Electronic document detection method and device, electronic equipment and storage medium

Abstract

The invention provides an electronic document detection method and device, electronic equipment and a storage medium, and belongs to the technical field of data processing. The method comprises the steps that a to-be-detected electronic document set is acquired; for each electronic document in the electronic document set, according to a key element dimension, a surface text dimension, a text style dimension, a deep semantic dimension, an image content dimension and a document structure homology dimension, respectively calculating the similarity of the electronic document and other electronic documents under each dimension, and obtaining a detection result of each dimension; the detection results of all the dimensions are integrated, comprehensive evaluation data are obtained, and the comprehensive evaluation data comprise information, obtained through detection of different dimensions, of the similarity between the electronic documents. The similarity is calculated based on multiple dimensions, the detection result is obtained, and subjectivity and low efficiency of manual detection are avoided.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to an electronic document detection method, device, electronic device and storage medium. Background Art

[0002] In public resource transactions, it is crucial to ensure the openness, fairness and justice of transactions, and real-time monitoring and early warning of abnormal transaction behaviors are required. At present, there are some problems in detecting abnormal behaviors in bidding documents in public resource transactions. For example, the detection methods are single, either focusing on the hardware physical feature code, or only performing surface text similarity detection, which makes it difficult to discover deep-seated problems such as semantics and style similarities; it is impossible to effectively detect non-text content such as pictures; it is impossible to detect complex abnormal situations such as DNA homology similarity; the high degree of manual participation leads to low efficiency and prone to misjudgment and missed judgment. Summary of the invention

[0003] The present application provides an electronic document detection method, device, electronic device and storage medium, aiming to solve the problems of limited detection dimension and drawbacks of manual detection in bidding document detection.

[0004] In a first aspect, the present application provides an electronic document detection method, the method comprising:

[0005] Obtaining the electronic document set to be tested;

[0006] For each electronic document in the electronic document set, the similarity between the electronic document and other electronic documents in each dimension is calculated according to the key element dimension, the surface text dimension, the text style dimension, the deep semantic dimension, the image content dimension and the document structure homology dimension, and the detection result of each dimension is obtained;

[0007] The detection results of the various dimensions are integrated to obtain comprehensive evaluation data, which includes information about similarities between electronic documents obtained from detections of different dimensions.

[0008] In a second aspect, the present application further provides an electronic document detection device, the device comprising:

[0009] A data acquisition module, used to acquire the electronic document set to be detected;

[0010] A multi-dimensional detection module is used to calculate, for each electronic document in the electronic document set, the similarity between the electronic document and other electronic documents in each dimension according to the key element dimension, the surface text dimension, the text style dimension, the deep semantic dimension, the image content dimension and the document structure homology dimension, and obtain the detection result of each dimension;

[0011] The result integration module is used to integrate the detection results of the various dimensions to obtain comprehensive evaluation data, wherein the comprehensive evaluation data includes information about the similarities between electronic documents obtained by detection of different dimensions.

[0012] In a third aspect, the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein when the processor executes the computer program, the steps of the electronic document detection method as described in any one of the first aspects are implemented.

[0013] In a fourth aspect, the present application provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the electronic document detection method as described in any one of the first aspects.

[0014] The electronic document detection method, device, electronic device and storage medium provided by the present application are intended to solve the problems of limited detection dimensions and manual detection in bidding document detection. The method adopts a multi-dimensional detection method, covering the key element dimension, surface text dimension, text style dimension, deep semantic dimension, image content dimension and document structure homology dimension. For example, the key element dimension goes deep into the uncommon expressions in the vocabulary of the professional field, the image content dimension considers other pictures except specific pictures, and the similarity between bidding documents is calculated separately under each dimension. For example, the deep semantic dimension deeply analyzes the semantics and logical order of large paragraphs of text, which greatly broadens the detection scope.

[0015] Therefore, this application calculates similarity based on multiple dimensions and obtains test results, avoiding the subjectivity and inefficiency of manual testing. At the same time, by integrating the test results of each dimension to obtain comprehensive evaluation data, it can systematically process information about the similarity of bidding documents in different dimensions, overcoming the problem that it is difficult for humans to comprehensively consider them. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the present application or the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0017] Figure 1 is a flow chart of the electronic document detection method provided by the present application;

[0018] Figure 2 It is a schematic diagram of the detection dimensions provided by this application;

[0019] Figure 3It is a schematic diagram similar to the key elements provided in the embodiments of the present application;

[0020] Figure 4 It is a schematic diagram similar to the surface text provided in the embodiment of the present application;

[0021] Figure 5 It is a schematic diagram similar to the text style provided in the embodiment of the present application;

[0022] Figure 6 is a schematic diagram with deep semantic similarity provided by the embodiments of the present application;

[0023] Figure 7 is a schematic diagram with similar image content provided in the embodiments of the present application;

[0024] Figure 8 It is a schematic diagram of a document structure similar to that provided by an embodiment of the present application;

[0025] Fig. 9 It is a structural schematic diagram of the electronic document detection device provided by the present application;

[0026] Fig.10 It is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solutions and advantages of this application clearer, the technical solutions in this application will be clearly and completely described below in conjunction with the drawings in this application. Obviously, the described embodiments are part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0028] The terms "first", "second", etc. in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the terms used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein.

[0029] In order to solve the problem of limited detection dimensions and manual detection drawbacks in bidding document detection, this application proposes an electronic document detection method, device, electronic device and storage medium. This method uses multi-dimensional (key elements, surface text, text style, deep semantics, image content, document structure homology dimension) detection to solve the problem of limited bidding document detection dimensions and comprehensively cover various characteristics of bidding documents. At the same time, this method can solve the drawbacks of manual detection. It can quickly calculate similarity to improve efficiency, enhance objectivity without being affected by subjective factors, and integrate multi-dimensional results to improve accuracy, thereby more comprehensively and accurately reflecting the similarity between bidding documents.

[0030] Combine the following Figure 1-Figure 10 The present application describes an electronic document detection method, device, electronic device and storage medium.

[0031] Please refer to Figure 1 , Figure 2 , Figure 1 is a flow chart of the electronic document detection method provided by this application, Figure 2 is a schematic diagram of the detection dimensions provided by this application. An electronic document detection method includes:

[0032] S110: Obtain an electronic document set to be detected.

[0033] For example, multiple bidding documents may be obtained from an external data source. These bidding documents may be in PDF format, and the multiple bidding documents constitute the electronic document set to be detected.

[0034] S120, for each electronic document in the electronic document set, the similarity between the electronic document and other electronic documents in each dimension is calculated according to the key element dimension, surface text dimension, text style dimension, deep semantic dimension, image content dimension and document structure homology dimension, and the detection result of each dimension is obtained.

[0035] For example, if the electronic document set contains 10 bidding documents, then each of the 10 bidding documents will be tested. Among them, key element similarity refers to the similarity of key content parts, such as the use of the same uncommon expressions in professional field vocabulary in the bidding documents. Surface text similarity refers to the similarity of surface text content, such as the existence of two or more consistent expressions in paragraph text in the bidding documents. Text style similarity refers to the similarity in text style characteristics, such as the consistency of subheadings in large paragraphs of text without specific format requirements in the bidding documents. Deep semantic similarity refers to the similarity at the semantic level, such as the existence of a small deviation in the text content of a large paragraph of text in the bidding documents, but the expression semantics and logical order of explanation are exactly the same. Image content similarity refers to the image similarity, such as the existence of highly similar pictures in the bidding documents except for seals, logos, and background images. File structure homology (also known as DNA homology) similarity refers to the similarity in file structure related content, such as the same content of the same reference file is cited multiple times in the bidding documents.

[0036] It is understandable that a bidding document is a complex document containing multiple types of information. This application uses six dimensions (i.e., key element dimension, surface text dimension, text style dimension, deep semantic dimension, image content dimension, and document structure homology dimension) to detect bidding documents from different angles, avoiding incomplete detection due to focusing on a single type of information. For example, detection only from the surface text dimension may ignore the deeper similarities of documents in semantics, structure, etc., while the combination of these six dimensions can more comprehensively cover the various characteristics of bidding documents. In addition, different bidding projects have different focuses and requirements. Some projects may pay more attention to the accuracy of key elements, while others may have special requirements for text style. These six dimensions can adapt to this diversity and ensure that bidding documents can be effectively detected in various situations.

[0037] S130, integrating the detection results of each dimension to obtain comprehensive evaluation data, where the comprehensive evaluation data includes information about similarities between electronic documents obtained by detection of different dimensions.

[0038] Specifically, corresponding weight values ​​can be set for the key element dimension, surface text dimension, text style dimension, deep semantic dimension, image content dimension, and document structure homology dimension; then, according to the weight value corresponding to each dimension, the detection results of each dimension that have been obtained are calculated to obtain a comprehensive score, which is the comprehensive evaluation data. If the comprehensive score exceeds the preset alarm threshold, it will be marked as an abnormal situation.

[0039] For example, the weight value of the key element dimension is set to 20, the weight value of the surface text dimension is set to 20, the weight value of the text style dimension is set to 15, the weight value of the deep semantic dimension is set to 20, the weight value of the image content dimension is set to 15, and the weight value of the document structure homology dimension is set to 10. The warning threshold is set to 50. Assume that the similarity scores of bid documents A and B in each dimension are as follows:

[0040] Key element dimension: Assume that the similarity score of bid document A and bid document B in the key element dimension is 70%. Since the weight value of the key element dimension is 20, the contribution value of this dimension in the comprehensive score is 70%×20=14.

[0041] Surface text dimension: If the similarity score in the surface text dimension is 50%, and its weight value is 20, the contribution value of this dimension is 50%×20=10.

[0042] Text style dimension: Assuming that the similarity score in the text style dimension is 60%, the weight value is 15, and its contribution value is 60%×15=9.

[0043] Deep semantic dimension: If the similarity score of the deep semantic dimension is 80%, the weight value is 20, and the contribution value is 80%×20=16.

[0044] Image content dimension: Assuming that the similarity score of the image content dimension is 40%, the weight value is 15, and the contribution value is 40%×15=6.

[0045] Document structure homology dimension: If the similarity score of the document structure homology dimension is 30%, the weight value is 10, and the contribution value is 30%×10=3.

[0046] Calculate the composite score:

[0047] The comprehensive score is the sum of the contribution values ​​of each dimension, that is, 14+10+9+16+6+3=58.

[0048] Compared with the alarm threshold:

[0049] The known warning threshold is 50. Since the comprehensive score of 58 is greater than 50, this indicates that there is a high degree of similarity between bid document A and bid document B, and further review is needed to determine whether there is plagiarism or unfair competition.

[0050] The following is a detailed description of the similarity calculation for the above six dimensions: key element dimension, surface text dimension, text style dimension, deep semantic dimension, image content dimension, and document structure homology dimension.

[0051] In some examples, in the above step 120, the steps of calculating the similarity between the electronic document and other electronic documents in each dimension according to the key element dimension and obtaining the detection result of each dimension include:

[0052] S120a, for the key element dimension, identify the professional domain vocabulary set in the electronic document, and based on the identified professional domain vocabulary set, detect whether the electronic documents have the same uncommon expressions. This includes:

[0053] S120a1, identifying predefined key elements in each electronic document, where the predefined key elements are uncommon expressions in professional domain vocabulary.

[0054] For example, see Figure 3 , Figure 3 It is a schematic diagram similar to the key elements provided in the embodiment of the present application. Assume that there are two electronic documents for engineering bidding, document 1 on the left and document 2 on the right. Some professional terms in the engineering field, such as "ecological restoration technology", "ISO9001 international quality standard", etc.

[0055] In Document 1, the predefined key elements (not commonly used expressions) identified include "adopting advanced ecological restoration technology" and "ISO9001 international quality standards".

[0056] In Document 2, the predefined key elements (not commonly used expressions) identified include “advanced ecological restoration technology”, “ISO9001 international quality standards”, etc.

[0057] S120a2, selecting key element content in one of the electronic documents, and comparing it with corresponding key element content in other electronic documents to calculate the similarity value of the key elements.

[0058] For example, the above-mentioned document 1 is expressed as "adopting advanced ecological restoration technology", and document 2 is expressed as "advanced ecological restoration technology". It can be seen that the expressions of the two are very similar, except that document 1 has the verb "adopt". Assuming that the similarity is determined by calculating the proportion of identical characters (excluding punctuation marks and irrelevant function words), it is assumed that the calculated similarity value is 0.86. For the key element of "ISO9001 International Quality Standard": the expressions of document 1 and document 2 are exactly the same, so their similarity value is 1. In order to obtain the comprehensive similarity of document 1 and document 2 in the key element dimension, the average method can be used. That is, add the similarities of the two key elements, and then divide them by the number of key elements, that is, (0.86+1) / 2=0.93. 0.93 is the comprehensive similarity value of document 1 and document 2 in the key element dimension.

[0059] S120a3, storing the similarity results of each electronic document with other electronic documents in the key element dimension in a preset data structure, and obtaining the detection result of the key element dimension.

[0060] For example, a two-dimensional array (a preset data structure) can be used to store the results. Assume that document 1 is labeled as D1 and document 2 is labeled as D2. The detection result of the key element dimension is [[1,0.93],[0.93,1]].

[0061] In some embodiments, in the above step 120, the step of calculating the similarity between the electronic document and other electronic documents in each dimension according to the surface text dimension, and obtaining the detection result of each dimension includes:

[0062] S120b, for the surface text dimension, extract paragraph text expressions from the electronic documents, and perform a search and match operation between the paragraph text expressions of each electronic document to determine whether there are at least two paragraph texts with the same expression. This includes:

[0063] S120b1, select the text content of one of the electronic documents and compare it with the text content of the remaining electronic documents to calculate the similarity value of the surface text; wherein, during the comparison process, if the proportion of the same parts reaches a preset threshold, it is determined to meet the similarity requirement of the surface text, and the preset threshold is the proportion corresponding to at least two paragraphs of text with consistent expressions.

[0064] For example, see Figure 4 , Figure 4 It is a schematic diagram similar to the surface text provided in the embodiment of the present application. Suppose there are two electronic documents for engineering bidding, Document 1 on the left and Document 2 on the right. The text content of Document 1 is "(10) Use machinery to clean up weeds, domestic garbage, silt and various undesirable soils on the surface, and discard them by vehicles as required. 1) The demolition of buildings and underground structures should be carried out according to their structural characteristics. When clearing the site, pay attention to protecting cultural relics and historical sites, and comply with relevant national laws and regulations; 2) If pipelines or other abnormal objects are found on the ground, underground or underwater in the construction site, the relevant departments should be requested to cooperate in advance." The text content of Document 2 is "(10) Use machinery to clean up weeds, domestic garbage, silt and various undesirable soils on the surface. And discard them by vehicles as required. 1) The demolition of buildings and underground structures should be carried out according to their structural characteristics. When clearing the site, pay attention to protecting cultural relics and historical sites, and comply with relevant national laws and regulations; 2) If pipelines or other abnormal objects are found on the ground, underground or underwater in the construction site, the relevant departments should be requested to cooperate in advance."

[0065] First, you need to determine the proportion of the same parts. You can determine the similarity by calculating the ratio of the number of identical characters to the total number of characters:

[0066] Document 1: about 69 characters;

[0067] Document 2: about 67 characters;

[0068] Same number of characters (punctuation removed): about 63 characters;

[0069] The ratio of the same parts is: Document 1: 63 / 69≈0.913 (calculated with the number of characters in Document 1 as the denominator); Document 2: 63 / 67≈0.940 (calculated with the number of characters in Document 2 as the denominator). Assuming the preset threshold is 0.8, since the calculated ratio is greater than the preset threshold, it is determined to meet the surface text similarity requirement.

[0070] S120b2, storing the similarity results of each electronic document with other electronic documents in the surface text dimension in a preset data structure, and obtaining the detection result of the surface text dimension.

[0071] For example, a two-dimensional array (a preset data structure) can be used to store the results. Assume that document 1 is labeled as D1 and document 2 is labeled as D2. The detection result of the surface text dimension is [[1,0.913], [0.913,1]].

[0072] In some embodiments, in the above step 120, the step of calculating the similarity between the electronic document and other electronic documents in each dimension according to the text style dimension, and obtaining the detection result of each dimension includes:

[0073] S120c, with respect to the text style dimension, extract subheadings from large texts without specific format requirements in the electronic document, and perform consistency determination on the extracted subheadings. This includes:

[0074] S120c1, select one of the electronic documents, analyze the subheading features in the large text without specific format requirements, and quantitatively compare the subheading features with the corresponding features of the remaining electronic documents to calculate the text style similarity value.

[0075] Among them, no specific format requirements means that when analyzing the characteristics of subheadings in large sections of electronic documents, subheadings have a great deal of freedom in terms of format. They are neither limited to specific typesetting styles, such as fonts, font sizes, bold or underline styles, nor are they constrained by specific structures. The structural position in the document can be flexible and changeable. At the same time, in terms of content presentation, the content of subheadings can be freely written according to the author's expression habits, and is not restricted by fixed expression patterns or specific vocabulary.

[0076] Please refer to Figure 5 , Figure 5It is a schematic diagram with a text style similar to that provided in the embodiment of the present application. Assume that there are two electronic documents for engineering bidding, Document 1 on the left and Document 2 on the right. The subtitle of Document 1 is "1. Construction period; 2. Guarantee system; 3. Key process; 4. Specific arrangements are as follows:". The subtitle of Document 2 is "1. Construction period; 2. Guarantee system; 3. Key process; 4. Specific arrangements are as follows:".

[0077] First, analyze the characteristics of the subheadings of Document 1. The subheadings of Document 1 are "1. Construction period; 2. Guarantee system; 2. Key process; 4. Specific arrangements are as follows:", with a total of 4 subheadings. Then, analyze the characteristics of the subheadings of Document 2 and make a quantitative comparison with Document 1. The subheadings of Document 2 are "1. Construction period; 2. Guarantee system; 3. Key process; 4. Specific arrangements are as follows:", also with 4 subheadings. Finally, when calculating the similarity value, you can consider multiple aspects, such as quantifying from three aspects: the number of titles, the content of the titles, and the accuracy of the title numbers (each aspect has the same weight, each accounting for 1 / 3). As follows:

[0078] In terms of the number of titles: Document 1 and Document 2 both have 4 subtitles, so the similarity of this part is 1.

[0079] In terms of title content: the title content is exactly the same, so the similarity of this part is 1.

[0080] In terms of the accuracy of the title number: the title numbers of the two documents are accurate, so the similarity of this part is 1.

[0081] The comprehensive calculation of text style similarity is: (1+1+1)÷3=1.

[0082] S120c2, storing the similarity results of each electronic document with other electronic documents in the text style dimension in a preset data structure, and obtaining the detection result of the text style dimension.

[0083] For example, a two-dimensional array is also used to store the results, document 1 is marked as D1, and document 2 is marked as D2. The detection result of the text style dimension is [[1,1],[1,1]].

[0084] In some examples, in the above step 120, the step of calculating the similarity between the electronic document and other electronic documents in each dimension according to the deep semantic dimension and obtaining the detection result of each dimension includes:

[0085] S120d, for deep semantic dimensions, locate large paragraphs of text in electronic documents, and for large paragraphs of text with a small amount of content deviation, analyze their semantic expressions and logical elaboration order to determine whether the two are consistent. This includes:

[0086] S120d1, preprocessing the large text paragraphs in each electronic document to identify key semantic elements and logical structure elements therein, and converting the preprocessed text of each electronic document into a semantic vector representation.

[0087] Please refer to Figure 6 , Figure 6 It is a schematic diagram with deep semantic similarity provided by the embodiment of the present application. Assume that there are two electronic documents for engineering bidding, document 1 on the left and document 2 on the right. The text of document 1 is: "During the implementation of the project, we will focus on green, ecological and sustainable design concepts to ensure that the green coverage rate reaches more than 80%.", "In the selection of soil and vegetation, we plan to use local varieties to adapt to local climatic conditions while reducing maintenance costs. Plant planting will combine seasonal flowers with evergreen plants to achieve landscape effects in all seasons.", "For hardened paving, we will use highly permeable materials to enhance the air permeability and rainwater recycling capacity of the ground." The text of document 2 is: "The engineering design will follow the principles of ecological priority and humanized design to ensure that the green area accounts for no less than 75%.", "In terms of plant configuration, the plan also tends to use native plants to ensure the adaptability of vegetation and the sustainability of growth. We will create a variety of visual effects through careful plant matching and seasonal plant matching.", "In ground paving, environmentally friendly materials, such as permeable bricks, will also be given priority to achieve natural infiltration of rainwater.".

[0088] First, the text is preprocessed and converted into a semantic vector representation:

[0089] Document 1:

[0090] For the first paragraph of the text in Document 1, "During the implementation of the project, emphasis will be placed on green, ecological, and sustainable design concepts to ensure that the green coverage rate reaches more than 80%.", the key semantic elements identified are "project implementation", "green ecological sustainable concept", and "green coverage rate of more than 80%". The logical structure is to explain the design concepts and green coverage rate targets in project implementation.

[0091] The second paragraph reads, "In terms of soil and vegetation selection, we plan to use local varieties to adapt to local climate conditions while reducing maintenance costs. Plant planting will combine seasonal flowers with evergreen plants to achieve landscape effects in all seasons." The key semantic elements include "soil and vegetation selection", "local varieties", "adapt to local climate", "reduce maintenance costs", "match seasonal flowers and evergreen plants", and "landscape effects in all seasons". The logical structure is about the considerations for soil and vegetation selection and the goals of plant planting matching.

[0092] The third paragraph reads "For hardened paving, we will use highly permeable materials to enhance the ground's air permeability and rainwater recycling capacity." The key semantic elements are "hardened paving," "permeable materials," "ground air permeability," and "rainwater recycling capacity." The logical structure is to explain the selection of hardened paving materials and their functions.

[0093] A semantic vector model (such as a pre-trained word vector model combined with certain semantic combination rules) can be used to convert document 1 into a semantic vector representation. Assume that this semantic vector is v1→.

[0094] Document 2:

[0095] The first paragraph reads "The engineering design will follow the principles of ecological priority and humanized design, ensuring that the green area accounts for no less than 75%." The key semantic elements are "engineering design", "ecological priority and humanized principles", and "green area accounts for more than 75%". The logical structure is the principles followed by the engineering design and the green area targets.

[0096] The second paragraph reads, "In terms of plant configuration, the plan also tends to use native plants to ensure the adaptability of vegetation and sustainable growth. We will create diverse visual effects through careful plant matching and seasonal plant matching." The key semantic elements include "plant configuration," "native plants," "vegetation adaptability," "seasonal plant matching," and "diversified visual effects." The logical structure is about the considerations and goals of plant configuration.

[0097] The third paragraph reads, "In ground paving, priority will be given to environmentally friendly materials, such as permeable bricks, to achieve natural infiltration of rainwater." The key semantic elements are "ground paving", "environmentally friendly materials", and "natural infiltration of rainwater". The logical structure is to explain the selection of ground paving materials and their functions.

[0098] The same semantic vector model can be used to convert document 2 into a semantic vector representation, assuming that this semantic vector is v2→.

[0099] S120d2, selecting a semantic vector of one of the electronic documents, and comparing it with the semantic vectors of the remaining electronic documents to calculate a similarity value of the semantic vectors.

[0100] For example, a suitable vector similarity calculation method (such as cosine similarity) is used. Assume that the function for calculating cosine similarity is cosine_similarity(v1→,v2→). The calculated similarity value is cosine_similarity(v1→,v2→)=0.85 (0.85 is a hypothetical calculation result, and the actual calculation will be based on the specific semantic vector and calculation method). This value indicates the similarity between document 1 and document 2 in the deep semantic dimension. The higher the value, the more similar the semantics.

[0101] S120d3, storing the similarity results of each electronic document with other electronic documents in the deep semantic dimension in a preset data structure, and obtaining the detection result of the deep semantic dimension.

[0102] For example, a two-dimensional array is used to store the results. Assume that document 1 is labeled D1 and document 2 is labeled D2. The detection result of the deep semantic dimension is [[1,0.85],[0.85,1]].

[0103] In some examples, in the above step 120, the step of calculating the similarity between the electronic document and other electronic documents in each dimension according to the image content dimension, and obtaining the detection result of each dimension includes:

[0104] S120e, in terms of image content, searches for images other than seals, logos, and background images in the electronic document, and performs feature analysis on the searched images to determine whether they are similar. This includes:

[0105] S120e1, for an electronic document containing image content, extract a target image except a seal, a logo, and a background image from the electronic document, and obtain a feature vector of the target image.

[0106] Please refer to Figure 7 , Figure 7 It is a schematic diagram with similar image content provided in the embodiment of the present application. Assume that there are two electronic documents for engineering bidding, document 1 on the left and document 2 on the right. Picture 1 of document 1 shows: "The reinforcement and plate reinforcement are tied tightly, reinforcement reinforcement is added on the top, and the bent steel bar is above the bottom reinforcement.". Picture 2 of document 2 also shows: "The reinforcement and plate reinforcement are tied tightly, reinforcement reinforcement is added on the top, and the bent steel bar is above the bottom reinforcement.".

[0107] First, extract the target image and obtain the feature vector. You can use an image feature extraction algorithm (such as a convolutional neural network) to obtain the feature vector of picture 1 in document 1. Assume that this feature vector is v1→, which contains the numerical representation of the image's features in terms of color, texture, shape, etc. The same image feature extraction algorithm is used to obtain the feature vector of picture 2 from document 2. Assume that this feature vector is v2→.

[0108] S120e2, selecting a feature vector of a target image in one of the electronic documents, comparing it with the feature vectors of the target images in the remaining electronic documents, and calculating a similarity value of the images.

[0109] For example, a suitable image feature vector similarity calculation method is used, such as Euclidean distance or cosine similarity. If the cosine similarity calculation method is used, the similarity calculation function is set to cosine_similarity(v1→,v2→). The calculated similarity value is cosine_similarity(v1→,v2→)=0.9 (0.9 is an assumed result here, and the actual result depends on the specific image content and feature extraction algorithm). This value indicates that the two images (image 1 and image 2) have a high similarity in features.

[0110] S120e3, storing the similarity results of each electronic document with other electronic documents in the image content dimension in a preset data structure, and obtaining the detection result of the image content dimension.

[0111] For example, a two-dimensional array is used to store the results. Assume that document 1 is labeled D1 and document 2 is labeled D2. The detection result of the image content dimension is [[1,0.9],[0.9,1]].

[0112] In some examples, in the above step 120, the steps of calculating the similarity between the electronic document and other electronic documents in each dimension according to the document structure homology dimension and obtaining the detection result in each dimension include:

[0113] S120f, with respect to the document structure homology dimension, parse the referenced part in the electronic document, and perform matching operations on the referenced content to detect whether the same content of the same reference file is cited multiple times. This includes:

[0114] S120f1, parsing the structure of each electronic document to identify the referenced part in the electronic document, and marking the referenced reference file and the specific content of the reference.

[0115] Please refer to Figure 8 , Figure 8It is a schematic diagram of the document structure homology provided in the embodiment of the present application. Suppose there are two electronic documents for engineering bidding, document 1 on the left and document 2 on the right. The reference documents cited by document 1 and document 2 are "Technical Code for Mechanical Connection of Steel Bars JGJ107-2016 (Soujianzhu.com)", "Rebar Worker Bidding Documents - Docin.com (Docin.com)", and "Rebar Worker Technical Documents ccll.org.cn [2020-09-04]".

[0116] First, parse the electronic document structure and mark the referenced parts:

[0117] Document 1:

[0118] Perform structural analysis on Document 1 to identify the referenced documents and specific referenced contents. For the referenced document "Technical Code for Mechanical Connection of Rebars JGJ107-2016 (Soujianzhu.com)", mark the specific location (such as chapter, paragraph, etc.) where the document is referenced in Document 1, as well as the specific technical terms or related explanatory contents referenced. For "Rebar Worker Bidding Documents - DocIn (DocIn)", also mark the referenced location and specific referenced contents, such as some standard terms, processes or technical requirements for rebar worker bidding. For "Rebar Worker Technical Documents ccll.org.cn [2020-09-04]", determine the parts of the document that are referenced in Document 1, such as specific process descriptions, quality control standards, etc., and mark them.

[0119] Document 2:

[0120] Parse the structure of Document 2 in the same way and identify the referenced parts. For "Technical Code for Mechanical Connection of Steel Bars JGJ107-2016 (Sou Jianzhu.com)", mark the referenced location and specific referenced content in Document 2. Also mark the referenced location and specific content for "Rebar Worker Bidding Documents - Docin.com (Docin.com)" and "Rebar Worker Technical Documents ccll.org.cn [2020-09-04]".

[0121] S120f2, based on the structural information obtained by the analysis, select the structural information of one of the electronic documents, and compare it with the structural information of the remaining electronic documents to calculate the similarity value of the electronic document structure.

[0122] For example, compare the structural information of Document 1 and Document 2. The similarity can be calculated from the types of referenced files, the number of citations of each referenced file, the similarity of the referenced content, etc. Assume that the types of referenced files are exactly the same (here there are 3 referenced files), the number of citations of each referenced file in the two documents is also the same (for example, each file is only cited once in Document 1 and Document 2), and the similarity of the referenced content is very high (assuming that after detailed comparison, it is found that the specific content of the reference is basically the same in key parts).

[0123] A calculation method can be set, for example: if the same type of cited file is scored 0.3 points, the same number of citations is scored 0.3 points, and high similarity of cited content is scored 0.4 points, then the similarity value between document 1 and document 2 in the document structure homology dimension is 0.3+0.3+0.4=1.

[0124] S120f3, storing the similarity results of each electronic document with other electronic documents in the document structure homology dimension in a preset data structure, and obtaining the detection result of the document structure homology dimension.

[0125] For example, a two-dimensional array is used to store the results. Assume that document 1 is labeled D1 and document 2 is labeled D2. The detection result of the document structure homology dimension is [[1,1],[1,1]].

[0126] Finally, the detection results of the key element dimension, the surface text dimension, the text style dimension, the deep semantic dimension, the image content dimension and the document structure homology dimension obtained above are integrated to obtain comprehensive evaluation data.

[0127] For example, the detection results of each dimension of document 1 and document 2 are:

[0128] The detection results of the key element dimensions are [[1,0.93],[0.93,1]];

[0129] The detection result of the surface text dimension is [[1,0.913],[0.913,1]];

[0130] The detection result of the text style dimension is [[1,1],[1,1]];

[0131] The detection results of the deep semantic dimension are [[1,0.85],[0.85,1]];

[0132] The detection result of the image content dimension is [[1,0.9],[0.9,1]];

[0133] The detection result of the document structure homology dimension is [[1,1],[1,1]].

[0134] Assume that the weight value of the key element dimension is set to 20, the weight value of the surface text dimension is set to 20, the weight value of the text style dimension is set to 15, the weight value of the deep semantic dimension is set to 20, the weight value of the image content dimension is set to 15, and the weight value of the document structure homology dimension is set to 10. The alarm threshold is set to 50.

[0135] Then, the specific calculation of the comprehensive evaluation data is as follows:

[0136] First, calculate the weighted score of each dimension:

[0137] (1) Key element dimensions:

[0138] The detection result of the key element dimension is [[1,0.93], [0.93,1]], and the similarity between document 1 and document 2 is 0.93. The weight value of this dimension is 20, and the weighted score = 0.93 × 20 = 18.6.

[0139] (2) Surface text dimension:

[0140] The detection result of the surface text dimension is [[1,0.913], [0.913,1]], and the similarity between document 1 and document 2 is 0.913. The weight value is 20, and the weighted score = 0.913 × 20 = 18.26.

[0141] (3) Text style dimension:

[0142] The detection result of the text style dimension is [[1,1],[1,1]], and the similarity between document 1 and document 2 is 1. The weight value is 15, and the weighted score = 1×15 = 15.

[0143] (4) Deep semantic dimension:

[0144] The detection result of the deep semantic dimension is [[1,0.85], [0.85,1]], and the similarity between document 1 and document 2 is 0.85. The weight value is 20, and the weighted score = 0.85 × 20 = 17.

[0145] (5) Image content dimension:

[0146] The detection result of the image content dimension is [[1,0.9],[0.9,1]], and the similarity between document 1 and document 2 is 0.9. The weight value is 15, and the weighted score = 0.9 × 15 = 13.5.

[0147] (6) Document structure homology dimension:

[0148] The detection result of the document structure homology dimension is [[1,1],[1,1]], and the similarity between document 1 and document 2 is 1. The weight value is 10, and the weighted score = 1×10 = 10.

[0149] Calculate the comprehensive evaluation data: The comprehensive evaluation data is the sum of the weighted scores of each dimension, that is, 18.6+18.26+15+17+13.5+10=92.36.

[0150] Since the comprehensive evaluation data 92.36>50 (warning threshold), this indicates that Document 1 and Document 2 are highly similar in multiple dimensions, and there may be certain correlation or similarity issues, which require further review (the specific review direction depends on the actual application scenario).

[0151] It should be noted that if the electronic document set to be detected contains more than two documents, the above method can also be used to perform pairwise comparison calculations, which will not be described in detail here.

[0152] It is understandable that the electronic document detection method described in this application can be based on a multimodal radar detection model, which has the ability to integrate multiple detection methods and can comprehensively analyze potential anomalies in the bidding documents. It can include the design of a multimodal radar detection model (such as key element dimensions, surface text dimensions, text style dimensions, deep semantic dimensions, image content dimensions, and document structure homology dimensions), specific implementation methods, and its application in bidding document anomaly detection. However, it does not mean that this application can only rely on a multimodal radar detection model. Other models or methods can also be used to achieve the same goal.

[0153] The electronic document detection device provided in the present application is described below. The electronic document detection device described below and the electronic document detection method described above can be referenced to each other.

[0154] Please refer to Fig. 9 , Fig. 9 9 is a schematic diagram of the structure of the electronic document detection device provided by the present application. An electronic document detection device 900 includes a data acquisition module 910 , a multi-dimensional detection module 920 and a result integration module 930 .

[0155] Exemplarily, the data acquisition module 910 is used to acquire a set of electronic documents to be detected.

[0156] Exemplarily, the multi-dimensional detection module 920 is used to calculate the similarity between each electronic document in the electronic document set and the remaining electronic documents in each dimension according to the key element dimension, surface text dimension, text style dimension, deep semantic dimension, image content dimension and document structure homology dimension, and obtain the detection result of each dimension.

[0157] Exemplarily, the result integration module 930 is used to integrate the detection results of the various dimensions to obtain comprehensive evaluation data, where the comprehensive evaluation data includes information about similarities between electronic documents obtained from detections of different dimensions.

[0158] Exemplarily, the multi-dimensional detection module 920 is further used for:

[0159] For the key element dimension, identify the professional domain vocabulary set in the electronic document, and based on the identified professional domain vocabulary set, detect whether the electronic documents have the same uncommon expressions;

[0160] With respect to the surface text dimension, paragraph text expressions are extracted from the electronic documents, and a search and matching operation is performed between the paragraph text expressions of each electronic document to determine whether there are at least two paragraph texts with the same expression;

[0161] In terms of text style, subheadings in large texts without specific format requirements in electronic documents are extracted, and consistency judgment is performed on the extracted subheadings;

[0162] In terms of deep semantics, we locate large sections of text in electronic documents. For large sections of text with small content deviations, we analyze their semantic expressions and logical order to determine whether the two are consistent.

[0163] In terms of image content, search for images other than seals, logos, and background images in the electronic document, and perform feature analysis on the searched images to determine whether they are similar;

[0164] In terms of document structure homology, the referenced parts in the electronic document are parsed, and the referenced contents are matched to detect whether the same content of the same reference file is cited multiple times.

[0165] Exemplarily, the multi-dimensional detection module 920 is further used for:

[0166] Identify predefined key elements in each electronic document, wherein the predefined key elements are uncommon expressions in professional domain vocabulary;

[0167] Selecting key element content in one of the electronic documents and comparing it with corresponding key element content in other electronic documents to calculate the similarity value of the key elements;

[0168] The similarity results of each electronic document with other electronic documents in the key element dimension are stored in a preset data structure to obtain the detection results of the key element dimension.

[0169] Exemplarily, the multi-dimensional detection module 920 is further used for:

[0170] Selecting the text content of one of the electronic documents and comparing it with the text content of the remaining electronic documents to calculate the similarity value of the surface text; wherein, during the comparison process, if the proportion of the same parts reaches a preset threshold, it is determined that the surface text similarity requirement is met, and the preset threshold is the proportion of at least two paragraphs of text with consistent expressions;

[0171] The similarity results of each electronic document in the surface text dimension with other electronic documents are stored in a preset data structure to obtain the detection result of the surface text dimension.

[0172] Exemplarily, the multi-dimensional detection module 920 is further used for:

[0173] Select one of the electronic documents, analyze the subheading features in the large text without specific format requirements, and quantitatively compare the subheading features with the corresponding features of the remaining electronic documents to calculate the text style similarity value;

[0174] The similarity results of each electronic document with other electronic documents in the text style dimension are stored in a preset data structure to obtain the detection result of the text style dimension.

[0175] Exemplarily, the multi-dimensional detection module 920 is further used for:

[0176] Preprocessing a large section of text in each electronic document to identify key semantic elements and logical structure elements therein, and converting the preprocessed text of each electronic document into a semantic vector representation;

[0177] Selecting a semantic vector of one of the electronic documents and comparing it with the semantic vectors of the remaining electronic documents to calculate a similarity value of the semantic vectors;

[0178] The similarity results of each electronic document with other electronic documents in the deep semantic dimension are stored in a preset data structure to obtain the detection result of the deep semantic dimension.

[0179] Exemplarily, the multi-dimensional detection module 920 is further used for:

[0180] For an electronic document containing image content, extract a target image other than a seal, a logo, and a background image from the electronic document, and obtain a feature vector of the target image;

[0181] Selecting a feature vector of a target image in one of the electronic documents, comparing it with feature vectors of target images in other electronic documents, and calculating a similarity value of the images;

[0182] The similarity results of each electronic document in the image content dimension with other electronic documents are stored in a preset data structure to obtain the detection result of the image content dimension.

[0183] Exemplarily, the multi-dimensional detection module 920 is further used for:

[0184] Parse the structure of each electronic document to identify the referenced parts within the electronic document and mark the referenced references and the specific content of the references;

[0185] According to the structural information obtained by the analysis, the structural information of one of the electronic documents is selected, and the structural information is compared with the structural information of the other electronic documents to calculate the similarity value of the electronic document structure;

[0186] The similarity results of each electronic document with other electronic documents in the document structure homology dimension are stored in a preset data structure to obtain the detection results of the document structure homology dimension.

[0187] Exemplarily, the result integration module 930 is further used for:

[0188] Set corresponding weight values ​​for the key element dimension, surface text dimension, text style dimension, deep semantic dimension, image content dimension, and document structure homology dimension respectively;

[0189] According to the weight value corresponding to each dimension, the test results of each dimension that have been obtained are calculated to obtain a comprehensive score, which is the comprehensive evaluation data;

[0190] If the combined score exceeds the preset alarm threshold, it is marked as an abnormal situation.

[0191] It should be noted here that the above-mentioned electronic document detection device provided in the embodiment of the present application can implement all the method steps implemented in the above-mentioned method embodiment, and can achieve the same technical effect. The parts and beneficial effects of this embodiment that are the same as those in the method embodiment will not be described in detail here.

[0192] Fig.10 is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application, such as Fig.10 As shown, the electronic device may include: a processor (Processor) 1010, a communication interface (Communications Interface) 1020, a memory (Memory) 1030 and a communication bus 1040, wherein the processor 1010, the communication interface 1020, and the memory 1030 communicate with each other through the communication bus 1040. The processor 1010 may call the logic instructions in the memory 1030 to execute the electronic document detection method.

[0193] In addition, the logic instructions in the above-mentioned memory 1030 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art, and the computer software product is stored in a storage medium, including several instructions to ensure that a computer device (which can be a personal computer, a server, or a network device, etc.) executes all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0194] On the other hand, the present application also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the electronic document detection method provided by the above methods.

[0195] On the other hand, the present application also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when the computer program is executed by a processor to execute the electronic document detection method provided above.

[0196] The embodiments of the present application provide an electronic device, a computer program product, and a processor-readable storage medium, on which the computer program stored enables the processor to implement all the method steps implemented in the above-mentioned method embodiments and to achieve the same technical effects. The parts and beneficial effects of this embodiment that are the same as those in the method embodiments will not be described in detail herein.

[0197] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0198] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions to ensure that a computer device (which can be a personal computer, a server, or a network device, etc.) executes the methods described in each embodiment or some parts of the embodiments.

[0199] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An electronic document detection method, characterized in that: The method comprises: Obtaining the electronic document set to be tested; For each electronic document in the electronic document set, the similarity between the electronic document and other electronic documents in each dimension is calculated according to the key element dimension, the surface text dimension, the text style dimension, the deep semantic dimension, the image content dimension and the document structure homology dimension, and the detection result of each dimension is obtained; The detection results of the various dimensions are integrated to obtain comprehensive evaluation data, which includes information about similarities between electronic documents obtained from detections of different dimensions.

2. The electronic document detection method according to claim 1, characterized in that: The step of calculating the similarity between the electronic document and other electronic documents in each dimension according to the key element dimension, the surface text dimension, the text style dimension, the deep semantic dimension, the image content dimension and the document structure homology dimension to obtain the detection result of each dimension includes: For the key element dimension, identify the professional domain vocabulary set in the electronic document, and based on the identified professional domain vocabulary set, detect whether the electronic documents have the same uncommon expressions; With respect to the surface text dimension, paragraph text expressions are extracted from the electronic documents, and a search and matching operation is performed between the paragraph text expressions of each electronic document to determine whether there are at least two paragraph texts with the same expression; In terms of text style, subheadings in large texts without specific format requirements in electronic documents are extracted, and consistency judgment is performed on the extracted subheadings; In terms of deep semantics, we locate large sections of text in electronic documents. For large sections of text with small content deviations, we analyze their semantic expressions and logical order to determine whether the two are consistent. In terms of image content, search for images other than seals, logos, and background images in the electronic document, and perform feature analysis on the searched images to determine whether they are similar; In terms of document structure homology, the referenced parts in the electronic document are parsed, and the referenced contents are matched to detect whether the same content of the same reference file is cited multiple times.

3. The electronic document detection method according to claim 2, characterized in that: The step of calculating the similarity between the electronic document and other electronic documents in each dimension according to the key element dimension to obtain the detection result of each dimension includes: Identify predefined key elements in each electronic document, wherein the predefined key elements are uncommon expressions in professional domain vocabulary; Selecting key element content in one of the electronic documents and comparing it with corresponding key element content in other electronic documents to calculate the similarity value of the key elements; The similarity results of each electronic document with other electronic documents in the key element dimension are stored in a preset data structure to obtain the detection results of the key element dimension.

4. The electronic document detection method according to claim 2, characterized in that: The step of calculating the similarity between the electronic document and other electronic documents in each dimension according to the surface text dimension and obtaining the detection result of each dimension comprises: Selecting the text content of one of the electronic documents and comparing it with the text content of the remaining electronic documents to calculate the similarity value of the surface text; wherein, during the comparison process, if the proportion of the same parts reaches a preset threshold, it is determined that the surface text similarity requirement is met, and the preset threshold is the proportion of at least two paragraphs of text with consistent expressions; The similarity results of each electronic document in the surface text dimension with other electronic documents are stored in a preset data structure to obtain the detection result of the surface text dimension.

5. The electronic document detection method according to claim 2, characterized in that: The step of calculating the similarity between the electronic document and other electronic documents in each dimension according to the text style dimension and obtaining the detection result of each dimension includes: Select one of the electronic documents, analyze the subheading features in the large text without specific format requirements, and quantitatively compare the subheading features with the corresponding features of the remaining electronic documents to calculate the text style similarity value; The similarity results of each electronic document with other electronic documents in the text style dimension are stored in a preset data structure to obtain the detection result of the text style dimension.

6. The electronic document detection method according to claim 2, characterized in that: The step of calculating the similarity between the electronic document and other electronic documents in each dimension according to the deep semantic dimension to obtain the detection result of each dimension includes: Preprocessing a large section of text in each electronic document to identify key semantic elements and logical structure elements therein, and converting the preprocessed text of each electronic document into a semantic vector representation; Selecting a semantic vector of one of the electronic documents and comparing it with the semantic vectors of the remaining electronic documents to calculate a similarity value of the semantic vectors; The similarity results of each electronic document with other electronic documents in the deep semantic dimension are stored in a preset data structure to obtain the detection result of the deep semantic dimension.

7. The electronic document detection method according to claim 2, characterized in that: The step of calculating the similarity between the electronic document and other electronic documents in each dimension according to the image content dimension to obtain the detection result of each dimension includes: For an electronic document containing image content, extract a target image other than a seal, a logo, and a background image from the electronic document, and obtain a feature vector of the target image; Selecting a feature vector of a target image in one of the electronic documents, comparing it with feature vectors of target images in other electronic documents, and calculating a similarity value of the images; The similarity results of each electronic document in the image content dimension with other electronic documents are stored in a preset data structure to obtain the detection result of the image content dimension.

8. The electronic document detection method according to claim 2, characterized in that: The steps of calculating the similarity between the electronic document and other electronic documents in each dimension according to the document structure homology dimension and obtaining the detection result in each dimension include: Parse the structure of each electronic document to identify the referenced parts within the electronic document and mark the referenced references and the specific content of the references; According to the structural information obtained by the analysis, the structural information of one of the electronic documents is selected, and the structural information is compared with the structural information of the other electronic documents to calculate the similarity value of the electronic document structure; The similarity results of each electronic document with other electronic documents in the document structure homology dimension are stored in a preset data structure to obtain the detection results of the document structure homology dimension.

9. The electronic document detection method according to claim 1, characterized in that: The step of integrating the detection results of the various dimensions to obtain comprehensive evaluation data includes: Set corresponding weight values ​​for the key element dimension, surface text dimension, text style dimension, deep semantic dimension, image content dimension, and document structure homology dimension respectively; According to the weight value corresponding to each dimension, the test results of each dimension that have been obtained are calculated to obtain a comprehensive score, which is the comprehensive evaluation data; If the combined score exceeds the preset alarm threshold, it is marked as an abnormal situation.

10. An electronic document detection device, characterized in that: The device comprises: A data acquisition module, used to acquire the electronic document set to be detected; A multi-dimensional detection module is used to calculate, for each electronic document in the electronic document set, the similarity between the electronic document and other electronic documents in each dimension according to the key element dimension, the surface text dimension, the text style dimension, the deep semantic dimension, the image content dimension and the document structure homology dimension, and obtain the detection result of each dimension; The result integration module is used to integrate the detection results of the various dimensions to obtain comprehensive evaluation data, wherein the comprehensive evaluation data includes information about the similarities between electronic documents obtained by detection of different dimensions.

11. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the steps of the electronic document detection method according to any one of claims 1 to 9 are implemented.

12. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the electronic document detection method according to any one of claims 1 to 9 are implemented.

Images