Large model question answering method and device based on cache mechanism, storage medium and equipment

Abstract

The invention discloses a large model question answering method and device based on a cache mechanism, a storage medium and equipment, and the method comprises the steps: receiving question description information inputted by a target question initiator, and querying whether first answer information matched with the question description information exists in a corresponding trust domain cache of the target question initiator; if the first answer information matched with the question description information exists in the trusted domain cache, performing answer feedback to the target question initiator based on the first answer information; and if the first answer information matched with the question description information does not exist in the trusted domain cache, calling a large model to generate second answer information corresponding to the question description information, and performing answer feedback to the target question initiator based on the second answer information, and associating the question description information with the second answer information and then storing the question description information and the second answer information in the trusted domain cache. And the response efficiency can be improved.

Description

Technical Field

[0001] The present application relates to the field of large model technology, and in particular to a large model question-answering method, apparatus, storage medium and device based on a cache mechanism. Background Art

[0002] In today's information society, artificial intelligence (AI) technology has demonstrated irreplaceable value in many fields, especially in intelligent question-answering systems, thanks to its excellent data processing and learning capabilities. AI models can quickly provide accurate and valuable information by deeply understanding and analyzing user questions, greatly improving the convenience of information acquisition. However, existing large AI models generally rely heavily on real-time Internet access when answering user questions, which brings many challenges.

[0003] The workflow of intelligent question-answering systems usually involves receiving user questions, parsing questions, accessing the Internet to obtain relevant information, generating answers, and returning them to users. This process ensures the timeliness and accuracy of answers, but it has exposed several deficiencies in actual applications. The instability of Internet access may cause system response delays, thereby reducing user experience; at the same time, frequent Internet access also increases operating costs, especially in areas with high data traffic costs. More importantly, for the same or similar questions frequently asked by users, the system needs to re-access the Internet to generate answers each time, which is not only inefficient, but also causes a huge waste of resources. This repeated Internet access not only further exacerbates the problem of inefficiency, but also leads to further waste of resources. Summary of the invention

[0004] In view of this, the embodiments of the present application provide a large model question and answer method, apparatus, storage medium and device based on a caching mechanism. By pre-storing previously processed questions and answers in the cache, the answers can be read directly from the cache without waiting for large model processing or Internet access. This can improve the system's response speed, improve processing efficiency, reduce latency, and thus improve user experience.

[0005] According to one aspect of the present application, a large model question answering method based on a cache mechanism is provided, the method comprising: Receiving question description information input by a target question initiator, and querying whether there is first answer information matching the question description information in a trust domain cache corresponding to the target question initiator; If the first answer information matching the question description information exists in the trust domain cache, then providing an answer feedback to the target question initiator based on the first answer information; If the first answer information matching the question description information does not exist in the trust domain cache, the large model is called to generate second answer information corresponding to the question description information, and the answer feedback is given to the target question initiator based on the second answer information. The question description information and the second answer information are associated and stored in the trust domain cache.

[0006] In an optional implementation manner, if the first answer information matching the question description information exists in the trust domain cache, then providing an answer feedback to the target question initiator based on the first answer information includes: If there is first answer information matching the question description information in the trust domain cache, the validity of the first answer information is verified, and an answer is fed back to the target question initiator based on the valid first answer information. When the first answer information is invalid, the large model is called to generate third answer information corresponding to the question description information, and an answer is fed back to the target question initiator based on the third answer information. The question description information and the third answer information are associated and stored in the trust domain cache.

[0007] In an optional implementation manner, verifying the validity of the first answer information includes: Verify the timeliness of the first answer information, and if the first answer information contains multiple answer contents, verify a preferred answer content among the multiple answer contents as the valid first answer information. In an optional implementation, before calling the large model to generate the second answer information corresponding to the question description information, it also includes: Parsing the problem description information through a parsing domain to determine a category chain to which the problem description information belongs, wherein the category chain includes multiple layers of classification categories corresponding to the problem description information; querying whether there is a first trust domain access address corresponding to the bottom classification category in the category chain in the resolution domain cache of the resolution domain; If the first trust domain access address corresponding to the bottom classification category in the category chain exists in the resolution domain cache, then the corresponding first target trust domain is accessed based on the first trust domain access address, and when the fourth answer information corresponding to the question description information exists in the cache of the first target trust domain, the answer is fed back to the target question initiator based on the fourth answer information, otherwise the step of calling the large model to generate the second answer information corresponding to the question description information is executed; If the first trust domain access address corresponding to the underlying classification category in the category chain does not exist in the resolution domain cache, then the pointing domain is accessed layer by layer based on the multi-layer classification categories of the category chain to obtain the second trust domain access address corresponding to the underlying classification category in the category chain, and the corresponding second target trust domain is accessed based on the second trust domain access address, and when the fifth answer information corresponding to the question description information exists in the cache of the second target trust domain, the answer is fed back to the target question initiator based on the fifth answer information, otherwise the step of calling the large model to generate the second answer information corresponding to the question description information is executed.

[0008] In an optional implementation manner, when the fourth answer information corresponding to the question description information exists in the cache of the first target trust domain, providing answer feedback to the target question initiator based on the fourth answer information includes: When the fourth answer information corresponding to the question description information exists in the cache of the first target trust domain, verify the validity of the fourth answer information, provide an answer feedback to the target question initiator based on the valid fourth answer information, and execute the step of calling the large model to generate the second answer information corresponding to the question description information when the fourth answer information is invalid; When the fifth answer information corresponding to the question description information exists in the cache of the second target trust domain, providing an answer feedback to the target question initiator based on the fifth answer information includes: When the fifth answer information corresponding to the question description information exists in the cache of the second target trust domain, the validity of the fifth answer information is verified, and the answer feedback is provided to the target question initiator based on the valid fifth answer information. When the fifth answer information is invalid, the step of calling the large model to generate the second answer information corresponding to the question description information is executed.

[0009] In an optional implementation, accessing the pointing domain layer by layer based on the multi-layer classification categories of the category chain to obtain the second trust domain access address corresponding to the bottom classification category in the category chain includes: Starting from the first layer of classification categories in the multi-layer classification categories of the category chain, the classification domains corresponding to the classification categories of each layer are accessed layer by layer in the pointing domain to obtain the classification domain access address corresponding to the next layer of classification categories, until the classification domain access address corresponding to the bottom layer of classification categories in the category chain is obtained as the second trust domain access address, wherein the pointing domain includes multiple layers of classification domains, the top-level classification domain is the central domain, the bottom-level classification domain is the trust domain, and each layer of classification domain stores the name and access address of the corresponding next layer of classification domain.

[0010] In an optional implementation, the target question initiator is any one of a plurality of question initiators, and the trust domains to which different question initiators belong are not exactly the same. The trust domain cache of each trust domain includes a local cache and an external domain cache. The local cache of the trust domain is used to store question description information and corresponding answer information belonging to the classification category of the trust domain proposed by the question initiator corresponding to the trust domain, and the external domain cache of the trust domain is used to store question description information and corresponding answer information that do not belong to the classification category corresponding to the trust domain proposed by the question initiator corresponding to the trust domain; the answer information includes the answer content or the address for obtaining the answer content, and at least one of the validity time, timestamp, and number of hits.

[0011] In an optional implementation, the method further includes: Counting the number of question description information corresponding to the same answer content in each trust domain cache, if the number of question description information meets the preset high-heat question condition, the question description information meeting the preset high-heat question condition and the corresponding answer information are associated and stored in the answer index cache; and / or, Counting the data size of each answer content in each trust domain cache, if the data size meets the preset large-length answer condition, then associating the question description information and answer information corresponding to the answer content meeting the preset large-length answer condition and storing them in the answer index cache; Correspondingly, if the first answer information matching the question description information does not exist in the trust domain cache, calling the large model to generate the second answer information corresponding to the question description information includes: If the first answer information matching the question description information does not exist in the trust domain cache, querying whether there is sixth answer information matching the question description information in the answer index cache; If there is sixth answer information matching the question description information in the answer index cache, an answer feedback is provided to the target question initiator based on the sixth answer information; otherwise, the step of calling the large model to generate second answer information corresponding to the question description information is executed.

[0012] In an optional implementation, the method further includes: Receiving trust domain management information, wherein the trust domain management information is at least one of trust domain creation information, trust domain modification information, and trust domain deletion information input by an administrator who has passed management authority authentication, the management authority of the administrator is valid after being applied for by the administrator and authorized by the central domain, and the management authority includes at least one of trust domain creation authority, at least one trust domain modification authority, and at least one trust domain deletion authority; Based on the trust domain management information, the corresponding trust domain is managed, wherein the trust domain creation information includes the classification name and description information of the newly created trust domain, the trust domain modification information includes the modified trust domain classification name and / or the modified trust domain description information, and the trust domain deletion information includes the name of the trust domain to be deleted.

[0013] According to another aspect of the present application, a large model question-answering device based on a cache mechanism is provided, the device comprising: A receiving module is used to: receive question description information input by a target question initiator, and query whether there is first answer information matching the question description information in a trust domain cache corresponding to the target question initiator; A feedback module is used to: if there is first answer information matching the question description information in the trust domain cache, then provide answer feedback to the target question initiator based on the first answer information; and If the first answer information matching the question description information does not exist in the trust domain cache, the large model is called to generate second answer information corresponding to the question description information, and the answer feedback is given to the target question initiator based on the second answer information. The question description information and the second answer information are associated and stored in the trust domain cache.

[0014] In an optional implementation, the feedback module is further used to: If there is first answer information matching the question description information in the trust domain cache, the validity of the first answer information is verified, and an answer is fed back to the target question initiator based on the valid first answer information. When the first answer information is invalid, the large model is called to generate third answer information corresponding to the question description information, and an answer is fed back to the target question initiator based on the third answer information. The question description information and the third answer information are associated and stored in the trust domain cache.

[0015] In an optional implementation, the feedback module is further used to: Verify the timeliness of the first answer information, and if the first answer information contains multiple answer contents, verify a preferred answer content among the multiple answer contents as the valid first answer information.

[0016] In an optional implementation, the feedback module is further used to: Parsing the problem description information through a parsing domain to determine a category chain to which the problem description information belongs, wherein the category chain includes multiple layers of classification categories corresponding to the problem description information; querying whether there is a first trust domain access address corresponding to the bottom classification category in the category chain in the resolution domain cache of the resolution domain; If the first trust domain access address corresponding to the bottom classification category in the category chain exists in the resolution domain cache, then the corresponding first target trust domain is accessed based on the first trust domain access address, and when the fourth answer information corresponding to the question description information exists in the cache of the first target trust domain, the answer is fed back to the target question initiator based on the fourth answer information, otherwise the step of calling the large model to generate the second answer information corresponding to the question description information is executed; If the first trust domain access address corresponding to the underlying classification category in the category chain does not exist in the resolution domain cache, then the pointing domain is accessed layer by layer based on the multi-layer classification categories of the category chain to obtain the second trust domain access address corresponding to the underlying classification category in the category chain, and the corresponding second target trust domain is accessed based on the second trust domain access address, and when the fifth answer information corresponding to the question description information exists in the cache of the second target trust domain, the answer is fed back to the target question initiator based on the fifth answer information, otherwise the step of calling the large model to generate the second answer information corresponding to the question description information is executed.

[0017] In an optional implementation, the feedback module is further used to: When the fourth answer information corresponding to the question description information exists in the cache of the first target trust domain, verify the validity of the fourth answer information, provide an answer feedback to the target question initiator based on the valid fourth answer information, and execute the step of calling the large model to generate the second answer information corresponding to the question description information when the fourth answer information is invalid; When the fifth answer information corresponding to the question description information exists in the cache of the second target trust domain, providing an answer feedback to the target question initiator based on the fifth answer information includes: When the fifth answer information corresponding to the question description information exists in the cache of the second target trust domain, the validity of the fifth answer information is verified, and the answer feedback is provided to the target question initiator based on the valid fifth answer information. When the fifth answer information is invalid, the step of calling the large model to generate the second answer information corresponding to the question description information is executed.

[0018] In an optional implementation, the feedback module is further used to: Starting from the first layer of classification categories in the multi-layer classification categories of the category chain, the classification domains corresponding to the classification categories of each layer are accessed layer by layer in the pointing domain to obtain the classification domain access address corresponding to the next layer of classification categories, until the classification domain access address corresponding to the bottom layer of classification categories in the category chain is obtained as the second trust domain access address, wherein the pointing domain includes multiple layers of classification domains, the top-level classification domain is the central domain, the bottom-level classification domain is the trust domain, and each layer of classification domain stores the name and access address of the corresponding next layer of classification domain.

[0019] In an optional implementation, the target question initiator is any one of a plurality of question initiators, and the trust domains to which different question initiators belong are not exactly the same. The trust domain cache of each trust domain includes a local cache and an external domain cache. The local cache of the trust domain is used to store question description information and corresponding answer information belonging to the classification category of the trust domain proposed by the question initiator corresponding to the trust domain, and the external domain cache of the trust domain is used to store question description information and corresponding answer information that do not belong to the classification category corresponding to the trust domain proposed by the question initiator corresponding to the trust domain; the answer information includes the answer content or the address for obtaining the answer content, and at least one of the validity time, timestamp, and number of hits.

[0020] In an optional implementation, the system further includes: a statistics module, which is used to: Counting the number of question description information corresponding to the same answer content in each trust domain cache, if the number of question description information meets the preset high-heat question condition, the question description information meeting the preset high-heat question condition and the corresponding answer information are associated and stored in the answer index cache; and / or, Counting the data size of each answer content in each trust domain cache, if the data size meets the preset large-length answer condition, then associating the question description information and answer information corresponding to the answer content meeting the preset large-length answer condition and storing them in the answer index cache; Accordingly, the feedback module is also used to: If the first answer information matching the question description information does not exist in the trust domain cache, querying whether there is sixth answer information matching the question description information in the answer index cache; If there is sixth answer information matching the question description information in the answer index cache, an answer feedback is provided to the target question initiator based on the sixth answer information; otherwise, the step of calling the large model to generate second answer information corresponding to the question description information is executed.

[0021] In an optional implementation, the system further includes: a management module, configured to: Receiving trust domain management information, wherein the trust domain management information is at least one of trust domain creation information, trust domain modification information, and trust domain deletion information input by an administrator who has passed management authority authentication, the management authority of the administrator is valid after being applied for by the administrator and authorized by the central domain, and the management authority includes at least one of trust domain creation authority, at least one trust domain modification authority, and at least one trust domain deletion authority; Based on the trust domain management information, the corresponding trust domain is managed, wherein the trust domain creation information includes the classification name and description information of the newly created trust domain, the trust domain modification information includes the modified trust domain classification name and / or the modified trust domain description information, and the trust domain deletion information includes the name of the trust domain to be deleted.

[0022] According to another aspect of the present application, a storage medium is provided, on which a computer program is stored, and when the program is executed by a processor, the large model question and answer method based on the cache mechanism is implemented.

[0023] According to another aspect of the present application, a computer device is provided, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor implements the above-mentioned large model question-answering method based on a cache mechanism when executing the program.

[0024] By means of the above technical solution, the embodiment of the present application provides a large model question-answering method, device, storage medium and equipment based on a cache mechanism, which first queries the answer to the question in the trust domain cache of the initiator of the target question asked by the user, and directly feeds back the answer if the cache hits, and if it does not hit, calls the large model to generate the answer and stores it in the cache, so as to give an answer to the question description information raised by the user. In the embodiment of the present application, since the cache stores the previously processed questions and answers, the answer can be read directly from the cache without waiting for the large model to process or access the Internet, which can improve the response speed of the system, improve the processing efficiency, reduce the delay, and thus improve the user experience. In addition, reducing the frequent access to the Internet means reducing the use of data traffic, especially in areas with high data traffic costs, this cost-saving effect is more obvious. At the same time, since the use of the cache reduces unnecessary computing resource consumption, the operating cost can also be reduced to a certain extent. In addition, the instability of Internet access may cause system response delays or failures. By introducing a cache mechanism, the dependence on the Internet can be reduced to a certain extent, thereby enhancing the stability and reliability of the system. Even if there is a problem with Internet access, the system can still provide answers from the cache to ensure the continuity of the service.

[0025] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings: Figure 1 A schematic diagram of a process flow of a large model question-answering method based on a cache mechanism provided in an embodiment of the present application is shown; Figure 2 A schematic diagram of a process flow of another large model question-answering method based on a cache mechanism provided in an embodiment of the present application is shown; Figure 3 A schematic diagram of a process flow of another large model question-answering method based on a cache mechanism provided in an embodiment of the present application is shown; Figure 4 A schematic diagram of a process for iteratively obtaining a trust domain access address provided in an embodiment of the present application is shown; Figure 5 A schematic diagram of the structure of a large model question-answering device based on a cache mechanism provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0027] The present application will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that the embodiments and features in the embodiments of the present application can be combined with each other without conflict.

[0028] As the user base continues to grow, repeated queries for common problems become more frequent, but existing systems fail to effectively use historical interaction data to optimize the response process. This repeated Internet access not only further exacerbates the problem of inefficiency, but also leads to further waste of resources. Therefore, in order to address these challenges, the present application provides a large model system that can combine Internet access and local data caching, which can significantly improve response speed, reduce operating costs, and significantly improve user experience.

[0029] In this embodiment, a large model question answering method based on a cache mechanism is provided. Figure 1 As shown, the method includes: Step 101: Receive question description information input by a target question initiator, and query whether there is first answer information matching the question description information in a trust domain cache corresponding to the target question initiator.

[0030] Step 102: If the first answer information matching the question description information exists in the trust domain cache, an answer is fed back to the target question initiator based on the first answer information.

[0031] Step 103: If the first answer information matching the question description information does not exist in the trust domain cache, call the large model to generate second answer information corresponding to the question description information, provide answer feedback to the target question initiator based on the second answer information, and associate the question description information and the second answer information and store them in the trust domain cache.

[0032] In the embodiment of the present application, when a user asks a question through the target question initiator, the question description information input by the user is first queried in the corresponding trust domain cache of the target question initiator. Among them, the target question initiator is any one of the multiple question initiators, and different users can use the question initiator with their own use authority to ask questions. Each question initiator corresponds to its own trust domain, and each trust domain corresponds to its own trust domain cache. The "trust domain cache" here can be understood as a local or distributed cache system specially set for the target question initiator, which is used to store previously processed questions and their answers. If there is answer information (called the first answer information) matching the current question description information in the trust domain cache, it means that this question has been processed before and the answer has been cached. The answer information can be a specific answer content or an address for obtaining the answer content. If the first answer information is a specific answer content, feedback can be directly based on the cached answer without calling the large model again or accessing the Internet. If the first answer information is an address for obtaining the answer content, the specific answer content can be obtained through the address and feedback can be given. If there is no matching answer information in the cache, it means that this is a new question (or the previous answer has expired and has been deleted from the cache). At this point, the big model is called to generate new answer information and the answer is fed back to the user. At the same time, the question description information and the generated answer information are associated to form a question-answer pair and stored in the cache so that the information can be reused in the future.

[0033] By applying the technical solution of this embodiment, the answer to the question is first queried in the trust domain cache of the initiator of the target question asked by the user. If the cache hits, the answer is directly fed back. If it does not hit, the large model is called to generate the answer and stored in the cache, so as to give an answer to the question description information raised by the user. In the embodiment of the present application, since the cache stores the previously processed questions and answers, the answer can be read directly from the cache without waiting for the large model processing or Internet access, which can improve the response speed of the system, improve the processing efficiency, reduce the delay, and thus improve the user experience. In addition, reducing the frequent access to the Internet means reducing the use of data traffic, especially in areas with high data traffic costs, this cost saving effect is more obvious. At the same time, since the use of the cache reduces unnecessary computing resource consumption, the operating cost can also be reduced to a certain extent. In addition, the instability of Internet access may cause system response delays or failures. By introducing a cache mechanism, the dependence on the Internet can be reduced to a certain extent, thereby enhancing the stability and reliability of the system. Even if there is a problem with Internet access, the system can still provide answers from the cache to ensure the continuity of the service.

[0034] Further, as a refinement and extension of the specific implementation of the above embodiment, in order to fully illustrate the specific implementation process of this embodiment, another large model question answering method based on a cache mechanism is provided, such as Figure 2 As shown, the method includes: Step 201: Receive question description information input by a target question initiator, and query whether there is first answer information matching the question description information in a trust domain cache corresponding to the target question initiator.

[0035] In this embodiment, if Figure 3 As shown, the question description information input by the user through the target question initiator is received. Then, the first answer information matching the question description information is queried in the trust domain cache corresponding to the target question initiator. In this way, by first querying the trust domain cache, questions that already have answers can be quickly responded to, thereby improving the system's response speed and user experience.

[0036] Step 202: If there is first answer information matching the question description information in the trust domain cache, verify the validity of the first answer information, provide answer feedback to the target question initiator based on the valid first answer information, and call the large model to generate third answer information corresponding to the question description information when the first answer information is invalid, provide answer feedback to the target question initiator based on the third answer information, and associate the question description information and the third answer information and store them in the trust domain cache.

[0037] In this embodiment, if there is first answer information matching the question description information in the trust domain cache, the validity of the first answer information can be further verified. If the answer is valid, it is directly fed back to the target question initiator. If the answer is invalid (for example, because the information is outdated or the cache is expired), the large model is called to generate new answer information (i.e., the third answer information) and the cache is updated. This step ensures that the user receives the latest and valid answer. At the same time, by updating the cache, a faster and more accurate response is provided for future queries.

[0038] In an embodiment of the present application, optionally, in step 202, verifying the validity of the first answer information includes: verifying the timeliness of the first answer information, and verifying a preferred answer content among the multiple answer contents as valid first answer information when the first answer information contains multiple answer contents.

[0039] In the above embodiment, the verification of the validity of the first answer information mainly includes two aspects of verification: on the one hand, the timeliness of the first answer information is verified: the timeliness verification is to ensure that the answer information is still accurate and relevant. In the rapidly changing Internet environment, a lot of information may become outdated quickly. Therefore, it can be determined whether it is still valid by checking the timestamp or other timeliness mark of the answer information. If the answer is outdated, the answer will be considered invalid, and a large model needs to be called to generate a new answer. On the other hand, a preferred answer content is verified in multiple answer contents: for the case of containing multiple answer contents, it is necessary to further determine which answer is the most accurate or most relevant. This can be achieved by a variety of methods, such as sorting or scoring based on factors such as the number of likes, the number of comments, the authoritative source mark, and the historical accuracy of the answer. The answer with the highest score is selected as the valid first answer information for feedback. In addition, if all answers do not meet certain quality standards, it can also be considered that no valid answer exists, and a large model is called to generate a new answer. The embodiment of the present application can ensure that the answer provided to the user is the latest and most relevant through the verification of timeliness and preferred answer content. The user receives the screened and verified answer, which increases the user's trust and satisfaction with the answer provided by the system.

[0040] Step 203: If the first answer information matching the problem description information does not exist in the trust domain cache, the problem description information is parsed through the resolution domain to determine the category chain to which the problem description information belongs, wherein the category chain includes multiple layers of classification categories corresponding to the problem description information; and the resolution domain cache of the resolution domain is queried to determine whether there is a first trust domain access address corresponding to the bottom-level classification category in the category chain.

[0041] In this embodiment, if Figure 3As shown, if there is no matching answer in the trust domain cache, the resolution domain can be further accessed to resolve the problem description information. Specifically, the category chain resolution can be implemented through keyword matching, semantic analysis, etc. to determine the category chain to which it belongs. This category chain represents the multi-layer classification structure corresponding to the problem description information. Then, the classification domain access address (i.e., the first trust domain access address) corresponding to the bottom-level classification category in the category chain is queried in the cache of the resolution domain. By determining the category chain, the embodiment of the present application can more accurately locate the field or subject to which the problem belongs, thereby more effectively searching for answers.

[0042] Step 204: If the first trust domain access address corresponding to the underlying classification category in the category chain exists in the resolution domain cache, then the corresponding first target trust domain is accessed based on the first trust domain access address, and when the fourth answer information corresponding to the question description information exists in the cache of the first target trust domain, the answer is fed back to the target question initiator based on the fourth answer information, otherwise execute step 206.

[0043] In this embodiment, if the address is found, the corresponding first target trust domain can be accessed directly through the first trust domain access address, and the answer (i.e., the fourth answer information corresponding to the question description information) can be found in its cache. By parsing the domain cache, the system can quickly locate the trust domain that may contain the answer, thereby reducing unnecessary search and calculation overhead. In addition, if the answer information corresponding to the question description information is not found in the first target trust domain cache, the question description information is answered through the large model.

[0044] In an embodiment of the present application, optionally, in step 204, when fourth answer information corresponding to the question description information exists in the cache of the first target trust domain, answer feedback is provided to the target question initiator based on the fourth answer information, including: when fourth answer information corresponding to the question description information exists in the cache of the first target trust domain, verifying the validity of the fourth answer information, providing answer feedback to the target question initiator based on the valid fourth answer information, and executing the step of calling the large model to generate second answer information corresponding to the question description information when the fourth answer information is invalid.

[0045] In the above embodiment, when the fourth answer information corresponding to the question description information is found in the cache of the first target trust domain, in order to ensure the validity of the fed-back answer, the validity of the fourth answer information can also be verified first, and the answer information is not immediately fed back to the target question initiator. The method for verifying the validity of the fourth answer information is consistent with the method for verifying the validity of the first answer information described above, and will not be repeated here. If the fourth answer information is confirmed to be valid after verification, then the answer feedback will be given to the target question initiator based on the fourth answer information that is verified to be valid. In this way, the target question initiator can obtain a quick and accurate answer. However, if the fourth answer information is determined to be invalid after verification, the step of calling the large model to generate the second answer information corresponding to the question description information will be executed. The embodiment of the present application verifies the validity of the answer in the cache, and even if the answer is found in the cache of the first target trust domain, it will ensure that the latest and most accurate answer is provided to the user, which further improves the accuracy and reliability of the large model question-answering method based on the cache mechanism.

[0046] Step 205: If the first trust domain access address corresponding to the underlying classification category in the category chain does not exist in the resolution domain cache, then access the pointing domain layer by layer based on the multi-layer classification categories of the category chain to obtain the second trust domain access address corresponding to the underlying classification category in the category chain, access the corresponding second target trust domain based on the second trust domain access address, and when the fifth answer information corresponding to the question description information exists in the cache of the second target trust domain, feedback the answer to the target question initiator based on the fifth answer information, otherwise execute step 206.

[0047] In this embodiment, if the first trust domain access address corresponding to the underlying classification category is not found in the resolution domain cache, the pointing domain will be accessed layer by layer until the second trust domain access address corresponding to the underlying classification category is found. Then the second target trust domain is accessed and the answer is searched in its cache. This step ensures that even if there is no direct trust domain access address, the possible source of the answer can be found by accessing the pointing domain layer by layer, which increases the flexibility and applicability of the system.

[0048] In an embodiment of the present application, optionally, in step 205, the pointing domain is accessed layer by layer based on the multi-layer classification categories of the category chain to obtain the second trust domain access address corresponding to the bottom-layer classification category in the category chain, including: starting from the first-layer classification category in the multi-layer classification categories of the category chain, accessing the classification domains corresponding to each layer of classification categories in the pointing domain layer by layer to obtain the classification domain access address corresponding to the next layer of classification categories, until the classification domain access address corresponding to the bottom-layer classification category in the category chain is obtained as the second trust domain access address, wherein the pointing domain includes multiple layers of classification domains, the top-level classification domain is the central domain, the bottom-level classification domain is the trust domain, and each layer of classification domain stores the name and access address of the corresponding next-layer classification domain.

[0049] In this embodiment, the pointing domain is a structured information base, which includes multiple layers of classification domains, each layer corresponding to a specific classification category. The top classification domain is usually regarded as the central domain, which represents the broadest classification range, while the bottom classification domain is a specific trust domain, which stores information or answers related to a specific category. In an optional implementation, Figure 3 As shown, the classification domain can include 4 layers, namely the central domain, the subdomain, the grandchild domain, and the trust domain. In the process of accessing the pointing domain layer by layer through the multi-layer classification categories of the category chain, first identify the top-level or first-level classification category to which the problem description information or the problem itself belongs. This category is the starting point of the category chain and defines the approximate scope or field of the problem. Starting from the top-level classification category, search for the classification domain corresponding to the category in the pointing domain. After accessing the classification domain of the current layer, find the name and access address of the next-level classification domain stored in the domain. This information is pre-defined and is used to guide the system on how to further go into a more specific classification level. Continue to repeat the above steps and access downward layer by layer until the classification domain corresponding to the bottom-level classification category in the category chain is reached. This domain is the second trust domain to be found in the end, which stores the information or answer most directly related to the problem. Finally, the access address of the classification domain corresponding to the bottom-level classification category is obtained, that is, the second trust domain access address. Through this address, the system can access the trust domain where specific answers or information are stored. The embodiment of the present application can more accurately locate the trust domain related to the problem through layer-by-layer classified access, more effectively utilize cache and storage resources, reduce unnecessary search and access operations, and improve the accuracy and relevance of the retrieved information.

[0050] In one example, if Figure 4 As shown, the detailed process of iteratively obtaining the final trust domain address (i.e., the second trust domain access address) is as follows: Accessing the central domain: The resolution domain sends domain A (classification name) to the central domain. The central domain queries the access address B of domain A and returns B to the resolution domain.

[0051] Access B: The resolution domain sends subdomain A1 (category name) to the central domain. The central domain queries the access address B1 of subdomain A1 and returns B1 to the resolution domain.

[0052] Access B1: The resolution domain sends subdomain A1-1 (classification name) to the central domain. The central domain queries the access address B1-1 of subdomain A1-1 and returns B1-1 to the resolution domain.

[0053] Access B1-1: The resolution domain sends the final trust domain (classification name) to the central domain. The central domain queries the access address of the final trust domain and returns the address to the resolution domain.

[0054] In an embodiment of the present application, optionally, in step 205, when fifth answer information corresponding to the question description information exists in the cache of the second target trust domain, answer feedback is provided to the target question initiator based on the fifth answer information, including: when fifth answer information corresponding to the question description information exists in the cache of the second target trust domain, verifying the validity of the fifth answer information, providing answer feedback to the target question initiator based on the valid fifth answer information, and executing the step of calling the large model to generate second answer information corresponding to the question description information when the fifth answer information is invalid.

[0055] In the above embodiment, when the fifth answer information corresponding to the question description information is found in the cache of the second target trust domain, in order to ensure the validity of the fed-back answer, the validity of the fifth answer information may be verified first, without immediately feeding the answer information back to the target question initiator. The method for verifying the validity of the fifth answer information is consistent with the method for verifying the validity of the first answer information described above, and will not be repeated here.

[0056] Step 206: call the large model to generate second answer information corresponding to the question description information, provide answer feedback to the target question initiator based on the second answer information, and associate the question description information and the second answer information and store them in the trust domain cache.

[0057] In this embodiment, if no matching answer is found in all the above steps, the large model is called to generate the second answer information corresponding to the question description information, and the information is fed back to the target question initiator. At the same time, the question description information and the answer information are associated to form a question-answer pair, which is stored in the corresponding trust domain cache. By calling the large model to generate answers, the embodiment of the present application can handle those questions that have not been encountered before and those whose answers have expired. At the same time, updating the cache ensures that future queries can get answers faster.

[0058] In an embodiment of the present application, optionally, the target question initiator is any one of a plurality of question initiators, and the trust domains to which different question initiators belong are not exactly the same. The trust domain cache of each trust domain includes a local cache and an external domain cache. The local cache of the trust domain is used to store question description information and corresponding answer information belonging to the classification category of the trust domain proposed by the question initiator corresponding to the trust domain, and the external domain cache of the trust domain is used to store question description information and corresponding answer information that do not belong to the classification category corresponding to the trust domain proposed by the question initiator corresponding to the trust domain; the answer information includes the answer content or the address for obtaining the answer content, and at least one of the validity time, timestamp, and number of hits.

[0059] In the above embodiment, the target question initiator is any one of the multiple question initiators, that is, the system can process query requests from different question initiators. Different question initiators may belong to different trust domains, that is, the trust domains corresponding to different question initiators may be the same or different. The trust domain can be understood here as a logical grouping, which is used to distinguish different question initiators and the scope or category of the questions they raise. In addition, each trust domain has its own trust domain cache, which includes two parts: local cache and external domain cache. The local cache is used to store the question description information and corresponding answer information raised by the question initiator corresponding to the trust domain and belonging to the trust domain classification category, which can also be understood as the local cache mainly storing questions and answers directly related to the trust domain. The external domain cache is used to store the question description information and corresponding answer information raised by the question initiator corresponding to the trust domain but not belonging to the trust domain classification category. This reflects the ability of the system to handle cross-trust domain problems, that is, it can store and retrieve answers to questions in other trust domains. Further, the answer information not only includes the answer content itself, but also may include the address for obtaining the answer content. In addition, the answer information also includes metadata such as valid time, timestamp, and number of hits. These metadata help manage cache content and ensure the timeliness of answers. The validity time indicates when the answer is invalid, helping to identify and remove outdated answers. The timestamp records the time when the answer is stored or updated, which helps track the freshness of the answer so that old answers can be considered invalid. Ensure that the answers provided to users are accurate, timely and relevant. The hit count records the number of times the answer is retrieved or used, providing the system with clues about the popularity of the answer.

[0060] In an embodiment of the present application, optionally, it also includes: counting the number of question description information corresponding to the same answer content in each trust domain cache, and if the number of question description information meets the preset high-heat question condition, the question description information that meets the preset high-heat question condition and the corresponding answer information are associated and stored in the answer index cache; and / or, counting the data size of each answer content in each trust domain cache, and if the data size meets the preset large-length answer condition, the question description information and answer information corresponding to the answer content that meets the preset large-length answer condition are associated and stored in the answer index cache; Correspondingly, if the first answer information matching the question description information does not exist in the trust domain cache, the large model is called to generate the second answer information corresponding to the question description information, including: if the first answer information matching the question description information does not exist in the trust domain cache, then querying whether there is the sixth answer information matching the question description information in the answer index cache; if the sixth answer information matching the question description information exists in the answer index cache, then based on the sixth answer information, an answer is fed back to the target question initiator, otherwise the step of calling the large model to generate the second answer information corresponding to the question description information is executed.

[0061] In the above embodiment, in order to further improve the efficiency and response speed of the system, especially for high-frequency questions and long-length answer content, the embodiment of the present application can also introduce an answer index cache mechanism. Specifically, the number of question description information corresponding to the same answer content in each trust domain cache can be counted. If the number of a certain question description information meets the preset high-heat question condition (for example, the number of question description information exceeds a certain threshold), this question description information and its corresponding answer information can be associated and stored in the answer index cache. Thus, when the same or similar question is raised again, the answer can be quickly retrieved directly from the answer index cache without calling the large model again or traversing the trust domain cache. In addition, the data size of each answer content in each trust domain cache can also be counted. If the data size of a certain answer content meets the preset large-length answer condition (for example, the answer length exceeds a certain number of characters), the question description information and answer information corresponding to this answer content can also be associated and stored in the answer index cache. Further, based on the original answer retrieval process, the following optimization can be performed: when a question description information is received and it is determined that there is no first answer information matching the question description information in the trust domain cache to which it belongs, first query whether there is answer information matching the question description information, that is, the sixth answer information, in the answer index cache. If there is a matching sixth answer information in the answer index cache, then the answer is directly fed back to the target question initiator based on the sixth answer information. If there is no matching answer information in the answer index cache, then continue to perform the step of parsing the question description information through the parsing domain in 203 above. In addition, for the answer information stored in the trust domain cache, the acquisition address of the answer content contained in the answer information can specifically point to the above-mentioned answer index cache, that is, the specific answer content is obtained in the answer index cache through the acquisition address of the answer content. Of course, the acquisition address of the answer content contained in the answer information can also point to other locations, which is not limited here. By introducing the answer index cache mechanism, the embodiment of the present application enables the system to retrieve frequently occurring questions and large-scale answer content more quickly, thereby reducing unnecessary calculations and resource consumption, so that users can obtain answers more quickly and improve the response speed of the system.

[0062] It should be noted that in order to further reduce cache pressure, for answer contents stored in the answer index cache, if these answer contents are also stored in some trust domain caches, the same answer contents in the trust domain cache can be replaced with answer address information pointing to the answer index cache. In this way, when the required answer address information is retrieved in the trust domain cache, the answer content in the answer index cache can be obtained through the answer address information, thereby achieving unified management of high-frequency and large-volume answer contents and reducing the pressure on the trust domain cache.

[0063] In an embodiment of the present application, optionally, it also includes: receiving trust domain management information, wherein the trust domain management information is at least one of trust domain creation information, trust domain modification information, and trust domain deletion information input by an administrator who has passed management authority authentication, the management authority of the administrator is valid after being applied for by the administrator and the central domain is authorized, and the management authority includes at least one of trust domain creation authority, modification authority of at least one trust domain, and deletion authority of at least one trust domain; based on the trust domain management information, managing the corresponding trust domain, wherein the trust domain creation information includes a classification name and description information of the newly created trust domain, the trust domain modification information includes a modified trust domain classification name and / or a modified trust domain description information, and the trust domain deletion information includes the name of the trust domain to be deleted.

[0064] In the above embodiment, in order to enhance the flexibility and manageability of the system, the embodiment of the present application also provides a trust domain management mechanism, which allows administrators who have passed the management authority authentication to perform operations such as creating, modifying and deleting trust domains. The system can receive trust domain management information, which is input by administrators with management authority. The management authority of the administrator is obtained by applying to the central domain and being authorized, which ensures the legality and security of the administrator's operation. The trust domain management information may include at least one of trust domain creation information, trust domain modification information and trust domain deletion information. These information correspond to the operations of creating, modifying and deleting trust domains respectively. Among them, the trust domain creation authority allows the administrator to create a new trust domain, including setting the classification name and description information of the newly created trust domain; the trust domain modification authority allows the administrator to modify at least one trust domain with management authority, including modifying the classification name, description information or other related attributes of the trust domain; the trust domain deletion authority allows the administrator to delete at least one trust domain with management authority, and the name of the trust domain to be deleted needs to be provided during the operation. Further, based on the received trust domain management information, the management operation is performed on the corresponding trust domain. Among them, for the trust domain creation information, a new trust domain is created, and its classification name and description information are set according to the provided information. For the trust domain modification information, the specified trust domain is modified accordingly according to the provided information. For the trust domain deletion information, the specified trust domain is deleted to ensure that the system's trust domain list remains up-to-date and accurate. The embodiment of the present application introduces a management authority authentication mechanism to ensure that only authorized administrators can perform management operations on the trust domain, thereby enhancing the security of the system.

[0065] In a specific application scenario, the detailed process of trust domain management is as follows: (1) Login and authority verification: The management role logs in to the management interface of the trust domain or trust entity through the security authentication mechanism. The system verifies the identity and authority of the management role to ensure that it has the management authority for the specific trust domain or trust entity.

[0066] (2) Classification rule setting: The management role enters the "Classification Management" module to view the current classification rule list.

[0067] Create a new category: The management role clicks "New Category", enters the category name and description, and the interface adds the new category to the rule list.

[0068] Edit category: The management role selects the category to be edited and modifies its name or description. The interface updates the corresponding rule configuration.

[0069] Delete category: The management role selects the category to be deleted. After confirming the operation, the interface removes the category from the rule list and adjusts the category affiliation of the relevant question and answer pairs.

[0070] (3) Apply for credit: The management role enters the "Credit Application" module, fills in the necessary application information, and the interface sends the application information to the central domain for review. The management role can view the application status in real time through the management interface.

[0071] The detailed processing flow of the central domain is as follows: (1) Login and authority verification: The management role logs in to the management interface of the central domain through a security authentication mechanism.

[0072] The system verifies the identity and permissions of the management role to ensure that it has management permissions for the central domain.

[0073] (2) Authentication and authorization: Review credit applications: The management role enters the "Credit Review" module to view the list of trust domains or trust entity applications to be reviewed.

[0074] The management role reviews the detailed application information and uploaded supporting documents for evaluation. The management role can choose to "approve" or "reject" the application. If approved, the interface generates a unique authorization code for the trust domain or trust body and adds it to the global trust list; if rejected, the interface notifies the applicant.

[0075] Revoke trust: The management role can select the trust domain or trust entity that has been granted trust. After confirming the operation, the interface removes the entity from the global trust list and stops its access to the interface.

[0076] By applying the technical solution of this embodiment, a multi-level DNS-like architecture answer cache is constructed, and by introducing the concept of trust domain, efficient management and utilization of question-answer pairs are achieved. This design not only solves the problem of resource waste caused by repeated queries in the prior art, but also improves the response speed and user experience of the system. In addition, the present invention expands information security and privacy protection, and can ensure the security and controllability of data in each trust domain or trust body through a credit mechanism and hierarchical management. Compared with the traditional AI question-and-answer system, this solution significantly reduces the reliance on large models, reduces computing power consumption and operating costs by introducing a local caching mechanism. At the same time, since most common problems can be solved locally, the user's waiting time is greatly shortened, and the interactive experience is greatly improved. More importantly, the multi-level architecture and trust domain management mechanism adopted by the present invention not only enhance the security and reliability of the system, but also provide possibilities for personalized services and community co-construction.

[0077] Further, as Figure 1 The specific implementation of the method, the embodiment of the present application provides a large model question-answering device based on a cache mechanism, such as Figure 5 As shown, the device comprises: A receiving module is used to: receive question description information input by a target question initiator, and query whether there is first answer information matching the question description information in a trust domain cache corresponding to the target question initiator; A feedback module is used to: if there is first answer information matching the question description information in the trust domain cache, then provide answer feedback to the target question initiator based on the first answer information; and If the first answer information matching the question description information does not exist in the trust domain cache, the large model is called to generate second answer information corresponding to the question description information, and the answer feedback is given to the target question initiator based on the second answer information. The question description information and the second answer information are associated and stored in the trust domain cache.

[0078] In an optional implementation, the feedback module is further used to: If there is first answer information matching the question description information in the trust domain cache, the validity of the first answer information is verified, and an answer is fed back to the target question initiator based on the valid first answer information. When the first answer information is invalid, the large model is called to generate third answer information corresponding to the question description information, and an answer is fed back to the target question initiator based on the third answer information. The question description information and the third answer information are associated and stored in the trust domain cache.

[0079] In an optional implementation, the feedback module is further used to: Verify the timeliness of the first answer information, and if the first answer information contains multiple answer contents, verify a preferred answer content among the multiple answer contents as the valid first answer information.

[0080] In an optional implementation, the feedback module is further used to: Parsing the problem description information through a parsing domain to determine a category chain to which the problem description information belongs, wherein the category chain includes multiple layers of classification categories corresponding to the problem description information; querying whether there is a first trust domain access address corresponding to the bottom classification category in the category chain in the resolution domain cache of the resolution domain; If the first trust domain access address corresponding to the bottom classification category in the category chain exists in the resolution domain cache, then the corresponding first target trust domain is accessed based on the first trust domain access address, and when the fourth answer information corresponding to the question description information exists in the cache of the first target trust domain, the answer is fed back to the target question initiator based on the fourth answer information, otherwise the step of calling the large model to generate the second answer information corresponding to the question description information is executed; If the first trust domain access address corresponding to the underlying classification category in the category chain does not exist in the resolution domain cache, then the pointing domain is accessed layer by layer based on the multi-layer classification categories of the category chain to obtain the second trust domain access address corresponding to the underlying classification category in the category chain, and the corresponding second target trust domain is accessed based on the second trust domain access address, and when the fifth answer information corresponding to the question description information exists in the cache of the second target trust domain, the answer is fed back to the target question initiator based on the fifth answer information, otherwise the step of calling the large model to generate the second answer information corresponding to the question description information is executed.

[0081] In an optional implementation, the feedback module is further used to: When the fourth answer information corresponding to the question description information exists in the cache of the first target trust domain, verify the validity of the fourth answer information, provide an answer feedback to the target question initiator based on the valid fourth answer information, and execute the step of calling the large model to generate the second answer information corresponding to the question description information when the fourth answer information is invalid; When the fifth answer information corresponding to the question description information exists in the cache of the second target trust domain, providing an answer feedback to the target question initiator based on the fifth answer information includes: When the fifth answer information corresponding to the question description information exists in the cache of the second target trust domain, the validity of the fifth answer information is verified, and the answer feedback is provided to the target question initiator based on the valid fifth answer information. When the fifth answer information is invalid, the step of calling the large model to generate the second answer information corresponding to the question description information is executed.

[0082] In an optional implementation, the feedback module is further used to: Starting from the first layer of classification categories in the multi-layer classification categories of the category chain, the classification domains corresponding to the classification categories of each layer are accessed layer by layer in the pointing domain to obtain the classification domain access address corresponding to the next layer of classification categories, until the classification domain access address corresponding to the bottom layer of classification categories in the category chain is obtained as the second trust domain access address, wherein the pointing domain includes multiple layers of classification domains, the top-level classification domain is the central domain, the bottom-level classification domain is the trust domain, and each layer of classification domain stores the name and access address of the corresponding next layer of classification domain.

[0083] In an optional implementation, the target question initiator is any one of a plurality of question initiators, and the trust domains to which different question initiators belong are not exactly the same. The trust domain cache of each trust domain includes a local cache and an external domain cache. The local cache of the trust domain is used to store question description information and corresponding answer information belonging to the classification category of the trust domain proposed by the question initiator corresponding to the trust domain, and the external domain cache of the trust domain is used to store question description information and corresponding answer information that do not belong to the classification category corresponding to the trust domain proposed by the question initiator corresponding to the trust domain; the answer information includes the answer content or the address for obtaining the answer content, and at least one of the validity time, timestamp, and number of hits.

[0084] In an optional implementation, the system further includes: a statistics module, which is used to: Counting the number of question description information corresponding to the same answer content in each trust domain cache, if the number of question description information meets the preset high-heat question condition, the question description information meeting the preset high-heat question condition and the corresponding answer information are associated and stored in the answer index cache; and / or, Counting the data size of each answer content in each trust domain cache, if the data size meets the preset large-length answer condition, then associating the question description information and answer information corresponding to the answer content meeting the preset large-length answer condition and storing them in the answer index cache; Accordingly, the feedback module is also used to: If the first answer information matching the question description information does not exist in the trust domain cache, querying whether there is sixth answer information matching the question description information in the answer index cache; If there is sixth answer information matching the question description information in the answer index cache, an answer feedback is provided to the target question initiator based on the sixth answer information; otherwise, the step of calling the large model to generate second answer information corresponding to the question description information is executed.

[0085] In an optional implementation, the system further includes: a management module, configured to: Receiving trust domain management information, wherein the trust domain management information is at least one of trust domain creation information, trust domain modification information, and trust domain deletion information input by an administrator who has passed management authority authentication, the management authority of the administrator is valid after being applied for by the administrator and authorized by the central domain, and the management authority includes at least one of trust domain creation authority, at least one trust domain modification authority, and at least one trust domain deletion authority; Based on the trust domain management information, the corresponding trust domain is managed, wherein the trust domain creation information includes the classification name and description information of the newly created trust domain, the trust domain modification information includes the modified trust domain classification name and / or the modified trust domain description information, and the trust domain deletion information includes the name of the trust domain to be deleted.

[0086] It should be noted that for other corresponding descriptions of the functional units involved in the large model question-answering device based on the cache mechanism provided in the embodiment of the present application, please refer to Figures 1 to 4 The corresponding description in the method will not be repeated here.

[0087] The embodiment of the present application also provides a computer device, which can be a personal computer, a server, a network device, etc. The computer device includes a bus, a processor, a memory and a communication interface, and can also include an input and output interface and a display device. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store location information. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the steps in each method embodiment are implemented.

[0088] Those skilled in the art will appreciate that the structure of the above-mentioned computer device is only a partial structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components, or combine certain components, or have a different arrangement of components.

[0089] In one embodiment, a computer-readable storage medium is provided. The computer-readable storage medium may be non-volatile or volatile, and stores a computer program thereon. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0090] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0091] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0092] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0093] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0094] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A large model question answering method based on a cache mechanism, characterized in that: The method comprises: Receiving question description information input by a target question initiator, and querying whether there is first answer information matching the question description information in a trust domain cache corresponding to the target question initiator; If the first answer information matching the question description information exists in the trust domain cache, then providing an answer feedback to the target question initiator based on the first answer information; If the first answer information matching the question description information does not exist in the trust domain cache, the large model is called to generate second answer information corresponding to the question description information, and the answer feedback is given to the target question initiator based on the second answer information. The question description information and the second answer information are associated and stored in the trust domain cache.

2. The large model question answering method based on cache mechanism according to claim 1 is characterized in that: If the first answer information matching the question description information exists in the trust domain cache, then providing an answer feedback to the target question initiator based on the first answer information, including: If there is first answer information matching the question description information in the trust domain cache, the validity of the first answer information is verified, and an answer is fed back to the target question initiator based on the valid first answer information. When the first answer information is invalid, the large model is called to generate third answer information corresponding to the question description information, and an answer is fed back to the target question initiator based on the third answer information. The question description information and the third answer information are associated and stored in the trust domain cache.

3. The large model question answering method based on cache mechanism according to claim 2 is characterized in that: Verifying the validity of the first answer information includes: Verify the timeliness of the first answer information, and if the first answer information contains multiple answer contents, verify a preferred answer content among the multiple answer contents as the valid first answer information.

4. The large model question answering method based on cache mechanism according to claim 1 is characterized in that: Before calling the large model to generate the second answer information corresponding to the question description information, the method further includes: Parsing the problem description information through a parsing domain to determine a category chain to which the problem description information belongs, wherein the category chain includes multiple layers of classification categories corresponding to the problem description information; querying whether there is a first trust domain access address corresponding to the bottom classification category in the category chain in the resolution domain cache of the resolution domain; If the first trust domain access address corresponding to the bottom classification category in the category chain exists in the resolution domain cache, then the corresponding first target trust domain is accessed based on the first trust domain access address, and when the fourth answer information corresponding to the question description information exists in the cache of the first target trust domain, the answer is fed back to the target question initiator based on the fourth answer information, otherwise the step of calling the large model to generate the second answer information corresponding to the question description information is executed; If the first trust domain access address corresponding to the underlying classification category in the category chain does not exist in the resolution domain cache, then the pointing domain is accessed layer by layer based on the multi-layer classification categories of the category chain to obtain the second trust domain access address corresponding to the underlying classification category in the category chain, and the corresponding second target trust domain is accessed based on the second trust domain access address, and when the fifth answer information corresponding to the question description information exists in the cache of the second target trust domain, the answer is fed back to the target question initiator based on the fifth answer information, otherwise the step of calling the large model to generate the second answer information corresponding to the question description information is executed.

5. The large model question answering method based on cache mechanism according to claim 4 is characterized in that: When the fourth answer information corresponding to the question description information exists in the cache of the first target trust domain, providing an answer feedback to the target question initiator based on the fourth answer information includes: When the fourth answer information corresponding to the question description information exists in the cache of the first target trust domain, verify the validity of the fourth answer information, provide an answer feedback to the target question initiator based on the valid fourth answer information, and execute the step of calling the large model to generate the second answer information corresponding to the question description information when the fourth answer information is invalid; When the fifth answer information corresponding to the question description information exists in the cache of the second target trust domain, providing an answer feedback to the target question initiator based on the fifth answer information includes: When the fifth answer information corresponding to the question description information exists in the cache of the second target trust domain, the validity of the fifth answer information is verified, and the answer feedback is provided to the target question initiator based on the valid fifth answer information. When the fifth answer information is invalid, the step of calling the large model to generate the second answer information corresponding to the question description information is executed.

6. The large model question answering method based on cache mechanism according to claim 4 is characterized in that: Accessing the pointing domain layer by layer based on the multi-layer classification categories of the category chain to obtain the second trust domain access address corresponding to the bottom classification category in the category chain includes: Starting from the first layer of classification categories in the multi-layer classification categories of the category chain, the classification domains corresponding to the classification categories of each layer are accessed layer by layer in the pointing domain to obtain the classification domain access address corresponding to the next layer of classification categories, until the classification domain access address corresponding to the bottom layer of classification categories in the category chain is obtained as the second trust domain access address, wherein the pointing domain includes multiple layers of classification domains, the top-level classification domain is the central domain, the bottom-level classification domain is the trust domain, and each layer of classification domain stores the name and access address of the corresponding next layer of classification domain.

7. The large model question answering method based on cache mechanism according to claim 6 is characterized in that: The target question initiator is any one of multiple question initiators, and the trust domains to which different question initiators belong are not exactly the same. The trust domain cache of each trust domain includes a local cache and an external domain cache. The local cache of the trust domain is used to store question description information and corresponding answer information belonging to the classification category of the trust domain proposed by the question initiator corresponding to the trust domain, and the external domain cache of the trust domain is used to store question description information and corresponding answer information that do not belong to the classification category corresponding to the trust domain proposed by the question initiator corresponding to the trust domain; the answer information includes the answer content or the address for obtaining the answer content, as well as at least one of the validity period, timestamp, and number of hits.

8. The large model question answering method based on cache mechanism according to claim 7 is characterized in that: The method further comprises: Counting the number of question description information corresponding to the same answer content in each trust domain cache, if the number of question description information meets the preset high-heat question condition, the question description information meeting the preset high-heat question condition and the corresponding answer information are associated and stored in the answer index cache; and / or, Counting the data size of each answer content in each trust domain cache, if the data size meets the preset large-length answer condition, then associating the question description information and answer information corresponding to the answer content meeting the preset large-length answer condition and storing them in the answer index cache; Correspondingly, if the first answer information matching the question description information does not exist in the trust domain cache, calling the large model to generate the second answer information corresponding to the question description information includes: If the first answer information matching the question description information does not exist in the trust domain cache, querying whether there is sixth answer information matching the question description information in the answer index cache; If there is sixth answer information matching the question description information in the answer index cache, an answer feedback is provided to the target question initiator based on the sixth answer information; otherwise, the step of calling the large model to generate second answer information corresponding to the question description information is executed.

9. The method according to claim 7, characterized in that: The method further comprises: Receiving trust domain management information, wherein the trust domain management information is at least one of trust domain creation information, trust domain modification information, and trust domain deletion information input by an administrator who has passed management authority authentication, the management authority of the administrator is valid after being applied for by the administrator and authorized by the central domain, and the management authority includes at least one of trust domain creation authority, at least one trust domain modification authority, and at least one trust domain deletion authority; Based on the trust domain management information, the corresponding trust domain is managed, wherein the trust domain creation information includes the classification name and description information of the newly created trust domain, the trust domain modification information includes the modified trust domain classification name and / or the modified trust domain description information, and the trust domain deletion information includes the name of the trust domain to be deleted.

10. A large model question-answering device based on a cache mechanism, characterized in that: The device comprises: A receiving module is used to: receive question description information input by a target question initiator, and query whether there is first answer information matching the question description information in a trust domain cache corresponding to the target question initiator; A feedback module is used to: if there is first answer information matching the question description information in the trust domain cache, then provide answer feedback to the target question initiator based on the first answer information; and If the first answer information matching the question description information does not exist in the trust domain cache, the large model is called to generate second answer information corresponding to the question description information, and the answer feedback is given to the target question initiator based on the second answer information. The question description information and the second answer information are associated and stored in the trust domain cache.

11. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 9 is implemented.

12. A computer device comprising a storage medium, a processor, and a computer program stored in the storage medium and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 9 is implemented.

Images