A keyword extraction-based document data recognition and classification method and system
By integrating a keyword extraction algorithm with vocabulary co-occurrence graph structure and semantic similarity information, and combining it with a custom hierarchical classification system, the problem of flexibility and accuracy in document classification in existing technologies is solved, achieving efficient and intelligent document management and retrieval.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BEIJING DEJUN TECH SERVICE CO LTD
- Filing Date
- 2026-02-25
- Publication Date
- 2026-06-02
Smart Images
Figure CN122132564A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of information processing technology, specifically to a document data identification and classification method and system based on keyword extraction. Background Technology
[0002] In today's information age, the number of electronic documents accumulated by various organizations and individuals has surged, with diverse formats and complex content. To achieve efficient management, accurate retrieval, and orderly organization of these massive amounts of documents, automated document classification technology has become crucial.
[0003] Existing automated classification methods are mainly divided into two categories. One category is based on predefined rules and keyword dictionaries. This method is difficult to adapt to new words, new fields, and dynamically changing terms, and has poor flexibility and maintainability. The other category is based on traditional machine learning models. These methods usually rely on statistical features such as bag-of-words models or TF-IDF. The feature vectors they generate are often high-dimensional and highly sparsity, failing to effectively distinguish between core keywords and general words in documents. This makes the classification model susceptible to noise interference, and its accuracy and generalization ability are limited.
[0004] Furthermore, existing technologies generally fail to effectively combine the deep semantic information and surface statistical patterns of documents, and lack an end-to-end solution that can flexibly adapt to customized classification needs in different industries and business scenarios. Therefore, this paper proposes a document data recognition and classification method and system based on keyword extraction. Summary of the Invention
[0005] To address the aforementioned technical problems, this invention provides a document data identification and classification method and system based on keyword extraction, which can effectively solve the challenges of managing, retrieving, and organizing massive amounts of multi-format documents, thereby improving the intelligence level and work efficiency of information processing.
[0006] The technical solution adopted in this invention is as follows: A document data identification and classification method based on keyword extraction includes the following steps: S1: Receive the input raw document file, identify the format type of the raw document file, and call the corresponding document parser according to the identified format type to extract the plain text content from the raw document file; S2: The plain text content is cleaned, segmented, and filtered for stop words to obtain a structured text word sequence; S3: Based on the text word sequence, a list of keywords representing the core theme of the document is extracted by a sorting algorithm that integrates the word co-occurrence graph structure and semantic similarity information; S4: Based on the predefined custom hierarchical classification system and its associated keyword-classification code mapping relationship, the keyword list is transformed into the document's associated feature vector for each classification code in the custom hierarchical classification system; S5: Based on the associated feature vector, determine one or more target classification codes to which the original document file belongs, and output the classification result.
[0007] Preferably, the custom hierarchical classification system is a tree or network classification structure pre-configured by the user according to business needs, where each leaf node or valid node corresponds to a unique classification identifier, which is a classification code.
[0008] Preferably, the operation of converting the keyword list into an associated feature vector in step S4 specifically includes: S41: A pre-defined custom hierarchical classification system and a keyword-classification code mapping library are provided, wherein the mapping library records the quantitative association weight between keywords and classification codes; S42: For each category code in the custom hierarchical classification system, query the mapping library to obtain the association weight of each keyword in the keyword list for that category code; S43: Aggregate the association weights of all keywords in the keyword list for the same category code to obtain the overall relevance score of the document to that category code; S44: The overall relevance score of the document to all classification codes constitutes the relevance feature vector.
[0009] Preferably, the keyword-classification code mapping library is constructed in the following way: analyze a batch of historical documents with labeled classification codes, statistically analyze the distribution information of each keyword in documents with different classification codes, and calculate the association weight between the keyword and each classification code accordingly.
[0010] Preferably, the determination of the target classification code in step S5 specifically involves: setting a global threshold θ, logically merging or selecting the classification codes in the associated feature vectors whose association scores exceed the global threshold θ in conjunction with the hierarchical structure of the custom hierarchical classification system, and finally determining the target classification code.
[0011] Preferably, determining the target classification code in step S5 specifically involves: inputting the associated feature vector into a trained multi-label classification model, wherein the multi-label classification model makes a decision by combining the hierarchical prior knowledge of the classification system, and outputs the target classification code and its probability.
[0012] Preferred training methods for multi-label classification models include: T1: Obtain a batch of training documents that have been labeled with accurate classification codes; T2: Perform steps S1 to S4 for each training document to obtain the associated feature vector of each training document; T3: Using the associated feature vector and its corresponding real classification code as training samples, supervise the training of the initial multi-label classification model to obtain the trained multi-label classification model.
[0013] Preferably, the document data identification and classification method based on keyword extraction further includes step S6, which specifically involves: storing the output target classification code as metadata associated with the original document file, and triggering subsequent automated document archiving, process routing, or retrieval indexing operations based on the target classification code.
[0014] Preferably, the ranking algorithm in step S3, which integrates the word co-occurrence graph structure and semantic similarity information, includes an adjustable parameter in its formula for iteratively updating the node importance score, which is used to balance the contribution ratio of co-occurrence graph statistical information and semantic similarity information.
[0015] This invention also provides a document data recognition and classification system based on keyword extraction, comprising: The document input and parsing module is used to execute step S1; The text preprocessing module, connected to the document input and parsing module, is used to execute step S2; The keyword extraction module, connected to the text preprocessing module, is used to execute step S3; The feature construction module is connected to the keyword extraction module and accesses the pre-stored custom hierarchical classification system and keyword-classification code mapping library to execute step S4; The classification decision module, connected to the feature construction module, is used to execute step S5; The classification system management module provides a configuration interface for users to define and maintain the custom hierarchical classification system.
[0016] The beneficial effects of this invention are: This invention significantly improves the ability to accurately characterize the core themes of documents by integrating a keyword extraction algorithm with vocabulary co-occurrence graph structure and semantic similarity information, overcoming the classification bias caused by inaccurate keyword extraction in traditional methods. Based on a custom hierarchical classification system and a data-driven keyword-classification code mapping library, the system can flexibly adapt to the classification needs of different industries and business scenarios, eliminating reliance on fixed rules and manually defined features, and possessing scalability and adaptability. The entire process achieves end-to-end automation from document input, parsing, feature extraction to classification decision. Through structured associated feature vectors and efficient threshold or model decision mechanisms, the efficiency and consistency of classification decisions are ensured. At the same time, the mapping library, built on historical labeled data and continuously optimized, enables the system to continuously learn and evolve from data, and the classification accuracy can be continuously improved with data accumulation. Finally, the output target classification code can directly drive downstream applications such as automated document archiving, intelligent retrieval, and process routing, effectively solving the problems of managing, retrieving, and organizing massive amounts of multi-format documents, and improving the intelligence level and work efficiency of information processing. Attached Figure Description
[0017] Figure 1 This is a flowchart of a document data identification and classification method based on keyword extraction according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating the mapping library construction process of an embodiment of the present invention; Figure 3 This is a flowchart illustrating the training process of the multi-label classification model according to an embodiment of the present invention. Figure 4 A flowchart illustrating the construction of associated feature vectors in an embodiment of the present invention; Figure 5 This is a block diagram of a document data recognition and classification system based on keyword extraction, according to an embodiment of the present invention. Detailed Implementation
[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0019] Example 1 like Figures 1-4 As shown, an embodiment of the present invention provides a document data identification and classification method based on keyword extraction, comprising the following steps: S1: Receives the input raw document file, identifies the format type of the raw document file, and calls the corresponding document parser according to the identified format type to extract the plain text content from the raw document file.
[0020] It receives raw document files as input via file upload interface, API call, or directory monitoring service. The format type of the raw document file is identified by analyzing its binary header information (MagicNumber) combined with the file extension.
[0021] For example, if the file header is "%PDF-" or the extension is ".pdf", it is determined to be in PDF format; if the file header is "PK" and the extension is ".docx", it is determined to be in DOCX format.
[0022] Based on the identified format type, the corresponding pre-integrated document parser is dispatched (e.g., PDFBox for PDF, Apache POI for Office documents, and BeautifulSoup for HTML). The parser executes its internal logic, stripping the document of formatting, styles, metadata, and non-text elements, and outputting a string of continuous plain text content contained in the original document file.
[0023] S2: Clean the plain text content, segment it, and filter out stop words to obtain a structured text word sequence.
[0024] The plain text content obtained in step S1 needs to be standardized. First, invisible control characters need to be removed, full-width characters need to be converted to half-width characters, and all letters need to be converted to lowercase. Then, a word segmentation tool needs to be selected according to the document language: for Chinese text, a word segmentation tool based on dictionary and statistical models (such as Jieba) is used; for space-separated languages such as English, a word segmenter based on spaces and punctuation is used.
[0025] After word segmentation, an initial vocabulary sequence is obtained. A predefined stop word list is loaded, containing high-frequency but low-information-entropy general words. Each word in the initial vocabulary sequence is compared with the stop word list, and all matching stop words are removed. Finally, an ordered, structured text word sequence composed of valid words is obtained, denoted as . ,in The number of valid words.
[0026] S3: Based on the text word sequence, a ranking algorithm that integrates the word co-occurrence graph structure and semantic similarity information is used to extract a list of keywords that represent the core theme of the document.
[0027] Candidate word generation and co-occurrence graph construction require sequence analysis. Part-of-speech tagging was performed to select nouns, verbs, adjectives, and other content words as candidate words. Set a fixed size as Sliding window (usually) )exist Swipe up. When two different candidate words and When they appear in the same window at the same time, it is considered that there is a co-occurrence between them.
[0028] Construct an undirected weighted graph in this way. , where the node set Corresponding candidate word set edge set weight Defined as candidate words and Co-occurrence counts throughout the text. Undirected weighted graphs model the lexical association structure at the surface level of the text.
[0029] Semantic similarity is calculated using a pre-trained word vector model or context embedding model, which integrates each candidate word... Mapped to a fixed dimension semantic vectors Calculate any two candidate words. and semantic similarity between The cosine similarity formula is used. all Constructing a semantic similarity matrix Similarity matrices model the intrinsic relationships between words in the semantic space.
[0030] The calculation of the fusion ranking and weights uses an improved TextRank algorithm to iteratively calculate each candidate word node. Importance score The algorithm's iterative update formula integrates co-occurrence graph structural information and semantic information: In the formula, Represents a node In the The score at the next iteration. This is the damping coefficient, typically set to 0.85, representing the probability of jumping from one node to a random neighbor. Adj Represents a node In the figure The set of all neighboring nodes. It is an adjustable fusion parameter ( This is used to balance the contribution ratio of co-occurrence graph statistics and semantic similarity information. When When the algorithm degenerates into the traditional TextRank algorithm, it is considered to be ineffective. At this time, the ranking depends entirely on semantic similarity. The formula contains two normalization factors. and Ensured from the neighbors Passed to The sum of weights is normalized. Initialize the scores for all nodes. Repeat the above iterations until the change in scores of all nodes is less than a preset threshold. (like The algorithm converged.
[0031] Ultimately, each candidate word The importance score is the score of the node after convergence. Value. 4. Generate a keyword list: based on the converged value. Values for all candidate words Sort in descending order. Select the top-ranked items. (For example The candidate words are used to form an ordered keyword list. ,in It has the highest importance score.
[0032] S4: Based on a predefined custom hierarchical classification system and its associated keyword-classification code mapping relationship, the keyword list is transformed into a document's associated feature vector for each classification code in the custom hierarchical classification system.
[0033] A custom hierarchical classification system is a tree or network structure pre-configured by the system administrator through the classification system management module. For example, the classification system of an enterprise knowledge base might be: Level 1: Technical, Marketing, and Management categories; Level 2 (Technical Category): Software Development, Hardware Design, Test Reports; Level 3 (under Software Development): Requirements Document, Design Document, Code Review. Each valid category node is assigned a unique category code.
[0034] The keyword-classification code mapping library stores the quantified correlation weights between keywords and various classification codes. The construction method involves analyzing a batch of historical documents already labeled with classification codes. For a given keyword... and classification code Its association weight can be calculated by word In the category The TF-IDF variants or other statistics are obtained from all documents. The calculation formula is: In the formula, It is a word In category Average frequency of appearance in the document It is the total number of categories. It is a word containing The number of categories. The higher the value, the stronger the word. For the category of identification The more representative it is.
[0035] For the document to be classified, associate the feature vector. The construction process is as follows, where It is the total number of valid classification codes in the classification system: For each classification code in the system The processing involves querying the mapping library to obtain a list of keywords. Each keyword Classification codes Association weight (If not recorded, the weight is 0). For The weights of all keywords in the text are aggregated, and a common aggregation method is summation: Score Or take the maximum value: Score Score That is, a document Classification codes The overall relevance score.
[0036] Then all The scores of each classification code are arranged in a fixed order, which constitutes the document's associated feature vector: This vector is A real-valued vector of dimensions, where each dimension directly quantifies the degree of matching between the document content and a specific classification code.
[0037] S5: Based on the associated feature vector, determine one or more target classification codes to which the original document file belongs, and output the classification result.
[0038] By setting a global threshold Filter out all that meet the Score The classification codes form the candidate code set. .
[0039] Logical merging or selection is performed based on the hierarchical structure of the custom hierarchical classification system: If If empty, output the most relevant classification codes. If the data contains classification codes with parent-child relationships, then according to the principle of proximity, the classification code of the most specific child node is retained first, and its parent node classification code is removed. If If a node contains multiple sibling node classification codes with similar scores, then the classification code of their common parent node is output. 3. The set of classification codes after hierarchical logical processing is the final target classification code. When outputting, its corresponding relevance score can be attached as a confidence level.
[0040] Furthermore, custom hierarchical classification The system is a tree or network classification structure pre-configured by the user according to business needs. Each leaf node or valid node corresponds to a unique classification identifier, which is a classification code.
[0041] Furthermore, step S4, which transforms the keyword list into associated feature vectors, specifically includes: S41: Pre-set custom hierarchical classification system and a keyword-classification code mapping library. The mapping library records the quantitative association weight between keywords and classification codes. S42: For each category code in the custom hierarchical classification system, query the mapping library to obtain the association weight of each keyword in the keyword list for that category code; S43: Aggregate the association weights of all keywords in the keyword list for the same category code to obtain the overall relevance score of the document to that category code; S44: The overall relevance score of the document to all classification codes constitutes the relevance feature vector.
[0042] Furthermore, the keyword-classification code mapping library is constructed in the following way: analyze a batch of historical documents with labeled classification codes, statistically analyze the distribution information of each keyword in documents with different classification codes, and calculate the association weight between the keyword and each classification code accordingly.
[0043] Furthermore, the determination of the target classification code in step S5 specifically involves: logically merging or selecting the classification codes in the associated feature vectors whose association scores exceed the global threshold θ, in conjunction with the hierarchical structure of the custom hierarchical classification system, to finally determine the target classification code.
[0044] Furthermore, the determination of the target classification code in step S5 specifically involves: inputting the associated feature vector into a trained multi-label classification model, which then makes a decision based on the hierarchical prior knowledge of the classification system, and outputs the target classification code and its probability.
[0045] Furthermore, training methods for multi-label classification models include: T1: Obtain a batch of training documents that have been labeled with accurate classification codes; T2: Perform steps S1 to S4 for each training document to obtain the associated feature vector of each training document; T3: Using the associated feature vectors and their corresponding real classification codes as training samples, supervise the training of the initial multi-label classification model to obtain the trained multi-label classification model.
[0046] Furthermore, the document data identification and classification method based on keyword extraction also includes step S6, which specifically involves: storing the output target classification code as metadata associated with the original document file, and triggering subsequent automated document archiving, process routing, or retrieval indexing operations based on the target classification code.
[0047] The target classification code output in step S5 is used as core metadata and associated with the original document file for storage. For example, the file can be moved or symbolically linked to the corresponding network directory or knowledge base node based on the classification code; in an office automation system, the document can be automatically pushed to the processing queue of the department or personnel responsible for that category; or the classification code can be added to the index field of a full-text search engine to support users in efficiently filtering and searching by category.
[0048] Furthermore, the ranking algorithm in step S3, which integrates the vocabulary co-occurrence graph structure and semantic similarity information, includes an adjustable parameter in its formula for iteratively updating the node importance score. This parameter is used to balance the contribution ratio of co-occurrence graph statistics and semantic similarity information.
[0049] Example 2 Corresponding to the methods in the above embodiments, the present invention also proposes a document data recognition and classification system based on keyword extraction.
[0050] like Figure 5 As shown, the document data recognition and classification system based on keyword extraction according to an embodiment of the present invention includes: The document input and parsing module is used to execute step S1; The text preprocessing module, connected to the document input and parsing module, is used to execute step S2; The keyword extraction module is connected to the text preprocessing module and is used to execute step S3; The feature construction module connects to the keyword extraction module and accesses the pre-stored custom hierarchical classification system and keyword-classification code mapping library to execute step S4; The classification decision module, connected to the feature construction module, is used to execute step S5; The classification system management module provides a configuration interface for users to define and maintain custom hierarchical classification systems.
[0051] The system's workflow begins with the document input and parsing module receiving the raw document file. This module automatically identifies the document format and calls the corresponding parser to extract the plain text content. Subsequently, the text preprocessing module cleans, segments, and filters stop words from the plain text, generating a structured text word sequence. Based on this sequence, the keyword extraction module calculates the weights of core words and generates an ordered keyword list using a ranking algorithm that integrates vocabulary co-occurrence graph structure and semantic similarity information. The feature construction module matches and aggregates this list with a pre-stored custom hierarchical classification system and a keyword-classification code mapping library, generating feature vectors reflecting the degree of association between the document and each category. The classification decision module, based on this vector, determines one or more target classification codes and their confidence levels for the document using preset threshold judgment rules or a trained multi-label classification model. The classification system management module stores the classification results as metadata associated with the document and can automatically trigger subsequent business operations such as archiving, routing, or retrieval, thus forming an end-to-end automated intelligent document classification and processing closed loop.
[0052] Example 3 The construction process of the keyword-classification code mapping library begins with collecting a batch of historical documents whose classification codes have been accurately labeled manually as a training set. Subsequently, each document in the training set is processed in accordance with steps S1 to S3 in claim 1, namely, plain text is extracted through format parsing, and text word sequences are obtained through cleaning, word segmentation and filtering. The same keyword extraction algorithm with fusion graph structure and semantics is used to generate a list of core keywords for each document. Then, the system merges the keywords of all documents to form a global dictionary, and calculates the universality and discriminative power of each keyword and each classification code based on term frequency-inverse class frequency (TF-ICF). At the same time, statistical methods such as chi-square test are used to evaluate the significance of the association, thereby comprehensively calculating the quantitative association weight of each keyword corresponding to each classification code.
[0053] Finally, the system persistently stores all keywords, classification codes, and their associated weights, forming a keyword-classification code mapping library that can be efficiently queried. This mapping library can be directly queried when the system performs real-time analysis on unclassified documents to quickly obtain the category weights of keywords, thereby supporting subsequent feature vector aggregation and classification decisions. Furthermore, this library supports periodic recalculation using incremental data to achieve continuous optimization.
[0054] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.
[0055] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as will be understood by those skilled in the art to which embodiments of the invention pertain.
[0056] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.
[0057] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.
[0058] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.
[0059] Furthermore, the functional units in the various embodiments of the present invention can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.
[0060] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.
Claims
1. A document data identification and classification method based on keyword extraction, characterized in that, Includes the following steps: S1: Receive the input raw document file, identify the format type of the raw document file, and call the corresponding document parser according to the identified format type to extract the plain text content from the raw document file; S2: The plain text content is cleaned, segmented, and filtered for stop words to obtain a structured text word sequence; S3: Based on the text word sequence, a list of keywords representing the core theme of the document is extracted by a sorting algorithm that integrates the word co-occurrence graph structure and semantic similarity information; S4: Based on the predefined custom hierarchical classification system and its associated keyword-classification code mapping relationship, the keyword list is transformed into the document's associated feature vector for each classification code in the custom hierarchical classification system; S5: Based on the associated feature vector, determine one or more target classification codes to which the original document file belongs, and output the classification result.
2. The document data identification and classification method based on keyword extraction according to claim 1, characterized in that, A custom hierarchical classification system is a tree or network classification structure pre-configured by the user according to business needs. Its leaf nodes or valid nodes correspond to unique classification identifiers, which are classification codes.
3. The document data identification and classification method based on keyword extraction according to claim 1, characterized in that, Step S4, which transforms the keyword list into an associated feature vector, specifically includes: S41: A pre-defined custom hierarchical classification system and a keyword-classification code mapping library are provided, wherein the mapping library records the quantitative association weight between keywords and classification codes; S42: For each category code in the custom hierarchical classification system, query the mapping library to obtain the association weight of each keyword in the keyword list for that category code; S43: Aggregate the association weights of all keywords in the keyword list for the same category code to obtain the overall relevance score of the document to that category code; S44: The overall relevance score of the document to all classification codes constitutes the relevance feature vector.
4. The document data identification and classification method based on keyword extraction according to claim 3, characterized in that, The keyword-classification code mapping library is constructed by analyzing a batch of historical documents labeled with classification codes, statistically analyzing the distribution information of each keyword in documents with different classification codes, and calculating the association weight between the keyword and each classification code accordingly.
5. The document data identification and classification method based on keyword extraction according to claim 1, characterized in that, The specific steps for determining the target classification code in step S5 are as follows: set a global threshold θ, and logically merge or select the classification codes in the associated feature vectors whose association scores exceed the global threshold θ, in conjunction with the hierarchical structure of the custom hierarchical classification system, to finally determine the target classification code.
6. The document data identification and classification method based on keyword extraction according to claim 1, characterized in that, The specific steps for determining the target classification code in step S5 are as follows: inputting the associated feature vector into a trained multi-label classification model, which makes a decision by combining the hierarchical prior knowledge of the classification system, and outputs the target classification code and its probability.
7. The document data identification and classification method based on keyword extraction according to claim 6, characterized in that, Training methods for multi-label classification models include: T1: Obtain a batch of training documents that have been labeled with accurate classification codes; T2: Perform steps S1 to S4 for each training document to obtain the associated feature vector of each training document; T3: Using the associated feature vector and its corresponding real classification code as training samples, supervise the training of the initial multi-label classification model to obtain the trained multi-label classification model.
8. The document data identification and classification method based on keyword extraction according to claim 1, characterized in that, The document data identification and classification method based on keyword extraction also includes step S6, which is: storing the output target classification code as metadata associated with the original document file, and triggering subsequent document automated archiving, process routing or retrieval indexing operations based on the target classification code.
9. The document data identification and classification method based on keyword extraction according to claim 1, characterized in that, The ranking algorithm in step S3, which integrates the vocabulary co-occurrence graph structure and semantic similarity information, includes an adjustable parameter in its formula for iteratively updating the node importance score. This parameter is used to balance the contribution ratio of co-occurrence graph statistics and semantic similarity information.
10. A document data recognition and classification system based on keyword extraction, characterized in that, include: The document input and parsing module is used to execute step S1; The text preprocessing module, connected to the document input and parsing module, is used to execute step S2; The keyword extraction module, connected to the text preprocessing module, is used to execute step S3; The feature construction module is connected to the keyword extraction module and accesses the pre-stored custom hierarchical classification system and keyword-classification code mapping library to execute step S4; The classification decision module, connected to the feature construction module, is used to execute step S5; The classification system management module provides a configuration interface for users to define and maintain the custom hierarchical classification system.