Semantic recognition service system for project document management

Abstract

The embodiment of the invention provides a semantic recognition service system for project document management, and belongs to the technical field of document processing. The semantic recognition service system comprises a storage layer which uses a Mysql database to store service information, supports backup of a master database and a slave database, and regularly exports the backup from the database; the core identification layer is used for identifying and analyzing various types of documents in the business information so as to extract all rich text information required in the documents; the AI algorithm middle table is used for carrying out semantic recognition and sequence labeling on the information extracted by the core recognition layer; and the business application layer is applied to project document management-oriented document classification, document extraction, risk auditing, document comparison and character recognition table analysis. According to the semantic recognition service system, deep understanding and analysis of documents can be realized, and document classification, document extraction, risk auditing, document comparison and character recognition table analysis of project documents can be completed.

Description

Technical Field

[0001] The present invention relates to the technical field of document processing, and in particular to a semantic recognition service system oriented to project document management. Background Art

[0002] At present, the project management system has long relied on manual review and verification for complex project documents. Although a set of specifications has been formed for document summarization, necessary steps such as compliance testing of project documents, project risk avoidance, and extraction of key information from documents have all adopted manual verification. Huge manpower is invested in document text extraction, document review, document classification, etc., and there is a lack of reliability. Summary of the invention

[0003] The purpose of an embodiment of the present invention is to provide a semantic recognition service system for project document management, which can achieve in-depth understanding and analysis of documents, and complete document classification, document extraction, risk review, document comparison, and text recognition table analysis of project documents.

[0004] In order to achieve the above object, an embodiment of the present invention provides a semantic recognition service system for project document management, the semantic recognition service system comprising:

[0005] The storage layer uses the MySQL database to store business information, supports master-slave database backup, and regular database export and backup;

[0006] The core recognition layer identifies and parses various types of documents in business information to extract all the rich text information required in the documents;

[0007] The AI ​​algorithm middle platform performs semantic recognition and sequence labeling on the information extracted by the core recognition layer;

[0008] The business application layer is used for document classification, document extraction, risk review, document comparison, and text recognition and table parsing for project document management.

[0009] Optionally, the core identification layer includes:

[0010] Node parsing module directly parses the PDF protocol, and the returned result is encapsulated as a Char object during rich text parsing, including coordinates, width and height;

[0011] The table parsing module identifies the corresponding table area from the PDF document and parses the table topology and content;

[0012] The chart parsing module extracts charts from documents through PDF feature extraction for static images and object detection services for generated images.

[0013] The document structure detection module, after merging the text lines, determines whether it is a header or footer according to certain rules, and merges paragraphs based on the left and right locking of the text lines, the character semantics at the beginning and end of the lines, and the font band if the content is not recognized as a header or footer.

[0014] Optionally, the core recognition layer supports the extraction of key information from Word, PDF, document scan images and unstructured documents.

[0015] Optionally, the AI ​​algorithm middle platform is used for:

[0016] Acquire data information extracted from the core identification layer;

[0017] Cleaning and standardizing the data information;

[0018] Extract keywords from the cleaned and standardized data to obtain a keyword list that meets the requirements;

[0019] Performing feature extraction on the keyword list, and performing text classification on the keywords after feature extraction;

[0020] The classified keywords are sequence labeled.

[0021] Optionally, the cleaned and standardized data information is subjected to keyword extraction to obtain a keyword list that meets the requirements, including:

[0022] Get the cleaned and standardized text;

[0023] According to formula (1), the frequency of each word in the text is calculated:

[0024]

[0025] Among them, TF(t,d) represents the frequency of word t in text d, f t,d represents the number of times word t appears in text d, f d Represents the total number of occurrences of all words in the text;

[0026] The importance of a word in a text is calculated according to formula (2):

[0027]

[0028] Among them, IDF(t,D) represents the inverse document frequency of word t in all text sets D, W t represents the number of texts containing word t;

[0029] The weight of each word is calculated according to formula (3):

[0030] TF-IDF(t,d,D)=TF(t,d)×IDF(t,D), formula (3)

[0031] Where TF-IDF(t,d,D) represents the TF-IDF value of word t in text d;

[0032] Sort each word according to the TF-IDF value, and select the top K words as keywords to get a keyword list that meets the requirements.

[0033] Optionally, performing feature extraction on the keyword list and performing text classification on the keywords after feature extraction includes:

[0034] Obtain all keywords in the keyword list;

[0035] Performing semantic recognition on the keywords to obtain corresponding semantics of each keyword;

[0036] The keywords with the largest range of semantics commonly represented by each keyword are grouped into a first group to complete the preliminary classification of the keywords;

[0037] Performing semantic analysis on the keywords in each first set after preliminary classification to obtain the number of categories in the first set, and grouping the keywords in the first set into corresponding categories to obtain a second set;

[0038] Performing semantic analysis on the keywords in the second set to obtain the number of categories in the second set, and grouping the keywords in the second set into corresponding categories to obtain a corresponding third set;

[0039] By analogy, the keywords are classified into the corresponding kth set;

[0040] Each of the keywords is labeled as a set from small to large categories.

[0041] Optionally, the AI ​​algorithm middle platform is used for:

[0042] Calculate the TF-IDF value of the words in each sentence according to formula (3) and convert each sentence into a feature vector;

[0043] Calculate the cosine similarity between the feature vector of each sentence and the feature vectors of the remaining sentences;

[0044] Two sentences whose cosine similarity is greater than a preset threshold are marked as similar texts.

[0045] Optionally, the AI ​​algorithm middle platform is used for:

[0046] Parse the risk information of the acquired risk project documents to obtain keywords and feature vectors containing risk information;

[0047] Determine a first similarity between the risk project document and the comparison project document according to a set of keywords in the risk project document and keywords in the comparison project document;

[0048] Determine a second similarity between the risk project document and the comparison project document according to the cosine similarity between the feature vector of the sentence of the risk project document and the feature vector of the sentence of the comparison project document;

[0049] Determine the comprehensive similarity between the risk project document and the comparison project document according to the first similarity and the second similarity;

[0050] Determine whether the compared project document has risks based on the comprehensive similarity.

[0051] Optionally, determining the first similarity between the risk project document and the comparison project document according to a set of keywords in the risk project document and keywords in the comparison project document includes:

[0052] Obtaining a classified set of keywords with risk information from the risk project document;

[0053] Obtaining a classification set of keywords of the comparison project document;

[0054] Determine the number of the classification set of keywords with risk information of the risk project document and the classification set of keywords of the comparison project document that belong to the same set;

[0055] According to the number of keywords belonging to the same classification set, each classification set is assigned a value and weighted addition is performed to obtain a first similarity.

[0056] Through the above technical scheme, a semantic recognition service system for project document management provided by the present invention includes a storage layer, which can use a Mysql database to store business information, support master-slave library backup, and regular database export backup. The core recognition layer can identify and parse various types of documents in business information, so that all rich text information required in the document can be extracted. The AI ​​core algorithm middle platform can perform semantic recognition and sequence labeling on the information extracted by the core recognition layer. The business application layer can be applied to document classification, document extraction, risk review, document comparison, and text recognition table analysis for project document management. The semantic recognition service system can provide a complete set of semantic recognition service systems for project managers and auditors. Other features and advantages of the embodiments of the present invention will be described in detail in the subsequent specific implementation method section. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following specific implementations, they are used to explain the embodiments of the present invention, but do not constitute a limitation on the embodiments of the present invention. In the accompanying drawings:

[0058] Figure 1 is a system connection diagram of a semantic recognition service system for project document management according to an embodiment of the present invention;

[0059] Figure 2 is a first flow chart of an AI algorithm middle station of a semantic recognition service system for project document management according to an embodiment of the present invention;

[0060] Figure 3 is a flowchart of obtaining a keyword list of a semantic recognition service system for project document management according to an embodiment of the present invention;

[0061] Figure 4 is a flowchart of text classification of keywords in a semantic recognition service system for project document management according to an embodiment of the present invention;

[0062] Figure 5 is a second flow chart of an AI algorithm middle station of a semantic recognition service system for project document management according to an embodiment of the present invention;

[0063] Figure 6 is a third flow chart of an AI algorithm middle station of a semantic recognition service system for project document management according to an embodiment of the present invention;

[0064] Figure 7 The present invention is a flowchart of obtaining a first similarity in a semantic recognition service system for project document management according to an embodiment of the present invention. DETAILED DESCRIPTION

[0065] The specific implementation of the embodiment of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the embodiment of the present invention, and is not used to limit the embodiment of the present invention.

[0066] Figure 1It is a system connection diagram of a semantic recognition service system for project document management according to an embodiment of the present invention. In the present invention, the semantic recognition service system may include: a storage layer, a core recognition layer, an AI algorithm middle platform and a business application layer. The storage layer can use a Mysql database to store business information, support master-slave library backup, and regularly export and backup databases, so that project document data can be stored. The AI ​​core recognition layer can identify and parse various types of documents in business information, such as .pdf, .doc, .docx and other documents, and extract all the rich text information required in the document through multiple steps of serial parsing, including text, title, paragraph, chapter structure, static pictures, generated graphs, tables and other information, generate feature information required for subsequent models, and facilitate subsequent model training and other operations. The business application layer can be applied to document classification, document extraction, risk review, document comparison and text recognition table analysis for project document management. The semantic recognition service system can provide a complete set of semantic recognition service systems for project managers and auditors.

[0067] In one embodiment of the present invention, the core recognition layer may include: a Node parsing module, a table parsing module, a chart parsing module, and a document structure detection module. The Node parsing module can directly parse the PDF protocol, and the returned result is encapsulated as a Char object during rich text parsing, including coordinates, width and height. The table parsing module can combine CV and NLP related technologies to identify the corresponding table area from the PDF document. The recognition results are serialized and saved in the json file obtained by document parsing. Table parsing includes two steps: wired table parsing and wireless table detection. The wired table can locate the table by detecting the lines through OpenCV; the wireless table detects the table through Faster R-CNN. The chart parsing module can extract PDF elements for static images. For generated images, such as bar charts, column charts, etc., the target detection service can be used to extract charts in the document. It contains information such as the coordinates, size, screenshot file path, and text lines in the chart. After merging the text lines, the document structure detection module can determine whether it is a header or footer according to certain rules. For content that is not recognized as headers and footers, paragraphs are merged based on the left and right indents of the text line, the semantics of the characters at the beginning and end of the line, font size, etc. Finally, the title is recognized through the features and the directory structure is generated.

[0068] In one embodiment of the present invention, the core recognition layer can support the extraction of key information from Word, PDF, document scan images and unstructured documents.

[0069] In one embodiment of the present invention, Figure 2 As shown, the first process of the AI ​​algorithm center may include:

[0070] In step S1, data information extracted from the core identification layer is obtained.

[0071] In step S2, the data information is cleaned and standardized.

[0072] In step S3, keywords are extracted from the cleaned and standardized data information to obtain a keyword list that meets the requirements.

[0073] In step S4, feature extraction is performed on the keyword list, and text classification is performed on the keywords after feature extraction.

[0074] In step S5, sequence labeling is performed on the classified keywords.

[0075] In the present invention, the AI ​​algorithm center can perform keyword classification and sequence labeling on project documents. When classifying and labeling, the data information extracted from the core recognition layer can be obtained, and then the data information can be cleaned and standardized. Then, keyword extraction is performed on the cleaned and standardized data information, so that a keyword list that meets the requirements can be obtained. Feature extraction is performed on the keyword list, and text classification can be performed on the keywords after feature extraction. After the classification is completed, the classified keywords can be sequence labeled.

[0076] In one embodiment of the present invention, Figure 3 As shown, the process of obtaining a keyword list may include:

[0077] In step S6, the cleaned and standardized text is obtained.

[0078] In step S7, the frequency of each word in the text is calculated according to formula (1):

[0079]

[0080] Among them, TF(t,d) represents the frequency of word t in text d, f t,d represents the number of times word t appears in text d, f d Represents the total number of occurrences of all words in the text.

[0081] In step S8, the importance of the word in the text is calculated according to formula (2):

[0082]

[0083] Among them, IDF(t,D) represents the inverse document frequency of word t in all text sets D, W t Represents the number of texts containing word t.

[0084] In step S9, the weight of each word is calculated according to formula (3):

[0085] TF-IDF(t,d,D)=TF(t,d)×IDF(t,D), formula (3)

[0086] Where TF-IDF(t,d,D) represents the TF-IDF value of word t in text d.

[0087] In step S10, each word is sorted according to the TF-IDF value, and the top K words are selected as keywords to obtain a keyword list that meets the requirements.

[0088] In the present invention, when extracting keywords from a text, the text can be first obtained after cleaning and standardization, including cleaning, word segmentation, and removal of stop words. After cleaning and standardization, the frequency of each word in the text can be calculated according to formula (1), and then the importance of the word in the text can be calculated according to formula (2). After obtaining the frequency and inverse document frequency of each word in the text, the weight of each word can be calculated according to formula (3). After obtaining the TF-IDF value of each word, which is also the weight of each word, each word can be sorted according to the TF-IDF value, and then the words ranked in the top K are selected as keywords, so that a keyword list that meets the requirements can be obtained.

[0089] In one embodiment of the present invention, Figure 4 As shown, the process of keyword text classification may include:

[0090] In step S11, all keywords in the keyword list are obtained.

[0091] In step S12, semantic recognition is performed on the keywords to obtain the corresponding semantics of each keyword.

[0092] In step S13, keywords with the largest range of semantics commonly represented by each keyword are grouped into a first group to complete a preliminary classification of the keywords.

[0093] In step S14, semantic analysis is performed on the keywords in each preliminarily classified set to obtain the number of categories in the first set, and the keywords in the first set are grouped into corresponding categories to obtain a second set.

[0094] In step S15, semantic analysis is performed on the keywords in the second set to obtain the number of categories in the second set, and the keywords in the second set are grouped into corresponding categories to obtain a corresponding third set.

[0095] In step S16, the keywords are classified into the corresponding kth set in this way.

[0096] In step S17, each keyword is labeled as a set from small to large categories.

[0097] In the present invention, when the text classification is performed on the keywords, all the keywords in the keyword list can be obtained. Then the keywords can be semantically identified, so that the semantics corresponding to each keyword can be obtained. According to the understanding of the semantics of the keywords, the keywords representing the same semantics can be classified into one category, and the keywords with the largest range of semantics represented by each keyword are collectively grouped into the first set. The first set contains the largest range, so that the preliminary classification of the keywords can be completed. After the preliminary classification of the keywords, the number of the first set can be multiple. The keywords in the first set can be semantically analyzed again, so that the keywords representing the same semantics can be classified into one category again according to the semantic understanding of the keywords in the first set. There can be multiple categories in the first set. The keywords in the first set are grouped into corresponding subcategories to obtain a second set. The second set can be smaller than the first set, and the semantics represented by the second set can more accurately describe the keywords relative to the semantics represented by the first set. The keywords in the second set are semantically analyzed, and the keywords representing the same semantics can be classified into one category again, so that the number of categories in the second set can be obtained, and then the keywords in the second set can be grouped into corresponding categories, so that the corresponding third set can be obtained. The semantics represented by the third set can describe the keywords more accurately than the semantics represented by the second set. Similarly, the keywords can be classified into the corresponding kth set, and then the keywords can be labeled as sets according to categories from small to large, thereby completing the classification of keywords.

[0098] In one embodiment of the present invention, Figure 5 As shown, the second process of the AI ​​algorithm center may include:

[0099] In step S18, the TF-IDF value of the words in each sentence is calculated according to formula (3) and each sentence is converted into a feature vector.

[0100] In step S19, the cosine similarity between the feature vector of each sentence and the feature vectors of the remaining sentences is calculated.

[0101] In step S20, two sentences whose cosine similarity is greater than a preset threshold are marked as similar texts.

[0102] In the present invention, after the TD-IDF value of the words in each word is calculated according to formula (3), each word can be converted into a feature vector according to the TF-IDF value. After the feature vector sum of each word is obtained, the cosine similarity between the feature vector of each word and the feature vector of the rest of the sentences can be calculated, and two sentences with a cosine similarity greater than a preset threshold can be marked as similar texts according to the cosine similarity.

[0103] In one embodiment of the present invention, Figure 6 As shown, the third process of the AI ​​algorithm center may include:

[0104] In step S21, the risk information of the acquired risk project document is parsed to obtain keywords and feature vectors having risk information.

[0105] In step S22, a first similarity between the risk project document and the comparison project document is determined based on a set of keywords in the risk project document and keywords in the comparison project document.

[0106] In step S23, a second similarity between the risk project document and the comparison project document is determined based on the cosine similarity between the feature vector of the sentence of the risk project document and the feature vector of the sentence of the comparison project document.

[0107] In step S24, the comprehensive similarity between the risk project document and the comparison project document is determined according to the first similarity and the second similarity.

[0108] In step S25, it is determined whether the compared project document has risks based on the comprehensive similarity.

[0109] In the present invention, when conducting a risk review of a project, the risk information of the acquired risk project document can be parsed, so that keywords and feature vectors with risk information can be obtained. Then the comparison project text that needs to be risk analyzed can be obtained. The first similarity between the risk project document and the comparison project document is determined based on the set of keywords between the risk project document and the comparison project document. The second similarity between the risk project document and the comparison project document can be determined based on the cosine similarity between the feature vector of the sentence of the risk project document and the feature vector of the sentence of the comparison project document. The comprehensive similarity between the risk project document and the comparison project document can be determined based on the first similarity and the second similarity. After obtaining the comprehensive similarity, it can be determined whether the comparison project has risks based on the comprehensive similarity.

[0110] In one embodiment of the present invention, Figure 7 As shown, the process of obtaining the first similarity may include:

[0111] In step S26, a classification set of keywords having risk information of the risk project document is obtained.

[0112] In step S27, a classification set of keywords of the comparison project document is obtained.

[0113] In step S28, the number of the classification set of keywords with risk information in the risk project document and the classification set of keywords in the comparison project document belonging to the same set is determined.

[0114] In step S29, according to the number of keywords belonging to the same classification set, each classification set is assigned a value and weightedly added to obtain a first similarity.

[0115] In the present invention, when obtaining the first similarity between the risk project document and the comparison project document, the classification set of keywords with risk information of the risk project document can be obtained, and then the classification set of keywords of the comparison project document can be obtained. For the classification sets of keywords of the risk project document and the comparison project document, determine the number of the classification sets of keywords with more prominent information of the risk project document and the classification sets of keywords of the comparison project document that belong to the same set. According to the number of the same classification sets of keyword data, the same classification set is determined, and then the classification sets belonging to the same can be assigned values ​​and weighted addition can be performed, so as to obtain the first similarity. After obtaining the first similarity and the second similarity, the first similarity and the second similarity can be weighted addition, so as to obtain the comprehensive similarity.

[0116] Through the above technical scheme, a semantic recognition service system for project document management provided by the present invention includes a storage layer, which can use a Mysql database to store business information, support master-slave library backup, and regular database export backup. The core recognition layer can identify and parse various types of documents in business information, so that all rich text information required in the document can be extracted. The AI ​​core algorithm middle platform can perform semantic recognition and sequence labeling on the information extracted by the core recognition layer. The business application layer can be applied to document classification, document extraction, risk review, document comparison, and text recognition table analysis for project document management. The semantic recognition service system can provide a complete set of semantic recognition service systems for project managers and auditors. Other features and advantages of the embodiments of the present invention will be described in detail in the subsequent specific implementation method section.

[0117] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0118] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0119] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0120] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0121] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0122] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0123] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0124] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0125] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A semantic recognition service system for project document management, characterized in that: The semantic recognition service system comprises: The storage layer uses the MySQL database to store business information, supports master-slave database backup, and regular database export and backup; The core recognition layer identifies and parses various types of documents in business information to extract all the rich text information required in the documents; The AI ​​algorithm middle platform performs semantic recognition and sequence labeling on the information extracted by the core recognition layer; The business application layer is used for document classification, document extraction, risk review, document comparison, and text recognition and table parsing for project document management.

2. The semantic recognition service system according to claim 1, characterized in that: The core identification layer includes: Node parsing module directly parses the PDF protocol, and the returned result is encapsulated as a Char object during rich text parsing, including coordinates, width and height; Table parsing module, identifies the corresponding table area from the PDF document, and parses the table topology and content; The chart parsing module extracts charts from documents through PDF feature extraction for static images and object detection services for generated images. The document structure detection module, after merging the text lines, determines whether it is a header or footer according to certain rules, and merges paragraphs according to the left and right locking of the text lines, the character semantics at the beginning and end of the lines, and the font band if the content is not recognized as a header or footer.

3. The semantic recognition service system according to claim 2, characterized in that: The core recognition layer supports the extraction of key information from Word, PDF, document scan images and unstructured documents.

4. The semantic recognition service system according to claim 1, characterized in that: The AI ​​algorithm platform is used for: Acquire data information extracted from the core identification layer; Cleaning and standardizing the data information; Extract keywords from the cleaned and standardized data to obtain a keyword list that meets the requirements; Performing feature extraction on the keyword list, and performing text classification on the keywords after feature extraction; The classified keywords are sequence labeled.

5. The semantic recognition service system according to claim 4, characterized in that: The cleaned and standardized data information is used for keyword extraction to obtain a keyword list that meets the requirements, including: Get the cleaned and standardized text; According to formula (1), the frequency of each word in the text is calculated: Among them, TF(t,d) represents the frequency of word t in text d, f t,d represents the number of times word t appears in text d, f d Represents the total number of occurrences of all words in the text; The importance of a word in a text is calculated according to formula (2): Among them, IDF(t,D) represents the inverse document frequency of word t in all text sets D, W t represents the number of texts containing word t; The weight of each word is calculated according to formula (3): TF-IDF(t,d,D)=TF(t,d)×IDF(t,D), formula (3) Where TF-IDF(t,d,D) represents the TF-IDF value of word t in text d; Sort each word according to the TF-IDF value, and select the top K words as keywords to get a keyword list that meets the requirements.

6. The semantic recognition service system according to claim 5, characterized in that: Extracting features from the keyword list and classifying the keywords after feature extraction includes: Obtain all keywords in the keyword list; Performing semantic recognition on the keywords to obtain corresponding semantics of each keyword; The keywords with the largest range of semantics commonly represented by each keyword are grouped into a first group to complete the preliminary classification of the keywords; Performing semantic analysis on the keywords in each first set after preliminary classification to obtain the number of categories in the first set, and grouping the keywords in the first set into corresponding categories to obtain a second set; Performing semantic analysis on the keywords in the second set to obtain the number of categories in the second set, and grouping the keywords in the second set into corresponding categories to obtain a corresponding third set; By analogy, the keywords are classified into the corresponding kth set; Each of the keywords is labeled as a set from small to large categories.

7. The semantic recognition service system according to claim 6, characterized in that: The AI ​​algorithm platform is used for: Calculate the TF-IDF value of the words in each sentence according to formula (3) and convert each sentence into a feature vector; Calculate the cosine similarity between the feature vector of each sentence and the feature vectors of the remaining sentences; Two sentences whose cosine similarity is greater than a preset threshold are marked as similar texts.

8. The semantic recognition service system according to claim 7, characterized in that: The AI ​​algorithm platform is used for: Parse the risk information of the acquired risk project documents to obtain keywords and feature vectors containing risk information; Determine a first similarity between the risk project document and the comparison project document according to a set of keywords in the risk project document and keywords in the comparison project document; Determine a second similarity between the risk project document and the comparison project document according to the cosine similarity between the feature vector of the sentence of the risk project document and the feature vector of the sentence of the comparison project document; Determine the comprehensive similarity between the risk project document and the comparison project document according to the first similarity and the second similarity; Determine whether the compared project document has risks based on the comprehensive similarity.

9. The semantic recognition service system according to claim 8, characterized in that: Determining the first similarity between the risk project document and the comparison project document according to the set of keywords in the risk project document and the comparison project document includes: Obtaining a classification set of keywords with risk information from the risk project document; Obtaining a classification set of keywords of the comparison project document; Determine the number of the classification set of keywords with risk information of the risk project document and the classification set of keywords of the comparison project document that belong to the same set; According to the number of keywords belonging to the same classification set, each classification set is assigned a value and weighted addition is performed to obtain a first similarity.

Images