Intelligent question-answering method and system based on information security protection

Abstract

The application discloses an intelligent question and answer method and system based on information security protection, and the method comprises the following steps: when a user question is received, it is judged whether external knowledge needs to be called to reply; if it is judged that the external knowledge needs to be called to reply, it is judged whether the user question meets a question layering condition; if it is judged that the user question meets the question layering condition, the user question is disassembled into multiple sub-questions by using a corresponding layering strategy; it is respectively judged whether confidential information is contained in each sub-question; if it is judged that the sub-question contains the confidential information, information leakage prevention processing is performed on the sub-question containing the confidential information; the sub-questions not containing the confidential information and the sub-questions after the information leakage prevention processing are replied. Therefore, the reliability of question reply is improved, and the safety of information in the question and answer process is greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent question answering, and in particular to an intelligent question answering method and system based on information security protection. Background Art

[0002] In related technologies, after receiving a question uploaded by a user, a reply is usually given based on a pre-set answer database. There is often a risk of confidential information leakage during the question-and-answer process, and the reliability of the reply is low because the question is only replied based on the answer database. Summary of the Invention

[0003] In order to solve the above technical problems, the present invention provides an intelligent question-answering method based on information security protection. When receiving a user question, the method responds to the user question by combining external knowledge, thereby improving the reliability of the question response. Moreover, when responding to the user question, the method first decomposes the user question, and then responds only after ensuring that the decomposed sub-questions do not contain confidential information, thereby greatly improving the security of information in the question-answering process.

[0004] The technical solution adopted in the present invention is as follows:

[0005] An intelligent question-answering method based on information security protection comprises the following steps: upon receiving a user question, determining whether it is necessary to call external knowledge for a reply; if it is determined that external knowledge is needed for a reply, determining whether the user question satisfies a question stratification condition; if it is determined that the user question satisfies the question stratification condition, employing a corresponding stratification strategy to decompose the user question into a plurality of sub-questions; determining whether each of the sub-questions contains confidential information; if it is determined that the sub-question contains confidential information, performing information leakage prevention processing on the sub-question containing confidential information; and replying to the sub-questions that do not contain confidential information and the sub-questions that have undergone information leakage prevention processing.

[0006] In one embodiment of the present invention, determining whether it is necessary to call the external knowledge for a reply specifically includes: determining whether the user question contains enterprise characteristic information; if the user question does not contain the enterprise characteristic information, determining that it is necessary to call the external knowledge for a reply; if the user question contains the enterprise characteristic information, performing feature matching and confidentiality matching on the user question and the internal mechanism, and calculating the reply confidence of the internal knowledge, and determining whether the reply confidence is less than a first preset value; if the match fails and the reply confidence is less than the first preset value, determining that it is necessary to call the external knowledge for a reply.

[0007] In one embodiment of the present invention, determining whether the user question satisfies the question stratification condition specifically includes: determining whether the user question contains iconic keywords, or determining whether the user question contains question structure features, or determining whether the user question contains separators, or determining whether the user question contains multiple information types; if it is determined that the user question contains the iconic keywords, or the user question contains the question structure features, or the user question contains the separators, or the user question contains multiple information types, then it is determined that the user question satisfies the question stratification condition.

[0008] In one embodiment of the present invention, determining whether the sub-problem contains the confidential information specifically includes: obtaining a corresponding first confidential keyword database according to the field to which the sub-problem belongs; performing confidential keyword identification on the document to be identified corresponding to the sub-problem based on the first confidential keyword database to obtain the first confidential keyword contained in the document to be identified; determining whether the first confidential keyword determined based on the first confidential keyword database has a single meaning in a preset language environment; if the first confidential keyword has a single meaning in the preset language environment, determining that the first confidential keyword is the confidential information in the sub-problem; if the first confidential keyword has multiple meanings in the preset language environment, determining whether the first confidential keyword is the confidential information in the sub-problem based on the document to be identified, a first meaning feature corresponding to the first confidential keyword, and a second meaning feature corresponding to the first confidential keyword.

[0009] In one embodiment of the present invention, the first confidential keyword database includes a first meaning feature and a second meaning feature corresponding to the confidential keyword, and determining whether the first confidential keyword is the confidential information in the sub-problem based on the document to be identified, the first meaning feature corresponding to the first confidential keyword, and the second meaning feature corresponding to the first confidential keyword specifically includes: obtaining the first meaning feature and the second meaning feature corresponding to the first confidential keyword from the first confidential keyword database based on the first confidential keyword; obtaining the third meaning feature of the first confidential keyword in the document to be identified; and determining whether the first confidential keyword is the confidential information in the sub-problem based on the first meaning feature, the second meaning feature, and the third meaning feature.

[0010] In one embodiment of the present invention, determining whether the first confidential keyword is the confidential information in the sub-problem based on the first meaning feature, the second meaning feature, and the third meaning feature specifically includes: calculating a first Euclidean distance between the third meaning feature and the first meaning feature, and calculating a second Euclidean distance between the third meaning feature and the second meaning feature; and determining whether the first confidential keyword is the confidential information in the sub-problem based on the first Euclidean distance and the second Euclidean distance.

[0011] An intelligent question-answering system based on information security protection comprises: a first judgment module, the first judgment module being used to judge whether it is necessary to call external knowledge for replying when receiving a user question; a second judgment module, the second judgment module being used to judge whether the user question meets the question stratification conditions when judging that external knowledge needs to be called for replying; a question decomposition module, the question decomposition module being used to adopt corresponding stratification strategies to decompose the user question into multiple sub-questions when judging that the user question meets the question stratification conditions; a third judgment module, the third judgment module being used to judge whether each of the sub-questions contains confidential information; an information processing module, the information processing module being used to perform anti-violation processing on the sub-questions containing confidential information when judging that the sub-questions contain confidential information; and a question reply module, the question reply module being used to reply to sub-questions that do not contain confidential information and sub-questions that have been processed for anti-violation.

[0012] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the intelligent question-answering method based on information security protection is implemented.

[0013] A non-temporary computer-readable storage medium stores a computer program, which, when executed by a processor, implements the above-mentioned intelligent question-answering method based on information security protection.

[0014] Beneficial effects of the present invention:

[0015] When receiving a user question, the present invention responds to the user question by combining external knowledge, thereby improving the reliability of the question response. Moreover, when responding to the user question, the user question is first decomposed, and then the response is made after ensuring that the decomposed sub-questions do not contain confidential information, thereby greatly improving the security of information in the question-and-answer process. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a flowchart of an intelligent question-answering method based on information security protection according to an embodiment of the present invention.

[0017] Figure 24 is a block diagram of an intelligent question-answering system based on information security protection according to an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0019] like Figure 1 As shown, the intelligent question-answering method based on information security protection according to an embodiment of the present invention may include the following steps:

[0020] S1, when receiving a user question, determines whether it is necessary to call external knowledge to respond.

[0021] In one embodiment of the present invention, determining whether external knowledge needs to be called for reply specifically includes the following steps:

[0022] S11, determining whether the user question contains enterprise feature information.

[0023] S12, if the user question does not contain enterprise feature information, it is determined that external knowledge needs to be called for to answer;

[0024] S13, if the user question contains enterprise feature information, perform feature matching and confidentiality level matching on the user question and the internal mechanism, calculate the response confidence of the internal knowledge, and determine whether the response confidence is less than a first preset value;

[0025] Specifically, if the user question does not contain information about related companies, it means that internal knowledge alone cannot support the response to the user question. Therefore, external knowledge needs to be called for response; if the user question contains information about related companies, further feature matching and confidentiality level matching are performed, that is, internal feature conformity judgment is performed, and whether the confidentiality level of the user question matches the preset confidentiality level is judged. Among them, if the confidentiality level of the user question matches the preset confidentiality level, then this part of the user question can only be answered using internal knowledge. At the same time, the response confidence of the internal knowledge can also be calculated, and it can be judged whether the response confidence is less than the first preset value.

[0026] Specifically, first, similar questions (questions of the same type as the current user's question) that are answered using internal knowledge in the historical records can be queried, and the maximum similarity between each similar question and the current user's question can be calculated, as well as the keyword coverage of the internal knowledge relative to the current user's question, and the probability value of the internal knowledge answering the current user's question based on the prediction model can be obtained. Then, the confidence of the internal knowledge response can be calculated based on the maximum similarity between each similar question and the current user's question, the keyword coverage of the internal knowledge relative to the current user's question, and the probability value of the internal knowledge answering the current user's question based on the prediction model.

[0027] The response confidence can be calculated using the following formula:

[0028]

[0029] Among them, ZD is the reply confidence, M is the maximum similarity between similar questions and the current user question, f1 is the keyword coverage of internal knowledge relative to the current user question, g1 is the probability value, a1, a2 and K are all constants greater than 0, α is the adjustment factor, k1 is the first weight value, k2 is the second weight value, and k3 is the third weight value.

[0030] S14: If the matching fails and the response confidence is less than a first preset value, it is determined that external knowledge needs to be called for response.

[0031] Among them, if the matching fails and the response confidence is less than the first preset value, it is determined that external knowledge needs to be called for response; if the matching is successful, or the response confidence is greater than the first preset value, it is determined that only internal knowledge is used to respond to the user question.

[0032] Specifically, when it is determined that only internal knowledge is used to respond to a user's question, the internally stored internal knowledge can be called upon. At this point, it can be determined whether permission to access the internal knowledge is granted. This can be determined by authenticating the user who entered the question. If permission to access the internal knowledge is granted, the internal response model is called upon for interaction, i.e., the knowledge in the cache or answer hotspot record area is called upon to respond. If permission to access the internal knowledge is not granted, a risk control audit is performed.

[0033] S2: If it is determined that external knowledge needs to be called for a response, then it is determined whether the user's question meets the question stratification conditions.

[0034] S3: If the user question is judged to meet the problem stratification conditions, the corresponding stratification strategy is adopted to decompose the user question into multiple sub-questions.

[0035] In one embodiment of the present invention, it is possible to determine whether the user question contains iconic keywords, for example, whether it contains multiple logical relationship words, such as "and", "or", "because", etc., or to determine whether the user question contains question structure features, such as a subject-predicate-object structure, or to determine whether the user question contains separators, or to determine whether the user question contains multiple information types, such as a cause analysis type and a competition situation type. If it is determined that the user question contains iconic keywords, or the user question contains question structure features, or the user question contains separators, or the user question contains multiple information types, then it is determined that the user question meets the question stratification conditions; otherwise, it is determined that the user question does not meet the question stratification conditions. When it is determined that the user question meets the question stratification conditions, the corresponding stratification strategy is adopted to decompose the user question into multiple sub-questions, that is, corresponding stratification strategies are adopted for different situations.

[0036] Therefore, by dividing user questions into multiple sub-questions for answering, it is possible to effectively avoid the situation where the question answering model used recognizes the complete question, thereby improving information security.

[0037] S4, determine whether each sub-question contains confidential information.

[0038] In one embodiment of the present invention, determining whether a sub-question contains confidential information specifically includes the following steps:

[0039] S41, obtaining a corresponding first confidential keyword database according to the field to which the sub-question belongs.

[0040] The confidential keywords in the first confidential keyword database may be confidential keywords pre-set by the relevant enterprise. It is understandable that the confidential keywords are different for different fields.

[0041] S42, performing confidential keyword recognition on the document to be recognized corresponding to the sub-question according to the first confidential keyword database to obtain a first confidential keyword contained in the document to be recognized;

[0042] The document to be identified corresponding to the sub-question can be identified using the confidential keywords stored in the first confidential keyword database, and the identified vocabulary is used as the first confidential keyword of the document to be identified.

[0043] However, the same word may have different meanings in different language contexts. For example, the confidential keyword "core" may be considered confidential information when used in the contexts of "core technology" or "core data," but non-confidential information when used in the context of "the enterprise should be centered around its employees." Therefore, if only the first confidential keyword is treated as confidential information using the above method, the accuracy of confidential keyword recognition will be significantly reduced. Therefore, in the present invention, after identifying the first confidential keyword contained in the document to be identified using the above method, step S43 is continued.

[0044] S43: Determine whether the first confidential keyword determined based on the first confidential keyword database has a single meaning in a preset language environment.

[0045] S44: If the first confidential keyword has a single meaning in the preset language environment, the first confidential keyword is determined to be confidential information in the sub-question.

[0046] The preset language environment may be a language environment pre-set by the relevant enterprise based on actual circumstances. For example, in the scenario of a company developing a new product, the preset language environment may be the language environment in which users learn about the product from the relevant enterprise. If the first confidential keyword has a single meaning in the preset language environment, for example, "This product is manufactured using a certain production process," where "a certain production process" has a single meaning in the preset language environment, it can be directly determined to be confidential information in the sub-question.

[0047] S45: If the first confidential keyword has more than one meaning in the preset language environment, determine whether the first confidential keyword is confidential information in the sub-question based on the document to be identified, the first meaning feature corresponding to the first confidential keyword, and the second meaning feature corresponding to the first confidential keyword.

[0048] The first confidential keyword database includes a first meaning feature and a second meaning feature corresponding to the confidential keyword. Specifically, the first meaning feature may be a confidential information feature, and the second meaning feature may be a non-confidential information feature.

[0049] In one embodiment of the present invention, determining whether the first confidential keyword is confidential information in the sub-problem based on the document to be identified, the first meaning feature corresponding to the first confidential keyword, and the second meaning feature corresponding to the first confidential keyword specifically includes: obtaining the first meaning feature and the second meaning feature corresponding to the first confidential keyword from a first confidential keyword database based on the first confidential keyword; obtaining the third meaning feature of the first confidential keyword in the document to be identified; and determining whether the first confidential keyword is confidential information in the sub-problem based on the first meaning feature, the second meaning feature, and the third meaning feature.

[0050] In one embodiment of the present application, determining whether the first confidential keyword is confidential information in the sub-problem according to the first meaning feature, the second meaning feature and the third meaning feature specifically comprises: calculating a first Euclidean distance between the third meaning feature and the first meaning feature, and calculating a second Euclidean distance between the third meaning feature and the second meaning feature; and determining whether the first confidential keyword is confidential information in the sub-problem according to the first Euclidean distance and the second Euclidean distance.

[0051] Specifically, the first Euclidean distance between the first confidential keyword and the third meaning feature of the to-be-identified document and the second Euclidean distance between the first confidential keyword and the second meaning feature of the to-be-identified document are calculated respectively, and then the first Euclidean distance is compared with the second Euclidean distance to determine whether the first confidential keyword is confidential information in the sub-problem according to the comparison result. If the first Euclidean distance is smaller than the second Euclidean distance, it indicates that the third meaning feature is closer to the first meaning feature, and thus it can be determined that the first confidential keyword is confidential information in the sub-problem. If the first Euclidean distance is greater than the second Euclidean distance, it indicates that the third meaning feature is closer to the second meaning feature, and thus it can be determined that the first confidential keyword is not confidential information in the sub-problem. It can be understood that in actual application, the first Euclidean distance is not equal to the second Euclidean distance.

[0052] Therefore, it can be more accurately determined whether the sub-problem contains confidential information.

[0053] S5, if it is determined that the sub-problem contains confidential information, performing information anti-leakage processing on the sub-problem containing the confidential information.

[0054] Specifically, different means can be used to perform information anti-leakage processing on different types of confidential information. Specifically, numerical noise is added to numerical information, for example, a floating value within a reasonable range is added before and after the specific number; time fuzzification processing is performed on time information. In addition, semantic association processing can also be used, for example, "sales" is replaced by "value A", and "customer segmentation" is replaced by "market segmentation"; and coding processing can also be used, for example, a company name is replaced by a code or a code.

[0055] It should be noted that it can also be determined whether the user problems contain confidential information after step S1, and according to actual needs, the user problems containing confidential information can also be subjected to information anti-leakage processing. Details can be referred to the above embodiments.

[0056] S6, replying to the sub-problems not containing confidential information and the sub-problems subjected to information anti-leakage processing.

[0057] Specifically, pre-trained models can be used to answer sub-questions that do not contain confidential information and those that have undergone information leakage prevention. This can be done through multi-engine search and multi-model question-and-answer methods. During the answering process, these methods can also be used to identify confidential information and perform information leakage prevention. Correction judgments and answer aggregation are performed on the answers output by each model. This answer aggregation specifically includes answer consistency verification, answer accuracy verification, answer security verification, internal answer priority mechanisms, and answer reorganization mechanisms.

[0058] In summary, according to the intelligent question-answering method based on information security protection of an embodiment of the present invention, when receiving a user question, it is determined whether it is necessary to call external knowledge for reply; if it is determined that it is necessary to call external knowledge for reply, it is determined whether the user question meets the question stratification conditions; if it is determined that the user question meets the question stratification conditions, the corresponding stratification strategy is adopted to decompose the user question into multiple sub-questions; it is determined whether each sub-question contains confidential information; if it is determined that the sub-question contains confidential information, the sub-question containing confidential information is processed for information leakage prevention; and the sub-questions that do not contain confidential information and the sub-questions that have been processed for information leakage prevention are replied. Therefore, when receiving a user question, the reliability of the question reply is improved by replying to the user question in combination with external knowledge, and when replying to the user question, the user question is first decomposed, and then the reply is made after ensuring that the decomposed sub-questions do not contain confidential information, thereby greatly improving the security of information in the question-answering process.

[0059] Corresponding to the intelligent question-answering method based on information security protection in the above embodiment, the present invention also proposes an intelligent question-answering system based on information security protection.

[0060] like Figure 2 As shown, the intelligent question-answering system based on information security protection of an embodiment of the present invention may include: a first judgment module 100, a second judgment module 200, a question decomposition module 300, a third judgment module 400, an information processing module 500 and a question reply module 600.

[0061] Among them, the first judgment module 100 is used to determine whether it is necessary to call external knowledge for a reply when receiving a user question; the second judgment module 200 is used to determine whether the user question meets the question stratification conditions when it is determined that external knowledge needs to be called for a reply; the question decomposition module 300 is used to use the corresponding stratification strategy to decompose the user question into multiple sub-questions when it is determined that the user question meets the question stratification conditions; the third judgment module 400 is used to determine whether each sub-question contains confidential information; the information processing module 500 is used to perform anti-violation processing on the sub-question containing confidential information when it is determined that the sub-question contains confidential information; the question reply module 600 is used to reply to sub-questions that do not contain confidential information and sub-questions that have been processed for anti-violation.

[0062] In one embodiment of the present invention, the first judgment module 100 is specifically used to: determine whether it is necessary to call external knowledge for a reply, specifically including: determining whether the user question contains enterprise characteristic information; if the user question does not contain enterprise characteristic information, determining that external knowledge needs to be called for a reply; if the user question contains enterprise characteristic information, performing feature matching and confidentiality matching, and calculating the reply confidence of the internal knowledge, and determining whether the reply confidence is less than a first preset value; if the match is successful and the reply confidence is less than the first preset value, determining that external knowledge needs to be called for a reply.

[0063] In one embodiment of the present invention, the second judgment module 200 is specifically used to: judge whether the user question contains iconic keywords, or judge whether the user question contains question structure features, or judge whether the user question contains separators, or judge whether the user question contains multiple information types; if it is judged that the user question contains iconic keywords, or the user question contains question structure features, or the user question contains separators, or the user question contains multiple information types, then it is judged that the user question meets the question stratification conditions.

[0064] In one embodiment of the present invention, the third judgment module 400 is specifically used to: obtain a corresponding first confidential keyword database according to the field to which the sub-problem belongs; perform confidential keyword recognition on the document to be identified corresponding to the sub-problem based on the first confidential keyword database to obtain the first confidential keyword contained in the document to be identified; determine whether the first confidential keyword determined based on the first confidential keyword database has a single meaning in a preset language environment; if the first confidential keyword has a single meaning in the preset language environment, determine that the first confidential keyword is confidential information in the sub-problem; if the first confidential keyword has a multiple meaning in the preset language environment, determine whether the first confidential keyword is confidential information in the sub-problem based on the document to be identified, the first meaning feature corresponding to the first confidential keyword, and the second meaning feature corresponding to the first confidential keyword.

[0065] In one embodiment of the present invention, the first confidential keyword database includes a first meaning feature and a second meaning feature corresponding to the confidential keyword, and the third judgment module 400 is specifically used to: obtain the first meaning feature and the second meaning feature corresponding to the first confidential keyword from the first confidential keyword database according to the first confidential keyword; obtain the third meaning feature of the first confidential keyword in the document to be identified; and determine whether the first confidential keyword is confidential information in the sub-problem based on the first meaning feature, the second meaning feature and the third meaning feature.

[0066] In one embodiment of the present invention, the third judgment module 400 is specifically used to: calculate the first Euclidean distance between the third meaning feature and the first meaning feature, and calculate the second Euclidean distance between the third meaning feature and the second meaning feature; and determine whether the first confidential keyword is confidential information of the sub-problem based on the first Euclidean distance and the second Euclidean distance.

[0067] It should be noted that for details not disclosed in the intelligent question-answering system based on information security protection in an embodiment of the present invention, please refer to the details disclosed in the intelligent question-answering method based on information security protection in an embodiment of the present invention, and the details will not be described in detail here.

[0068] According to the intelligent question-answering system based on information security protection of an embodiment of the present invention, when receiving a user question, the first judgment module determines whether it is necessary to call external knowledge for replying, and when the second judgment module determines that it is necessary to call external knowledge for replying, it determines whether the user question meets the question stratification conditions, and when the question decomposition module determines that the user question meets the question stratification conditions, it adopts the corresponding stratification strategy to decompose the user question into multiple sub-questions, and when the third judgment module determines whether each sub-question contains confidential information, the information processing module performs anti-violation processing on the sub-question containing confidential information when it determines that the sub-question contains confidential information, and the question reply module replies to the sub-question that does not contain confidential information and the sub-question that has been subjected to anti-violation processing. Therefore, when receiving a user question, the reliability of the question reply is improved by replying to the user question in combination with external knowledge, and when replying to the user question, the user question is first decomposed, and then the reply is made after ensuring that the decomposed sub-questions do not contain confidential information, thereby greatly improving the security of information in the question-answering process.

[0069] Corresponding to the above embodiment, the present invention further provides a computer device.

[0070] The computer device of an embodiment of the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the intelligent question-answering method based on information security protection of the above embodiment is implemented.

[0071] According to the computer device of an embodiment of the present invention, when receiving a user question, the computer device responds to the user question by combining external knowledge, thereby improving the reliability of the question response. Moreover, when responding to the user question, the computer device first decomposes the user question and then responds after ensuring that the decomposed sub-questions do not contain confidential information, thereby greatly improving the security of information during the question-and-answer process.

[0072] Corresponding to the above embodiment, the present invention further proposes a non-transitory computer-readable storage medium.

[0073] The non-transitory computer-readable storage medium of an embodiment of the present invention stores a computer program, which, when executed by a processor, implements the above-mentioned intelligent question-answering method based on information security protection.

[0074] According to the non-transitory computer-readable storage medium of an embodiment of the present invention, when receiving a user question, the user question is replied by combining external knowledge, thereby improving the reliability of the question reply. Moreover, when replying to the user question, the user question is first broken down, and then the reply is made after ensuring that the broken down sub-questions do not contain confidential information, thereby greatly improving the security of information in the question-and-answer process.

[0075] In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, a feature specified as "first" or "second" may explicitly or implicitly include one or more of such features. "Multiple" means two or more, unless otherwise specifically defined.

[0076] In the present invention, unless otherwise expressly specified or limited, the terms "mounted," "connected," "connect," "fixed," etc. should be understood broadly. For example, they may refer to fixed connection, detachable connection, or integration; mechanical connection or electrical connection; direct connection or indirect connection through an intermediate medium; internal communication between two components or interaction between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0077] In the present invention, unless otherwise expressly specified or limited, when a first feature is "above" or "below" a second feature, it may mean that the first and second features are in direct contact, or that the first and second features are in indirect contact through an intermediary. Furthermore, when a first feature is "above," "above," or "above" a second feature, it may mean that the first feature is directly above or diagonally above the second feature, or simply means that the first feature is at a higher level than the second feature. When a first feature is "below," "below," or "below" a second feature, it may mean that the first feature is directly below or diagonally below the second feature, or simply means that the first feature is at a lower level than the second feature.

[0078] In the description of this specification, the reference terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0079] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing module, or each unit may exist physically separately, or two or more units may be integrated into a single module. The aforementioned integrated modules may be implemented in the form of hardware or in the form of software functional modules. If the integrated modules are implemented in the form of software functional modules and sold or used as independent products, they may also be stored in a computer-readable storage medium.

[0080] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. An intelligent question-answering method based on information security protection, characterized in that: The following steps are involved: When receiving a user question, determine whether external knowledge needs to be called to respond; If it is determined that external knowledge needs to be called for a response, then whether the user question meets the question stratification conditions is determined; If it is determined that the user question meets the problem stratification conditions, the corresponding stratification strategy is adopted to decompose the user question into multiple sub-questions; Determine whether each of the sub-questions contains confidential information. Determine whether the sub-questions contain confidential information specifically including: Acquire a corresponding first confidential keyword database according to the field to which the sub-question belongs; Performing confidential keyword recognition on the document to be recognized corresponding to the sub-question according to the first confidential keyword database to obtain a first confidential keyword contained in the document to be recognized; determining whether the first confidential keyword determined based on the first confidential keyword database has a single meaning in a preset language environment; If the first confidential keyword has a single meaning in the preset language environment, determining that the first confidential keyword is the confidential information in the sub-question; If the first confidential keyword has multiple meanings in the preset language environment, determining whether the first confidential keyword is the confidential information in the sub-problem based on the document to be identified, the first meaning feature corresponding to the first confidential keyword, and the second meaning feature corresponding to the first confidential keyword; wherein the first confidential keyword database includes the first meaning feature and the second meaning feature corresponding to the confidential keyword, and determining whether the first confidential keyword is the confidential information in the sub-problem based on the document to be identified, the first meaning feature corresponding to the first confidential keyword, and the second meaning feature corresponding to the first confidential keyword specifically includes: obtaining the first meaning feature and the second meaning feature corresponding to the first confidential keyword from the first confidential keyword database based on the first confidential keyword; obtaining the third meaning feature of the first confidential keyword in the document to be identified; and determining whether the first confidential keyword is the confidential information in the sub-problem based on the first meaning feature, the second meaning feature, and the third meaning feature; If it is determined that the sub-question contains confidential information, then performing information leakage prevention processing on the sub-question containing the confidential information; Respond to sub-questions that do not contain confidential information and sub-questions that have been processed to prevent information leakage.

2. The intelligent question-answering method based on information security protection according to claim 1 is characterized in that: Determining whether it is necessary to call the external knowledge for reply specifically includes: Determining whether the user question contains enterprise feature information; If the user question does not contain the enterprise feature information, it is determined that the external knowledge needs to be called for reply; If the user question contains the enterprise characteristic information, then perform feature matching and confidentiality level matching on the user question and the internal mechanism, calculate the response confidence of the internal knowledge, and determine whether the response confidence is less than a first preset value; If the matching fails and the response confidence is less than the first preset value, it is determined that the external knowledge needs to be called for response.

3. The intelligent question-answering method based on information security protection according to claim 1, characterized in that: Determining whether the user question meets the question stratification conditions specifically includes: Determining whether the user question contains a landmark keyword, or whether the user question contains a question structure feature, or whether the user question contains a separator, or whether the user question contains multiple information types; If it is determined that the user question contains the iconic keywords, or the user question contains the question structure features, or the user question contains the separator, or the user question contains multiple information types, then it is determined that the user question meets the question stratification conditions.

4. The intelligent question-answering method based on information security protection according to claim 1, characterized in that: Determining whether the first confidential keyword is the confidential information in the sub-question according to the first meaning feature, the second meaning feature, and the third meaning feature specifically includes: calculating a first Euclidean distance between the third meaning feature and the first meaning feature, and calculating a second Euclidean distance between the third meaning feature and the second meaning feature; It is determined whether the first confidential keyword is the confidential information of the sub-problem according to the first Euclidean distance and the second Euclidean distance.

5. An intelligent question-answering system based on information security protection, characterized in that: include: The first judgment module is used to determine whether it is necessary to call external knowledge to respond when receiving a user question; The second judgment module is used to judge whether the user question meets the question stratification conditions when it is judged that external knowledge needs to be called for a reply; A question decomposition module is used to decompose the user question into multiple sub-questions by adopting a corresponding stratification strategy when it is determined that the user question meets the question stratification conditions; The third judgment module is used to judge whether each of the sub-problems contains confidential information; wherein the third judgment module is specifically used to: obtain a corresponding first confidential keyword database according to the field to which the sub-problem belongs; perform confidential keyword recognition on the document to be identified corresponding to the sub-problem according to the first confidential keyword database to obtain the first confidential keyword contained in the document to be identified; judge whether the first confidential keyword determined based on the first confidential keyword database has a single meaning in a preset language environment; if the first confidential keyword has a single meaning in the preset language environment, determine that the first confidential keyword is the confidential information in the sub-problem; if the first confidential keyword has a multiple meaning in the preset language environment, Then, based on the document to be identified, the first meaning feature corresponding to the first confidential keyword, and the second meaning feature corresponding to the first confidential keyword, it is determined whether the first confidential keyword is the confidential information in the sub-problem; wherein, the first confidential keyword database includes the first meaning feature and the second meaning feature corresponding to the confidential keyword, and the third judgment module is specifically used to: obtain the first meaning feature and the second meaning feature corresponding to the first confidential keyword from the first confidential keyword database according to the first confidential keyword; obtain the third meaning feature of the first confidential keyword in the document to be identified; and determine whether the first confidential keyword is the confidential information in the sub-problem based on the first meaning feature, the second meaning feature, and the third meaning feature; An information processing module, the information processing module is used to perform anti-violation processing on the sub-question containing confidential information when it is determined that the sub-question contains confidential information; The question response module is used to respond to sub-questions that do not contain confidential information and sub-questions that have been processed for anti-violation.

6. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the intelligent question-answering method based on information security protection according to any one of claims 1 to 4 is implemented.

7. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the intelligent question-answering method based on information security protection according to any one of claims 1 to 4 is implemented.

Images