取原始输入行为记录集中的字符输入序列，并构建拼音片段流；

[0107] 本实施例中，从原始输入行为记录集中提取出按键事件序列，以字符形式还原出用户连续拼音输入的字符串，例如从以下原始行为数据：{"user_id":"U20250620","key_seq":["n","i","h","a","o","z","h","o","n","g","g","u","o"],"timestamp":[100,120,150,170,190,250,270,290,310,330,350,370,390]}，拼接字符流为原始输入串"nihaozhongguo"。为便于后续切分处理，该字符串被转化为拼音片段流["n","i","h","a","o","z","h","o","n","g","g","u","o"]，并对每个字符赋予序号索引，用于后续起止位置定位与切分边界跟踪。

[0108] 步骤S122：根据预设的声母-韵母-音调三元组匹配逻辑对拼音片段流进行拼音子串切分，得到候选拼音段集合；

[0109] 本实施例中，加载预设的声母-韵母-音调组合表，该表覆盖常用拼音组合形式(共收录423个常用拼音项)，每项包括声母结构、韵母模式、音调容差范围。例如"zhong"的结构为{"shengm u":"zh","yunmu":"ong","tone":null}。系统扫描拼音片段流，结合最大匹配长度(不超过6个字符)和优先声母匹配的逻辑，从首字符起逐步生成拼音子串候选段，例如：["ni","hao","zhong","guo"]，每个候选拼音段被记录其起止位置、切分组合与结构匹配类型，作为后续置信统计基础。

[0110] 步骤S123：对候选拼音段集合进行拼音分布统计，并根据拼音分布统计结果计算音节置信度值；

[0111] 本实施例中，基于候选拼音段集合进行拼音频次统计，加载平台词频统计表(基于用户行为构建，包含拼音使用频次与错误率分布)，对每个拼音段的音节置信度进行评估。以"zhong"为例，其在近30天用户输入中频次为43212，错误率为0.0089，置信度计算结果为1-误差率×调整系数，设定调整系数为1.2，得到：conf("zhong")＝1-0.0089×1.2≈0.9893；该值被记录于拼音分布统计表中，并附加在对应拼音段结构中供后续加权评分使用。

[0112] 步骤S124：基于候选拼音段集合中的拼音长度、输入连续性及对应的音节置信度值进行加权评分，选取评分最高的子串切分组合作为分节边界规则，并利用分节边界规则对候选拼音段集合进行分节切分，得到拼音切分边界序列；

[0113] 本实施例中，基于每个候选拼音段的长度L、输入连续性C(无中断次数比例)和置信度P，计算其分节评分，评分函数如下：Score＝0.4×L_norm+0.3×C+0.3×P；其中L_norm为长度归一化值，最大设定为6，C值基于输入时序判断(如当前段无延迟即为1)，P为步骤S123所得置信度。以"zhong"为例：L＝5、C＝1、P＝0.9893，得分为：Score＝0.4×(5 / 6)+0.3×1+0.3×0.9893≈0.966；保留所有得分≥0.7的段落组合方案作为可用边界，并以得分最高者构建分节规则表，例如：[{"segment":"ni","start":0,"end":1,"score":0.87},{"segment":"hao","start":2,"end":4,"score":0.91},{"segment":"zhong","start":5,"end":9,"score":0.966},{"segment":"guo","start":10,"end":12,"score":0.924}]。

[0114] 步骤S125：对拼音切分边界序列进行拼音单元纠错扩展，从而构建拼音动作项集合；

[0115] 本实施例中，加载预设拼音纠错映射表，其中包含三类规则：声母替换(如)、韵母归并及音调误差容差(±1级音调调整)。在此基础上，系统对分节拼音段执行结构扩展，例如"zhong”生成纠错候选：["zhong","zong","zheng","zhuang"]。对片段流前后±2个拼音段构建上下文窗口，提取词首词尾拼音对的联合概率，如：{"hao→zhong":0.87,"zhong→guo":0.92}，将上下文置信加权与结构误差指数进行组合评价，设定重构保留阈值为0.6，仅保留："zhong"→{"zong":0.61,"zheng":0.72,"zhong":0.98}，最终保留置信度最高的"zhong”作为拼音动作项。

[0116] 步骤S126：将拼音动作项集合与拼音片段流中对应的起止位置、纠错状态与拼音类型进行对齐，生成拼音动作序列矩阵。

[0117] 本实施例中，将纠错后的拼音动作项与原始拼音片段流进行对应，按起止位置标记拼音段范围，同时标注纠错状态(如是否为原始匹配、是否为结构调整结果)，并对拼音类型(全拼、缩写、混合拼)进行分类。在上述案例中：[{"pinyin":"ni","start":0,"end":1,"corrected":false,"type":"full"},{"pinyin":"hao","start":2,"end":4,"corrected":false,"type":"full"},{"pinyin":"zhong","start":5,"end":9,"corrected":true,"type":"full"},{"pinyin":"guo","start":10,"end":12,"corrected":fals e,"type":"full"}]，将以上结构组织为拼音动作序列矩阵每行包含拼音内容、起止索引、纠错标志及拼音类型，用于驱动后续拼音-汉字映射及语义联动。

[0118] 可选地，步骤S125包括：

[0119] 提取拼音切分边界序列中的拼音子串，构建基础拼音单元列表；

[0120] 本实施例中，从已获得的拼音切分边界序列中，逐一提取所有拼音子串，形成基础拼音单元列表。例如，对输入序列["ni","hao","zhong","guo"]，基础拼音单元列表即为该数组。列表结构采用二维字符数组表示：拼音单元列表P_base＝[["ni"],["hao"],["zhong"],["guo"]]，其中每个元素对应一个拼音子串，列表长度即拼音单元数量，索引从0开始，便于后续定位。

[0121] 加载平台内置的拼音纠错映射表，包括声母替换规则、韵母归并规则及音调容差表；

[0122] 本实施例中，加载内置拼音纠错映射表。该映射表包括三大类规则，分别为声母替换、韵母归并和音调容差表。表结构以字典形式存储，如：声母替换规则S_rule＝{"n":["l"],"zh":["z","j"],...}；韵母归并规则Y_rule＝{"eng":["en"],"in":["ing"],...}；音调容差表T_rule＝{1:[1,2],2:[1,2,3],...}，其中声母替换规则覆盖常见混淆对，韵母归并规则兼顾发音相近归类，音调容差表允许音调上下浮动范围设定为±1级。

[0123] 根据拼音纠错映射表对基础拼音单元列表中各拼音子串进行声母-韵母错误匹配扩展，得到拼音纠错候选集；

[0124] 本实施例中，对基础拼音单元列表中每个拼音子串执行声母和韵母的错误匹配扩展。例如对拼音"zhong”，先分解为声母"zh”和韵母"ong”，分别查询S_rule和Y_rule，得到扩展集合：扩展声母集S_expand("zhong")＝["zh","z","j"]；扩展韵母集Y_expand("zhong")＝["ong","eng"]；拼接生成扩展拼音候选项集合：拼音纠错候选集C_candidat es("zhong")＝["zhong","zheng","zong","jeng"]，该扩展操作遵循严格长度匹配原则，确保新候选拼音长度与原拼音长度差异不超过1。

[0125] 提取拼音片段流中左右上下文区间±2个拼音单元的词首词尾拼音搭配概率，结合拼音纠错候选集中每个候选项的自身结构误差计算输入上下文误判概率，并根据输入上下午误判概率计算置信重构值，保留置信重构值≥0.6的候选项；

[0126] 本实施例中，提取拼音片段流中每个拼音单元在上下文区间±2范围内的词首词尾拼音搭配概率。该概率由预先统计的用户输入大数据词频库提供，库内存储格式如下：上下文搭配概率矩阵m为拼音单元数量。M_context[i,j]＝词首拼音i到词尾拼音j的搭配概率，取值范围[0,1]；例如，针对"hao”与"zhong”间的搭配，概率为0.87，若概率低于0.1则视为不常用搭配，影响候选置信度。结合拼音纠错候选集内各项自身的结构误差值(取值范围[0,1]，0表示完全匹配，0.3为最大容错阈值)，计算每个候选拼音的输入上下文误判概率。输入上下文误判概率计算公式内隐包含权重参数：输入上下文误判概率＝0.7×(1-词首词尾搭配概率)+0.3×结构误差值；置信重构值为该概率的补数，定义为：置信重构值＝1-输入上下文误判概率；例如，"zhong”候选"zheng”对应词首词尾概率为0.4，结构误差值0.2，则输入上下文误判概率为：＝0.7×(1-0.4)+0.3×0.2＝0.7×0.6+0.06＝0.42+0.06＝0.48，置信重构值＝1-0.48＝0.52<0.6，故该候选项被剔除。最后，保留置信重构值≥0.6的候选项作为有效纠错选项集合。

[0127] 将各拼音子串与相对应的拼音纠错候选集中置信重构值最高的候选项进行匹配，得到标准化拼音动作项集合。

[0128] 本实施例中，在标准化阶段，为每个拼音子串匹配其对应的候选集中置信重构值最高的项。该映射存储于如下矩阵结构：标准化拼音动作项集合每行包含：[原拼音子串,纠错后拼音,置信重构值]。例如：[["ni","ni",0.98],["hao","hao",0.95],["zhong","zhong",0.97],["guo","guo",0.99]]。该集合作为后续拼音动作序列构建的核心输入，保障输入拼音标准化一致性，有效提升了系统对拼音误输入的纠正能力与用户体验。

[0129] 可选地，步骤S2中汉字结构路径重建包括：

[0130] 利用字词结构索引表对拼音-汉字映射关系中每个汉字进行结构路径查询与定位，得到结构记录序列；

[0131] 本实施例中，利用预设的字词结构索引表对拼音-汉字映射关系中的每个汉字进行结构路径查询与定位。该索引表采用哈希映射结构存储，键为汉字Unicode编码，值为对应的结构路径记录。结构路径记录以二维矩阵形式表示，例如，汉字"林”(Unicode：U+6797)的结构路径记录序列如下：结构记录序列L_林＝[[主部件,子部件1,子部件2],["木”,"木”,null]]。其中，主部件和子部件均以汉字部件编码表示，null表示无下级部件。查询时，系统对输入的每个汉字检索该表，得到对应的结构路径矩阵，矩阵维度为n×3，n为部件层级数。

[0132] 基于结构记录序列进行结构层级展开，从而构建汉字结构拓扑图序列；

[0133] 本实施例中，以邻接矩阵形式构造拓扑图，矩阵维度m×m，m为部件总数。邻接矩阵元素为0或1，表示部件间的连接关系，例如：A_桥＝[[0,1,1],[0,0,0],[0,0,0]]，表示"木”指向"乔”及无其他子部件。矩阵的行和列分别对应汉字"桥”的部件节点，假设第1个节点是"木”，第2个是"乔”，第3个是另一个部件(比如"丿”或无具体部件，用null表示)。矩阵中的元素A_{ij}表示节点i是否指向节点j：如果A_{ij}＝1，表示从部件i指向部件j，即部件j是部件i的子部件或依附部件。如果A_{ij}＝0，表示部件i与部件j之间没有直接的层级连接关系。第一行[0,1,1]，具体含义是：A1,1＝0：部件"木”不指向自己(没有自环)；A1,2＝1：部件"木”直接连接指向"乔”，说明"乔”是"木”的子部件或者依赖部件；A1,3＝1：部件"木”也直接连接指向第三个部件，表明"木”与该部件也存在层级或依附关系。这样一来，"木”作为主部件，连接了"乔”和另一个部件，因此第一行是[0,1,1]，表示"木”指向这两个子部件。第二行和第三行全为0，表示"乔”和第三部件没有再往下的子部件。此拓扑图序列覆盖所有汉字的部件结构层级，体现部件间的父子关系，构成输入字词的结构骨架。

[0134] 将汉字结构拓扑图序列中各图从主部件向次部件的遍历路径进行线性编码，得到结构路径编码序列；

[0135] 本实施例中，采用深度优先遍历策略，按访问顺序将部件节点编码为字符串路径。编码格式如"M1->M2”，其中M1、M2为部件编号或简写。示例如下：编码序列＝["木->乔","乔->丿"]，编码序列以数组形式保存，长度取决于部件层级和数量，便于后续快速比较和匹配。

[0136] 对结构路径编码序列进行语义压缩，形成结构路径片段集合；

[0137] 本实施例中，设定相似度阈值α＝0.7，利用编辑距离计算编码序列之间的相似性，若相似度大于该阈值，则将两条路径合并为单一片段。压缩后的路径片段集合去除冗余，显著降低存储空间需求。示例压缩结果：片段集合＝["木->乔"]，结构路径片段以字符串数组形式存储，支持快速索引。

[0138] 将结构路径片段集合与拼音对齐输入帧中的用户输入行为ID绑定，生成用户输入字词结构路径集。

[0139] 本实施例中，用户输入行为ID采用32位整型唯一标识，建立关联矩阵M_bind(维度k×2)，第一列为行为ID，第二列为结构路径片段索引，实现一对多映射关系。例如：关联矩阵M_bind＝[[行为ID_1,片段索引_1],[行为ID_2,片段索引_1],...]；该矩阵用于快速定位用户输入行为对应的结构路径片段，构成完整的用户输入字词结构路径集，便于后续内容匹配和学习轨迹分析。

[0140] 可选地，步骤S3中拆分内容片段结构单元包括：

[0141] 根据平台用户交互数据中的用户输入行为ID和课程编号，通过在线汉语言学习平台获取对应课程内容集合，并加载对应课程内容集合中课程文本单元，从而得到学习内容数据；

[0142] 本实施例中，依据平台用户交互数据中的用户输入行为ID和课程编号，通过在线汉语言学习平台访问课程管理系统，调用课程内容数据库接口，获取对应课程编号下的完整课程内容集合。课程内容集合包括多层次文本资源，例如章节标题、自然段落、词汇表及示例句等结构化文本单元，所有文本单元均存储为UTF-8编码的字符串，统一格式方便后续解析。系统基于课程内容集合中的自然段落进行后续处理，保证文本完整性和上下文关联，加载得到学习内容数据。

[0143] 对学习内容数据中的自然段落进行语义切分，构建语义片段单元；

[0144] 本实施例中，针对学习内容数据中每个自然段落，通过语言理解模型内置的文本分割框架，将自然段落切分成语义连贯且主题一致的片段单元。该切分依据句法标点符号和句子主谓宾结构，同时结合上下文词向量相似度阈值0.65进行分割，确保语义片段长度保持在20至80个汉字范围内，避免片段过短导致语义碎片化或过长影响处理效率，最终构建出完整的语义片段单元集合。

[0145] 根据语义片段单元执行字词标注与结构特征标记，得到结构化标注结果；

[0146] 本实施例中，对构建的语义片段单元，执行逐字词级的标注与结构特征标记，应用预训练的汉语词性标注模型及词汇依存关系标注框架，获得包括词性类别(名词、动词、形容词等)、词组边界及层级依存关系矩阵(以邻接矩阵形式存储，形如A＝\begin{bmatrix}0&1\\0&0\end{bmatrix})的结构化标注结果。该矩阵用于指示词语之间的依存连接关系，便于后续语义分析和内容理解。

[0147] 对结构化标注结果进行语义紧密度分析，将语义紧密度≥0.75的语义片段标记为知识焦点，得到知识焦点集合；

[0148] 本实施例中，针对上述结构化标注结果，基于词向量加权平均方法计算每个语义片段的内在关联强度，设定阈值为0.75，筛选出语义紧密度高于该阈值的语义片段，标记为知识焦点。知识焦点集合通过节点列表形式存储，每个节点包含片段标识符、起止位置及紧密度得分，用于精确定位重点内容。

[0149] 根据知识焦点集合将语义片段单元进行内容片段单元划分，并对各内容片段单元进行时间戳重定位，得到内容片段结构单元。

[0150] 本实施例中，根据知识焦点集合，对原始语义片段单元进行重新划分，合并相邻且均属知识焦点的片段形成内容片段单元，同时将片段间时间戳基于用户学习行为日志进行校正，精确重定位每个内容片段的起止时间，时间戳精度达到毫秒级，确保内容片段结构单元在时间轴上的连续性和准确性。最终内容片段结构单元以包含文本内容、结构标识及时间戳的多字段数据结构保存，便于后续多模态嵌入与分析使用。

[0151] 可选地，步骤S3中多模态内容节点嵌入包括：

[0152] 将用户输入字词结构路径集进行结构拓扑向量化编码，得到结构向量编码集；

[0153] 本实施例中，用户输入字词结构路径集经过结构拓扑向量化编码时，首先基于每个汉字的部件层级关系，利用预设的部件编码字典将部件类别转换为向量形式。部件编码字典是一种预先构建的结构化数据表，将汉字的基本部件映射为唯一的数字编码，以便于计算机进行结构化分析和特征提取。该部件编码字典基于汉字部首表和标准汉字字典，通过形态学分析和图像特征提取相结合的方法，为每个部件生成统一的特征向量编码，包含形状、笔画和结构位置信息，确保编码的准确性和一致性。部件"口”：编码为[0.12,0.45,0.78,0.33,0.09,…](长度固定为64的浮点数向量)，表示其典型的方形轮廓、四笔画结构及在字中的常见位置特征。部件"氵”(三点水)：编码为[0.05,0.67,0.23,0.55,0.11,…]，体现其波浪形状特征、三点笔画数及偏旁位置属性。部件"亻”(单人旁)：编码为[0.22,0.31,0.48,0.39,0.15,…]，反映其竖笔形态、两笔结构和在字中的左侧偏旁位置。该字典支持动态更新与版本管理，为汉字结构解析和向量化表示提供了基础数据保障。每个部件编码为长度为128的定长特征序列，序列中的每个元素均为0-1之间的小数，反映部件激活权重。编码过程中，采用最大部件层级深度为5层，编码过程中部件权重衰减系数设为0.85，保证深层次部件贡献逐渐降低，避免冗余信息。编码结果保存为结构向量编码集，按字词顺序依次排列，形成列表形式，便于后续检索和匹配。

[0154] 根据内容片段结构单元提取对应课程内容集合中的文本词义嵌入、配套图文中汉字部件图形纹理特征、片段配音音频中的音素序列特征及文本语法标签分布特征，从而构建多模态片段特征集；

[0155] 本实施例中，对于内容片段结构单元，分别从文本、图形、音频及语法四个维度提取特征。文本维度采用预训练语言模型输出的词义嵌入，词向量维度为768维；图形维度通过图像处理提取汉字部件的纹理特征，采用64维向量表示；音频维度使用MFCC提取音素特征，长度固定为50维；语法维度统计文本中常见语法标签的分布，编码成30维多热向量。所有特征合并后构成多模态片段特征集，以列表形式存储，方便快速索引。每个特征维度均在0-1范围内，且提取过程中的归一化系数为0.99，防止极端值影响匹配。

[0156] 对结构向量编码集及多模态片段特征集执行双向匹配，得到关联匹配矩阵；

[0157] 本实施例中，对结构向量编码集与多模态片段特征集进行双向匹配时，采用相似度评分机制，具体为将两个向量组分别遍历，每对向量根据内积结果计算相似度得分，得分范围限定在0至1之间。匹配结果以"字词编号-内容片段编号-相似度得分”的三元组形式存储，例如(字词3，片段2，0.85)。设定阈值为0.7，只有得分高于该阈值的匹配关系才被保存，以保证匹配的精准度和实用性。

[0158] 从关联匹配矩阵中筛选匹配度≥0.7的匹配对映对，构建内容节点匹配集合；

[0159] 本实施例中，根据筛选出的匹配对，构建内容节点匹配集合。该集合以哈希表方式实现，键为字词结构路径编号，值为对应内容片段编号列表及其匹配得分。此结构便于快速查询字词对应的相关内容片段，实现内容与输入行为的高效关联。匹配对的数目根据具体输入量变化，一般单次会话匹配对数量控制在100以内，确保处理效率。

[0160] 根据内容节点匹配集合进行内容片段归组聚合，生成内容联动片段簇。

[0161] 本实施例中，根据内容节点匹配集合，执行内容片段归组聚合。将匹配度较高且时间戳接近的内容片段组合成簇，每个簇包含不超过10个片段，保证聚合的语义紧密性和计算效率。聚合规则基于时间差阈值0.5秒以内视为同簇，且同簇内片段的匹配得分均不低于0.7，避免无关内容混入。生成的内容联动片段簇以列表形式保存，包含簇内片段标识及其对应的匹配强度信息，便于后续拓扑轨迹构建和分析。

[0162] 可选地，步骤S4包括：

[0163] 步骤S41：提取内容联动片段簇中每个簇内的片段列表，并依据每个内容片段单元的触发时间对片段列表进行时间排序，得到初始片段轨迹序列；

[0164] 本实施例中，从内容联动片段簇里取出每个簇包含的所有学习内容片段，这些片段有各自的触发时间。接着，按照时间先后顺序对这些片段进行排列，确保学习轨迹顺序清晰，方便后续分析。时间戳格式采用标准的毫秒数表示，比如"1623456789123”，保证时间排序精准无误。通过这一步，可以得到用户在一次学习会话中，内容访问的时间序列，反映出学习的先后顺序。

[0165] 步骤S42：对初始片段轨迹序列进行语义相邻性分析，生成片段转移关系图；

[0166] 本实施例中，根据刚排好序的内容片段，分析每个片段与相邻片段之间的语义相似性。具体来说，会用已有的语义模型，将每个片段的文本转换成向量表示，再计算相邻片段向量的相似程度，数值范围0～1。设置一个阈值，比如0.6，超过这个阈值则认为两片段语义相关。还会结合课程知识点之间的关联度，综合判断片段间的转移关系，最终生成一个带权重的有向图，图中节点代表内容片段，边代表片段间的语义关联。

[0167] 步骤S43：计算片段转移关系图中的边集合的转移路径权重，并对转移路径权重<0.45的边进行剪枝，得到优化学习子轨迹集；

[0168] 本实施例中，对转移关系图中的每条边的权重进行定量计算，权重来源为语义相似度(权重0.6)和知识点关联度(0.4)的加权平均，权重阈值设定为0.45。低于该阈值的边被视为弱连接，予以剪除，以降低噪声影响和复杂度，从而得到简化且更具代表性的学习路径子集。此举有助于提高后续图结构聚合的准确性和计算效率。

[0169] 步骤S44：将优化学习子轨迹集进行有向图结构聚合，得到单次会话学习拓扑轨迹图；

[0170] 本实施例中，采用有向图聚合方法，将剪枝后的学习子轨迹集进行节点合并与路径整合，构建单次会话的学习拓扑轨迹图。图中节点仍为内容片段，边代表有效学习路径关系，整体反映用户在该会话中知识点访问及学习内容的拓扑结构。图结构的构建支持后续路径排序和轨迹分析，节点数一般控制在100以内以保持图的可读性与处理效率。

[0171] 步骤S45：对单次会话学习拓扑轨迹图进行双向拓扑排序，得到学习轨迹拓扑图。

[0172] 本实施例中，针对构建的学习拓扑轨迹图，分别执行正向和反向的节点序列扫描，形成双向拓扑排序。正向扫描依赖于节点入度为零的起点，反向扫描则以出度为零的终点开始，双向排序序列覆盖了所有可能的学习路径。最终，合并两方向排序结果，剔除冗余路径，形成完整且连贯的学习轨迹拓扑图，节点排序反映内容学习的时间及逻辑顺序，支持后续状态聚合和行为分析。

[0173] 尤其重要的是，步骤S45包括：

[0174] 步骤S451：对单次会话学习拓扑轨迹图进行前向拓扑扫描，生成正向片段路径集合；

[0175] 本实施例中，针对单次会话学习拓扑轨迹图从起始节点开始，依次遍历所有可达后续片段，形成正向片段路径集合。每条路径以内容片段的时间序列为基础，路径长度限定在不超过20个片段，确保路径完整同时避免计算资源过度消耗，形成结构化的有序路径列表。路径节点信息包括片段ID、触发时间及对应的语义标签，路径集合以列表形式存储，便于后续处理。

[0176] 步骤S452：对单次会话学习拓扑轨迹图进行逆向拓扑回溯，生成反向片段路径集合；

[0177] 本实施例中，从单次会话学习拓扑轨迹图中终止节点逆向遍历拓扑轨迹图，沿着可达的前驱节点依次回溯，生成反向片段路径集合。该集合与正向路径集合对应，最长路径同样限定为20个片段，确保前后路径长度一致性。路径数据结构同样包含片段ID、时间戳与语义标签，形成反向有序列表，便于交叉比对。

[0178] 步骤S453：根据正向片段路径集合及反向片段路径集合进行路径交并比对，筛选路径节点相交且重合比例≥0.6的路径对，得到一致路径子集；

[0179] 本实施例中，对正向片段路径集合及反向片段路径集合进行比对，针对每对路径计算节点集合的交集比例。若交集节点占比达到或超过60％(即相交比例≥0.6)，则判定该对路径为一致路径。该判断以节点ID集合形式执行，保证对比的准确性与效率。筛选出的路径对组成一致路径子集，作为下一步处理的基础。

[0180] 步骤S454：将一致路径子集进行顺序融合，消除冗余分支，保留路径主干，从而构建轨迹主路径链；

[0181] 本实施例中，将一致路径子集内的路径按时间序列顺序进行合并，消除重复节点和分支，保留路径中最具代表性的主干部分。合并过程中，依据节点的触发时间和语义标签保持路径逻辑连续性和语义一致性。最终形成轨迹主路径链，该链以序列结构保存，代表用户学习过程中的主要认知轨迹。

[0182] 步骤S455：将正向片段路径集合及反向片段路径集合中非属于轨迹主路径链的路径作为辅助路径，并将辅助路径的语义特征映射至轨迹主路径链中的相交节点中，得到学习轨迹拓扑图。

[0183] 本实施例中，将正向与反向路径集合中未包含于轨迹主路径链的路径识别为辅助路径，这些路径提供补充的学习轨迹信息。辅助路径中的语义特征被映射至主路径链中相交节点，丰富主路径的语义表达。映射操作通过节点ID匹配实现，确保语义特征准确对应，生成的结果构成完整的学习轨迹拓扑图，反映用户多层次的学习行为动态。

[0184] 可选地，步骤S5包括：

[0185] 步骤S51：对学习轨迹拓扑图进行语义阶段划分，生成语义阶段片段组；

[0186] 本实施例中，针对学习轨迹拓扑图中记录的片段，依据其语义特征进行阶段划分，形成若干语义阶段片段组。划分过程主要通过分析片段内关键词共现频率与主题一致性，设定相似度阈值为0.7，确保片段组内内容语义连贯。各阶段片段按时间顺序排列，且其对应的时间区间长度一般控制在1分钟以内，形成结构化的阶段列表，包含阶段标识、时间起止、主要语义标签等信息，便于后续处理和索引。

[0187] 步骤S52：依据语义阶段片段组中的起止时间间隔与结构跳跃度进行阶段聚合，构建学习状态演化链；

[0188] 本实施例中，通过对上述语义阶段片段组的时间起止点和结构跳跃度进行聚合，构建学习状态演化链。时间起止间隔阈值设为不超过30秒，结构跳跃度作为衡量连续两个阶段间内容差异的指标，其取值范围0～1，限定阈值不超过0.3以保证阶段间连贯。该演化链以有序时间段列表形式体现，反映学习过程中文本语义与结构的连续发展轨迹，为后续压缩和索引提供基础数据支持。

[0189] 步骤S53：基于学习状态演化链，提取主成分特征字段，从而构建汉语言状态压缩帧；

[0190] 本实施例中，基于学习状态演化链，对每个阶段片段组的多维特征数据进行主成分特征提取，目的是从高维语义及结构信息中提炼出最具代表性的核心指标。首先，将阶段组中所有包含的主题密度、结构复杂度、学习时长等原始特征整理成特征矩阵，矩阵行表示各个阶段片段，列对应各项指标。然后，针对该矩阵应用主成分选择机制，通过分析各特征的方差贡献率和相关性，筛选出方差贡献率累计超过85％的特征子集，确保压缩后特征集既保留大部分信息，又大幅减少冗余。所选特征主要包括主题密度(关键词出现频率比例)、结构复杂度(分层嵌套层数)、学习时长段(秒级时间长度)等。接着，对筛选出的特征进行加权融合，结合用户行为偏好权重(例如主题密度权重0.5，结构复杂度权重0.3，学习时长权重0.2)，计算出每个阶段的综合主成分特征值。最后，将各阶段的主成分特征字段以结构化数据形式保存，形成汉语言状态压缩帧，为后续多维索引建立和行为模式识别提供精准、简洁的特征基础。

[0191] 步骤S54：提取汉语言状态压缩帧中的主题密度、结构复杂度与学习时长段，将主题密度、结构复杂度与学习时长段转换为多主键组合，从而生成主键索引集合；

[0192] 本实施例中，对压缩帧中的主题密度、结构复杂度及学习时长段三项指标进行组合，形成多主键索引集合。索引格式采用三元组表示，例如(0.8,3,150)，对应主题密度0.8、结构复杂度3和学习时长150秒。所有索引均存储于多维索引表，方便基于多属性的快速查询和匹配，提升系统检索效率。

[0193] 步骤S55：将主键索引集合中具有同类主题密度与结构复杂度的主键聚合成行为模式，建立触发映射表；

[0194] 本实施例中，依据多主键索引集合，将具有相近主题密度(允许±0.05误差)和结构复杂度(允许±1误差)的索引项聚合，形成行为模式组。每组行为模式表示用户在特定语义和结构环境下的典型学习轨迹，如语法细节重点学习或词汇反复练习。基于这些行为模式，建立触发映射表，将模式映射到具体的学习触发事件，如学习提醒或重点推送，映射表采用键值对结构，支持实时响应。

[0195] 步骤S56：将主键索引集合与触发映射表进行归并打包，得到动态索引分布块，并将动态索引分布块传输至在线汉语言学习平台，以部署机制编排任务。

[0196] 本实施例中，将多主键索引集合与触发映射表进行归并打包，构成动态索引分布块。该分布块采用JSON格式封装，包含完整的索引数据和触发规则，便于在平台间传输和解析。动态索引分布块实时传输至在线汉语言学习平台，支持根据行为模式动态调度学习资源和任务，实现个性化教学管理，有效提升学习效率与体验。

[0197] 尤其重要的是，步骤S55具体为：

[0198] 提取主键索引集合中每个主键中包括主题密度、结构复杂度与学习时长段的三维属性，并基于主题密度对三维属性执行密度聚类，得到主题密度簇；

[0199] 本实施例中，从主键索引集合中抽取包含主题密度、结构复杂度与学习时长段的三维属性数据。针对主题密度维度，采用基于密度的空间聚类方法对样本点进行分类。具体地，设置邻域半径ε为0.05，最小邻域点数MinPts为6，确保聚类能够识别紧密的主题密度高聚集区。通过遍历样本数据，依次判定每个数据点是否属于核心点、边界点或噪声点，形成密度可达的点簇集合。聚类过程中，自动识别具有显著主题密度集中趋势的簇，剔除主题密度分布稀疏的异常数据。聚类完成后，基于主题密度簇的划分结果，将对应的三维属性分组归类，为后续结合结构复杂度及学习时长进行更细粒度的行为模式提取打下基础。此步骤有效地过滤了主题密度维度的噪声影响，提高整体聚合的准确性和稳定性。

[0200] 对主题密度簇中的索引项，按照结构复杂度进行K均值聚合，得到结构复杂度子簇；

[0201] 本实施例中，针对每个主题密度簇，继续根据结构复杂度属性划分子集。结构复杂度通过分析汉字部件层级数获得，数值范围为1至5，其中1代表结构简单(如单一偏旁)，5代表结构复杂(多层嵌套部件)。系统在主题密度簇内部，依据结构复杂度的数值区间(简单1-2级、中等3级、复杂4-5级)对主键进一步分组。该分组过程保证每个子集中的内容在结构层级上相对一致，有助于识别不同复杂度对学习行为的影响，实现更精准的行为模式划分。

[0202] 对结构复杂度子簇执行用户行为路径聚合，形成行为轨迹模板；

[0203] 本实施例中，在结构复杂度分组的基础上，利用用户学习行为轨迹数据，对同一子组内的学习路径进行聚合。学习轨迹数据以时间序列形式存储，包含内容访问顺序及对应停留时间，例如[(内容A,100秒),(内容B,200秒)]。系统采用路径相似度度量(设定阈值为0.8)对行为路径进行比对，识别相似轨迹，将相似行为轨迹归并为典型行为模板。该模板代表该类内容条件下用户的主要学习模式，便于后续的行为分析和个性化推荐设计。

[0204] 基于行为轨迹模板中的轨迹模版与统计指标，定义行为触发规则；

[0205] 本实施例中，结合行为模板及统计分析结果，定义行为触发规则。触发规则包括匹配度阈值(≥0.85)、访问频率限制和停留时间阈值(如300秒以上视为深度学习)。系统根据历史数据反馈动态调整阈值，确保规则既不过于宽松导致误触，也不过于严格导致漏判。该规则用于识别用户何时进入重点学习阶段，触发对应的辅助教学或资源推荐，提升学习效率和体验。

[0206] 将行为轨迹模板中的用户行为模式与对应行为触发规则进行映射，构建触发映射表。

[0207] 本实施例中，将行为模板与触发规则结构化整合，构建触发映射表。映射表采用JSON或表格形式存储，内容包含行为模式ID、轨迹模板内容ID序列、匹配度和触发条件参数等信息。例如：{

[0208] "模式ID”:"Pattern_001”,"轨迹内容”:["课文1段落3”,"课文1段落4”],"触发条件”:{"匹配度”:0.85,"停留时间”:300},"触发动作”:"推送练习题”}，该表支持实时查询与触发机制调用，保障平台智能反馈系统的高效运作，实现个性化动态学习指导。

[0209] 可选地，本说明书还提供一种在线汉语言学习平台的数据信息储存系统，用于执行如上所述的在线汉语言学习平台的数据信息储存方法，该在线汉语言学习平台的数据信息储存系统包括：

[0210] 汉字结构对齐模块，用于获取平台用户交互数据，并根据平台用户交互数据进行拼音输入轨迹解析与汉字部件结构编码，得到拼音对齐输入帧；

[0211] 映射关系分析模块，用于提取拼音对齐输入帧中的拼音-汉字映射关系，利用预设的字词结构索引表对拼音-汉字映射关系进行汉字结构路径重建，得到用户输入字词结构路径集；

[0212] 内容片段嵌入模块，用于通过在线汉语言学习平台调取与平台用户交互数据所关联的学习内容数据，并根据学习内容数据拆分内容片段结构单元；将用户输入字词结构路径集与内容片段结构单元进行多模态内容节点嵌入，得到内容联动片段簇；

[0213] 学习轨迹分析模块，用于根据内容联动片段簇构建单次会话学习拓扑轨迹，并对单次会话学习拓扑轨迹进行双向拓扑排序，得到学习轨迹拓扑图；

[0214] 学习状态聚合模块，用于对学习轨迹拓扑图进行学习状态分段聚合，生成汉语言状态压缩帧；根据汉语言状态压缩帧建立多维主键索引及行为触发机制，得到动态索引分布块，并将动态索引分布块传输至在线汉语言学习平台，以部署机制编排任务。

[0215] 因此，无论从哪一点来看，均应将实施例看作是示范性的，而且是非限制性的，本发明的范围由所附权利要求而不是上述说明限定，因此旨在将落在申请文件的等同要件的含义和范围内的所有变化涵括在本发明内。

[0216] 以上所述仅是本发明的具体实施方式，使本领域技术人员能够理解或实现本发明。对这些实施例的多种修改对本领域的技术人员来说将是显而易见的，本文中所定义的一般原理可以在不脱离本发明的精神或范围的情况下，在其它实施例中实现。因此，本发明将不会被限制于本文所示的这些实施例，而是要符合与本文所发明的原理和新颖特点相一致的最宽的范围。

Claims

1. A method for storing data information in an online Chinese language learning platform, characterized in that, Includes the following steps: Step S1: Obtain platform user interaction data, and perform pinyin input trajectory analysis and Chinese character component structure encoding based on the platform user interaction data to obtain pinyin aligned input frames; Step S2: Extract the pinyin-Chinese character mapping relationship in the pinyin-aligned input frame, and use the preset word structure index table to reconstruct the Chinese character structure path of the pinyin-Chinese character mapping relationship to obtain the user input word structure path set; Step S3: Retrieve learning content data associated with user interaction data from the online Chinese language learning platform, and decompose content fragment structure units based on the learning content data; embed the user input word structure path set and content fragment structure units into multimodal content nodes to obtain content linkage fragment clusters; Step S4: Construct a single-session learning topology trajectory based on the content linkage fragment clusters, and perform bidirectional topology sorting on the single-session learning topology trajectory to obtain the learning trajectory topology map; Step S5: Perform learning state segmentation and aggregation on the learning trajectory topology map to generate Chinese language state compressed frames; A multidimensional primary key index and behavior triggering mechanism are established based on the Chinese language state compression frame to obtain a dynamic index distribution block. The dynamic index distribution block is then transmitted to the online Chinese language learning platform to deploy the mechanism and orchestrate tasks.

2. The data information storage method for the online Chinese language learning platform according to claim 1, characterized in that, Step S1 includes: Step S11: Obtain platform user interaction data and encode it with a unified timestamp to obtain the original input behavior record set; Step S12: Perform dynamic pinyin segmentation on the original input behavior record set and mark the pinyin segment boundaries to obtain the pinyin action sequence matrix; Step S13: Call the preset local dictionary index to perform pinyin-Chinese character mapping alignment on the pinyin action sequence to obtain the pinyin-Chinese character alignment table; Step S14: Encode the Chinese character component structure in the Pinyin-Chinese character alignment table to generate a component structure vector set; Step S15: Associate and fuse the Pinyin action sequence matrix and the component structure vector set according to the alignment relationship in the Pinyin-Chinese character alignment table to generate a Pinyin aligned input frame.

3. The data information storage method for the online Chinese language learning platform according to claim 2, characterized in that, Step S12 includes: Step S121: Extract the character input sequence from the original input behavior record set and construct the pinyin fragment stream; Step S122: According to the preset initial consonant-final vowel-tone triplet matching logic, the pinyin segment stream is segmented into pinyin substrings to obtain a set of candidate pinyin segments; Step S123: Perform pinyin distribution statistics on the candidate pinyin segment set, and calculate the syllable confidence value based on the pinyin distribution statistics results; Step S124: Based on the pinyin length, input continuity and corresponding syllable confidence value in the candidate pinyin segment set, a weighted score is performed, the substring segmentation combination with the highest score is selected as the segmentation boundary rule, and the candidate pinyin segment set is segmented using the segmentation boundary rule to obtain the pinyin segmentation boundary sequence; Step S125: Perform pinyin unit error correction expansion on the pinyin segmentation boundary sequence to construct a set of pinyin action items; Step S126: Align the set of pinyin action items with the corresponding start and end positions, error correction status and pinyin type in the pinyin segment stream to generate a pinyin action sequence matrix.

4. The data information storage method for the online Chinese language learning platform according to claim 3, characterized in that, Step S125 includes: Extract pinyin substrings from the pinyin segmentation boundary sequence and construct a list of basic pinyin units; Load the platform's built-in Pinyin error correction mapping table, including initial consonant replacement rules, final vowel merging rules, and tone tolerance table; Based on the pinyin error correction mapping table, each pinyin substring in the basic pinyin unit list is expanded by initial consonant-final vowel error matching to obtain a pinyin error correction candidate set; Extract the word-initial and word-final pinyin collocation probabilities of the left and right context intervals ±2 pinyin units in the pinyin fragment stream, calculate the input context misjudgment probability by combining the self-structural error of each candidate option in the pinyin error correction candidate set, and calculate the confidence reconstruction value based on the input morning and afternoon misjudgment probabilities, and retain the candidate options with confidence reconstruction values ​​≥0.6; Each pinyin substring is matched with the candidate with the highest confidence reconstruction value in the corresponding pinyin error correction candidate set to obtain a standardized pinyin action item set.

5. The data information storage method for the online Chinese language learning platform according to claim 1, characterized in that, Step S2, the reconstruction of the Chinese character structure path, includes: The structural path query and location of each Chinese character in the pinyin-Chinese character mapping relationship are performed using the word structure index table to obtain the structural record sequence; The structural hierarchy is expanded based on the structural record sequence, thereby constructing a sequence of Chinese character structural topology graphs; Linearly encode the traversal path from the main component to the secondary component in each graph of the Chinese character structure topology graph sequence to obtain the structure path encoding sequence; Semantic compression is performed on the structure path encoded sequence to form a set of structure path segments; The set of structural path fragments is bound to the user input behavior ID in the pinyin-aligned input frame to generate a set of user input word structural paths.

6. The data information storage method for the online Chinese language learning platform according to claim 1, characterized in that, Step S3 involves splitting the content fragment structure unit, which includes: Based on the user input behavior ID and course number in the platform's user interaction data, the corresponding course content set is obtained through the online Chinese language learning platform, and the course text units in the corresponding course content set are loaded to obtain the learning content data; Semantic segmentation is performed on natural paragraphs in the learning content data to construct semantic fragment units; Based on semantic fragment units, word and structural feature annotations are performed to obtain structured annotation results; Semantic tightness analysis is performed on the structured annotation results, and semantic segments with semantic tightness ≥ 0.75 are marked as knowledge focuses to obtain a set of knowledge focuses; The semantic fragment units are divided into content fragment units based on the knowledge focus set, and the timestamps of each content fragment unit are relocated to obtain the content fragment structure unit.

7. The data information storage method for the online Chinese language learning platform according to claim 1, characterized in that, Step S3, the embedding of multimodal content nodes, includes: The user-input word structure path set is encoded using structural topology vectorization to obtain a structure vector encoding set; Based on the structural units of the content fragments, extract the text semantic embeddings, graphic texture features of Chinese character components in the accompanying graphics and text, phoneme sequence features in the audio dubbing of the fragments, and text grammatical tag distribution features from the corresponding course content set to construct a multimodal fragment feature set; Bidirectional matching is performed on the structure vector encoding set and the multimodal fragment feature set to obtain the association matching matrix; Filter matching pairs with a matching degree ≥ 0.7 from the association matching matrix to construct a content node matching set; Based on the content node matching set, content fragments are grouped and aggregated to generate content-linked fragment clusters.

8. The data information storage method for the online Chinese language learning platform according to claim 1, characterized in that, Step S4 includes: Step S41: Extract the list of segments within each cluster of content-linked segments, and sort the segment list by time according to the trigger time of each content segment unit to obtain the initial segment trajectory sequence; Step S42: Perform semantic adjacency analysis on the initial segment trajectory sequence to generate a segment transition relationship graph; Step S43: Calculate the transition path weights of the edge set in the fragment transition graph, and prune the edges with transition path weights < 0.45 to obtain the optimized learning sub-trajectory set; Step S44: Aggregate the optimized learning sub-trajectory set into a directed graph structure to obtain a single-session learning topology trajectory graph; Step S45: Perform bidirectional topological sorting on the single-session learning topology trajectory graph to obtain the learning trajectory topology graph.

9. The data information storage method for the online Chinese language learning platform according to claim 8, characterized in that, Step S5 includes: Step S51: Divide the learning trajectory topology graph into semantic stages and generate semantic stage segment groups; Step S52: Aggregate stages based on the start and end time intervals and structural jumps in the semantic stage fragment group to construct a learning state evolution chain; Step S53: Based on the learned state evolution chain, extract the principal component feature fields to construct a compressed frame of Chinese language state; Step S54: Extract the topic density, structural complexity, and learning duration segments from the Chinese language state compression frame, and convert the topic density, structural complexity, and learning duration segments into multiple primary key combinations to generate a primary key index set; Step S55: Aggregate primary keys with similar topic density and structural complexity in the primary key index set into behavior patterns and establish a trigger mapping table; Step S56: Merge and package the primary key index set and the trigger mapping table to obtain a dynamic index distribution block, and transmit the dynamic index distribution block to the online Chinese language learning platform to deploy the orchestration task.

10. A data information storage system for an online Chinese language learning platform, characterized in that, For executing the data information storage method of the online Chinese language learning platform as described in claim 1, the data information storage system of the online Chinese language learning platform includes: The Chinese character structure alignment module is used to acquire platform user interaction data, and to perform pinyin input trajectory analysis and Chinese character component structure encoding based on the platform user interaction data to obtain pinyin aligned input frames; The mapping relationship analysis module is used to extract the pinyin-Chinese character mapping relationship in the pinyin-aligned input frame, and to reconstruct the Chinese character structure path of the pinyin-Chinese character mapping relationship using a preset word structure index table to obtain the user input word structure path set; The content fragment embedding module is used to retrieve learning content data associated with user interaction data from the online Chinese language learning platform, and to split the content fragment structure units according to the learning content data; the user input word structure path set and the content fragment structure units are embedded with multimodal content nodes to obtain content linkage fragment clusters; The learning trajectory analysis module is used to construct a single-session learning topology trajectory based on content-linked fragment clusters, and to perform bidirectional topological sorting on the single-session learning topology trajectory to obtain a learning trajectory topology map. The learning state aggregation module is used to aggregate the learning state segments of the learning trajectory topology map and generate Chinese language state compressed frames. Based on the Chinese language state compressed frames, a multi-dimensional primary key index and behavior triggering mechanism are established to obtain dynamic index distribution blocks. The dynamic index distribution blocks are then transmitted to the online Chinese language learning platform to deploy the mechanism and orchestrate tasks.

Images