Enterprise knowledge retrieval method, system and equipment based on knowledge base

Abstract

The invention provides an enterprise knowledge retrieval method, system and equipment based on a knowledge base, and belongs to the technical field of artificial intelligence. The method comprises the following steps: in response to a query request for system questions and answers input by a user, determining user permission information corresponding to the query request, and generating a corresponding double-knowledge retrieval dictionary based on the user permission information; the double knowledge retrieval dictionary comprises knowledge base metadata and knowledge metadata. Determining first context retrieval information and second context retrieval information corresponding to the query request according to a system retrieval text corresponding to the query request; according to the first context retrieval information and the second context retrieval information, sequentially screening a retrieval knowledge base and each corresponding retrieval knowledge block from the double knowledge retrieval dictionary; and matching each retrieval knowledge block with the system retrieval text to determine a target knowledge block corresponding to the query request according to a first matching result, and sending the target knowledge block to the user terminal and displaying the target knowledge block on a user interface.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a knowledge retrieval method, system and device based on a knowledge base for enterprise knowledge retrieval. Background Technology

[0002] In the daily operations and management of enterprises, it is crucial to efficiently and accurately retrieve and obtain relevant corporate policies, regulations, processes, and other related knowledge. Employees often need to quickly look up and understand relevant corporate policies when performing tasks, making decisions, or answering questions from customers or colleagues.

[0003] Currently, existing enterprise knowledge retrieval methods typically employ unstructured searches, lacking precise filtering dimensions and resulting in low retrieval efficiency across massive datasets. Furthermore, traditional retrieval processes often treat permission verification as a separate step, failing to adapt to dynamic business scenarios and provide accurate knowledge retrieval results. Summary of the Invention

[0004] This application provides a knowledge base-based enterprise knowledge retrieval method, system, and device to address the technical problems of low efficiency in enterprise knowledge retrieval, high difficulty in retrieval under dynamic business scenarios, and poor accuracy of retrieval results.

[0005] On the one hand, embodiments of this application provide a knowledge base-based enterprise knowledge retrieval method, which includes:

[0006] In response to a user's query request for questions about regulations, the system determines the user's permission information corresponding to the query request and generates a corresponding dual-knowledge retrieval dictionary based on the user's permission information; the dual-knowledge retrieval dictionary includes knowledge base metadata and knowledge metadata.

[0007] Based on the policy retrieval text corresponding to the query request, determine the first context retrieval information and the second context retrieval information corresponding to the query request;

[0008] Based on the first context search information and the second context search information, the search knowledge base and the corresponding search knowledge blocks are sequentially filtered from the dual knowledge search dictionary;

[0009] Each of the search knowledge blocks is matched with the system search text to determine the target knowledge block corresponding to the query request based on the first matching result, and the target knowledge block is sent to the user terminal and displayed on the user interface.

[0010] In one implementation of this application, determining the user permission information corresponding to the query request specifically includes:

[0011] Based on the user identifier in the query request, determine the login account corresponding to the user identifier;

[0012] Based on the login account and the pre-stored registration information table, the user's organizational attribute code is determined; wherein, the organizational attribute code includes at least the following attribute dimensions: user department, user job level, and user office area;

[0013] The user permission information is generated based on the organization attribute code and the preset system retrieval permission lookup table.

[0014] In one implementation of this application, generating a corresponding dual-knowledge retrieval dictionary based on the user permission information specifically includes:

[0015] The user permission information is encoded into a permission feature vector;

[0016] The permission feature vector is matched with each first-level index vector to determine the matching knowledge base label based on the second matching result; wherein, the first-level index vector is constructed based on the first permission mapping rule of the knowledge base label; one knowledge base label corresponds to one first-level index vector;

[0017] Determine each second-level index vector corresponding to the matching knowledge base tag; the second-level index vector is constructed based on the second permission mapping rule of knowledge block tagging;

[0018] The permission feature vector is matched with the second-level index vector to determine the matching knowledge block tag based on the third matching result;

[0019] The dual knowledge retrieval dictionary is constructed based on the matching knowledge base tags and the matching knowledge block markers; wherein the matching knowledge base tags are the parent filtering fields of the matching knowledge block markers.

[0020] In one implementation of this application, before determining the first context search information and the second context search information corresponding to the query request based on the system search text in the query request, the method further includes:

[0021] The initial policy-related question and answer statements in the query request are segmented into words to obtain policy retrieval keywords.

[0022] The policy retrieval keywords are input into a pre-trained enterprise terminology conversion model to determine the policy retrieval text based on the model output; wherein, the enterprise terminology conversion model is trained based on several natural language question-and-answer statements.

[0023] In one implementation of this application, determining the first context search information and the second context search information corresponding to the query request based on the system search text in the query request specifically includes:

[0024] Based on the retrieved text of the aforementioned system and the preset business dimension text library, keywords for each business area are determined to construct knowledge base tag filtering conditions, and the knowledge base tag filtering conditions are used as the first context retrieval information; wherein, the business area keywords include at least one or more of the following classification dimensions: finance, human resources, and operations;

[0025] According to the system, the text is retrieved and a preset micro-attribute feature text library is used to determine the corresponding retrieval micro-attribute features and generate knowledge block filtering conditions, so as to use the knowledge block filtering conditions as the second context retrieval information; the preset micro-attribute feature text library includes at least the following dimension feature attributes: timeliness, permission level, and regional attribute.

[0026] In one implementation of this application, the first context retrieval information includes one or more of the following filtering conditions: logical combination of business domain keywords based on AND and OR operations, equality comparison, and inclusion comparison;

[0027] The second contextual retrieval information includes one or more of the following filtering conditions: logical combination of retrieval micro-attribute features based on AND and OR operations, numerical type comparison, string type comparison, and time type comparison.

[0028] In one implementation of this application, based on the first context retrieval information and the second context retrieval information, the retrieval knowledge base and corresponding retrieval knowledge blocks are sequentially filtered from the dual knowledge retrieval dictionary, specifically including:

[0029] The retrieval knowledge base is obtained by filtering from the dual knowledge retrieval dictionary based on the first context retrieval information;

[0030] In the obtained retrieval knowledge base, knowledge blocks are filtered using the second context retrieval information to determine each retrieval knowledge block based on the filtering results.

[0031] In one implementation of this application, each of the retrieved knowledge blocks is matched with the system retrieval text to determine the target knowledge block corresponding to the query request based on the matching results, specifically including:

[0032] The similarity between the knowledge block vector corresponding to each retrieval knowledge block and the vector corresponding to the system retrieval text is calculated using a vector matching algorithm.

[0033] The vector similarity is compared with a first preset threshold.

[0034] If the vector similarity is greater than the first preset threshold, the corresponding retrieval knowledge block is determined as the target knowledge block.

[0035] Secondly, embodiments of this application also provide an enterprise knowledge retrieval system based on a knowledge base, the system comprising:

[0036] The first determining module is used to respond to a user's query request for a policy-related question and answer, determine the user permission information corresponding to the query request, and generate a corresponding dual knowledge retrieval dictionary based on the user permission information.

[0037] The second determining module is used to determine the first contextual retrieval information and the second contextual retrieval information corresponding to the query request based on the system retrieval text in the query request.

[0038] The filtering module is used to sequentially filter the retrieval knowledge base and corresponding retrieval knowledge blocks from the dual knowledge retrieval dictionary based on the first context retrieval information, the second context retrieval information and preset filtering rules.

[0039] The matching module is used to match each of the search knowledge blocks with the system search text, so as to determine the target knowledge block corresponding to the query request based on the first matching result, and send the target knowledge block to the user terminal for display on the user interface.

[0040] Thirdly, embodiments of this application also provide an enterprise knowledge retrieval device based on a knowledge base, the device comprising:

[0041] At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform a knowledge base-based enterprise knowledge retrieval method as described above.

[0042] Compared with the prior art, the significant advantages of this application are as follows:

[0043] The above scheme constructs a metadata dictionary corresponding to user permissions, turning permissions into internal variables of the retrieval logic and enabling deep dynamic coupling between permissions and the retrieval process. Simultaneously, it generates first and second contextual retrieval information to avoid misunderstandings of user searches and performs dual metadata filtering, achieving high-precision and high-efficiency matching of knowledge retrieval results. Furthermore, through three-stage matching—searching the knowledge base, searching knowledge blocks, and targeting knowledge blocks—it achieves refined retrieval from the database level to the block level. This solves the current technical problems of low efficiency in enterprise knowledge retrieval, high retrieval difficulty in dynamic business scenarios, and poor accuracy of retrieval results. Attached Figure Description

[0044] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0045] Figure 1 This is a flowchart illustrating a knowledge base-based enterprise knowledge retrieval method in an embodiment of this application.

[0046] Figure 2 This is a schematic diagram illustrating the process of configuring tags and labels in a knowledge base-based enterprise knowledge retrieval method according to an embodiment of this application;

[0047] Figure 3 This is a schematic diagram of the filtering conditions corresponding to the first context retrieval information in the embodiments of this application;

[0048] Figure 4 This is a schematic diagram of the filtering conditions corresponding to the second context retrieval information in the embodiments of this application;

[0049] Figure 5 This is a schematic diagram of the structure of a knowledge retrieval device corresponding to a knowledge base-based enterprise knowledge retrieval method in an embodiment of this application.

[0050] Figure 6 This is a schematic diagram of the structure of a knowledge base-based enterprise knowledge retrieval system according to an embodiment of this application;

[0051] Figure 7 This is a schematic diagram of the structure of a knowledge base-based enterprise knowledge retrieval device in an embodiment of this application. Detailed Implementation

[0052] To make the purpose, technical solutions, and advantages of this application more clear, the technical solutions of this application will be clearly and completely described below in conjunction with the specific embodiments of this application and the corresponding drawings. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0053] Existing enterprise knowledge retrieval systems typically employ unstructured searches, lacking precise filtering dimensions and exhibiting low retrieval efficiency across massive datasets. Furthermore, traditional retrieval processes often treat authorization verification as a separate step, failing to adapt to dynamic business scenarios and provide accurate knowledge retrieval results.

[0054] Based on this, the embodiments of this application provide a knowledge base-based enterprise knowledge retrieval method, system, and device to solve the technical problems of low efficiency in enterprise knowledge retrieval, high difficulty in retrieval under dynamic business scenarios, and poor accuracy of retrieval results.

[0055] The various embodiments of this application are described in detail below with reference to the accompanying drawings.

[0056] This application provides a knowledge base-based enterprise knowledge retrieval method, such as... Figure 1 As shown, the method may include steps S101-S104:

[0057] S101, the server responds to the user's query request for policy-related questions and answers, determines the user's permission information corresponding to the query request, and generates a corresponding dual knowledge retrieval dictionary based on the user's permission information.

[0058] The aforementioned dual knowledge retrieval dictionary includes knowledge base metadata and knowledge metadata.

[0059] It should be noted that the server, as the executor of the knowledge retrieval method based on the knowledge base, exists only as an example, and the executor is not limited to the server. This application does not make any specific limitation in this regard.

[0060] Users can communicate with the server through user terminals, which can be devices such as mobile phones and computers, without specific limitations. Users can be internal employees of an enterprise, with policy requirements such as reimbursement procedures and limits when returning from business trips. Users can send query requests to the server according to these policy requirements.

[0061] The server can pre-store user permission information for several users. This information includes, but is not limited to, the relationship between attributes such as the user's department, job level, and work area, and the knowledge base and knowledge. The knowledge base stores knowledge that answers user questions. The server can deploy multiple knowledge bases, each storing different categories of policy knowledge. Knowledge base tag categories can be created based on macro-level classification dimensions such as business domains. Each category represents a classification dimension. The main process involves standardizing the definition of each confirmed classification dimension to ensure uniqueness and readability. Adding, deleting, modifying, and querying tag categories is supported. Corresponding tags are created based on defined tag categories, including tag name, tag description, creator, and creation time. Adding, deleting, modifying, and querying tags is also supported.

[0062] The knowledge (knowledge blocks) in the knowledge base can be understood as specific institutional content, which can be set according to expert experience and actual use scenarios. This application does not make specific limitations on this.

[0063] in, Figure 2 This application provides a flowchart of a tagging and tag configuration process for tagging a knowledge base and marking knowledge items, as shown in the embodiments of this application. Figure 2As shown, users can manage tag types, tag knowledge blocks, and mark the document knowledge within knowledge blocks with numerical, string, or time types.

[0064] In this embodiment of the application, the determination of the user permission information corresponding to the query request specifically includes:

[0065] Based on the user identifier in the query request, determine the login account corresponding to the user identifier. Based on the login account and the pre-stored registration information table, determine the user's organizational attribute code. The organizational attribute code includes at least the following attribute dimensions: user department, user job level, and user office area. Based on the organizational attribute code and the preset system retrieval permission lookup table, generate user permission information.

[0066] In other words, the query request carries a user identifier representing the user's identity, and this user identifier is associated with the login account. This login account can be understood as the user account used to log in to the policy-based question-and-answer system corresponding to the knowledge base-based enterprise knowledge retrieval method of this application. The server can pre-store a registration information table containing the correspondence between login accounts and organizational attribute codes. The organizational attribute code represents the user's identity attribute within the enterprise. Simultaneously, the server pre-stores a preset system retrieval permission lookup table, which records the correspondence between different organizational attribute codes and different user permissions. This preset system retrieval permission lookup table can be maintained by a designated person to promptly correct the knowledge base and knowledge accessible to different users when personnel or job positions change.

[0067] Furthermore, in one embodiment of this application, generating a corresponding dual-knowledge retrieval dictionary based on user permission information specifically includes:

[0068] User permission information is encoded into permission feature vectors. These feature vectors are matched against each first-level index vector to determine matching knowledge base tags based on a second matching result. The first-level index vectors are constructed based on the first permission mapping rule for knowledge base tags. One knowledge base tag corresponds to one first-level index vector. Second-level index vectors corresponding to the matching knowledge base tags are determined. These second-level index vectors are constructed based on the second permission mapping rule for knowledge block tags. The permission feature vectors are matched against the second permission mapping rule to determine matching knowledge block tags based on a third matching result. A dual knowledge retrieval dictionary is constructed based on the matching knowledge base tags and the matching knowledge block tags. The matching knowledge base tags serve as the parent filter field for the matching knowledge block tags.

[0069] In other words, this application pre-sets a first-level index vector for each knowledge base tag in the knowledge base. This vector can contain the permission rules for the corresponding knowledge base tag, i.e., the first permission mapping rule, which refers to the relationship between permissions and knowledge base tags. The server encodes the user permission information and calculates the second vector similarity between the encoded permission feature vector and each first-level index vector. This second vector similarity can be cosine similarity or can be calculated using other similarity algorithms; no specific limitation is made here. The second vector similarity is compared with a second preset threshold. If the second vector similarity is greater than the second preset threshold, then the knowledge base tag corresponding to the current first-level index vector is used as the matching knowledge base tag. Otherwise, the second vector similarity of the next first-level index vector is calculated.

[0070] Subsequently, the server will calculate the third vector similarity between the permission feature vector and the second-level index vector. This second-level index vector can be understood as an encoding vector of the relationship between knowledge blocks and permissions. The third vector similarity is compared with a third preset threshold. If the third vector similarity is greater than the third preset threshold, the corresponding knowledge block tag is used as the matching knowledge block tag. Otherwise, the third vector similarity is calculated for the next second-level index vector. Then, the server constructs a dual knowledge retrieval dictionary based on the obtained matching knowledge base tags and matching knowledge block tags. The aforementioned second and third preset thresholds can be set by experts or by users according to actual usage scenarios; no specific limitations are imposed here.

[0071] The dual knowledge retrieval dictionary includes both knowledge base tags and knowledge tags, which are knowledge base metadata and knowledge metadata, respectively.

[0072] The above scheme constructs a metadata dictionary closely related to user permissions, enabling permissions to be deeply integrated into the policy retrieval process itself. This allows for precise permission adaptation during policy retrieval, improving the accuracy of retrieval results.

[0073] S102, the server determines the first context search information and the second context search information corresponding to the query request based on the policy retrieval text corresponding to the query request.

[0074] In this embodiment of the application, before determining the first context search information and the second context search information corresponding to the query request based on the system search text in the query request, the method further includes:

[0075] The initial policy-related question-and-answer statements in the query request are segmented to obtain policy retrieval keywords. These keywords are then input into a pre-trained enterprise terminology conversion model to determine the policy retrieval text based on the model's output. The enterprise terminology conversion model is trained using several natural language question-and-answer statements.

[0076] In other words, the statements users input into the system may be natural language, which may not be accurate enough for direct use in knowledge blocks and knowledge block matching. In this case, the server can first perform word segmentation on the initial policy question-and-answer statement input by the user, thereby obtaining one or more policy search keywords. Subsequently, the server processes the policy search keywords using a pre-trained enterprise terminology conversion model, which can then convert the policy search text into enterprise terminology.

[0077] For example, the keyword for policy retrieval corresponding to the user's input statement might be "taxi fare." After processing by the enterprise terminology conversion model, the resulting policy retrieval text would be "travel and transportation expenses." The enterprise terminology conversion model can be a machine learning model or a deep learning model. This application can also set up an enterprise terminology dictionary to replace the enterprise terminology conversion model to determine the policy retrieval text. This application does not make any specific limitations on this.

[0078] In this embodiment of the application, the above-mentioned determination of the first contextual retrieval information and the second contextual retrieval information corresponding to the query request based on the system retrieval text in the query request specifically includes:

[0079] Based on the policy retrieval text and a pre-defined business dimension text library, keywords for each business area are determined to construct knowledge base tag filtering conditions, which are then used as the first context retrieval information. The business area keywords must include at least one or more of the following classification dimensions: finance, human resources, and operations. Based on the policy retrieval text and a pre-defined micro-attribute feature text library, corresponding retrieval micro-attribute features are determined, and knowledge block filtering conditions are generated, which are then used as the second context retrieval information. The pre-defined micro-attribute feature text library must include at least the following dimension features: timeliness, access level, and regional attributes.

[0080] In other words, a business-dimensional text library is pre-set in the server. This library records the correspondence between keywords in different business areas and search texts of different regulations. By matching the search text of a regulation with the pre-set business-dimensional text library, the keywords of each business area corresponding to the search text of the regulation can be obtained. That is, this application maps the search text of a regulation to a macro-business dimension, thereby constructing knowledge block tag filtering conditions. For example, based on the search text of a regulation, the obtained business area keywords are: {"reimbursement": "finance", "leave": "human resources", "procurement": "operations"}. The constructed knowledge base tag filtering conditions are: matching the knowledge base tags corresponding to the business area keywords, and the applicable user department is "sales department". Combining knowledge block tags and user permissions, the first context search information is constructed.

[0081] Simultaneously, the server pre-stores a preset micro-attribute feature text library, recording the retrieval micro-attribute features of knowledge blocks and the corresponding relationships between several different institutional retrieval texts. By matching the institutional retrieval text with the preset micro-attribute feature text library, knowledge block filtering conditions can also be constructed to obtain second context retrieval information. For example, the second context retrieval information is: Timeliness == 'Valid in 2024' && Access Level >= P7 && Topic == 'Reimbursement'; P7 represents the user's level, and && represents sum.

[0082] In one embodiment of this application, the first contextual retrieval information includes one or more of the following filtering conditions: logical combination of business domain keywords based on AND and OR operations, equality comparison, and inclusion comparison. The second contextual retrieval information includes one or more of the following filtering conditions: logical combination of retrieval micro-attribute features based on AND and OR operations, numerical type comparison, string type comparison, and time type comparison.

[0083] The first contextual retrieval information can include two parts, such as... Figure 3 As shown, one part consists of conditional rules, and the other part consists of comparison rules. Conditional rules include logical combinations of business domain keywords with or with the operation. Comparison rules include equality comparisons and inclusion comparisons. An example of equality comparison is: "label" == "Operations and Maintenance"; an example of inclusion comparison is: "category" in ["Finance", "Human Resources"]. And a conditional rule is: "label" == "Operations and Maintenance" && "category" in ["Finance", "Human Resources"].

[0084] Similarly, the second contextual retrieval information also includes multiple filtering conditions, such as Figure 4 As shown, one part consists of conditional rules, and the other part consists of comparison rules, which include comparisons of numeric types, string types, and time types. Specifically, for numeric type comparisons, the rules include equals, greaterThan, lessThan, between, and isNull; for string type comparisons, the rules include equals, contains, isEmpty, startsWith, and endsWith; and for time type comparisons, the rules include equals, before, after, between, and isNull. In other words, for specific knowledge blocks, detailed searches can be performed using numeric, string, and time criteria.

[0085] S103, the server sequentially filters the search knowledge base and corresponding search knowledge blocks from the dual knowledge search dictionary based on the first context search information and the second context search information.

[0086] In this embodiment of the application, the above-mentioned method of sequentially filtering the retrieval knowledge base and corresponding retrieval knowledge blocks from the dual knowledge retrieval dictionary based on the first context retrieval information and the second context retrieval information specifically includes:

[0087] Based on the first contextual retrieval information, a retrieval knowledge base is obtained by filtering from the dual knowledge retrieval dictionary. Within the obtained retrieval knowledge base, knowledge blocks are further filtered using the second contextual retrieval information to determine each retrieval knowledge block based on the filtering results.

[0088] In other words, this application utilizes first contextual retrieval information for knowledge base matching and second contextual retrieval information for knowledge block querying, thus employing dual contextual data for system retrieval. It fully leverages contextual understanding for metadata matching to enable knowledge retrieval and improve retrieval efficiency.

[0089] Specifically, the server can use the first context retrieval information to filter knowledge block tags that meet the first context retrieval information in the dual knowledge retrieval dictionary, thereby determining the retrieval knowledge base corresponding to the user's current retrieval needs. Then, using the second context retrieval information, the server can further filter out the knowledge blocks that meet the user's current retrieval needs from the obtained retrieval knowledge base.

[0090] S104, the server matches each search knowledge block with the system search text to determine the target knowledge block corresponding to the query request based on the first matching result, and sends the target knowledge block to the user terminal for display on the user interface.

[0091] In this embodiment of the application, the above-mentioned matching of each retrieval knowledge block with the system retrieval text to determine the target knowledge block corresponding to the query request based on the matching results specifically includes:

[0092] The similarity between the knowledge block vector corresponding to each retrieved knowledge block and the vector corresponding to the system retrieval text is calculated using a vector matching algorithm. The vector similarity is then compared to a first preset threshold. If the vector similarity is greater than the first preset threshold, the corresponding retrieved knowledge block is determined as the target knowledge block.

[0093] The vector matching algorithm can be a cosine similarity algorithm or other similarity calculation algorithms; this application does not specifically limit this. The server can calculate the vector similarity between the knowledge block vector constructed from the retrieved knowledge block and the system retrieval text. Then, it compares this vector similarity with a first preset threshold. If the vector similarity is greater than the first preset threshold, it indicates that the retrieved knowledge block corresponding to this vector similarity is the target knowledge block; otherwise, it is not the target knowledge block. This allows for the filtering of system-related knowledge blocks that meet the user's search needs, which are then displayed on the user interface for the user to view.

[0094] To better understand the above embodiments, this application also provides a schematic diagram of the structure of a knowledge retrieval device corresponding to a knowledge base-based enterprise knowledge retrieval method, as shown below. Figure 5 As shown, the knowledge retrieval device includes: an input module, a knowledge base filtering module, and a knowledge filtering module. The input module is used for user question input (initial system question and answer statement) and identifies contextual information (labels + markers). The knowledge base filtering module uses the contextual information (labels), i.e., the first contextual retrieval information, to filter knowledge bases from the labels for knowledge base matching. The knowledge bases may include knowledge base A, knowledge base B, ..., knowledge base G. If knowledge base A is matched, the knowledge filtering module uses the contextual information (marks), i.e., the second contextual retrieval information, to filter knowledge blocks from the markers for knowledge block matching. The knowledge blocks may include knowledge block-1, knowledge block-2, ..., knowledge block-50, a total of 50 knowledge blocks. Subsequently, based on the knowledge blocks and the user question, vector matching is performed to obtain the final similar knowledge blocks, which are the target knowledge blocks.

[0095] The above scheme constructs a metadata dictionary corresponding to user permissions, turning permissions into internal variables of the retrieval logic and enabling deep dynamic coupling between permissions and the retrieval process. Simultaneously, it generates first and second contextual retrieval information to avoid misunderstandings of user searches and performs dual metadata filtering, achieving high-precision and high-efficiency matching of knowledge retrieval results. Furthermore, through three-stage matching—searching the knowledge base, searching knowledge blocks, and targeting knowledge blocks—it achieves refined retrieval from the database level to the block level. This solves the current technical problems of low efficiency in enterprise knowledge retrieval, high retrieval difficulty in dynamic business scenarios, and poor accuracy of retrieval results.

[0096] Figure 6 This is a schematic diagram of the structure of a knowledge base-based enterprise knowledge retrieval system provided in an embodiment of this application. The knowledge base-based enterprise knowledge retrieval system adopts the knowledge base-based enterprise knowledge retrieval method described above. Figure 6 As shown, the knowledge base-based enterprise knowledge retrieval system 600 includes:

[0097] The first determining module 601 is used to respond to a user's query request for policy-related questions and answers, determine the user's permission information corresponding to the query request, and generate a corresponding dual-knowledge retrieval dictionary based on the user's permission information. The second determining module 602 is used to determine the first context retrieval information and the second context retrieval information corresponding to the query request based on the policy retrieval text in the query request. The filtering module 603 is used to sequentially filter the retrieval knowledge base and corresponding retrieval knowledge blocks from the dual-knowledge retrieval dictionary according to the first context retrieval information, the second context retrieval information, and preset filtering rules. The matching module 604 is used to match each retrieval knowledge block with the policy retrieval text to determine the target knowledge block corresponding to the query request based on the first matching result, and send the target knowledge block to the user terminal for display on the user interface.

[0098] Figure 7 A schematic diagram of the structure of a knowledge base-based enterprise knowledge retrieval device provided in this application embodiment is shown below. Figure 7 As shown, the device includes:

[0099] At least one processor; and a memory communicatively connected to the at least one processor. The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to:

[0100] In response to a user's query request regarding regulations, the system determines the user's permission information and generates a dual-knowledge retrieval dictionary based on this information. The dual-knowledge retrieval dictionary includes knowledge base metadata and knowledge metadata. Based on the regulation retrieval text corresponding to the query request, the system determines the first and second context retrieval information. Based on the first and second context retrieval information, the system sequentially filters the knowledge base and corresponding retrieval knowledge blocks from the dual-knowledge retrieval dictionary. Each retrieval knowledge block is matched with the regulation retrieval text to determine the target knowledge block corresponding to the query request based on the first matching result. The target knowledge block is then sent to the user terminal and displayed on the user interface.

[0101] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system and device embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the description of the method embodiments.

[0102] The systems, devices, and methods provided in this application are one-to-one correspondences. Therefore, the systems and devices also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the systems and devices will not be repeated here.

[0103] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0104] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A knowledge base-based enterprise knowledge retrieval method, characterized in that, The method includes: In response to a user's query request for questions about regulations, the system determines the user's permission information corresponding to the query request and generates a corresponding dual-knowledge retrieval dictionary based on the user's permission information; the dual-knowledge retrieval dictionary includes knowledge base metadata and knowledge metadata. Based on the policy retrieval text corresponding to the query request, determine the first context retrieval information and the second context retrieval information corresponding to the query request; Based on the first context search information and the second context search information, the search knowledge base and the corresponding search knowledge blocks are sequentially filtered from the dual knowledge search dictionary; Each of the search knowledge blocks is matched with the system search text to determine the target knowledge block corresponding to the query request based on the first matching result, and the target knowledge block is sent to the user terminal and displayed on the user interface.

2. The enterprise knowledge retrieval method based on a knowledge base according to claim 1, characterized in that, Determining the user permission information corresponding to the query request specifically includes: Based on the user identifier in the query request, determine the login account corresponding to the user identifier; Based on the login account and the pre-stored registration information table, the user's organizational attribute code is determined; wherein, the organizational attribute code includes at least the following attribute dimensions: user department, user job level, and user office area; The user permission information is generated based on the organization attribute code and the preset system retrieval permission lookup table.

3. The enterprise knowledge retrieval method based on a knowledge base according to claim 1, characterized in that, Based on the user permission information, a corresponding dual-knowledge retrieval dictionary is generated, specifically including: The user permission information is encoded into a permission feature vector; The permission feature vector is matched with each first-level index vector to determine the matching knowledge base label based on the second matching result; wherein, the first-level index vector is constructed based on the first permission mapping rule of the knowledge base label; one knowledge base label corresponds to one first-level index vector; Determine each second-level index vector corresponding to the matching knowledge base tag; the second-level index vector is constructed based on the second permission mapping rule of knowledge block tagging; The permission feature vector is matched with the second-level index vector to determine the matching knowledge block tag based on the third matching result; The dual knowledge retrieval dictionary is constructed based on the matching knowledge base tags and the matching knowledge block markers; wherein the matching knowledge base tags are the parent filtering fields of the matching knowledge block markers.

4. The enterprise knowledge retrieval method based on a knowledge base according to claim 1, characterized in that, Before determining the first context search information and the second context search information corresponding to the query request based on the policy search text in the query request, the method further includes: The initial policy-related question and answer statements in the query request are segmented into words to obtain policy retrieval keywords. The policy retrieval keywords are input into a pre-trained enterprise terminology conversion model to determine the policy retrieval text based on the model output; wherein, the enterprise terminology conversion model is trained based on several natural language question-and-answer statements.

5. The enterprise knowledge retrieval method based on a knowledge base according to claim 4, characterized in that, Based on the policy retrieval text in the query request, determine the first context retrieval information and the second context retrieval information corresponding to the query request, specifically including: Based on the retrieved text of the aforementioned system and the preset business dimension text library, keywords for each business area are determined to construct knowledge base tag filtering conditions, and the knowledge base tag filtering conditions are used as the first context retrieval information; wherein, the business area keywords include at least one or more of the following classification dimensions: finance, human resources, and operations; According to the system, the text is retrieved and a preset micro-attribute feature text library is used to determine the corresponding retrieval micro-attribute features and generate knowledge block filtering conditions, so as to use the knowledge block filtering conditions as the second context retrieval information; the preset micro-attribute feature text library includes at least the following dimension feature attributes: timeliness, permission level, and regional attribute.

6. The enterprise knowledge retrieval method based on a knowledge base according to claim 5, characterized in that, The first contextual retrieval information includes one or more of the following filtering conditions: logical combination of business domain keywords based on AND and OR operations, equality comparison, and inclusion comparison; The second contextual retrieval information includes one or more of the following filtering conditions: logical combination of retrieval micro-attribute features based on AND and OR operations, numerical type comparison, string type comparison, and time type comparison.

7. The enterprise knowledge retrieval method based on a knowledge base according to claim 6, characterized in that, Based on the first contextual retrieval information and the second contextual retrieval information, the retrieval knowledge base and corresponding retrieval knowledge blocks are sequentially filtered from the dual knowledge retrieval dictionary, specifically including: The retrieval knowledge base is obtained by filtering from the dual knowledge retrieval dictionary based on the first context retrieval information; In the obtained retrieval knowledge base, knowledge blocks are filtered using the second context retrieval information to determine each retrieval knowledge block based on the filtering results.

8. The enterprise knowledge retrieval method based on a knowledge base according to claim 1, characterized in that, Matching each of the aforementioned knowledge blocks with the system retrieval text to determine the target knowledge block corresponding to the query request based on the matching results, specifically including: The similarity between the knowledge block vector corresponding to each retrieval knowledge block and the vector corresponding to the system retrieval text is calculated using a vector matching algorithm. The vector similarity is compared with a first preset threshold. If the vector similarity is greater than the first preset threshold, the corresponding retrieval knowledge block is determined as the target knowledge block.

9. A knowledge base-based enterprise knowledge retrieval system, characterized in that, The system includes: The first determining module is used to respond to a user's query request for a policy-related question and answer, determine the user permission information corresponding to the query request, and generate a corresponding dual knowledge retrieval dictionary based on the user permission information. The second determining module is used to determine the first contextual retrieval information and the second contextual retrieval information corresponding to the query request based on the system retrieval text in the query request. The filtering module is used to sequentially filter the retrieval knowledge base and corresponding retrieval knowledge blocks from the dual knowledge retrieval dictionary based on the first context retrieval information, the second context retrieval information and preset filtering rules. The matching module is used to match each of the search knowledge blocks with the system search text, so as to determine the target knowledge block corresponding to the query request based on the first matching result, and send the target knowledge block to the user terminal for display on the user interface.

10. A knowledge base-based enterprise knowledge retrieval device, characterized in that, The device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform a knowledge base-based enterprise knowledge retrieval method as described in any one of claims 1-8.

Images