A material review processing method and system based on AI model

Abstract

The present invention discloses a material review processing method and system based on an AI model, the method comprising obtaining a to-be-reviewed file input by a user, dividing the to-be-reviewed file into a digital file, an image file, and a text file according to the file type; generating a target element label according to the image file, associating the text fragment segmented from the text file with the corresponding data information in the digital file with the corresponding target element label, and forming a plurality of to-be-reviewed information groups; conducting a step-by-step review of the to-be-reviewed information groups according to the corresponding review conditions in the review standard database, and generating an electronic review report according to the review results of each to-be-reviewed information group. The present invention uses different types of to-be-reviewed materials submitted by users to be comprehensively evaluated, and extracts the key content of the to-be-reviewed materials in combination with the information of the image file, the text file, and the digital file, so as to improve the reliability of the review results.

Description

Technical Field

[0001] The present invention relates to the field of material review technology, and in particular to a material review processing method and system based on an AI model. Background Art

[0002] With the continuous development of project evaluation technology, review platforms that automatically evaluate project materials are gradually being used in project evaluation.

[0003] In existing evaluation platforms, the text content in the documents submitted by users is often detected and analyzed, so that a feasibility evaluation report is issued according to the review direction required by the user. However, as the projects to be evaluated gradually increase, the evaluation materials provided by users are not limited to text materials. It is difficult to comprehensively evaluate the feasibility of a project by only evaluating text materials, which can easily lead to inaccurate evaluation and cause project losses.

[0004] It can be seen that how to conduct a more accurate and comprehensive review of project materials has become a technical problem that technical personnel in this field need to solve urgently. Summary of the invention

[0005] The present invention provides an AI model-based material review processing method and system for performing a comprehensive evaluation on review materials submitted by users.

[0006] In order to solve the above technical problems, an embodiment of the present invention provides a material review processing method based on an AI model, comprising:

[0007] The files to be reviewed input by the user are obtained, and the files to be reviewed are divided into digital files, image files and text files according to the file types.

[0008] The image file is identified and analyzed by using an AI image recognition algorithm to obtain content information contained in the image file.

[0009] Semantic recognition is performed on the content information, and element labels corresponding to each content information are generated according to the recognition result, and element labels with similar semantics are merged to obtain each target element label.

[0010] The text content in the text file is semantically segmented to obtain text fragments with different semantics, and each of the text fragments and the corresponding data information in the digital file is associated with the corresponding target element label to form a plurality of information groups to be reviewed; the information groups to be reviewed are designed to be associated step by step according to the information entropy of the corresponding text fragments to form an information tree directory.

[0011] The review content is determined according to the review objectives of the documents to be reviewed and the target element labels of each information group to be reviewed, and the corresponding review conditions in the review standard database are obtained according to the review content. The information groups to be reviewed in the information tree directory are reviewed step by step according to the review conditions, and an electronic review report is generated according to the review results of each information group to be reviewed.

[0012] Furthermore, the image file is identified and analyzed by the AI ​​image recognition algorithm to obtain content information contained in the image file, including:

[0013] The image data in the image file is divided into dynamic images and static images.

[0014] The difference change rate between adjacent continuous frames of the dynamic image is calculated, and key frames are extracted according to the difference change rate to obtain a plurality of dynamic image key frames.

[0015] The image edge enhancement is performed on the dynamic image key frame according to the adjacent frame images of the dynamic image key frame.

[0016] Through the AI ​​image recognition algorithm, the text information in the dynamic image key frame and the static image after image edge enhancement is extracted as a plurality of first content information.

[0017] The key frame of the dynamic image after image edge enhancement and the graphic information in the static image are identified to obtain the second content information in the image file.

[0018] The correlation between each of the first content information and the second content information is calculated, and the first content information whose correlation with each of the second content information is lower than a preset correlation threshold is filtered out to obtain the content information contained in the image file.

[0019] Furthermore, the semantic segmentation of the text content in the text file to obtain text segments with different semantics includes:

[0020] The text content in the text file is extracted, and semantic segmentation is performed on the text content to obtain a plurality of initial text segments.

[0021] Semantic analysis and recognition are performed on each of the initial text segments to obtain a first semantic recognition result.

[0022] Each of the initial text segments and adjacent initial text segments are combined for analysis and recognition to obtain a second semantic recognition result.

[0023] According to the cosine similarity between the first semantic recognition result and the second semantic recognition result, the initial text segment is semantically improved to obtain a plurality of text segments with different semantics.

[0024] Furthermore, the semantic recognition of the content information is performed, and element labels corresponding to each content information are generated according to the recognition result, and element labels with similar semantics are merged to obtain each target element label, including:

[0025] The semantics of the content information are identified, and element tags corresponding to each of the content information are generated according to the identification result.

[0026] The word vector of each of the element labels is calculated, and the element labels with similar semantics are fused according to the distance between the word vectors to obtain the target element label.

[0027] Furthermore, the text segments and the corresponding data information in the digital file are associated with the corresponding target element tags to form a plurality of information groups to be reviewed, including:

[0028] The word vectors of each of the text segments and each of the target element labels are calculated, and each of the text segments is matched with a corresponding target element label according to the word vectors of each of the text segments and the target element label.

[0029] The data information in the digital file is extracted, and each of the data information is associated with a corresponding text segment to form a plurality of information groups to be reviewed.

[0030] Furthermore, after obtaining a plurality of text segments with different semantics, the method further includes:

[0031] Select the corresponding language probability distribution model according to the field of the document to be reviewed.

[0032] Each of the text segments is segmented, and the probability of each segmented word in each of the text segments relative to the language probability distribution model is calculated to obtain the information entropy of each of the text segments.

[0033] The information entropy density per unit text length in each of the text segments is calculated based on the text length and information entropy of each of the text segments.

[0034] According to the information entropy density of each of the text segments, the text segments whose information entropy density is higher than a preset information entropy density standard value are screened out.

[0035] Furthermore, the calculating of the word vectors of each of the text segments and each of the target element labels, and matching each of the text segments with a corresponding target element label according to the word vectors of each of the text segments and the target element label, includes:

[0036] The word vectors of each of the text segments are calculated in sequence, and based on the word vectors of each of the text segments and the word vectors of each of the target element labels, the similarity sim between each of the text segments and each of the target element labels is calculated according to the following formula:

[0037]

[0038] Among them, L(c i ) is the number of overlapping words between the text segment and the target element label, α is the overlapping word weighting coefficient, and v s is the word vector of the text segment, v t is the word vector of the target feature label, l is the number of overlapping words, n is the dimension of the word vector of the text segment and the target feature label, and m is the vector v s and v t The subscript value of , k1 and k2 are the adjustment factors of vector distance similarity and overlapping word similarity respectively.

[0039] According to the similarity between each text segment and each target element label, it is determined whether there is a matching target element label among the target element labels. If so, the text segment is matched with the corresponding target element label. Otherwise, a target element label is generated according to the content of the text segment.

[0040] Furthermore, generating a target element label according to the content of the text segment includes:

[0041] The LDA model is used to obtain the topic distribution in the text segment, and the target element label is selected according to the topic distribution.

[0042] Furthermore, the text segments and the corresponding data information in the digital file are associated with the corresponding target element tags to form a plurality of information groups to be reviewed, including:

[0043] According to the information entropy of each text fragment, an information tree directory is set up, and each text fragment is associated with a node of the information tree directory step by step; the data information in the digital file is associated with the text fragment under the corresponding node in the information tree directory.

[0044] The word vector of each text segment is calculated, and the corresponding target element label is associated with the word vector of each text segment to form a plurality of information groups to be reviewed, and a directory index of the information tree directory is established according to each target element label.

[0045] Another embodiment of the present invention provides a material review processing system based on an AI model, comprising:

[0046] The file acquisition module is used to acquire the files to be reviewed input by the user, and classify the files to be reviewed into digital files, image files and text files according to the file types.

[0047] The image recognition module is used to identify and analyze the image file through an AI image recognition algorithm to obtain content information contained in the image file.

[0048] The label generation module is used to perform semantic recognition on the content information, generate element labels corresponding to each content information according to the recognition result, and merge element labels with similar semantics to obtain each target element label.

[0049] The text segmentation module is used to perform semantic segmentation on the text content in the text file to obtain text fragments with different semantics, and associate each of the text fragments and the corresponding data information in the digital file with the corresponding target element label to form a plurality of information groups to be reviewed; the information groups to be reviewed are designed to be associated step by step according to the information entropy of the corresponding text fragments to form an information tree directory.

[0050] The content review module is used to determine the review content based on the review objectives of the documents to be reviewed and the target element labels of each information group to be reviewed, and obtain the corresponding review conditions in the review standard database based on the review content, and conduct a level-by-level review of the information groups to be reviewed in the information tree directory based on the review conditions, and generate an electronic review report based on the review results of each information group to be reviewed.

[0051] Compared with the prior art, the embodiments of the present invention have the following advantages:

[0052] (1) Extract key contents of the review materials by combining information from image files, text files, and digital files to improve the accuracy and reliability of the review results;

[0053] (2) Improve and supplement the semantics of each text segment based on its adjacent text segments to ensure the logical relationship and order of the initial text and improve the accuracy of the review direction;

[0054] (3) By establishing a hierarchical information tree directory based on the information entropy of the text fragments, it is possible to distinguish text fragments with different information contents, conduct reviews at each level, and generate a comprehensive review report with more detailed review content. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 A flowchart of the steps of the material review processing method based on the AI ​​model provided in an embodiment of the present invention;

[0056] Figure 2 A structural block diagram of the material review and processing system based on the AI ​​model provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0057] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0058] In the description of this application, the terms "first", "second", "third", etc. are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first", "second", "third", etc. may explicitly or implicitly include one or more of the feature. In the description of this application, unless otherwise specified, "plurality" means two or more.

[0059] In the description of the present application, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be a connection between the two elements. The terms "vertical", "horizontal", "left", "right", "upper", "lower" and similar expressions used herein are only for illustrative purposes, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0060] In the description of this application, it should be noted that, unless otherwise defined, all technical and scientific terms used in the present invention have the same meaning as those commonly understood by those skilled in the art. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood by specific circumstances.

[0061] An embodiment of the present invention provides a material review processing method based on an AI model. For details, see Figure 1 , Figure 1The flowchart of the material review processing method based on the AI ​​model in one embodiment of the present invention includes steps S11 to S15:

[0062] S11, obtaining the to-be-reviewed files input by the user, and classifying the to-be-reviewed files into digital files, image files and text files according to the file types.

[0063] In the application of various projects, it is often necessary to evaluate the project content and issue an evaluation report to provide a basis and guidance for subsequent project adjustments and implementation. For example, scientific research projects, teaching reform projects, digital teaching materials projects, curriculum training, professional skills school competitions, quality course selection, teaching ability competitions, dual-qualified teacher certification, dual innovation competitions, new engineering and new liberal arts practices, etc. All these projects require prior evaluation. Therefore, it is necessary to analyze the evaluation materials provided by users.

[0064] The files to be reviewed input by the user are obtained, and the files to be reviewed are divided into digital files, image files and text files according to the file types. For comprehensive files, such as files containing images and text content, they are divided into image files and text files.

[0065] Among them, the types of digital files, image files and text files are not fixed to numbers, images and text. For example, Excel files can be used as digital files, and the specific type of content recorded in the file shall prevail.

[0066] S12, using an AI image recognition algorithm, identifying and analyzing the image file to obtain content information contained in the image file.

[0067] The image data in the image file may include dynamic images and static images. Therefore, before performing recognition analysis on the image file, the image data in the image file needs to be divided into dynamic images and static images.

[0068] For dynamic images, this embodiment adopts the key frame extraction method to convert the dynamic images into static images. Specifically, since the principle of dynamic images presenting dynamics is to perceive dynamics through the difference changes between adjacent frames and the visual persistence phenomenon of the human eye, when the difference between two adjacent frames reaches a certain threshold, it is considered that the scene has changed.

[0069] Since the workload of image recognition for consecutive frames with little difference is huge and the recognition results are highly repetitive, this embodiment calculates the difference changes between each consecutive frame of the dynamic image to obtain a continuous difference change rate. When the difference change rate between two frames of images is significantly increased compared with the difference change rate between the previous consecutive frames, the dynamic image is cut off from the frame, thereby dividing each dynamic image into several dynamic image segments.

[0070] Key frames are extracted from each dynamic image segment as the basis for subsequent image recognition. Since image frames in dynamic images usually have image blur, resulting in unclear edge contours of objects in the image, in order to enhance the clarity of image contours, the similarity of adjacent frames in the same dynamic image segment as the key frame can be used for analysis, and similar feature points in adjacent frames can be extracted. The similar feature points are used to enhance the image contour of the key frame to obtain several dynamic image key frames to improve the accuracy of subsequent image recognition.

[0071] Since images involved in project review often contain project-related text content, it is necessary to first identify and separate the text information in the image. Specifically, this embodiment uses optical character recognition (OCR) technology to process dynamic image key frames and static images, detects the text graphics therein and saves them as a number of first content information.

[0072] After the text in the image is collected, the key frames of the dynamic image and the elements in the static image are continuously recognized to obtain a plurality of second content information.

[0073] Since content irrelevant to the current review item may be identified in the text information of the image, it is necessary to screen out irrelevant content information. Specifically, for English character strings in the content information, the edit distance method is used to calculate the correlation between each content information and other content information. For Chinese character strings in the content information, the word vector difference between each content information and other content information is calculated as the correlation. When the difference in the word vectors of two texts is large enough, it means that there is a large difference in the semantics of the two texts.

[0074] To avoid accidental deletion, the word vectors of each content information are calculated. When the correlation between a first content information and other second content information is less than a preset correlation threshold, the first content information is filtered out as irrelevant information.

[0075] S13, performing semantic recognition on the content information, generating element labels corresponding to each content information according to the recognition result, and fusing element labels with similar semantics to obtain each target element label.

[0076] In actual project files, due to the large number of pictures and high content overlap and similarity, the identified content information has a lot of overlapping content. In order to avoid content duplication and reduce the workload of subsequent classification, in addition to filtering out content information with large word vector differences, it is also necessary to delete the same content information and merge similar content information. Specifically, the method adopted in this embodiment is:

[0077] The semantics of each content information is identified, and the element label corresponding to each content information is generated according to the identification result; the word vector of each element label is calculated. When the distance between the word vectors is close enough, the content represented by the two is highly similar or essentially the same. According to the distance between the word vectors, the element labels with similar semantics are fused to obtain the target element label.

[0078] S14, performing semantic segmentation on the text content in the text file to obtain text segments with different semantics, and associating each of the text segments and the corresponding data information in the digital file to the corresponding target element label to form a plurality of information groups to be reviewed.

[0079] Extract text content from a text file, perform semantic segmentation on the text content, and obtain several initial text segments.

[0080] In the existing text semantic segmentation, the text is often segmented into independent paragraphs and sentence blocks. However, the contents of the materials to be reviewed often have a high correlation. Therefore, only semantic recognition of each initial text segment may lead to an incorrect interpretation of the meaning of the text segment, or a one-sided interpretation of the meaning. Therefore, this embodiment uses the contents of the initial text segments adjacent to each initial text segment to semantically improve the current initial text segment content. Specifically, the process of obtaining the text segment is as follows:

[0081] Semantic analysis and recognition are performed on each initial text to obtain a first semantic recognition result, and then semantic analysis and recognition are performed on initial text segments before and after the initial text segment to obtain a second semantic recognition result.

[0082] The cosine similarity between the first semantic recognition result and the second semantic recognition result is calculated. If the two have a certain cosine similarity, the initial text segment is supplemented and improved according to the second semantic recognition result to improve the logical relationship and order of the initial text segment to obtain a text segment.

[0083] The information entropy of each text fragment is calculated. When the information entropy of a text fragment is much higher than the normal information entropy value, it indicates that the information in this text fragment is difficult to predict and determine. The appearance of each character, word or phrase seems to have no obvious rules or patterns to follow. It is difficult to infer the subsequent content based on the existing information. Therefore, text fragments with information entropy density higher than the standard value are screened out.

[0084] Specifically, the screening process is:

[0085] Select the corresponding language probability distribution model according to the field of the document to be reviewed. For example, for a robotics scientific research and innovation project, select language probability distribution models for data analysis, fault diagnosis, predictive maintenance, etc.

[0086] For each text fragment, it is segmented and the information entropy contribution of each word relative to the language probability distribution model is calculated, so as to sum up and obtain the information entropy of the text fragment; then, the information entropy content per unit text length is calculated according to the information entropy and text length of the text fragment to obtain the information entropy density of the text fragment.

[0087] Repeat the above steps to calculate the information entropy density of all text segments, and filter out text segments with information entropy density higher than a preset information entropy density standard value according to the information entropy density of each text segment.

[0088] The word vectors of each text segment are calculated in turn, and the similarity between each text segment and each target element label is calculated based on the word vectors of each text segment and the word vectors of each target element label, wherein the vector text similarity between the text segment and the target element label is:

[0089]

[0090] Among them, v s is the word vector of the text segment, v t is the word vector of the target feature label, n is the dimension of the word vector of the text segment and the target feature label, and m is the vector v s and v t The subscript value of .

[0091] The overlapping word similarity between the text segment and the target feature label is:

[0092]

[0093] Among them, L(c i ) is the number of overlapping words between the text segment and the target element label, α is the weighting coefficient of the overlapping words. When the number of overlapping times is 1, α has no weighting effect, and l is the number of overlapping words.

[0094] Therefore, the similarity between the text segment and each target feature label is:

[0095]

[0096] Among them, k1 and k2 are adjustment factors for vector distance similarity and overlapping word similarity respectively. The values ​​of k1 and k2 are determined according to the overlap of material contents in the actual review project.

[0097] For each text fragment, based on its similarity with each target element label, determine whether there is a matching target element label among the target element labels. If so, match the text fragment to the corresponding target element label. Otherwise, generate a target element label based on the content of the text fragment.

[0098] Target feature labels whose word vector differences with the text fragment are within a certain range are selected from the existing target feature labels, and the target feature labels are associated with the text fragment.

[0099] If the difference between the word vector of the text segment and the word vector of any element label is not within the preset range, a new target element label is generated according to the content of the text segment, and the text segment is associated with the target element label. Specifically, the method of generating a new target element label in this embodiment is:

[0100] Use the LDA model to obtain the topic distribution in the text segment, select topics based on the topic distribution, and save them as feature labels.

[0101] In actual semantic segmentation, the amount of information contained in each text segment obtained by segmentation is not the same, and there may be a situation where a text segment is associated with multiple element labels. In order to facilitate the review and generate a clearer and more intuitive review report, in the process of associating element labels, this embodiment adopts the form of setting an information tree directory according to the information entropy in each text segment to classify and manage the numerous information extracted from the files to be reviewed. Specifically, according to the size of the information entropy of each text segment, the text segments are associated with the nodes of the information tree directory in order from the most to the least information, which is convenient for the subsequent step-by-step review of the materials and also provides a basis for generating a clear and visual project framework in the subsequent review report.

[0102] In addition, data is also an important basis for project review. Therefore, after associating the text fragments with the information tree directory, the data information in the data file is analyzed, and the data information is associated with the text fragments under the corresponding nodes in the information tree directory to form several groups of information groups to be reviewed. Each group of information groups to be reviewed contains an element label and the text fragments corresponding to the element label. Some of the information groups to be reviewed contain data information.

[0103] Preferably, in order to facilitate the flexible review of project materials, this embodiment also establishes a directory index of the information tree directory according to each target element tag, which is used to quickly retrieve the information group to be reviewed in a certain direction and conduct a specific evaluation. For example, in the evaluation of the dual-qualified teacher certification project, not only can the comprehensive quality of each teacher be evaluated, but also the content related to the teaching skill item can be quickly pulled out according to the directory index to conduct a single evaluation of the teaching skill, or the content related to the practical experience can be pulled out to conduct a single evaluation of the teaching practical experience of each teacher.

[0104] S15, determining the review content according to the review objectives of the documents to be reviewed and the target element labels of each information group to be reviewed, and obtaining the corresponding review conditions in the review standard database according to the review content, reviewing the information groups to be reviewed in the information tree directory step by step according to the review conditions, and generating an electronic review report according to the review results of each information group to be reviewed.

[0105] The AI ​​model-based material review processing method of the present invention combines the information of image files, text files and data files to extract the key content of the materials to be reviewed, thereby improving the accuracy and reliability of the review results; a hierarchical information tree directory is established according to the information entropy of text fragments, which can distinguish text fragments with different information contents, conduct reviews at each level, and generate a comprehensive review report with more detailed review content.

[0106] The embodiment of the present invention further provides a material review processing system based on an AI model, which is used to execute the material review processing method based on an AI model as described above. Figure 2 This is a structural block diagram of a material review processing system based on an AI model according to an embodiment of the present invention, and the system includes:

[0107] The file acquisition module 21 is used to acquire the files to be reviewed input by the user, and classify the files to be reviewed into digital files, image files and text files according to the file types.

[0108] The image recognition module 22 is used to identify and analyze the image file through an AI image recognition algorithm to obtain content information contained in the image file.

[0109] The label generation module 23 is used to perform semantic recognition on the content information, generate element labels corresponding to each content information according to the recognition result, and merge element labels with similar semantics to obtain each target element label.

[0110] The text segmentation module 24 is used to perform semantic segmentation on the text content in the text file to obtain text fragments with different semantics, and associate each of the text fragments and the corresponding data information in the digital file to the corresponding target element label to form a plurality of information groups to be reviewed; the information groups to be reviewed are designed to be associated step by step according to the information entropy of the corresponding text fragments to form an information tree directory.

[0111] The content review module 25 is used to determine the review content based on the review objectives of the document to be reviewed and the target element labels of each information group to be reviewed, and obtain the corresponding review conditions in the review standard database based on the review content, and conduct a level-by-level review of the information groups to be reviewed in the information tree directory based on the review conditions, and generate an electronic review report based on the review results of each information group to be reviewed.

[0112] The technical features and technical effects of the system proposed in the embodiment of the present invention are the same as those of the method proposed in the embodiment of the present invention, and are not described in detail here. Each module in the above system can be implemented in whole or in part by software, hardware, and a combination thereof. Each of the above modules can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each of the above modules.

[0113] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A material review processing method based on an AI model, characterized in that: include: Obtaining the files to be reviewed input by the user, and classifying the files to be reviewed into digital files, image files and text files according to the file types; Using an AI image recognition algorithm, the image file is identified and analyzed to obtain content information contained in the image file; Performing semantic recognition on the content information, generating element labels corresponding to each content information according to the recognition result, and fusing element labels with similar semantics to obtain each target element label; Performing semantic segmentation on the text content in the text file to obtain text segments with different semantics, and associating each of the text segments and the corresponding data information in the digital file with the corresponding target element label to form a plurality of information groups to be reviewed; The information group to be reviewed is designed to be associated step by step according to the information entropy of the corresponding text fragments to form an information tree directory, specifically including: selecting a corresponding language probability distribution model according to the field of the document to be reviewed; segmenting each of the text fragments, calculating the probability of each segmented word in each of the text fragments relative to the language probability distribution model, to obtain the information entropy of each of the text fragments; The review content is determined according to the review objectives of the documents to be reviewed and the target element labels of each information group to be reviewed, and the corresponding review conditions in the review standard database are obtained according to the review content. The information groups to be reviewed in the information tree directory are reviewed step by step according to the review conditions, and an electronic review report is generated according to the review results of each information group to be reviewed.

2. The material review processing method based on the AI ​​model according to claim 1, characterized in that: The image file is identified and analyzed by the AI ​​image recognition algorithm to obtain content information contained in the image file, including: Dividing the image data in the image file into dynamic images and static images; Calculating the difference change rate between adjacent continuous frames of the dynamic image, extracting key frames according to the difference change rate, and obtaining a plurality of dynamic image key frames; Performing image edge enhancement on the dynamic image key frame according to adjacent frame images of the dynamic image key frame; Extracting the key frames of the dynamic image and the text information in the static image after image edge enhancement as a plurality of first content information through an AI image recognition algorithm; Identifying the key frame of the dynamic image after image edge enhancement and the graphic information in the static image to obtain second content information in the image file; The correlation between each of the first content information and the second content information is calculated, and the first content information whose correlation with each of the second content information is lower than a preset correlation threshold is filtered out to obtain the content information contained in the image file.

3. The material review processing method based on the AI ​​model according to claim 1, characterized in that: The semantic segmentation of the text content in the text file to obtain text segments with different semantics includes: Extracting text content from the text file, performing semantic segmentation on the text content, and obtaining a plurality of initial text segments; Performing semantic analysis and recognition on each of the initial text segments to obtain a first semantic recognition result; Merging and analyzing each of the initial text segments and adjacent initial text segments to obtain a second semantic recognition result; According to the cosine similarity between the first semantic recognition result and the second semantic recognition result, the initial text segment is semantically improved to obtain a plurality of text segments with different semantics.

4. The material review processing method based on the AI ​​model according to claim 1, characterized in that: The semantic recognition of the content information is performed, and the element labels corresponding to each content information are generated according to the recognition result, and the element labels with similar semantics are merged to obtain each target element label, including: Identify the semantics of the content information, and generate element labels corresponding to each of the content information according to the identification results; The word vector of each of the element labels is calculated, and the element labels with similar semantics are fused according to the distance between the word vectors to obtain the target element label.

5. The material review processing method based on the AI ​​model as claimed in claim 3, characterized in that: The step of associating the corresponding data information in each of the text segments and the digital file with the corresponding target element tag to form a plurality of information groups to be reviewed includes: Calculating the word vectors of each of the text segments and each of the target element labels, and matching each of the text segments with a corresponding target element label according to the word vectors of each of the text segments and the target element label; The data information in the digital file is extracted, and each of the data information is associated with a corresponding text segment to form a plurality of information groups to be reviewed.

6. The material review processing method based on the AI ​​model as claimed in claim 3, characterized in that: After obtaining a plurality of text segments with different semantics, the method further includes: Calculating the information entropy density per unit text length in each of the text segments according to the text length and information entropy of each of the text segments; According to the information entropy density of each of the text segments, the text segments whose information entropy density is higher than a preset information entropy density standard value are screened out.

7. The material review processing method based on the AI ​​model as claimed in claim 5, characterized in that: The step of calculating the word vectors of each of the text segments and each of the target element labels, and matching each of the text segments with a corresponding target element label according to the word vectors of each of the text segments and the target element label, comprises: The word vectors of each text segment are calculated in sequence, and the similarity between each text segment and each target element label is calculated according to the following formula: : ； in, is the number of overlapping words between the text segment and the target feature label, is the weight coefficient of overlapping words, is the word vector of the text fragment, is the word vector of the target feature label, is the number of overlapping words, is the dimension of the word vectors for the text fragment and target feature label, For vector and The subscript value of and They are the adjustment factors for vector distance similarity and overlapping word similarity respectively; According to the similarity between each text segment and each target element label, it is determined whether there is a matching target element label among the target element labels. If so, the text segment is matched with the corresponding target element label. Otherwise, a target element label is generated according to the content of the text segment.

8. The material review processing method based on the AI ​​model according to claim 7, characterized in that: The step of generating a target element label according to the content of the text segment includes: The LDA model is used to obtain the topic distribution in the text segment, and the target element label is selected according to the topic distribution.

9. The material review processing method based on the AI ​​model according to claim 6, characterized in that: The step of associating the corresponding data information in each of the text segments and the digital file with the corresponding target element tag to form a plurality of information groups to be reviewed includes: According to the information entropy of each of the text fragments, an information tree directory is set, and each text fragment is associated with a node of the information tree directory step by step; the data information in the digital file is associated with the text fragment under the corresponding node in the information tree directory; The word vector of each text segment is calculated, and the corresponding target element label is associated with the word vector of each text segment to form a plurality of information groups to be reviewed, and a directory index of the information tree directory is established according to each target element label.

10. A material review processing system based on AI model, characterized in that: The system comprises: A file acquisition module, used to acquire the files to be reviewed input by the user, and classify the files to be reviewed into digital files, image files and text files according to the file types; An image recognition module is used to identify and analyze the image file through an AI image recognition algorithm to obtain content information contained in the image file; A label generation module is used to perform semantic recognition on the content information, generate element labels corresponding to each content information according to the recognition result, and merge element labels with similar semantics to obtain each target element label; The text segmentation module is used to perform semantic segmentation on the text content in the text file to obtain text segments with different semantics, and associate the corresponding data information in each text segment and the digital file with the corresponding target element label to form a plurality of information groups to be reviewed; the information groups to be reviewed are designed to be associated step by step according to the information entropy of the corresponding text segments to form an information tree directory, specifically including: selecting a corresponding language probability distribution model according to the field of the file to be reviewed; segmenting each text segment, and calculating the probability of each segment in each text segment relative to the language probability distribution model to obtain the information entropy of each text segment; The content review module is used to determine the review content based on the review objectives of the documents to be reviewed and the target element labels of each information group to be reviewed, and obtain the corresponding review conditions in the review standard database based on the review content, and conduct a level-by-level review of the information groups to be reviewed in the information tree directory based on the review conditions, and generate an electronic review report based on the review results of each information group to be reviewed.

Images