Secure text checking method based on positive and negative dot information

Abstract

The application discloses a secret text checking method based on positive and negative point information, which comprises the following steps: extracting sample file text content information; dividing the sample file text content information into a plurality of independent information pieces; training a secret checking neural network to obtain a positive and negative semantic feature library; dividing text content information of a file to be judged into N independent information pieces; matching Hash values of the information pieces of the file to be judged with Hash values of feature information pieces in the semantic feature library; assigning a similarity value of 100% to the information pieces of the file to be judged and the feature information pieces which are successfully matched; extracting semantic features of the information pieces of the file to be judged which are not successfully matched, and calculating the similarity of the information pieces of the file to be judged and the feature information pieces; scoring the information pieces of the file to be judged; calculating a score of the file to be judged; and judging whether the file to be judged involves a secret according to the score of the file to be judged.

Description

Technical Field

[0001] The present invention belongs to the technical field of confidential information inspection, and in particular relates to a confidential text inspection method based on positive and negative secret point information. Background Art

[0002] Currently, due to confidentiality inspection business needs, it is necessary to conduct confidentiality inspections on massive files in office terminals to determine whether the files are confidential files illegally stored in office terminals. The existing confidentiality inspection method is based on sensitive keyword recognition and feedback of the context information of the matched sensitive keyword to the confidentiality inspection administrator for judgment. This method requires manual judgment, and its judgment result cannot have a positive or negative impact on the judgment of the next file. In the confidentiality inspection process for massive files, the number of files that match sensitive keyword words is huge, and each information requires manual judgment, which has low inspection efficiency and cannot adapt to the confidentiality inspection needs in the rapid development of office digitalization. Summary of the Invention

[0003] In order to solve the problems existing in the prior art, the present invention provides a confidential text checking method based on positive and negative secret point information, which can quickly check whether the text is confidential.

[0004] The technical problem to be solved by the present invention is achieved through the following technical solutions:

[0005] A confidential text checking method based on positive and negative secret point information is provided, including:

[0006] Extracting text content information of sample files;

[0007] Divide the extracted sample file text content information into several independent information pieces;

[0008] The semantic feature training of the confidentiality check neural network is performed through the information pieces of the sample file text content information to obtain the positive and negative semantic feature library;

[0009] Calculate the hash value of the feature information piece in the positive and negative semantic feature library;

[0010] Divide the text content information of the file to be determined into N independent information pieces;

[0011] Calculate the hash value of each piece of information in the file to be judged;

[0012] Match the hash value of each information piece of the file to be judged with the hash value of the feature information piece in the positive and negative semantic feature library, and assign the positive similarity value or negative similarity value of the successfully matched information piece of the file to be judged and the feature information piece to 100%;

[0013] Semantic feature extraction is performed on the information piece of the unmatched file to be determined, and the positive similarity value and negative similarity value between the information piece of the file to be determined and the feature information piece are obtained based on the extracted semantic features and the positive and negative semantic feature library;

[0014] Scoring the information piece of the file to be determined based on the positive similarity value and negative similarity value between the information piece of the file to be determined and the feature information piece;

[0015] Calculate the score of the file to be determined according to the score of the information piece of the file to be determined;

[0016] Determine whether the file to be judged is confidential based on its score.

[0017] Furthermore, the scoring of the information piece of the file to be judged includes:

[0018] The positive density point score of the information piece of the file to be judged is obtained by formula (1):

[0019] (1)

[0020] in, is a positive density point score, Indicates the The weight of the information piece of the file to be judged;

[0021] Similarly, the negative density point score is obtained by formula (2):

[0022] (2)

[0023] Calculate the density point score of the information piece of the file to be judged according to formula (3)

[0024] (3)

[0025] Indicates the The key point score of the information piece of the file to be judged.

[0026] Furthermore, calculating the score of the file to be judged includes:

[0027] Calculate the total score of the file to be judged according to formula (4)

[0028] (4)

[0029] in, is the score of the file to be judged, is the total score of the file to be judged, The number of pieces into which the file to be determined is divided.

[0030] Furthermore, the step of determining whether the file to be determined is confidential based on the score of the file to be determined includes:

[0031] Determine the score range: ;

[0032] If the score of the file to be judged is in [0, a), it is judged as a non-confidential file; if the score is in , it is determined to be a confidential document; if the score is , a second review will be conducted manually.

[0033] Furthermore, if it is determined to be a confidential file, the information piece contained in the confidential file will be used as the positive feedback input of the positive and negative semantic feature library; if it is determined to be a non-confidential file, the information piece contained in the confidential file will be used as the negative feedback input of the positive and negative semantic feature library.

[0034] Furthermore, the positive and negative semantic features in the positive and negative semantic feature library must satisfy orthogonality.

[0035] The beneficial effects of the present invention include: (1) the file confidentiality check of the present invention is to check the text information in fragments, which is different from the traditional whole article check. By quickly matching the positive and negative hash values ​​of the fragmented text information, the efficiency of the file confidentiality check is improved and the granularity of the confidentiality check is improved;

[0036] (2) The present invention enhances the essential inspection of text semantics and expression meaning by training the semantic features of sample text fragment information and calculating the semantic similarity of the sample text fragment information to be tested, thereby improving the accuracy of confidentiality inspection;

[0037] (3) The present invention constructs a positive and negative feedback secret point information database based on positive and negative sample information, and performs positive and negative feedback on the secret point information database through the results of each confidentiality check. While continuously improving the accuracy of confidentiality checks, it can also adapt to dynamic confidentiality checks of secret point information in different periods and different business scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 The flowchart of the confidential text checking method based on positive and negative secret point information of the present invention. DETAILED DESCRIPTION

[0039] In order to further illustrate the technical features and effects of the present invention, the present invention is further described below with reference to the accompanying drawings and specific embodiments.

[0040] like Figure 1 As shown, a confidential text checking method based on positive and negative secret point information is provided. The steps included in the present invention are mainly as follows:

[0041] 1) Extracting sample file text content information

[0042] For formatted documents, text parsing is used; for image files, text information is extracted through OCR recognition.

[0043] 2) Divide the extracted sample file text content information into several independent information pieces

[0044] The sample file converted into text is segmented according to the main idea of ​​the paragraphs, and the main idea of ​​each paragraph is ensured to be independent of each other.

[0045] 3) Build a positive and negative semantic feature library

[0046] The confidentiality personnel classify the information pieces of the sample files after segmentation, including the subject scope and whether it is confidential.

[0047] The confidentiality check neural network is trained using labeled sample file information pieces to obtain a library of positive and negative semantic features. The library includes positive feature information pieces (confidential, positive confidential points) and negative feature information pieces (non-confidential, negative confidential points), and one feature information piece is used as a confidential point feature.

[0048] 4) Calculate the hash value of the feature information piece in the positive and negative semantic feature library

[0049] 5) Divide the text content of the file to be judged into N independent pieces of information, and calculate the hash value of each piece of information of the file to be judged;

[0050] 6) Match the hash value of each information piece of the file to be judged with the hash value of the feature information piece in the positive and negative semantic feature library. If it matches the positive feature information piece successfully, the positive similarity between the two is assigned to 100%. If it matches the negative feature information piece successfully, the negative similarity between the two is assigned to 100%.

[0051] 7) Semantic features are extracted from the information pieces of the unmatched files to be determined. Based on the extracted semantic features and the positive and negative semantic feature libraries, the positive and negative similarity values ​​between the information pieces of the files to be determined and the feature information pieces are obtained.

[0052] 8) Score the information piece of the document to be judged;

[0053] First, the positive and negative similarity values ​​between the information piece and the feature information piece of the file to be judged are calculated, and then the scores are given according to the weight coefficient of each dense point feature (feature information piece), that is,

[0054] (1)

[0055] in, is a positive density point score, Indicates the The weight of the information piece of the file to be judged;

[0056] Similarly, the negative density point score is obtained by formula (2):

[0057] (2)

[0058] Then, the density point score of the information piece of the file to be judged is calculated according to formula (3):

[0059] (3)

[0060] Indicates the The key point score of the information piece of the file to be judged.

[0061] 9) Calculate the score of the file to be determined based on the score of the information piece of the file to be determined;

[0062] Calculate the total score of the file to be judged according to formula (4)

[0063] (4)

[0064] in, is the score of the file to be judged, is the total score of the file to be judged, The number of pieces into which the file to be determined is divided.

[0065] 10) Determine whether the file to be judged is confidential based on its score

[0066] For scores in the range of [0, a), they are used as negative feedback inputs for the positive and negative dense point libraries. For scores in the range of (b, 100], they are used as positive feedback inputs for the positive and negative semantic feature libraries. For scores in the range of [a, b], a secondary analysis is performed manually to determine whether it is positive or negative feedback. While performing positive and negative feedback inputs, the dense point feature information should also be calculated to determine whether it meets the orthogonality of the positive and negative semantic feature libraries to ensure that the positive and negative semantic feature libraries always meet orthogonality after feedback (i.e., the overlap between any two dense points in the feature library is less than 20%).

[0067] Explanation of terms

[0068] Positive and negative confidential points: Positive confidential points refer to key content information points of confidential files, and negative confidential points refer to key content information points of non-confidential files.

[0069] Text segmentation: select one or several consecutive paragraphs from the entire text and segment them relatively independently according to their meaning. In the actual implementation process, the text information can be segmented according to the first-level title, second-level title or N-level title.

[0070] Semantic orthogonality: Indicates that the semantic similarity between two pieces of text information is very low. The similarity value can be calculated. If it is lower than a certain value (for example, 20%), it is considered to meet semantic orthogonality.

[0071] 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 take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0072] 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 block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks 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 produce 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.

[0073] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0074] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0075] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. A confidential text checking method based on positive and negative secret point information, characterized in that: include: Extracting text content information of sample files; Divide the extracted sample file text content information into several independent information pieces; Classify information pieces, including subject matter and whether they are classified; The confidentiality check neural network is trained with semantic features using the labeled sample file text content information pieces to obtain a positive and negative semantic feature library; the positive and negative semantic feature library includes positive feature information pieces and negative feature information pieces; the positive feature information pieces indicate confidentiality, and the negative feature information pieces indicate non-confidentiality; Calculate the hash value of the feature information piece in the positive and negative semantic feature library; Divide the text content information of the file to be determined into N independent information pieces; Calculate the hash value of each piece of information in the file to be judged; Match the hash value of each information piece of the file to be determined with the hash value of the feature information piece in the positive and negative semantic feature library, and assign a positive similarity value or a negative similarity value of 100% to the information piece of the file to be determined and the feature information piece that is successfully matched; the positive similarity represents the similarity with the positive feature information piece, and the negative similarity represents the similarity with the negative feature information piece; Semantic feature extraction is performed on the information piece of the unmatched file to be determined, and the positive similarity value and negative similarity value between the information piece of the file to be determined and the feature information piece are obtained based on the extracted semantic features and the positive and negative semantic feature library; Scoring the information piece of the file to be determined based on the positive similarity value and negative similarity value between the information piece of the file to be determined and the feature information piece; Calculate the score of the file to be determined according to the score of the information piece of the file to be determined; Determine whether the file to be judged is confidential based on its score.

2. A confidential text checking method based on positive and negative secret point information according to claim 1, characterized in that: The scoring of the information piece of the file to be judged includes: The positive density point score of the information piece of the file to be judged is obtained by formula (1): （1） in, is a positive density point score, Indicates the The weight of the information piece of the file to be judged; Similarly, the negative density point score is obtained by formula (2): （2） Calculate the density point score of the information piece of the file to be judged according to formula (3) （3） Indicates the The key point score of the information piece of the file to be judged.

3. The method for checking confidential text based on positive and negative secret point information according to claim 2, characterized in that: Calculating the score of the file to be judged includes: Calculate the total score of the file to be judged according to formula (4) （4） in, is the score of the file to be judged, is the total score of the file to be judged, The number of pieces into which the file to be determined is divided.

4. The method for checking confidential text based on positive and negative secret point information according to claim 3, characterized in that: The determining whether the file to be determined is confidential based on the score of the file to be determined includes: Determine the score range: ; If the score of the file to be judged is in [0, a), it is judged as a non-confidential file; if the score is in , it is determined to be a confidential document; if the score is , a second review will be conducted manually.

5. The method for checking confidential text based on positive and negative secret point information according to claim 4, characterized in that: If it is determined to be a confidential file, the information piece contained in the confidential file will be used as the positive feedback input of the positive and negative semantic feature library. If it is determined to be a non-confidential file, the information piece contained in the confidential file will be used as the negative feedback input of the positive and negative semantic feature library.

6. The method for checking confidential text based on positive and negative secret point information according to claim 1, characterized in that: The positive and negative semantic features in the positive and negative semantic feature library must satisfy orthogonality.

Images