An industry important data recognition method and system based on a large language model

Abstract

The application relates to the technical field of natural language processing, and proposes an industry important data recognition method and system based on a large language model.The method comprises the following steps: acquiring an industry data classification and grading standard file; establishing a Chinese feature library corresponding to feature attributes and categories by using an inverted index, and / or establishing an English feature library corresponding to feature attributes and categories by using a Trie tree; constructing a question set Q for an industry file to be recognized; inputting questions in the question set Q into a large language model one by one to ask the industry file to be recognized, and outputting an answer set A; based on the Chinese feature library and / or the English feature library, performing feature attribute matching on data in the industry file to be recognized in combination with the answer set A, obtaining weights of various feature attribute categories, and performing importance recognition on the industry file to be recognized based on the weights, and outputting an importance level of the industry file to be recognized.

Description

Technical Field

[0001] This invention relates to the field of natural language processing technology, and more specifically, to a method and system for identifying important industry data based on a large language model. Background Technology

[0002] To strengthen the protection of critical data, identifying and protecting critical data plays a crucial role in supporting data security regulatory objectives. With the explosive growth of data volume and the proliferation of multimodal data, identifying, classifying, and effectively protecting critical data faces unprecedented challenges. The development of technologies for identifying critical industry data is not only significant for implementing data security laws, but also provides powerful tools and methods for promoting industry data analysis, enhancing data security supervision, and protecting personal privacy.

[0003] Currently, key data identification technologies in the industry primarily focus on processing structured data, employing methods such as regular expressions to identify specific field attributes. For example, in identifying key industry data, the application of regular expressions allows for the rapid and accurate filtering and extraction of data items that meet specific format requirements, such as ID numbers, phone numbers, and email addresses, which are often key criteria for identifying important data. However, when dealing with semi-structured and unstructured data, relying solely on regular expression matching techniques has significant limitations. For semi-structured data, such as XML or JSON files, although they contain a certain degree of structure, the flexibility and complexity of their data organization make it difficult for simple regular expressions to efficiently identify and parse them. Unstructured data, such as long text and images, lacks a fixed format and structure, making it difficult for traditional methods to effectively extract the information required for "industry important data classification and grading standards." Summary of the Invention

[0004] To overcome the shortcomings of existing technologies for identifying important industry data, which are unable to efficiently identify and parse semi-structured and unstructured data, this invention provides a method and system for identifying important industry data based on a large language model.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0006] A method for identifying industry-critical data based on a large language model includes the following steps:

[0007] Obtain industry data classification and grading standard documents;

[0008] Based on industry data classification and grading standard documents, an inverted index is used to build a Chinese feature library corresponding to feature attributes and categories, and / or a Trie tree is used to build an English feature library corresponding to feature attributes and categories.

[0009] Construct a set of questions Q for the industry documents to be identified;

[0010] The questions in the question set Q are input one by one into the large language model to ask questions in the industry document to be recognized, and the answer set A is output.

[0011] Based on the Chinese feature library and / or English feature library, and combined with the answer set A, feature attribute matching is performed on the data in the industry documents to be identified to obtain the weight of each feature attribute category. Based on the weight, the importance of the industry documents to be identified is identified, and the importance level of the industry documents to be identified is output.

[0012] Furthermore, this invention proposes an industry-critical data identification system based on a large language model, applying the industry-critical data identification method based on a large language model proposed in this invention. The system includes:

[0013] The input module is used to obtain industry data classification and grading standard files and industry files to be identified;

[0014] The feature library construction module is used to build a Chinese feature library corresponding to feature attributes and categories based on industry data classification and grading standard files, using an inverted index, and / or to build an English feature library corresponding to feature attributes and categories using a Trie tree;

[0015] The question generation module is used to construct a set of questions Q for the industry documents to be identified;

[0016] The answer generation module, which is equipped with a large language model, is used to input any question from the question set Q into the large language model to ask questions in the industry document to be identified, and output the answer set A;

[0017] The importance identification module is used to perform feature attribute matching on the data in the industry documents to be identified based on the Chinese feature library and / or English feature library, combined with the answer set A, to obtain the weight of each feature attribute category, and to identify the importance of the industry documents to be identified based on the weights, and output the importance level of the industry documents to be identified.

[0018] Furthermore, the present invention also proposes a computer device including a memory and a processor, wherein the memory stores computer-readable instructions, which, when executed by the processor, cause the processor to perform the steps of the industry-critical data identification method as described in the present invention.

[0019] Furthermore, the present invention also proposes a storage medium storing computer-readable instructions thereon, which, when executed by a processor, implement the steps of the industry-critical data identification method as described in the present invention.

[0020] Compared with existing technologies, the beneficial effects of the technical solution of this invention are as follows: This invention uses a large language model to identify important industry data, which can more accurately understand and interpret the content in semi-structured or unstructured data and can adapt to data of various formats and structures; This invention uses inverted indexes and Trie trees to establish Chinese feature libraries and English feature libraries, and uses efficient index structures to achieve feature matching for identifying the importance of documents, thereby improving the accuracy and efficiency of identification. Attached Figure Description

[0021] Figure 1 This is a flowchart of the industry-critical data identification method based on a large language model according to the present invention.

[0022] Figure 2 This is a schematic diagram of the Trie tree in Example 1.

[0023] Figure 3 This is an architecture diagram of the industry-critical data recognition system based on a large language model, as presented in this invention. Detailed Implementation

[0024] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the invention.

[0025] It will be understood by those skilled in the art that some well-known descriptions may be omitted in the accompanying drawings.

[0026] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0027] Example 1

[0028] This embodiment proposes a method for identifying important industry data based on a large language model, such as... Figure 1 The diagram shown is a flowchart of the industry-important data identification method based on a large language model in this embodiment.

[0029] The industry-critical data identification method based on a large language model proposed in this embodiment includes the following steps:

[0030] S1. Obtain industry data classification and grading standard documents;

[0031] S2. Based on industry data classification and grading standard documents, use inverted indexes to build a Chinese feature library corresponding to feature attributes and categories, and / or use Trie trees to build an English feature library corresponding to feature attributes and categories.

[0032] S3. Construct a set of questions Q for the industry documents to be identified;

[0033] S4. Input the questions in the question set Q into the large language model one by one to ask questions in the industry document to be recognized, and output the answer set A;

[0034] S5. Based on the Chinese feature library and / or English feature library, and in conjunction with the answer set A, perform feature attribute matching on the data in the industry documents to be identified, obtain the weight of each feature attribute category, and identify the importance of the industry documents to be identified based on the weights, and output the importance level of the industry documents to be identified.

[0035] This embodiment uses a large language model to identify important industry data, which can more accurately understand and interpret the content of semi-structured or unstructured data and can adapt to data of various formats and structures.

[0036] Furthermore, this invention utilizes inverted indexes and Trie trees to establish Chinese and English feature databases, and employs efficient index structures to achieve feature matching for identifying the importance of files, thereby improving the accuracy and efficiency of identification.

[0037] For Chinese text, inverted indexes are better suited for handling the need for fast retrieval of feature attributes, making them suitable for large-scale document collections. Trie trees are suitable for storing and searching strings in character order, and are particularly efficient when handling prefix matching queries. In English text, because English words naturally have character-level prefix relationships, Trie trees can handle this situation very well.

[0038] In an optional embodiment, step S2 involves establishing a Chinese feature library corresponding to feature attributes and categories using an inverted index, including the following steps:

[0039] Feature extraction is performed on the industry data classification and grading standard file to construct a data category set S and a feature attribute set F;

[0040] An inverted index table is built based on the mapping I between data categories and feature attributes, and it is represented as follows:

[0041] I: F→P(S)

[0042] Where P(S) represents the power set of the data category set S, and f represents all possible subsets of set S; for each feature attribute f∈F, the mapping I(f) produces a subset. S f For all categories containing feature f;

[0043] Based on the inverted index I(f), all categories S corresponding to feature attribute f are... f The features are categorized and stored to obtain a Chinese feature library.

[0044] As an example, let's define a data category set S = {"Financial Personal Information", "Medical Personal Information", "Contact Information"}; and a feature attribute set F = {"Name", "Gender", "Nationality", "Home Address", "ID Card Number", "Health Card Number", "Mobile Phone Number", "Email Address"}. The mapping will then be established as follows:

[0045] I("Name") = {"Financial Personal Information", "Medical Personal Information"};

[0046] I("Gender") = {"Financial Personal Information"};

[0047] I("Nationality") = {"Financial Personal Information"};

[0048] I("Home Address") = {"Financial Personal Information"};

[0049] I("ID card") = {"Medical Personal Information"};

[0050] I("Health Card Number") = {"Medical Personal Information"};

[0051] I("Mobile phone") = {"Contact information"};

[0052] I("email address") = {"contact information"}.

[0053] To find all categories containing a specific feature f ∈ F, querying the mapping I(f) allows for a quick search of all categories S corresponding to any feature attribute f. f For example, querying I("Name") will return {"Financial Personal Information", "Medical Personal Information"}, indicating that the "Name" feature appears in both data categories.

[0054] In an optional embodiment, step S2 involves constructing an English feature library corresponding to feature attributes and categories using a Trie tree, including the following steps:

[0055] Feature extraction is performed on the industry data classification and grading standard document to obtain an English feature set;

[0056] Initialize the Trie tree; traverse the English feature set and insert each English feature into the Trie tree to obtain the English feature library;

[0057] In this Trie tree, the root node does not contain any character, and each intermediate node represents a character; any node and the nodes between the root node and its terminal node form an English feature related to the feature attributes; the terminal node stores the corresponding data category.

[0058] As an example, a Trie tree contains all features, such as "name," "gender," "personal income status," etc., and each terminal node (i.e., the end of each word) is associated with the data category to which that feature belongs, such as "Personal information," etc. Figure 2 The diagram shown is a schematic of the Trie tree in this embodiment.

[0059] When a new word is inserted, such as "status of real estate owned", the Trie tree will create or traverse nodes along each character until the end of the word, and mark the last node as the terminal node, recording the data category to which the feature belongs: "Personal property data".

[0060] When performing a search, the system traverses character by character from the root node. If the entire word can be traversed and the last node is marked as a terminal node, the category information stored in that terminal node is returned. If a character is missing along the way, it means that the word is not in the tree.

[0061] In an optional embodiment, step S3 involves constructing a set of queries Q for the industry document to be identified, including the following steps:

[0062] Based on industry data classification and grading rules, question templates are set using a structured query language format.

[0063] The question template is converted into a natural language question, forming a question set Q.

[0064] For example, Table 1 below shows an example of industry data classification and grading rules in this embodiment.

[0065] Table 1 Examples of Industry Data Classification and Grading Rules

[0066]

[0067]

[0068] Furthermore, as an example, the question template is set based on a structured query language (SQL) format, as follows:

[0069] Select[All|Distinct]<feature attributes>[, ...

[0070] From <filename>[, ...

[0071] [Where<conditional expression>];

[0072] This question template indicates that a new set of records should be extracted from a specified file, consisting of records within a specified range that meet certain conditions and have specified characteristic attributes.

[0073] For example, the question template is set as follows:

[0074] SELECT Name, Gender, Nationality, Ethnicity, MaritalStatus, DocumentType, DocumentNumber, IssueDate, ExpiryDate, Address

[0075] FROM Document

[0076] WHERE DocumentID = 'C';

[0077] The template above indicates how to extract basic information such as name, gender, and nationality from document C.

[0078] Furthermore, it is transformed into a natural language problem, optionally using natural language templates, natural language processing (NLP) techniques, knowledge base dictionaries, machine learning methods, etc.

[0079] For example, the above question template can be converted into a natural language question as follows: "Please extract the following named entities from document C: name, gender, nationality, ethnicity, marital status, document type, document number, document effective date, document expiry date, and home address".

[0080] In an optional embodiment, step S4 involves inputting each question in the question set Q into the large language model to extract importance indicator features from the industry documents to be identified, including the following steps:

[0081] Input the natural language questions from the question set Q into the large language model, and output the generative answer a. i ; where, for question q i The answer a is obtained through a function based on the industry document C to be identified. i Its expression is:

[0082] F i (C,q i ) = a i

[0083] Among them, F i (·) indicates that information about question q is extracted from the industry document C to be identified. i Named entities; for all questions q in the question set Q. i After feature extraction is completed, the result is obtained from the answer a. i The set of answers A is formed.

[0084] In this embodiment, the large language model provides a generative answer by understanding the question and analyzing the document content.

[0085] For example, the input industry document C to be identified is: "Zhang San is a male, a Chinese citizen, and currently married. His ID number is 110101198012345678, and his permanent address is located at No. D, Street C, District B, City A. Regarding personal assets, Zhang San demonstrates strong economic strength and social security. He currently works as a senior project manager in a multinational company, with an annual total income of approximately 2 million yuan, covering a fixed salary, performance bonuses, and equity incentive income. Regarding real estate, Zhang San and his spouse each own a property in Haidian District, Beijing, and Sanya City, Hainan Province, with a total market value of up to 17 million yuan. In terms of tax performance, Zhang San's total tax payment in the past three years has exceeded 2 million yuan, maintaining a good record of no tax arrears. In addition, he strictly complies with national regulations, having accumulated 1.5 million yuan in social security contributions and a housing provident fund account balance of 800,000 yuan."

[0086] The generated question set Q is:

[0087] q1: "Please extract the following named entities from file C: name, gender, nationality, ethnicity, marital status, document type, document number, document effective date, document expiry date, and home address";

[0088] q2: "Please extract the following named entities from file C: personal income status, real estate ownership status, vehicle ownership status, tax payment, housing provident fund contribution amount, and personal social security and medical insurance contribution amount";

[0089] q3: "Please extract the following named entities from file C: medical insurance payment information, transaction amount, transaction record, insurance status, and insurance amount."

[0090] Then define the recognition function F. i for:

[0091] F1(C, q1) = {"Name: Zhang San", ...}

[0092] Gender: Male

[0093] Nationality: China

[0094] Marital Status: Married

[0095] "Document Type: Resident Identity Card"

[0096] "ID Number: 1101011980012345678"

[0097] Home address: No. D, Street C, District B, City A;

[0098] F2(C, q2) = {"Personal income status: Zhang San works as a senior project manager in a multinational company, with an annual total income of approximately 2 million yuan, including fixed salary, performance bonuses, and equity incentive income",}

[0099] "Real estate holdings: Zhang San and his spouse own one property each in Haidian District, Beijing and Sanya City, Hainan Province, with a total market value of up to 17 million yuan."

[0100] "Tax payment: In the past three years, Zhang San's total tax payment has exceeded 2 million yuan."

[0101] "Housing provident fund contribution amount: The balance in the housing provident fund account has reached 800,000 yuan."

[0102] "Personal social security and medical insurance contributions: The personal social security account has accumulated 1.5 million yuan in contributions."

[0103]

[0104] Then we get the answer set A = F1(C, q1)∪F2(C, q2)∪F3(C, q3).

[0105] Further, in an optional embodiment, in step S5, feature attribute matching is performed on the data in the industry document to be identified based on the answer set A to obtain the weight of each feature attribute category, and the importance of the industry document to be identified is identified based on the weight, including the following steps:

[0106] The answer set A is processed to retain the names of the feature attributes in the answer set A, thus obtaining the answer feature attribute set;

[0107] Based on the set of answer feature attributes, a data category retrieval is performed in the Chinese feature database and / or the English feature database to obtain the data category to which any answer feature attribute belongs;

[0108] Calculate the weight G of the data category to which the answer feature attribute belongs; where the weight G of the data category is initially 0, and when an answer feature attribute matches a data category, the weight G of that data category is incremented by 1;

[0109] The importance of industry documents to be identified is determined based on weights, and the expression is as follows:

[0110] Importance_level=U(C,H,G)

[0111] Where U(·) is the evaluation function, C is the industry document to be identified, G is the security level set by the data category in the industry data classification and grading standard, and Importance_level represents the importance value of the data category;

[0112] The highest value among the importance values ​​of each data category is taken as the importance level of the industry document to be identified.

[0113] For example, processing the above answer set A, retaining the names of the feature attributes and removing the specific values, yields the answer feature attribute set A' = {name, gender, nationality, marital status, document type, document number, home address, personal income, real estate ownership, tax amount, housing provident fund contribution amount, personal social security and medical insurance contribution amount}. This results in the feature attributes, as shown in set A', contained in the industry document C to be identified.

[0114] Furthermore, categories are quickly retrieved based on feature attributes using a Chinese feature library. For example, a Chinese feature library built using an inverted index is represented as follows:

[0115] inverted_index={

[0116] "Name":{"Financial Personal Information", "Medical Personal Information"},

[0117] "Gender":{"Financial Personal Information"}

[0118] "Nationality":{"Financial Personal Information"}

[0119] "Ethnicity":{"Financial Personal Information"}

[0120] "Marital Status":{"Financial Personal Information"}

[0121] "Document Type":{"Financial Personal Information"}

[0122] "ID Number":{"Financial Personal Information"}

[0123] "Document Effective Date":{"Financial Personal Information"}

[0124] "Document Expiry Date":{"Financial Personal Information"}

[0125] "Current Location":{"Geographic Location Data"}

[0126] "Personal Income Status": {"Financial Personal Assets"}

[0127] "Status of real estate holdings": {"Financial personal assets"}

[0128] "Vehicle Ownership Status": {"Financial Personal Assets"}

[0129] "Tax Payment": {"Personal Financial Assets"}

[0130] "Amount of Housing Provident Fund Contributions": {"Personal Financial Assets"}

[0131] "Individual social security and medical insurance contributions": {"Personal financial assets"}

[0132] "Medical insurance payment information": {"medical payment data"}

[0133] "Transaction Amount":{"Medical Payment Data"}

[0134] "Transaction Records":{"Medical Payment Data"}

[0135] "Insurance Status":{"Medical Payment Data"}

[0136] "Insured Amount":{"Medical Payment Data"}

[0137] };

[0138] By querying the aforementioned Chinese feature database, the data category to which any answer feature attribute belongs in the answer feature attribute set A' can be determined.

[0139] Further calculate the weight H of the data category to which the answer's feature attribute belongs; initially, assume that the weight of each data category is 0. When a feature attribute matches a category C, then the weight H of category C is... C Add 1. For example, if the set of answer feature attributes A' contains the feature attribute "name", then the weights of the categories "financial personal information" and "medical personal information" are both increased by 1.

[0140] For example, based on the feature attributes in the above set A', the weight statistics for each category are as follows:

[0141] The weight of "Financial Personal Information" is 7; the weight of "Financial Personal Assets" is 5; and the weight of "Medical Personal Information" is 1.

[0142] The weights mentioned above reflect the number of data categories that match the feature attributes in set A'.

[0143] Furthermore, the importance of the industry documents to be identified is determined based on the evaluation function. The evaluation function may optionally employ machine learning methods such as Bayesian classifiers, SVMs, or deep learning models, or it may be based on manually defined rules.

[0144] Example 2

[0145] This embodiment proposes an industry-critical data identification system based on a large language model, applying the industry-critical data identification method based on a large language model proposed in Embodiment 1, such as... Figure 3 The diagram shown is an architecture diagram of the industry-critical data recognition system based on a large language model in this embodiment.

[0146] The industry-critical data identification system based on a large language model proposed in this embodiment includes:

[0147] The input module is used to obtain industry data classification and grading standard files and industry files to be identified;

[0148] The feature library construction module is used to build a Chinese feature library corresponding to feature attributes and categories based on industry data classification and grading standard files, using an inverted index, and / or to build an English feature library corresponding to feature attributes and categories using a Trie tree;

[0149] The question generation module is used to construct a set of questions Q for the industry documents to be identified;

[0150] The answer generation module, which is equipped with a large language model, is used to input any question from the question set Q into the large language model to ask questions in the industry document to be identified, and output the answer set A;

[0151] The importance identification module is used to perform feature attribute matching on the data in the industry documents to be identified based on the Chinese feature library and / or English feature library, combined with the answer set A, to obtain the weight of each feature attribute category, and to identify the importance of the industry documents to be identified based on the weights, and output the importance level of the industry documents to be identified.

[0152] It is understood that the system in this embodiment corresponds to the method in Embodiment 1 above, and the options in Embodiment 1 above are also applicable to this embodiment, so they will not be described again here.

[0153] Example 3

[0154] This embodiment proposes a computer device, including a memory and a processor. The memory stores computer-readable instructions, which, when executed by the processor, cause the processor to perform the steps of the industry-important data recognition method based on a large language model as proposed in Embodiment 1.

[0155] Example 4

[0156] This embodiment proposes a storage medium storing computer-readable instructions, wherein when the computer-readable instructions are executed by a processor, they implement the steps of the industry-important data identification method based on a large language model proposed in Embodiment 1.

[0157] By way of example, the storage medium includes, but is not limited to, USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks or optical disks, and other media capable of storing program code.

[0158] By way of example, the instructions, programs, code sets, or instruction sets may be implemented using conventional programming languages.

[0159] By way of example, the processor includes, but is not limited to, smartphones, personal computers, servers, network devices, etc., for performing all or part of the steps of the industry-critical data identification method based on a large language model as described in Example 1.

[0160] The terminology used in the accompanying drawings is for illustrative purposes only and should not be construed as limiting the invention.

[0161] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art can make other variations or modifications based on the above description. It is neither necessary nor possible to exhaustively describe all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A method for identifying industry-critical data based on a large language model, characterized in that, Includes the following steps: Obtain industry data classification and grading standard documents; Based on industry data classification and grading standard documents, an inverted index is used to build a Chinese feature library corresponding to feature attributes and categories, and / or a Trie tree is used to build an English feature library corresponding to feature attributes and categories. Construct a set of questions Q for the industry documents to be identified; The questions in the question set Q are input one by one into the large language model to extract the importance index features of the industry documents to be identified, and the answer set A is obtained. Based on the Chinese feature library and / or English feature library, the data in the industry documents to be identified are matched with feature attributes in combination with the answer set A to obtain the weight of each feature attribute category, and the importance of the industry documents to be identified is identified based on the weight, and the importance level of the industry documents to be identified is output. The step of establishing a Chinese feature library corresponding to feature attributes and categories using an inverted index includes the following steps: Feature extraction is performed on the industry data classification and grading standard document to construct a data category set. S and feature attribute set F ; Based on the mapping between data categories and feature attributes I An inverted index table is created, represented as follows: in, Represents a collection of data categories S The power set, representing the set S All possible subsets; for each feature attribute f ∈ F , mapping I ( f Produce a subset , S f For all included features f Category; Based on inverted index I ( f ), to feature attributes f All corresponding categories Categorize and store the data to obtain a Chinese feature library; The method of constructing an English feature library corresponding to feature attributes and categories using a Trie tree includes the following steps: Feature extraction is performed on the industry data classification and grading standard document to obtain an English feature set; Initialize the Trie tree; Traverse the English feature set and insert each English feature into a Trie tree to obtain the English feature library; In this Trie tree, the root node does not contain any character, and each intermediate node represents a character; any node and the nodes between the root node and its terminal node form an English feature related to the feature attributes; the terminal node stores the corresponding data category.

2. The method for identifying important industry data according to claim 1, characterized in that, The construction of the query set Q for the industry document to be identified includes the following steps: Based on industry data classification and grading rules, question templates are set using a structured query language format. The question template is converted into a natural language question, forming a question set Q.

3. The method for identifying important industry data according to claim 1, characterized in that, The process of inputting each question in the question set Q into the large language model to extract importance indicator features from the industry documents to be identified includes the following steps: The natural language questions in the question set Q are input into the large language model, and generative answers are output. Among them, regarding the problem q i Based on the industry document to be identified C The answer is obtained through the function. a i Its expression is: in, Indicates from the industry document to be identified C Extract information about the question q i Named entities; for all questions in the question set Q. q i After feature extraction is completed, the answer is obtained. a i The set of answers A is formed.

4. The method for identifying important industry data according to any one of claims 1 to 3, characterized in that, The data in the industry documents to be identified are matched with the answer set A to obtain the weights of each feature attribute category. Based on these weights, the importance of the industry documents to be identified is determined, including the following steps: The answer set A is processed to retain the names of the feature attributes in the answer set A, thus obtaining the answer feature attribute set; Based on the set of answer feature attributes, a data category retrieval is performed in the Chinese feature database and / or the English feature database to obtain the data category to which any answer feature attribute belongs; Calculate the weight of the data category to which the feature attribute of the answer belongs. H Among them, the weight of data categories H The initial value is 0. When an answer's feature attribute matches a data category, the weight of that data category is... H Add 1; The importance of industry documents to be identified is determined based on weights, and the expression is as follows: Importance_level = U ( C ， H ， G ) Where U(·) is the evaluation function, C For the industry document to be identified, G The security level set for data categories in the industry data classification and grading standards; Importance_level This indicates the importance value of the data category; The highest value among the importance values ​​of each data category is taken as the importance level of the industry document to be identified.

5. The method for identifying important industry data according to claim 4, characterized in that, The evaluation function includes a Bayesian classifier, an SVM, or a deep learning model.

6. A system for identifying important industry data based on a large language model, employing the method for identifying important industry data according to any one of claims 1 to 5, characterized in that, include: The input module is used to obtain industry data classification and grading standard files and industry files to be identified; The feature library construction module is used to build a Chinese feature library corresponding to feature attributes and categories based on industry data classification and grading standard files, using an inverted index, and / or to build an English feature library corresponding to feature attributes and categories using a Trie tree; The question generation module is used to construct a set of questions Q for the industry documents to be identified; The answer generation module, which is equipped with a large language model, is used to input any question from the question set Q into the large language model to ask questions in the industry document to be identified, and output the answer set A; The importance identification module is used to perform feature attribute matching on the data in the industry documents to be identified based on the Chinese feature library and / or English feature library, combined with the answer set A, to obtain the weight of each feature attribute category, and to identify the importance of the industry documents to be identified based on the weights, and output the importance level of the industry documents to be identified.

7. A computer device comprising a memory and a processor, wherein the memory stores computer-readable instructions, characterized in that, When the computer-readable instructions are executed by the processor, the processor performs the steps of the industry-critical data identification method as described in any one of claims 1 to 5.

8. A storage medium having computer-readable instructions stored thereon, characterized in that, When the computer-readable instructions are executed by a processor, they implement the steps of the industry-critical data identification method as described in any one of claims 1 to 5.

Images