Knowledge base updating method and apparatus

Abstract

The application provides a knowledge base updating method and device, the method comprising: obtaining a plurality of question information processed by a task processing model without relying on a knowledge base and a satisfactory result corresponding to the question information, the satisfactory result being used to indicate a user's satisfaction degree for feedback information output by the task processing model in processing the question information; determining at least one target document in the knowledge base matched with the question information; determining a comprehensive satisfaction degree of the user for feedback information corresponding to at least one question information associated with the target document based on the satisfactory result corresponding to each question information associated with the target document; and if the comprehensive satisfaction degree corresponding to the target document exceeds a set threshold, performing compression processing on the target document.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a method and apparatus for updating a knowledge base. Background Technology

[0002] With the continuous development of artificial intelligence technology, various task processing models such as large language models can more accurately determine the feedback information corresponding to the user's input questions by leveraging knowledge bases.

[0003] However, if the knowledge base contains a large amount of knowledge, the process of using the task processing model to process the query information will inevitably consume a lot of memory resources. It is easy to encounter a situation where the electronic device deploying the task processing model cannot effectively determine the feedback information due to insufficient memory resources. Summary of the Invention

[0004] On the one hand, this application provides a knowledge base updating method, including:

[0005] The task processing model obtains multiple query information processed without relying on a knowledge base and the corresponding satisfaction results of the query information. The satisfaction results are used to indicate the user's satisfaction with the feedback information output by the task processing model in processing the query information.

[0006] Identify at least one target document in the knowledge base that matches the query information;

[0007] Based on the satisfaction results corresponding to each question associated with the target document, determine the overall satisfaction level of the user with the feedback information corresponding to at least one question associated with the target document;

[0008] If the overall satisfaction level of the target document exceeds a set threshold, the target document is compressed.

[0009] In one possible implementation, the knowledge base includes a document library and a vector library, wherein the document library includes at least one document segment from at least one document, and the vector library includes vectors of each document segment in the document library;

[0010] The step of determining at least one target document in the knowledge base that matches the question information includes:

[0011] Determine at least one target vector in the vector library that matches the information features of the question information;

[0012] Determine at least one target document to which the document fragment corresponding to the at least one target vector belongs;

[0013] The compression process of the target document includes at least one of the following:

[0014] The vectors in the vector library that are related to the target document are compressed.

[0015] The document fragments related to the target document in the document library are compressed.

[0016] In another possible implementation, the step of compressing the target document if the overall satisfaction level of the target document exceeds a set threshold includes:

[0017] If the overall satisfaction level of the target document exceeds a set threshold, at least one target knowledge object in the knowledge base that is associated with the target document and meets the importance requirement is identified, and all knowledge objects in the knowledge base that are associated with the target document but do not belong to the target knowledge object are removed.

[0018] Alternatively, if the overall satisfaction level of the target document exceeds a set threshold, remove all knowledge objects associated with the target document from the knowledge base;

[0019] The knowledge object represents at least one of a document fragment and a vector of the document fragment.

[0020] In another possible implementation, if the overall satisfaction level corresponding to the target document exceeds a set threshold, determining at least one target knowledge object in the knowledge base that is associated with the target document and meets the importance requirement, and removing all knowledge objects in the knowledge base that are associated with the target document but do not belong to the target knowledge object, includes:

[0021] If the overall satisfaction level of the target document exceeds a first set threshold and is lower than a second set threshold, determine at least one target knowledge object in the knowledge base that is associated with the target document and meets the importance requirement, and remove all knowledge objects in the knowledge base that are associated with the target document but do not belong to the target knowledge object.

[0022] Alternatively, if the overall satisfaction level corresponding to the target document exceeds a set threshold, removing all knowledge objects associated with the target document from the knowledge base includes:

[0023] If the overall satisfaction level of the target document is not lower than the second set threshold, remove all knowledge objects associated with the target document from the knowledge base.

[0024] In another possible implementation, determining at least one target knowledge object in the knowledge base that is associated with the target document and meets the importance requirement includes:

[0025] Cluster the vectors of each document fragment associated with the target document in the knowledge base to determine the target vector of at least one target document fragment as the cluster center;

[0026] The removal of knowledge objects from the knowledge base that are associated with the target document but do not belong to the target knowledge object includes:

[0027] Remove all knowledge objects in the knowledge base that are associated with the target document but do not belong to the target document fragment or target vector.

[0028] Another possible implementation includes:

[0029] Determine the total number of query messages associated with the target document;

[0030] If the overall satisfaction level of the target document exceeds a set threshold, the target document is compressed, including:

[0031] If the total number is greater than the set number and the overall satisfaction level of the target document exceeds the set threshold, the target document is compressed.

[0032] Another possible implementation includes:

[0033] Obtain the question information entered by the user;

[0034] If it is confirmed that the question information is selected as the question information for testing, the task processing model is used to output the feedback information of the question information without relying on the knowledge base;

[0035] Obtain satisfactory results from the user's feedback information regarding the question;

[0036] The question information is stored as the question information processed by the task processing model without relying on the knowledge base, and the corresponding satisfactory result is stored.

[0037] If the satisfactory result corresponding to the question information is unsatisfactory, the task processing model is instructed to use the knowledge base to re-output the feedback information of the question information;

[0038] The process of obtaining multiple query messages processed by the task processing model without relying on a knowledge base, and the corresponding satisfactory results for the query messages, includes:

[0039] When the set knowledge base update time is reached or the data volume of the knowledge base exceeds the set data volume, the task processing model obtains multiple query information and the corresponding satisfactory results of the query information without relying on the knowledge base.

[0040] Another possible implementation includes:

[0041] Obtain at least one pending document fragment to be stored;

[0042] The task processing model is used to generate test question information for the pending document fragment, and the task processing model is used to generate test feedback information for the test question information without the pending document fragment, to obtain test results. The test results are used to indicate the generation status of the test question information and the test feedback information for the pending document fragment.

[0043] If the test results indicate that the task processing model can generate test question information for the pending document fragment, and the task processing model cannot generate test feedback information for the test question information without the pending document fragment, then at least one of the pending document fragment and the corresponding test feedback information is added to the knowledge base.

[0044] Another possible implementation includes:

[0045] If the test results indicate that the task processing model can generate test question information for the pending document fragment, and the task processing model can generate test feedback information for the test question information without the pending document fragment, or if the task processing model cannot generate test question information for the pending document fragment, then the knowledge base will not include the pending document fragment.

[0046] Furthermore, this application also provides a knowledge base updating device, comprising:

[0047] The information acquisition unit is used to acquire multiple question information processed by the task processing model without relying on a knowledge base, as well as the satisfaction results corresponding to the question information. The satisfaction results are used to indicate the user's satisfaction with the feedback information output by the task processing model in processing the question information.

[0048] A document determination unit is used to determine at least one target document in the knowledge base that matches the query information;

[0049] The comprehensive determination unit is used to determine the overall satisfaction level of the user with the feedback information corresponding to at least one question associated with the target document based on the satisfaction results corresponding to each question associated with the target document.

[0050] A compression processing unit is used to compress the target document if the overall satisfaction level of the target document exceeds a set threshold. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0052] Figure 1 This paper illustrates a flowchart of a knowledge base update method provided in an embodiment of this application.

[0053] Figure 2 This illustration shows another flowchart of the knowledge base update method provided in an embodiment of this application;

[0054] Figure 3 This illustration shows a schematic diagram of the implementation principle framework of the knowledge base update method provided in an embodiment of this application;

[0055] Figure 4 This illustration shows another flowchart of the knowledge base update method provided in an embodiment of this application;

[0056] Figure 5 This illustration shows a schematic diagram of an implementation process for controlling the addition of document fragments to the knowledge base in an embodiment of this application;

[0057] Figure 6 This illustration shows a schematic diagram of the implementation principle framework for controlling the addition of document fragments to the knowledge base in an embodiment of this application;

[0058] Figure 7 This illustration shows a schematic diagram of the composition structure of a knowledge base updating device provided in an embodiment of this application;

[0059] Figure 8 A schematic diagram of the composition structure of an electronic device provided in an embodiment of this application is shown. Detailed Implementation

[0060] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0061] like Figure 1 This illustration shows a flowchart of a knowledge base update method provided in an embodiment of this application. The method of this embodiment can be applied to any electronic device, especially to electronic devices with relatively limited memory or other hardware resources, such as mobile phones, laptops, or desktop computers.

[0062] The method in this embodiment may include:

[0063] S101, obtain multiple query information and corresponding satisfactory results of the task processing model without relying on the knowledge base.

[0064] In this application, the task processing model is a machine learning model used to output feedback information based on the question information provided by the user.

[0065] For example, the task processing model can be a large language model. Users can input information query requests, dialogue statements, or task instructions into the large language model, and the large language model can provide feedback information such as query results, reply statements, or task execution results based on the user's input.

[0066] Of course, task processing models can also be used for other large-scale machine learning models, which will not be elaborated on here.

[0067] The knowledge base stores a large number of knowledge objects that the task processing model needs to query or refer to when processing query information. In this application, the knowledge objects in the knowledge base can at least represent document fragments switched from documents. Based on the information contained in the document fragments represented by each knowledge object in the knowledge base, the task processing module can provide relevant knowledge basis for analyzing and processing query information.

[0068] Among them, the satisfaction result corresponding to the question information is used to indicate the user's satisfaction with the feedback information output by the task processing model in processing the question information.

[0069] For example, a satisfactory result can indicate whether the feedback information output by the task processing model in processing the question is satisfactory. Accordingly, a satisfactory result can be either satisfactory or unsatisfactory.

[0070] For example, a satisfaction result can be the user's level of satisfaction or level of satisfaction with the feedback information output by the task processing model after processing the query information. The satisfaction level can be divided into several levels, such as very satisfied, fairly satisfied, neutral, and dissatisfied. Satisfaction can also be expressed as a satisfaction score, which can be categorized as 100 points, 80 points, 60 points, or below 60 points. Alternatively, satisfaction can be expressed as a percentage, such as 100% satisfied, 80% satisfied, 60% satisfied, and 20% satisfied, without any restrictions.

[0071] It is understandable that these multiple question messages are historical inputs into the task processing model, and that the task processing model processes the question messages without relying on a knowledge base.

[0072] To ensure the task processing model can properly handle the user-inputted questions, in this application, the multiple question messages obtained can be selected from the user's historical input and used for testing. These test questions are used to test the validity of documents in the knowledge base.

[0073] S102, for each query information, identify at least one target document in the knowledge base that matches the query information.

[0074] In this context, matching the target document with the query information indicates that among the document fragments segmented from the target document, there exists at least one candidate document fragment whose candidate knowledge object matches the query information. In this application, for ease of distinction, the document matching the query information is referred to as the target document.

[0075] It is understood that a knowledge base includes multiple knowledge objects, and the document fragments represented by these knowledge objects originate from documents. Based on this, this application can determine at least one target document to which each candidate knowledge object in the knowledge base that matches the query information belongs.

[0076] Among them, matching candidate knowledge objects with question information can be achieved when the similarity between the candidate knowledge objects and the question information exceeds a set similarity threshold.

[0077] It should be noted that the target document matching the query information in the knowledge base can be determined in real-time from the knowledge base when updates are needed. Alternatively, after obtaining the query information, the target document matching the query information can be pre-determined, and the document information (such as document name or other document identifiers) of at least one matching target document can be stored. Based on this, this application can directly determine the target document corresponding to the query information based on the stored matching relationships between different query information and target documents.

[0078] S103, for each target document, based on the satisfaction results corresponding to each question information associated with the target document, determine the overall satisfaction level of the user with the feedback information corresponding to at least one question information associated with the target document.

[0079] The overall satisfaction level reflects the user's overall satisfaction with the feedback information output by the task processing model in processing at least one question related to the target document.

[0080] In this application, for each target document, the overall satisfaction level can be determined by comprehensively considering the satisfaction results corresponding to all the questions associated with the target document; or it can be determined by comprehensively considering the satisfaction results corresponding to some of the questions associated with the target document. There is no limitation on either approach.

[0081] In this application, there are several possible implementations for determining the overall satisfaction level based on the satisfaction results corresponding to the query information associated with the target document. The specific implementation method for determining the overall satisfaction level will also differ depending on the specific form of the satisfaction results corresponding to the query information; no limitation is imposed on this. For ease of understanding, the following explanations, in conjunction with several possible forms of satisfaction results, illustrate several possible scenarios for determining the overall satisfaction level.

[0082] In one possible scenario, if the satisfaction result corresponding to the question information can be divided into satisfactory or unsatisfactory, then for a target document, the proportion of question information with a satisfactory result can be counted among at least one question information associated with the target document, and the overall satisfaction level corresponding to the target document can be determined by this proportion.

[0083] In another possible scenario, if the satisfaction result corresponding to the question information is a satisfaction score used to characterize the degree of satisfaction, then for a target document, the average of the satisfaction scores corresponding to at least one question information associated with the target document can be calculated, and the calculated average value can be determined as the overall degree of satisfaction.

[0084] In another possible scenario, if the satisfaction result corresponding to the question information is a satisfaction level, then for a target document, the correspondence between satisfaction level and satisfaction score can be used to convert the satisfaction level corresponding to each question information associated with the target document into the satisfaction score corresponding to the question information. Then, the average value of the satisfaction scores corresponding to at least one question information associated with the target document is calculated, and the calculated average value is determined as the overall satisfaction level.

[0085] Of course, there are other possibilities for the specific form of the satisfactory result of the question information, and correspondingly, there are other possibilities for the specific implementation of determining the overall satisfaction level of the target document, which will not be elaborated here.

[0086] S104. If the overall satisfaction level of the target document exceeds the set threshold, the target document is compressed.

[0087] The threshold value can be set as needed and is not restricted.

[0088] Understandably, the overall satisfaction level of the target document reflects the user's overall satisfaction with the feedback information output by the task processing model for the various questions associated with the target document. Since the task processing model processes the questions without relying on a knowledge base, a high overall satisfaction level for the target document indicates that the task processing model can accurately output the feedback information associated with the questions without relying on the knowledge objects related to the target document in the knowledge base.

[0089] Therefore, the higher the overall satisfaction level of the target document, the less helpful the knowledge information contained in the target document is to the task processing model in processing the query information. In other words, the knowledge information contained in the target document has a smaller impact on the accurate processing of the query information by the task processing model, which also means that the knowledge objects related to the target document in the knowledge base are less meaningful. Based on this, in order to reduce the amount of data in the knowledge base and the amount of knowledge base data that the task processing model needs to load when processing query information, this application can compress target documents whose overall satisfaction level exceeds a set threshold.

[0090] The purpose of compressing the target document is to reduce the number of knowledge objects related to the target document in the knowledge base. For example, compressing the target document can involve compressing document fragments belonging to the target document in the knowledge base, such as partially or completely removing document fragments that were segmented from the target document in the knowledge base.

[0091] As described above, this application identifies target documents in the knowledge base that match each query message processed by the task processing model, and determines the overall satisfaction level for each target document. Since the overall satisfaction level of a target document reflects the user's overall satisfaction with the feedback information output by the task processing model for each query message associated with that target document, and since the task processing model processes query information without relying on the knowledge base, a high overall satisfaction level for a target document indicates that the task processing model can accurately output the feedback information corresponding to the query message associated with that target document even without relying on the knowledge objects related to that target document in the knowledge base. This naturally means that the knowledge information contained in the target document has a relatively small impact on the task processing model's accurate processing of query information. Based on this, by compressing target documents whose overall satisfaction level exceeds a set threshold, it is possible to effectively reduce the number of knowledge objects in the knowledge base that have a small impact on the task processing model. This reduces the amount of knowledge information loaded into memory during the task processing model's operation, reduces memory resource consumption, and thus effectively reduces situations where the task processing model cannot effectively utilize query information due to insufficient memory resources of electronic devices.

[0092] It is understood that, in the embodiments of this application, if the number of query information associated with the target document is small, the satisfactory results corresponding to the query information associated with the target document cannot accurately reflect the impact of the target document on the task processing model's processing of query information.

[0093] Based on this, this application can also determine the total number of questions associated with the target document. Only when this total number is greater than a set number, and the overall satisfaction level corresponding to the target document exceeds a set threshold, will the target document be compressed.

[0094] It is understandable that the specific form of the knowledge objects in the knowledge base in this application can be varied.

[0095] For example, in one possible scenario, a knowledge base may include a document library containing multiple document fragments, each of which can be considered a knowledge object.

[0096] In another possible scenario, to improve the data processing speed of electronic devices, vector representation is commonly used. Based on this, knowledge objects in a knowledge base can also represent vectors of document fragments. For example, a knowledge base might include a vector library containing vectors of different document fragments, each vector representing a document fragment. Each vector can be considered a knowledge object; therefore, a knowledge object can also represent a document fragment.

[0097] Of course, a knowledge base can also include both a document library and a vector library, without any restrictions.

[0098] Accordingly, in this application, compressing the target document may include at least one of the following:

[0099] Compress the vectors in the vector library that are related to the target document;

[0100] Compress document fragments in the document library that are related to the target document.

[0101] Understandably, considering that vectors based on document fragments can more efficiently determine whether a document fragment matches the query information, this application can determine the target document that matches the query information based on the vectors of each document fragment in the vector library. The following is an example of one possible implementation.

[0102] like Figure 2 This illustration shows a flowchart of a knowledge base update method provided in an embodiment of this application. The method in this embodiment may include:

[0103] S201, obtain multiple query messages processed by the task processing model without relying on a knowledge base, and the satisfactory result corresponding to each query message.

[0104] Among them, the satisfaction result corresponding to the question information is used to indicate the user's satisfaction with the feedback information output by the task processing model in processing the question information.

[0105] In this embodiment, for ease of understanding, the knowledge base includes a document library and a vector library as an example. The document library includes at least one document segment from at least one document, and the vector library includes vectors of each document segment in the document library.

[0106] S202, for each query information, determine at least one target vector in the vector library that matches the information features of the query information.

[0107] Among them, the information features of the question information are the extracted feature information used to characterize the question information.

[0108] For example, the question information can be vector-encoded to obtain a feature vector that represents the question information.

[0109] The target vector that matches the information features of the question can be a vector whose similarity to the information features of the question is less than a set matching threshold.

[0110] For example, the information features of a question are its feature vectors. The distance between the feature vectors of the question and each vector in the vector library can be calculated, and vectors whose distance from the feature vectors of the question is less than a set matching threshold can be identified as target vectors.

[0111] S203, determine at least one target document to which the document fragment corresponding to the at least one target vector belongs.

[0112] For example, based on the mapping relationship between document fragments in the document library and vectors in the vector library, the document fragment represented by the target vector is determined. Then, based on the metadata recorded in the document library, the target document to which the document fragment represented by the target vector belongs is queried.

[0113] In this embodiment, the vectors of each document fragment in the vector library are matched with the information features of the question information, which can more efficiently and accurately determine the document fragment represented by the vector that matches the question information.

[0114] It should be noted that this embodiment is illustrated by taking the example of determining at least one target document that matches each question information in real time when updating the knowledge base.

[0115] In practical applications, in order to improve the efficiency of knowledge base updates, this application may also determine at least one target document that matches the question information before updating the knowledge base and after each time the question information is obtained. However, the determination of the at least one target document can also be done in the manner of steps S202 and S203, which will not be elaborated here.

[0116] S204, For each target document, based on the satisfaction results corresponding to each question associated with the target document, determine the overall satisfaction level of the user with the feedback information corresponding to at least one question associated with the target document.

[0117] This step can be referred to in the relevant description of the previous embodiments, and there are no limitations thereon.

[0118] S205, If the overall satisfaction level of the target document exceeds the set threshold, compress the vectors related to the target document in the vector library.

[0119] For example, removing some or all vectors related to the target document from the vector library.

[0120] Understandably, considering that in most application scenarios, task processing models rely on document fragment vectors for knowledge retrieval to improve the efficiency of processing query information, the knowledge base loaded into memory in most scenarios is a vector library containing the corresponding vectors of each document fragment. Therefore, to reduce the amount of data loaded into the knowledge base, the vectors in the vector library need to be compressed. Based on this, this embodiment uses the compression of vectors in the vector library as an example.

[0121] It is understandable that if the task processing model uses a document library for knowledge retrieval during the processing of query information, then this embodiment can also compress document fragments related to the target document in the document library.

[0122] Of course, in order to ensure a one-to-one mapping between documents in the document library and vectors in the vector library, this embodiment can also compress the vectors in the vector library related to the target document while simultaneously compressing the document fragments in the document library related to the target document.

[0123] In any of the above embodiments of this application, there are multiple possible implementations for compressing the knowledge base based on the target document. Several possible implementations will be described below.

[0124] In one possible implementation, for each target document, if the overall satisfaction level corresponding to the target document exceeds a set threshold, at least one target knowledge object in the knowledge base that is associated with the target document and meets the importance requirement is identified, and all knowledge objects in the knowledge base that are associated with the target document but do not belong to the target knowledge object are removed.

[0125] For example, remove vectors from the vector library that are associated with the target document but do not meet the importance requirements.

[0126] For example, remove documents from the document library that are associated with the target document but do not meet the importance requirements.

[0127] The requirement for meeting the importance level can be set as needed.

[0128] For example, a target knowledge object that meets the importance requirement can be at least one segment of the target document or a vector of the target document segment that can represent the subject meaning of the target document.

[0129] For example, a target knowledge object that meets the importance requirement can be a representative target knowledge object among the various knowledge objects associated with the target document.

[0130] To determine representative target knowledge objects, in one optional approach, this application can cluster the vectors of each document fragment associated with the target document in the knowledge base, and determine the target vector of at least one target document fragment as the cluster center. Correspondingly, knowledge objects associated with the target document but not belonging to the target vector or the target document fragment represented by the target vector can be removed from the knowledge base.

[0131] For example, a knowledge base may include at least one of a document library and a vector library. For the document library, documents that are associated with the target document but do not belong to the target document fragments may be removed. For the vector library, vectors that are associated with the target document but do not belong to the target vector may be removed.

[0132] In another possible implementation, if the overall satisfaction level of the target document exceeds a set threshold, all knowledge objects associated with the target document in the knowledge base can be removed. In this implementation, all knowledge objects associated with the target document in the knowledge base can be deleted or moved from the knowledge base to other storage areas.

[0133] It is understandable that, in practical applications, this application may choose either of the two implementation methods above to compress and process the knowledge objects associated with the target document in the knowledge base, according to actual needs.

[0134] Furthermore, considering that the overall satisfaction level of the target document varies, the degree of influence of the target document on the task processing model in processing the query information will also vary. Therefore, this application can also adopt different compression processing methods based on the specific value of the overall satisfaction level of the target document.

[0135] For example, if the overall satisfaction level of the target document exceeds the first set threshold but is lower than the second set threshold, at least one target knowledge object in the knowledge base that is associated with the target document and meets the importance requirements is identified, and all knowledge objects in the knowledge base that are associated with the target document but do not belong to the target knowledge object are removed.

[0136] If the overall satisfaction level of the target document is not lower than the second set threshold, remove all knowledge objects associated with the target document from the knowledge base.

[0137] It is understood that in this application, the operation of updating the knowledge base by combining multiple query information and the satisfactory results of each query information can be performed when a user's triggering operation is detected, or it can be performed under certain conditions.

[0138] For example, in one possible implementation, when the set knowledge base update time is reached or the amount of data in the knowledge base exceeds the set amount of data, the task processing model obtains multiple query information and the corresponding satisfactory results for each query information without relying on the knowledge base.

[0139] The knowledge base update time can be determined according to the set knowledge base update cycle, or it can be a knowledge base update time preset by the user, without any restrictions.

[0140] The amount of data can be set as needed, such as by considering the memory resources of the electronic device, without any restrictions.

[0141] Understandably, before updating the knowledge base, this application can select a portion of the user-inputted question information as test questions. Correspondingly, after obtaining the user-inputted question information, if it is determined that the question information has been selected for testing, this application will utilize a task processing model to output feedback information for that question information without relying on the knowledge base, and obtain the user's satisfactory feedback result for that question information. Based on this, this application will store the question information used for testing and the corresponding satisfactory result, so that when the knowledge base needs to be updated, it can obtain multiple question information processed by the task processing model without relying on the knowledge base.

[0142] Furthermore, considering that users expect the task processing model to output accurate feedback when processing query information, in order to reduce the impact of updating the knowledge base on the accuracy of the feedback information output by the task processing model, and to reduce the impact of the query information being used for testing on the user experience, this application can also instruct the task processing model to re-output the feedback information of the query information using the knowledge base when the satisfactory result corresponding to the query information is unsatisfactory.

[0143] For easier understanding, please refer to Figure 3 The diagram illustrates a schematic framework of one implementation principle of the solution in this application.

[0144] Depend on Figure 3It can be seen that if question information is obtained before the knowledge base needs to be updated, and the question information is used for testing, different processing strategies will be adopted after the question information is input into the task processing model, depending on whether the user is satisfied with the feedback information output by the task processing model.

[0145] Regardless of whether the user is satisfied with the feedback information output by the task processing model, this application, after determining the information features of the question information (such as feature vectors), will search a vector library based on the information features of the question information, and determine at least one target document that matches the question information based on the detected vectors. Then, it will store the satisfaction result corresponding to the question information and the information of at least one target document that matches the question information.

[0146] If a user is dissatisfied with the feedback information output by the task processing model, in order to avoid affecting the user experience, this application will also use the task processing model to re-output the feedback information of the question, relying on the knowledge base.

[0147] For example, based on the information features of the question information, at least one vector in the vector library that matches the information features of the question information is determined, and the at least one vector is input into the task processing model as the prompt information of the question information, so that the task processing model can re-output feedback information based on the question information and the prompt information.

[0148] Based on the above, if the knowledge base needs to be updated, the total number of times each target document is matched and the overall satisfaction level of the document can be calculated. The total number of times a target document is matched is the total number of times the target document is matched with different query information, which is the total number of query information matches the target document.

[0149] Accordingly, based on the correspondence between the overall satisfaction level of the document and different set thresholds, different compression methods can be used to compress the knowledge base.

[0150] exist Figure 3 Based on this, the following will combine Figure 4 Let's take one implementation method as an example to illustrate.

[0151] like Figure 4 This illustration shows a flowchart of a knowledge base update method provided in an embodiment of this application. The method in this embodiment may include:

[0152] S401, obtain the question information input by the user.

[0153] S402, if it is confirmed that the question information is selected as the question information for testing, the task processing model is used to output the feedback information of the question information without relying on the knowledge base.

[0154] Understandably, human task processing models typically rely on knowledge bases to process query information in order to improve the accuracy of the identified feedback. In this application, however, to determine the target document requiring compression, the task processing model needs to process the selected query information for testing without relying on a knowledge base.

[0155] Whether the question information is used as the question information for testing can be determined according to the pre-set selection strategy for question information, and there are no restrictions on this.

[0156] For example, some questions can be randomly selected as test questions. Therefore, if a question is randomly selected, it is determined that the question is for testing.

[0157] For example, you can set the frequency at which test questions are obtained. If the set frequency is reached, the question is designated as a test question. For instance, every set number of questions, one question is designated as a test question.

[0158] Of course, there are other ways to implement the question information used for testing, and there are no restrictions on this.

[0159] S403, obtain a satisfactory result from the user's feedback information input regarding the question.

[0160] S404, determine at least one target vector in the vector library of the knowledge base that matches the information features of the question information, and based on the association relationship between each document fragment and document in the document library, determine the target document associated with each document fragment represented by the target vector, and obtain at least one target document that matches the question information.

[0161] For details on determining the target vector and the specific implementation of determining the target document based on the target vector, please refer to the previous sections. Figure 2 The details of the embodiments will not be repeated here.

[0162] S405, the question information is stored as the question information processed by the task processing model without relying on the knowledge base, and the satisfactory result corresponding to the question information and the document identifier of at least one target document matched by the question information are stored.

[0163] The document identifier of the target document can be the document name, document code, or other identification information, without any restrictions.

[0164] It should be noted that this embodiment takes the determination of the target document that matches the question information when the question information for testing is obtained as an example. Of course, the determination of the target document that matches the question information can also be performed during the subsequent update of the knowledge base. For details, please refer to the relevant descriptions in the previous embodiments, which will not be repeated here.

[0165] In particular, such as Figure 3 It is understood that if the satisfactory result corresponding to the question information is unsatisfactory, this application can also instruct the task processing model to re-output the feedback information of the question information using a knowledge base (such as a vector library).

[0166] It should be noted that steps S401 to S405 above are preparatory work before updating the knowledge base. They are not operations that need to be performed every time the knowledge base is updated. They are only introduced for ease of understanding, describing the process of obtaining the question information and its corresponding satisfactory results.

[0167] S406, when the set knowledge base update time is reached or the data volume of the knowledge base exceeds the set data volume, obtain multiple query information processed by the task processing model without relying on the knowledge base and the corresponding satisfactory results of each query information.

[0168] S407, For each query message, identify at least one target document that matches the query message.

[0169] For example, based on the correspondence between stored query information and the document identifier of at least one target document, at least one target document that matches the query information can be determined.

[0170] It is understandable that each target document can be matched with one or more query messages, therefore, each target document can be associated with at least one query message.

[0171] S408, for each target document, based on the satisfaction results corresponding to each question information associated with the target document, determine the overall satisfaction level of the user with the feedback information corresponding to at least one question information associated with the target document.

[0172] This step can be referred to in the relevant description of the previous embodiments, and will not be repeated here.

[0173] S409, For each target document, determine the total number of query messages associated with that target document.

[0174] The total number of query messages associated with the target document is the total number of times the target document is matched as a document for query message matching.

[0175] S410, for each target document, if the total number is greater than a set number, the overall satisfaction level of the target document exceeds a first set threshold but is lower than a second set threshold, determine at least one target vector in the vector library that is associated with the target document and meets the importance requirement, remove all vectors in the vector library that are associated with the target document but do not belong to the target vector, and remove all document fragments in the document library that are associated with the target document but do not belong to the target document fragment represented by the target vector.

[0176] like Figure 3 As shown, taking a set number of times (equivalent to a set quantity) of 10 as an example, if the total number of times the target document is matched is no more than 10, then there is no need to compress the vectors and document fragments associated with the target document. Only when the total number of times the target document is matched is greater than 10, this application further determines the threshold range that the overall satisfaction level of the target document meets, and performs different compression processes.

[0177] exist Figure 3 Taking a first threshold of 60% and a second threshold of 80% as an example, it can be understood that if the overall satisfaction level of the target document is greater than 60% but less than 80%, it indicates that the target document still has some influence on the task processing model's handling of the query information. In this case, it is not appropriate to remove all vectors and document fragments associated with the target document. Based on this, this application only removes vectors and document fragments that are associated with the target document and have relatively low importance.

[0178] Among them, the target vector that meets the importance requirement (i.e., has a relatively high importance) can be determined by clustering the vectors of each document segment of the target document, as described above, and will not be repeated here.

[0179] S411, for each target document, if the overall satisfaction level corresponding to the target document is not lower than the second set threshold, remove each document fragment associated with the target document from the document library, and remove each vector associated with the target document from the vector library.

[0180] like Figure 3 Taking the second set threshold of 80% as an example, if the overall satisfaction level of the target document is not less than 80%, it means that the knowledge information contained in the target document has no effect on the task processing model in processing the question information. Therefore, all vectors and document fragments associated with the target document can be directly removed to minimize the amount of data in the vector library and document library.

[0181] Understandably, users of electronic devices can add document fragments to the document library as needed. To minimize invalid data in the knowledge base that has little effect on the task processing model's handling of query information, this application can also determine whether a document fragment is invalid data that is useless to the task processing model before adding it to the knowledge base. Only if the document fragment is valid data will it be added to the knowledge base (such as the document library).

[0182] like Figure 5 This illustration shows a schematic diagram of an implementation process for controlling the addition of document fragments to the knowledge base in an embodiment of this application. This embodiment may include:

[0183] S501, Obtain at least one pending document fragment to be stored.

[0184] For example, it can obtain at least one pending document that a user is about to upload, and obtain at least one pending document fragment from the pending document.

[0185] Of course, users can also directly upload at least one pending document fragment associated with different documents, without any restrictions.

[0186] S502, for each pending document fragment, use the task processing model to generate test question information for the pending document fragment, and use the task processing model to generate test feedback information for the test question information without the pending document fragment, and obtain the test result.

[0187] Specifically, the test question information for the pending document fragment refers to the question information generated by the task processing model using the pending document fragment. This question information is non-generalized and related to the pending document fragment. Correspondingly, the feedback information corresponding to the test question information should also be related to the content of the pending document fragment.

[0188] The test result is used to indicate the generation status of the test questions and feedback information for the pending document fragment.

[0189] For example, the test results can indicate whether the task processing model generated the test question information for the pending document fragment; and, given that the task processing model generated the test question information for the pending document fragment, whether the task processing model can generate feedback information for the test question information without the pending document fragment. For ease of distinction, the feedback information output by the task processing model in response to the test question information is called test feedback information.

[0190] S503, if the test result shows that the task processing model can generate test question information for the pending document fragment and the task processing model cannot generate test feedback information for the test question information without the help of the pending document fragment, add at least one of the pending document fragment and the test feedback information corresponding to the pending document fragment to the knowledge base.

[0191] Understandably, if the task processing model cannot generate test questions from a pending document fragment, it means that the fragment cannot provide the necessary knowledge for processing the questions. Therefore, the information in the fragment is invalid and irrelevant to the questions the model can handle. For example, the fragment might be table of contents or other information that doesn't contain valid knowledge. Based on this, if the model cannot generate test questions from a pending document fragment, it will not be added to the knowledge base to avoid increasing data volume by storing invalid data.

[0192] Similarly, if the task processing model generates test question information using the pending document fragment, but the task processing model can also generate test feedback information for the test question information without the pending document fragment, then it means that the knowledge base already contains similar knowledge. Even without the pending document fragment, it will not have any impact on the feedback information for determining the test question information. Based on this, the pending document fragment will not be added to the knowledge base.

[0193] Therefore, for each pending document fragment, if the test result shows that the task processing model cannot generate test question information for the pending document fragment, or if the test result shows that the task processing model can generate test question information for the pending document fragment and the task processing model can generate test feedback information for the test question information without the aid of the pending document fragment, the knowledge base will not include the pending document fragment.

[0194] Correspondingly, if the task processing model can generate test question information for the pending document fragment, and the task processing model cannot generate test feedback information for the test question information without the help of the pending document fragment, then it means that the pending document fragment is useful and effective information for the task processing model, and the pending document fragment will be added to the knowledge base.

[0195] In cases where the knowledge base includes a document library, this application can add the document fragment to be determined to the document library. Furthermore, if the knowledge base also includes a vector library, in this case, the vector corresponding to the document fragment can be generated simultaneously with or after adding the document fragment to the document library, and the generated vector can be added to the vector library.

[0196] There are several possible implementations for obtaining the test results using the task processing model in this application.

[0197] For example, in one possible implementation, the task processing model can be controlled to generate test question information and test feedback information in steps.

[0198] Specifically, firstly, for each pending document fragment, the task processing model can be instructed to generate test questions for that fragment, thus obtaining the task processing model's question generation result. This result characterizes whether the task processing model successfully generated the test questions.

[0199] For example, an electronic device may run a task processing program that is configured with the task processing model. The task processing program can input the pending document fragment into the task processing model and instruct the task processing model to generate test question information for the pending document fragment.

[0200] Secondly, if the question generation result indicates that the task processing model can generate the test question information, the task processing model is instructed to generate test feedback information for the test question information without the aid of the pending document fragment, and the feedback generation result is obtained.

[0201] The feedback generation result is used to characterize whether the task processing model has successfully generated the test feedback information for the test question.

[0202] Correspondingly, the query generation results corresponding to the pending document fragment indicate that the task processing model can generate test query information for the pending document fragment, and the feedback generation results indicate that the task processing model cannot generate test feedback information for the test query information without the help of the pending document fragment, so the pending document fragment is added to the knowledge base.

[0203] To more intuitively understand the process of adding document fragments to the knowledge base, please refer to... Figure 6 , Figure 6 This diagram illustrates a principle framework for controlling the addition of document fragments to a knowledge base.

[0204] exist Figure 6 The example given is that the obtained undetermined document fragments are obtained by segmenting user-uploaded documents.

[0205] Depend on Figure 6 As can be seen, after obtaining the document uploaded by the user, the document is segmented into at least one document fragment. For each document fragment, this application uses a task processing model to generate test question information for that document fragment. If the task processing model cannot generate the document fragment, then the document fragment will not be added to the knowledge base.

[0206] Provided that the task processing model is able to generate the test question information, it also needs to generate test feedback information for the test question information without relying on the pending document fragment. If test feedback information can be generated, the document fragment will not be added to the knowledge base; otherwise, if test feedback information cannot be generated and feedback generation results cannot be obtained, the document fragment will be added to the knowledge base.

[0207] Corresponding to the knowledge base updating method of this application, this application also provides a knowledge base updating device.

[0208] like Figure 7 The diagram illustrates a possible structural composition of a knowledge base updating device provided in this embodiment. The device in this embodiment may include:

[0209] The information acquisition unit 701 is used to acquire multiple question information processed by the task processing model without relying on a knowledge base and the satisfaction results corresponding to the question information. The satisfaction results are used to indicate the degree of user satisfaction with the feedback information output by the task processing model in processing the question information.

[0210] The document determination unit 702 is used to determine at least one target document in the knowledge base that matches the question information;

[0211] The comprehensive determination unit 703 is used to determine the overall satisfaction level of the user with the feedback information corresponding to at least one question associated with the target document based on the satisfaction results corresponding to each question associated with the target document.

[0212] The compression processing unit 704 is used to compress the target document if the overall satisfaction level of the target document exceeds a set threshold.

[0213] In one possible implementation, the knowledge base includes a document library and a vector library, wherein the document library includes at least one document segment from at least one document, and the vector library includes vectors of each document segment in the document library;

[0214] This document identifies units, including:

[0215] A vector determination subunit is used to determine at least one target vector in the vector library that matches the information features of the query information;

[0216] The document determination subunit is used to determine at least one target document to which the document fragment corresponding to the at least one target vector belongs;

[0217] The compression processing unit performs compression processing on the target document, including at least one of the following:

[0218] The vectors in the vector library that are related to the target document are compressed.

[0219] The document fragments related to the target document in the document library are compressed.

[0220] In yet another possible implementation, the compression processing unit includes:

[0221] The first compression processing subunit is used to determine at least one target knowledge object in the knowledge base that is associated with the target document and meets the importance requirement if the overall satisfaction level corresponding to the target document exceeds a set threshold, and remove each knowledge object in the knowledge base that is associated with the target document but does not belong to the target knowledge object.

[0222] Alternatively, the second compression processing subunit is configured to remove each knowledge object associated with the target document from the knowledge base if the overall satisfaction level corresponding to the target document exceeds a set threshold, wherein the knowledge object represents at least one of document fragments and vectors of document fragments.

[0223] In another possible implementation, the first compression processing subunit is specifically used to determine at least one target knowledge object in the knowledge base that is associated with the target document and meets the importance requirement if the overall satisfaction level corresponding to the target document exceeds a first set threshold and is lower than a second set threshold, and to remove each knowledge object in the knowledge base that is associated with the target document but does not belong to the target knowledge object.

[0224] Alternatively, the second compression processing subunit is specifically configured to remove each knowledge object associated with the target document from the knowledge base if the overall satisfaction level corresponding to the target document is not lower than a second preset threshold.

[0225] In another possible implementation, when the first compression processing subunit determines at least one target knowledge object in the knowledge base that is associated with the target document and whose importance meets the requirements, it is specifically used to cluster the vectors of each document fragment associated with the target document in the knowledge base, and determine the target vector of at least one target document fragment as the cluster center.

[0226] When removing knowledge objects in the knowledge base that are associated with the target document but do not belong to the target knowledge object, the first compression processing subunit is specifically used to remove knowledge objects in the knowledge base that are associated with the target document but do not belong to the target document fragment or target vector.

[0227] In yet another possible implementation, the device further includes:

[0228] A quantity determination unit is used to determine the total number of query messages associated with the target document;

[0229] Specifically, the compression processing unit is used to compress the target document if the total number is greater than a set number and the overall satisfaction level of the target document exceeds a set threshold.

[0230] In yet another possible implementation, the device further includes:

[0231] The question acquisition unit is used to obtain the question information input by the user;

[0232] The feedback output unit is used to output feedback information of the question information without relying on the knowledge base if it is confirmed that the question information has been selected as the question information for testing.

[0233] A satisfaction result obtaining unit is used to obtain the satisfaction result input by the user in response to the question information;

[0234] An information storage unit is used to store the question information as the question information processed by the task processing model without relying on a knowledge base, and to store the satisfactory result corresponding to the question information;

[0235] The feedback re-output unit is used to instruct the task processing model to re-output the feedback information of the question information using the knowledge base when the satisfactory result corresponding to the question information is unsatisfactory.

[0236] The information acquisition unit is specifically used to obtain multiple query information processed by the task processing model without relying on the knowledge base, as well as the satisfactory results corresponding to the query information, when the set knowledge base update time is reached or the data volume of the knowledge base exceeds the set data volume.

[0237] In yet another possible implementation, the device further includes:

[0238] A fragment acquisition unit is used to acquire at least one undetermined document fragment to be stored;

[0239] The test processing unit is used to generate test question information for the pending document fragment using the task processing model, and to generate test feedback information for the test question information using the task processing model without the aid of the pending document fragment, to obtain test results. The test results are used to indicate the generation status of the test question information and the test feedback information for the pending document fragment.

[0240] The document entry unit is used to add at least one of the pending document fragment and the corresponding test feedback information to the knowledge base if the test results show that the task processing model can generate test question information for the pending document fragment and the task processing model cannot generate test question information without the pending document fragment.

[0241] Furthermore, the device also includes:

[0242] The non-entry control unit is configured to prevent the knowledge base from adding the pending document fragment if the test results indicate that the task processing model can generate test question information for the pending document fragment, and the task processing model can generate test feedback information for the test question information without the pending document fragment, or the task processing model cannot generate test question information for the pending document fragment.

[0243] Furthermore, this application also provides an electronic device, such as Figure 8 As shown, it illustrates a schematic diagram of the composition structure of the electronic device. The electronic device can be any type of electronic device, and the electronic device includes at least a processor 801 and a memory 802.

[0244] The processor 801 is used to execute the knowledge base update method as described in any of the above embodiments.

[0245] The memory 802 is used to store programs required by the processor to perform operations.

[0246] It is understood that the electronic device may also include a display unit 803 and an input unit 804.

[0247] Of course, the electronic device can also have more than Figure 8 There are no restrictions on the number of components, whether more or fewer.

[0248] On the other hand, this application also provides a computer-readable storage medium storing at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by a processor to implement the knowledge base update method as described in any of the above embodiments.

[0249] This application also proposes a computer program comprising computer instructions stored in a computer-readable storage medium. When run on an electronic device, the computer program performs the knowledge base update method as described in any of the above embodiments.

[0250] It is understood that in this application, the terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar parts and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in a sequence other than that illustrated herein.

[0251] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. Furthermore, the features described in the various embodiments of this specification can be substituted or combined with each other, enabling those skilled in the art to implement or use this application. For apparatus embodiments, since they are basically similar to method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0252] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0253] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0254] The above are merely preferred embodiments of this application. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A knowledge base update method, comprising: The task processing model obtains multiple query information processed without relying on a knowledge base and the corresponding satisfaction results of the query information. The satisfaction results are used to indicate the user's satisfaction with the feedback information output by the task processing model in processing the query information. Identify at least one target document in the knowledge base that matches the query information; Based on the satisfaction results corresponding to each question associated with the target document, determine the overall satisfaction level of the user with the feedback information corresponding to at least one question associated with the target document; If the overall satisfaction level of the target document exceeds a set threshold, the target document is compressed.

2. The knowledge base update method according to claim 1, wherein the knowledge base includes a document library and a vector library, the document library includes at least one document segment from at least one document, and the vector library includes vectors of each document segment in the document library; The step of determining at least one target document in the knowledge base that matches the question information includes: Determine at least one target vector in the vector library that matches the information features of the question information; Determine at least one target document to which the document fragment corresponding to the at least one target vector belongs; The compression process of the target document includes at least one of the following: The vectors in the vector library that are related to the target document are compressed. The document fragments related to the target document in the document library are compressed.

3. The knowledge base update method according to claim 1, wherein if the overall satisfaction level corresponding to the target document exceeds a set threshold, the target document is compressed, comprising: If the overall satisfaction level of the target document exceeds a set threshold, at least one target knowledge object in the knowledge base that is associated with the target document and meets the importance requirement is identified, and all knowledge objects in the knowledge base that are associated with the target document but do not belong to the target knowledge object are removed. Alternatively, if the overall satisfaction level of the target document exceeds a set threshold, remove all knowledge objects associated with the target document from the knowledge base; The knowledge object represents at least one of a document fragment and a vector of the document fragment.

4. The knowledge base update method according to claim 3, wherein if the overall satisfaction level corresponding to the target document exceeds a set threshold, determining at least one target knowledge object in the knowledge base that is associated with the target document and meets the importance requirement, and removing all knowledge objects in the knowledge base that are associated with the target document but do not belong to the target knowledge object, includes: If the overall satisfaction level of the target document exceeds a first set threshold and is lower than a second set threshold, determine at least one target knowledge object in the knowledge base that is associated with the target document and meets the importance requirement, and remove all knowledge objects in the knowledge base that are associated with the target document but do not belong to the target knowledge object. Alternatively, if the overall satisfaction level corresponding to the target document exceeds a set threshold, removing each knowledge object associated with the target document from the knowledge base includes: If the overall satisfaction level of the target document is not lower than the second set threshold, remove all knowledge objects associated with the target document from the knowledge base.

5. The knowledge base updating method according to claim 3 or 4, wherein determining at least one target knowledge object in the knowledge base that is associated with the target document and whose importance meets the requirements includes: Cluster the vectors of each document fragment associated with the target document in the knowledge base to determine the target vector of at least one target document fragment as the cluster center; The removal of knowledge objects from the knowledge base that are associated with the target document but do not belong to the target knowledge object includes: Remove all knowledge objects in the knowledge base that are associated with the target document but do not belong to the target document fragment or target vector.

6. The knowledge base update method according to claim 1 further includes: Determine the total number of query messages associated with the target document; If the overall satisfaction level of the target document exceeds a set threshold, the target document is compressed, including: If the total number is greater than the set number and the overall satisfaction level of the target document exceeds the set threshold, the target document is compressed.

7. The knowledge base updating method according to claim 1 further includes: Obtain the question information entered by the user; If it is confirmed that the question information is selected as the question information for testing, the task processing model is used to output the feedback information of the question information without relying on the knowledge base; Obtain satisfactory results from the user's feedback information regarding the question; The question information is stored as the question information processed by the task processing model without relying on the knowledge base, and the corresponding satisfactory result is stored. If the satisfactory result corresponding to the question information is unsatisfactory, the task processing model is instructed to use the knowledge base to re-output the feedback information of the question information; The process of obtaining multiple query messages processed by the task processing model without relying on a knowledge base, and the corresponding satisfactory results for the query messages, includes: When the set knowledge base update time is reached or the data volume of the knowledge base exceeds the set data volume, the task processing model obtains multiple query information and the corresponding satisfactory results of the query information without relying on the knowledge base.

8. The knowledge base updating method according to claim 1 further includes: Obtain at least one pending document fragment to be stored; The task processing model is used to generate test question information for the pending document fragment, and the task processing model is used to generate test feedback information for the test question information without the pending document fragment, to obtain test results. The test results are used to indicate the generation status of the test question information and the test feedback information for the pending document fragment. If the test results indicate that the task processing model can generate test question information for the pending document fragment, and the task processing model cannot generate test feedback information for the test question information without the pending document fragment, then at least one of the pending document fragment and the corresponding test feedback information is added to the knowledge base.

9. The knowledge base updating method according to claim 8 further includes: If the test results indicate that the task processing model can generate test question information for the pending document fragment, and the task processing model can generate test feedback information for the test question information without the pending document fragment, or the task processing model cannot generate test question information for the pending document fragment, then the knowledge base will not include the pending document fragment.

10. A knowledge base updating device, comprising: The information acquisition unit is used to acquire multiple question information processed by the task processing model without relying on a knowledge base, as well as the satisfaction results corresponding to the question information. The satisfaction results are used to indicate the user's satisfaction with the feedback information output by the task processing model in processing the question information. A document determination unit is used to determine at least one target document in the knowledge base that matches the query information; The comprehensive determination unit is used to determine the overall satisfaction level of the user with the feedback information corresponding to at least one question associated with the target document based on the satisfaction results corresponding to each question associated with the target document. A compression processing unit is used to compress the target document if the overall satisfaction level of the target document exceeds a set threshold.

Images