Enterprise document classification method and device
By extracting structural attribute vectors and target detection and matching of models for enterprise documents, reconstructing the document layout, the efficiency and accuracy of traditional document management methods in massive, diverse and rapidly growing document management is solved, and high Accurate document classification and management.
Patent Information
- Application Number
- CN202411538148.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-31
- Publication Date
- 2025-02-18
AI Technical Summary
When traditional document management methods face massive, diverse and rapidly growing corporate documents, it is difficult to achieve efficient and accurate classification, resulting in high costs, low efficiency and error-prone.
By obtaining the structural attribute vectors corresponding to each semantic entity in the pending enterprise document, using pre-trained models for object detection, structure analysis and matching, and reconstructing the document layout, thereby achieving high-precision document classification.
It significantly improves the recognition accuracy of handwritten documents, complex typesetting documents and technical documents in professional fields, and has broad applicability and can meet the efficient and accurate needs of modern enterprises for document management.
Smart Images
Figure CN119474373A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of computer technology, and in particular to a method and device for classifying enterprise documents. Background Art
[0002] Document management plays a vital role. Whether it is contracts, reports, policy documents, or emails and memos in daily communication, documents are an important part of corporate knowledge assets. However, as the scale of enterprises expands and the scope of business expands, the number and types of documents are also growing rapidly, bringing unprecedented challenges to document management. The massive data, diversity and rapid growth make traditional document management methods seem inadequate. Manual classification is not only costly and inefficient, but also prone to errors, and it is difficult to meet the needs of modern enterprises for efficient and accurate document management.
[0003] It should be noted that the above introduction to the technical background is only for the convenience of providing a clear and complete description of the technical solutions of the present application and for the convenience of understanding by those skilled in the art. It cannot be considered that the above technical solutions are well known to those skilled in the art simply because they are described in the background technology section of the present application. Summary of the invention
[0004] The embodiments of the present application provide a method and device for classifying enterprise documents.
[0005] The first embodiment of the present application proposes a method for classifying enterprise documents, including:
[0006] Obtaining the structural attribute vector corresponding to each semantic entity in the enterprise document to be processed;
[0007] Performing target detection and structural analysis on the structural attribute vector based on a pre-trained first model to obtain a feature set of the structural attribute vector and a hierarchical relationship between different structural attribute vectors;
[0008] The feature set and the hierarchical structure relationship are matched based on a pre-trained second model, and the layout is obtained to obtain a reconstructed target document.
[0009] The second embodiment of the present application proposes an enterprise document classification device, including:
[0010] A first acquisition module, the first acquisition module is used to acquire a structural attribute vector corresponding to each semantic entity in the enterprise document to be processed;
[0011] A second acquisition module, the second acquisition module is used to perform target detection and structural analysis on the structural attribute vector based on the pre-trained first model, and obtain a feature set of the structural attribute vector and a hierarchical relationship between different structural attribute vectors;
[0012] The third acquisition module is used to match the feature set and the hierarchical structure relationship based on the pre-trained second model, and acquire the layout to obtain the reconstructed target document.
[0013] The third aspect embodiment of the present application proposes an electronic device, comprising: a processor; a memory for storing instructions executable by the processor; wherein the processor is configured to execute the instructions to implement the enterprise document classification method proposed in the first aspect embodiment of the present application.
[0014] The fourth aspect embodiment of the present application proposes a non-temporary computer-readable storage medium. When the instructions in the storage medium are executed by the processor of an electronic device, the electronic device can execute the method proposed in the first aspect embodiment of the present application.
[0015] The fifth aspect embodiment of the present application proposes a computer program product, including a computer program, which implements the method proposed in the first aspect embodiment of the present application when executed by a processor in a communication device.
[0016] The technical solution provided by the embodiments of the present application brings at least the following beneficial effects:
[0017] The first model performs target detection and structural analysis on the structural attribute vectors corresponding to each semantic entity in the enterprise document to be processed, and obtains the corresponding feature set and the hierarchical relationship between each structural attribute vector; the second model matches the feature set and the hierarchical relationship to obtain the layout and reconstruct the target document. This greatly improves the recognition accuracy of handwritten documents, documents with complex layouts, and technical documents in professional fields, and has wide applicability.
[0018] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS
[0019] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:
[0020] Figure 1 A flowchart of a method for classifying enterprise documents provided in an embodiment of the present application;
[0021] Figure 2 A schematic diagram of a second method for classifying enterprise documents provided in an embodiment of the present application;
[0022] Figure 3 A schematic diagram of a third method for classifying enterprise documents provided in an embodiment of the present application;
[0023] Figure 4 A schematic diagram of the structure of an enterprise document classification device provided in an embodiment of the present application;
[0024] Figure 5 A schematic diagram of the structure of an electronic device provided according to an embodiment of the present application;
[0025] Figure 6 It is a schematic diagram of the structure of another electronic device provided according to an embodiment of the present application. DETAILED DESCRIPTION
[0026] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the embodiments of the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the embodiments of the present application as detailed in the appended claims.
[0027] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the embodiments of the present application. The singular forms of "a" and "the" used in the embodiments of the present application and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.
[0028] It should be understood that, although the terms first, second, third, etc. may be used to describe various information in the embodiments of the present application, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the embodiments of the present application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the words "if" and "if" as used herein may be interpreted as "at..." or "when..." or "in response to determination".
[0029] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.
[0030] It should be noted that the enterprise document classification method provided in any embodiment of the present application can be executed alone, or in combination with possible implementation methods in other embodiments, or in combination with any technical solution in related technologies.
[0031] The enterprise document classification method and device thereof according to an embodiment of the present application will be described below with reference to the accompanying drawings.
[0032] Figure 1 This is a flow chart of a method for classifying enterprise documents provided in an embodiment of the present application. Figure 1 As shown, the method includes but is not limited to the following steps:
[0033] S101, obtaining a structural attribute vector corresponding to each semantic entity in the enterprise document to be processed.
[0034] In a feasible implementation, the enterprise documents to be processed can be preprocessed first, including removing noise (such as irrelevant characters, copyright notices, etc.); performing operations such as word segmentation and stop word removal on the enterprise documents; and converting the format of the enterprise documents according to the preset format. For example, the enterprise documents (including handwritten documents, documents with complex typeset, professional technical documents, etc.) can be converted into text format. Then, using named entity recognition technology, key semantic entities are identified from the enterprise documents. These semantic entities usually include names of people, places, names of organizations, dates and times, numbers, and proper nouns, etc.; and the starting and ending positions of the semantic entities are determined, and unified annotation specifications are formulated, including annotation formats, label selection, etc., to ensure the consistency and accuracy of annotations; if the enterprise documents contain semantic entities in specific fields, a custom semantic entity dictionary can be constructed.
[0035] Next, according to predefined rules, the structural attributes of semantic entities are extracted from corporate documents in text format. For example, regular expressions can be used to match the establishment date, registered capital and other information after the organization name. The extracted structural attributes are then converted into vector representations, for example, through word embedding technology, which can convert the vocabulary of corporate documents in text format into fixed-length vectors that can capture the semantic relationship between words. The identified semantic entities are then integrated with their corresponding structural attribute vectors, and the integrated information is output in an appropriate form, such as JSON, CSV, etc.
[0036] It should be noted that when processing corporate documents, attention should be paid to data privacy and security issues; and according to the task requirements and the characteristics of corporate documents, the appropriate framework and attribute extraction method of named entity recognition technology should be selected; at the same time, the erroneous structural attribute vectors should be corrected or re-labeled accordingly.
[0037] S102, performing target detection and structural analysis on the structural attribute vector based on the pre-trained first model, and obtaining a feature set of the structural attribute vector and a hierarchical relationship between different structural attribute vectors.
[0038] In a feasible implementation, a pre-trained model is selected as the first model, and the first model has strong feature extraction and target detection capabilities. For example, an OCR model based on deep learning (full name Optical Character Recognition, optical character recognition, which is used to convert text content in pictures or scans into editable text) can be selected; a convolutional network model based on deep learning can also be selected, or a residual network model can be selected to classify, detect objects, and semantically segment structural attribute vectors through identity mapping; or a combination of deep separable convolution and stationary point convolution can be used to extract and calculate features of structural attribute vectors.
[0039] Next, the data containing the structural attribute vector is input into the first model. This data can be images, text or other forms of data, depending on the application scenario. The convolutional layer, pooling layer and other structures in the first model are used to extract the features of the data. These features can be local, such as edges, textures, etc.; they can also be global, such as shape, color, etc. Based on the extracted features, the detection layer in the first model can be used to locate the structural attribute vector in the data. This usually involves scoring and screening operations on candidate areas to determine the exact location and range of the structural attribute vector. The features extracted in the target detection phase are sorted to form a feature set of the structural attribute vector. These feature sets contain various attribute information of the target, such as shape, size, color, etc.
[0040] Next, the features can be further analyzed through cluster analysis, association rule mining, tree structure, etc. to determine the hierarchical relationship between different structural attribute vectors, classify structural attribute vectors with similar attributes into the same category, and determine the parent-child relationship or brother relationship between them. The hierarchical relationship obtained by parsing can be presented in a visual way, such as a tree diagram, a network diagram, etc.; the hierarchical relationship can also be output in the form of text or table. The hierarchical relationship describes the hierarchy and association between the elements in the document layout. For example, the relationship between paragraphs, the relationship between paragraphs and titles, the relationship between images and text, etc.
[0041] It should be noted that the training of the first model is a key step. By selecting appropriate training data and training strategies, it is ensured that the first model can accurately extract features and detect targets. Before inputting data, appropriate data preprocessing work, such as denoising and normalization, is performed to improve the detection accuracy and analysis effect of the first model. When performing structural analysis, the algorithm used by the first model can be optimized and adjusted according to actual needs to improve the accuracy and efficiency of analysis.
[0042] S103, matching the feature set and the hierarchical structure relationship based on the pre-trained second model, acquiring the layout to obtain the reconstructed target document.
[0043] In a feasible implementation, a pre-trained model is selected as the second model, and the second model can handle the matching task of feature sets and hierarchical relationships. For example, a model based on deep learning can be selected; an NLP model (full name Natural Language Processing, that is, a natural language processing model, which is a model that trains and learns machines through a large amount of text data so that they can understand the structure, grammatical rules and semantic information of a language) can also be used; or a model for document layout analysis, such as LayoutLM, can be selected. Then, the pre-trained second model can be fine-tuned according to the specific task requirements. For example, a labeled training data set (the label can map the feature set and the corresponding hierarchical relationship) can be used to train the model, and the model parameters can be fine-tuned according to the training results.
[0044] Next, the feature set and the hierarchical relationship are input into the second model to calculate the similarity between the feature and the feature set, which can be calculated by cosine similarity, Euclidean distance and other metrics; according to the similarity results, the features and hierarchical relationships are matched, and the matching includes regrouping, sorting and other operations to produce the expected layout. After the matching is completed, the enterprise document can be reconstructed according to the matching results, and the reconstruction includes adjusting the position, size, shape, etc. of the semantic entity, and adding or deleting metadata to produce the expected layout and format requirements, so as to obtain the target document.
[0045] It should be noted that the reconstructed target document may also need to be optimized to improve readability and aesthetics, including adjusting font size, color, spacing, etc. Verification of the reconstructed document can be achieved through manual review.
[0046] In summary, the first model performs target detection and structural analysis on the structural attribute vectors corresponding to each semantic entity in the enterprise document to be processed, and obtains the corresponding feature set and the hierarchical relationship between each structural attribute vector; the second model matches the feature set and the hierarchical relationship to obtain the layout and reconstruct the target document. This greatly improves the recognition accuracy of handwritten documents, documents with complex layouts, and technical documents in professional fields, and has wide applicability.
[0047] Figure 2 This is a flow chart of the second enterprise document classification method provided in the embodiment of the present application. Figure 2 As shown, the method includes but is not limited to the following steps:
[0048] S201, pre-processing the enterprise document to be processed according to the preset confidentiality level, and obtaining the structural attribute vector corresponding to each semantic entity.
[0049] Optionally, as an example, the corporate document to be processed is preprocessed according to a preset confidentiality level to obtain the first intermediate information. Specifically, the corporate document to be processed is classified according to the preset confidentiality level to ensure that corporate documents of different confidentiality levels are not confused. Then, a content review is performed on the corporate document to be processed, and based on the results of the content review, the information in the corporate document is annotated to obtain a first text that meets the confidentiality level. For example, for confidential information, encrypted storage, restricted access and other measures may be taken; for sensitive information, desensitization, blurring and other measures may be taken. The results of the content review and information annotation are integrated to form a review opinion; according to the instructions of the review opinion, the corporate document is revised, including deleting all information that violates confidentiality regulations, modifying parts that may leak sensitive information, etc.; the revised part is the first text.
[0050] After obtaining the first text, the first text is formatted according to the preset document format to obtain the second text, wherein the preset document format requirements are clearly defined, including the document's page layout, paragraph format, table and image processing methods, etc.; then the first text is formatted to obtain the second text, and it is ensured that the name and storage method of the second text comply with the regulations of the organization. Then the second text is subjected to irrelevant information removal to obtain the first intermediate information, for example, redundant descriptions that do not increase the value of the document or the difficulty of understanding are removed; if certain details are not directly related to the subject or purpose of the document and do not affect the overall understanding, they can be deleted; paragraphs or sentences that are not related to the subject can be deleted or reorganized to ensure the coherence and focus of the content.
[0051] After obtaining the first intermediate information, the second intermediate information is obtained based on segmentation and part-of-speech tagging of the first intermediate information, wherein the first intermediate information is first segmented according to sentences, each sentence is used as the basic unit of processing, and this operation can be performed through punctuation marks; the sentences are further segmented into words or phrases, and this operation can be based on preset rules, or based on the statistics of proprietary vocabulary, or based on deep learning operations. According to the segmentation results, part-of-speech tagging sets can be used for tagging, or natural language toolkits and proper noun segmenters can be used for tagging; after the part-of-speech tagging is completed, the tagging results are checked and the wrong or uncertain standards are corrected. The segmented results and their corresponding part-of-speech tags are integrated once to form the second intermediate information.
[0052] After obtaining the second intermediate information, the second intermediate information is feature-converted according to the preset semantic entity types and the association relationship between the semantic entities to obtain the structural attribute vector. Specifically, the semantic entity types that need to be identified and extracted are determined, such as names of people, places, names of organizations, time, date, numbers, etc. These types should be defined according to specific application scenarios and requirements; the possible association relationships between different semantic entities are clarified, such as parent-child relationships, parallel relationships, attribute relationships, etc. These relationships help to understand the hierarchy and interdependence between entities. In the second intermediate information, all semantic entities are identified and extracted according to predefined semantic entity types; based on the identified semantic entities and the association relationships between them, an entity association graph is constructed, which can be directed or undirected, depending on the nature of the association relationship; according to the semantic entity types and association relationships, a feature space is defined, which can accommodate all possible entity types and relationships, and a unique feature vector is assigned to each entity and relationship; for each identified semantic entity, a feature vector is assigned in the feature space according to its type and attributes (such as name, location, time, etc.); for the association relationship between entities, a feature vector also needs to be assigned in the feature space, and this vector should be able to reflect the type and strength of the relationship; the entity feature vector and the relationship feature vector are combined together to form a structural attribute vector, which should be able to comprehensively represent the semantic entities in the second intermediate information and the association relationships between them.
[0053] For further introduction to step S201, please refer to the relevant contents in the above embodiment, which will not be repeated here.
[0054] S202, performing target detection and structural analysis on the structural attribute vector based on the pre-trained first model, and obtaining a feature set of the structural attribute vector and a hierarchical relationship between different structural attribute vectors.
[0055] Optionally, as an example, based on preprocessing the structural attribute vector, a first vector matching the input requirement of the pre-trained first model is obtained, wherein a data set of the structural attribute vector is first obtained; based on preprocessing and feature engineering operations on the data set, a first vector matching the input requirement of the pre-trained first model is obtained.
[0056] Then, based on the first model, convolution and pooling are performed on the first vector to obtain a feature set reflecting the document content, document category and document location, wherein an appropriate number and size of convolution kernels are selected, and the convolution kernels are used to slide on the first vector, and the dot product of the convolution kernel energy and the local area of the first vector is calculated to generate a corresponding feature map; the corresponding pooling type (maximum pooling or average pooling) is set, and the pooling operation is applied to each feature map to reduce the dimension of the feature map while retaining the most important features; all feature maps obtained after the convolution and pooling operations are combined together to form a feature set, which contains information on the document content, document category and document location.
[0057] After obtaining the feature set, hierarchical clustering is performed on the first vector based on the first model to obtain the hierarchical relationship between the different structural attribute vectors. The first vector can be properly preprocessed, such as normalization or dimension reduction; then agglomerative hierarchical clustering or divisive hierarchical clustering can be selected to perform similarity measurement on the first vector to obtain the distance or similarity between the first vectors. According to the result of the distance operation, a hierarchical tree is constructed, which shows the nested relationship between clusters; then, by analyzing the hierarchical tree, the hierarchical relationship between different structural attribute vectors is identified.
[0058] For further introduction to step S202, please refer to the relevant contents in the above embodiment, which will not be repeated here.
[0059] S203, matching the feature set and the hierarchical structure relationship based on the pre-trained second model, acquiring the layout to obtain the reconstructed target document.
[0060] Optionally, as an example, according to business needs, document classification labels are obtained. In response to the mapping operation of the document classification labels, the feature set and the hierarchical relationship are substituted into the pre-trained second model to obtain the matching result between the feature set and the hierarchical relationship, wherein each document is mapped to its corresponding classification label, which can be achieved using a hash table, dictionary or other data structure; ensuring that each feature in the feature set is associated with one or more documents (and their classification labels), which can be achieved by retaining the document index or ID during the feature extraction process. The feature set is then substituted into the input layer of the second model, which usually involves using the feature vector as the input tensor of the model; the hierarchical relationship is substituted into the model in an appropriate form, which may involve converting the hierarchical tree into a graph structure and processing it using a graph neural network, or encoding the hierarchical relationship into a vector or matrix and using it as an additional input to the model.
[0061] Run the second model to obtain the matching results between the feature set and the hierarchical relationship, which may be a classification task (such as determining whether two feature sets belong to the same category), a regression task (such as predicting the similarity between two feature sets), or a more complex task (such as predicting the type of relationship in the hierarchy); interpret the matching results based on the output of the model. This may involve decoding the output, calculating indicators (such as precision, recall, etc.), or performing visualization to present the results more intuitively.
[0062] After obtaining the matching results between the feature set and the hierarchical relationship, the second model generates the layout of the target document according to the matching results, and based on the layout, obtains the layout to obtain the reconstructed target document. It should be noted that the matching results may include the classification labels of the various parts of the document, the hierarchical relationship between them, and possible layout suggestions; according to the matching results and business requirements, a suitable layout template is selected, which should be able to accommodate all parts of the document and meet the expected visual effects and reading experience; the various parts of the document (such as titles, paragraphs, pictures, tables, etc.) are positioned to the corresponding positions in the layout template according to the matching results and hierarchical relationships, which may involve calculating the size, position and alignment of each element; according to the classification labels and possible layout suggestions, appropriate styles (such as fonts, colors, margins, etc.) are applied to each part of the document, which can ensure the consistency and readability of the document; the generated layout and style are applied to the reorganized content to create the final target document.
[0063] For further introduction to step S202, please refer to the relevant contents in the above embodiment, which will not be repeated here.
[0064] In summary, the first model performs target detection and structural analysis on the structural attribute vectors corresponding to each semantic entity in the enterprise document to be processed, and obtains the corresponding feature set and the hierarchical relationship between each structural attribute vector; the second model matches the feature set and the hierarchical relationship to obtain the layout and reconstruct the target document. This greatly improves the recognition accuracy of handwritten documents, documents with complex layouts, and technical documents in professional fields, and has wide applicability.
[0065] Figure 3 This is a flow chart of the third enterprise document classification method provided in the embodiment of the present application. Figure 3 As shown, the method includes but is not limited to the following steps:
[0066] S301, obtaining specific requirements for enterprise document management.
[0067] For further introduction to step S301, please refer to the relevant contents in the above embodiment, which will not be repeated here.
[0068] S302: Determine the technology stack of the first model and the second model according to specific requirements.
[0069] It should be noted that the first model is mainly responsible for document classification and feature extraction, and its technology stack may include OCR model training, deep learning technology, feature engineering, clustering analysis, etc.; the second model is mainly responsible for layout generation and document reconstruction, and its technology stack may include natural language processing, generative adversarial networks, reinforcement learning, content extraction and reorganization, style application, etc.
[0070] For further introduction to step S302, please refer to the relevant contents in the above embodiment, which will not be repeated here.
[0071] S303, collect various enterprise documents, perform data cleaning and data labeling on the enterprise documents, and obtain enterprise documents to be processed.
[0072] For further introduction to step S303, please refer to the relevant contents in the above embodiment, which will not be repeated here.
[0073] S304, substituting the structural attribute vector corresponding to each semantic entity in the enterprise document to be processed into the first model for training, and obtaining a feature set of the structural attribute vector and a hierarchical relationship between each structural attribute vector.
[0074] For further introduction to step S304, please refer to the relevant contents in the above embodiment, which will not be repeated here.
[0075] S305, substitute the feature set and the hierarchical structure relationship into the second model for training, obtain the matching result between the feature set and the hierarchical structure relationship, and construct the page layout and classification rules of the target document according to the matching result.
[0076] For further introduction to step S305, please refer to the relevant contents in the above embodiment, which will not be repeated here.
[0077] S306, creating a target document according to the page layout and classification rules.
[0078] For further introduction to step S306, please refer to the relevant contents in the above embodiment, which will not be repeated here.
[0079] S307, according to step S301 to step S306, develop an enterprise document management system, and perform document management based on the enterprise document management system.
[0080] It should be noted that the development of an enterprise document management system includes demand analysis, business process streamlining, system architecture design, etc.; document management includes document uploading and storage, classification and label management, version control, etc.
[0081] For further introduction to step S307, please refer to the relevant contents in the above embodiment, which will not be repeated here.
[0082] Figure 4 This is a schematic diagram of the structure of a corporate document classification device provided in an embodiment of the present application. Figure 4 As shown, the enterprise document classification device 400 includes:
[0083] A first acquisition module 401, the first acquisition module 401 is used to acquire a structural attribute vector corresponding to each semantic entity in the enterprise document to be processed;
[0084] A second acquisition module 402, the second acquisition module 402 is used to perform target detection and structural analysis on the structure attribute vector based on the pre-trained first model, and obtain a feature set of the structure attribute vector and a hierarchical relationship between different structure attribute vectors;
[0085] The third acquisition module 403 is used to match the feature set and the hierarchical structure relationship based on the pre-trained second model, and obtain the layout to obtain the reconstructed target document.
[0086] Figure 5 The present invention is a schematic diagram of the structure of an electronic device provided according to an embodiment of the present application. Figure 5 The electronic device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.
[0087] like Figure 5 As shown, the electronic device 500 includes a processor 501, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 502 or a program loaded from a memory 506 to a random access memory (RAM) 503. In the RAM 503, various programs and data required for the operation of the electronic device 500 are also stored. The processor 501, the ROM 502, and the RAM 503 are connected to each other via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.
[0088] The following components are connected to the I / O interface 505: a memory 506 including a hard disk, etc.; and a communication part 507 including a network interface card such as a LAN (Local Area Network) card, a modem, etc., and the communication part 507 performs communication processing via a network such as the Internet; a drive 508 is also connected to the I / O interface 505 as needed.
[0089] In particular, according to an embodiment of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 507. When the computer program is executed by the processor 501, the above-mentioned functions defined in the method of the present application are executed.
[0090] In an exemplary embodiment, a storage medium including instructions is also provided, such as a memory including instructions, and the instructions can be executed by the processor 501 of the electronic device 500 to complete the above method. Optionally, the storage medium can be a non-transitory computer-readable storage medium, for example, the non-transitory computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.
[0091] In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, apparatus, or device. In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in combination with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium may be transmitted using any suitable medium, including but not limited to: wireless, wire, optical cable, RF, etc., or any suitable combination of the above.
[0092] Figure 6 It is a schematic diagram of the structure of another electronic device provided according to an embodiment of the present application. Figure 6 The electronic device shown is only an example and should not limit the functions and scope of use of the embodiments of the present application. Figure 6 As shown, the electronic device 600 includes a processor 601 and a memory 602. The memory 602 is used to store program codes, and the processor 601 is connected to the memory 602 to read the program codes from the memory 602 to implement the enterprise document classification method in the above embodiment.
[0093] Optionally, the number of processors 601 may be one or more.
[0094] Optionally, the electronic device may further include an interface 603, and the number of the interface 603 may be multiple. The interface 603 may be connected to an application program and may receive data from an external device such as a sensor.
[0095] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art that are not disclosed in this application. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present application are indicated by the following claims.
[0096] It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.
Claims
1. A method for classifying enterprise documents, characterized in that: include: Obtaining the structural attribute vector corresponding to each semantic entity in the enterprise document to be processed; Performing target detection and structural analysis on the structural attribute vector based on a pre-trained first model to obtain a feature set of the structural attribute vector and a hierarchical relationship between different structural attribute vectors; The feature set and the hierarchical structure relationship are matched based on a pre-trained second model, and the layout is obtained to obtain a reconstructed target document.
2. The method according to claim 1, characterized in that The step of obtaining the structural attribute vector corresponding to each semantic entity in the enterprise document to be processed includes: Pre-process the enterprise documents to be processed according to the preset confidentiality level to obtain the first intermediate information; acquiring second intermediate information based on segmenting and part-of-speech tagging the first intermediate information; According to the preset semantic entity type and the association relationship between the semantic entities, feature conversion is performed on the second intermediate information to obtain a structural attribute vector.
3. The method according to claim 2, characterized in that The step of pre-processing the enterprise document to be processed according to the preset confidentiality level to obtain the first intermediate information includes: According to the preset confidentiality level, the content of the enterprise documents to be processed is reviewed and information is marked to obtain the first text that meets the confidentiality level requirements; Formatting the first text according to a preset document format to obtain a second text; Based on removing irrelevant information from the second text, first intermediate information is obtained.
4. The method according to any one of claims 1 to 3, characterized in that The performing target detection and structural analysis on the structural attribute vector based on the pre-trained first model to obtain a feature set of the structural attribute vector and a hierarchical relationship between different structural attribute vectors includes: Based on preprocessing the structural attribute vector, obtaining a first vector matching a pre-trained first model input requirement; Convolving and pooling the first vector based on the first model to obtain a feature set reflecting document content, document category, and document location; Hierarchical clustering is performed on the first vector based on the first model to obtain a hierarchical structure relationship between different structural attribute vectors.
5. The method according to claim 4, characterized in that The step of preprocessing the structural attribute vector to obtain a first vector matching the input requirement of a pre-trained first model comprises: Obtaining a data set of the structural attribute vector; Based on preprocessing and feature engineering operations on the data set, a first vector matching an input requirement of a pre-trained first model is obtained.
6. The method according to claim 1, characterized in that The method of matching the feature set and the hierarchical structure relationship based on the pre-trained second model to obtain the layout to obtain the reconstructed target document includes: Obtain document classification labels based on business needs; In response to the mapping operation of the document classification label, the feature set and the hierarchical structure relationship are substituted into a pre-trained second model to obtain a matching result between the feature set and the hierarchical structure relationship; According to the matching result, the second model generates a page layout of the target document, and based on the page layout, the page layout is acquired to obtain a reconstructed target document.
7. An enterprise document classification device, characterized in that: include: A first acquisition module, the first acquisition module is used to acquire a structural attribute vector corresponding to each semantic entity in the enterprise document to be processed; A second acquisition module, the second acquisition module is used to perform target detection and structural analysis on the structural attribute vector based on the pre-trained first model, and obtain a feature set of the structural attribute vector and a hierarchical relationship between different structural attribute vectors; The third acquisition module is used to match the feature set and the hierarchical structure relationship based on the pre-trained second model, and acquire the layout to obtain the reconstructed target document.
8. An electronic device, characterized in that: include: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the method according to any one of claims 1 to 6.
9. A non-transitory computer-readable storage medium, characterized in that: When the instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the method as claimed in any one of claims 1 to 6.
10. A computer program product, characterized in that The invention comprises a computer program, which implements the method according to any one of claims 1 to 6 when being executed by a processor.