A management method of a multi-modal enterprise knowledge base system

Abstract

The application provides a management method of a multimodal enterprise knowledge base system, comprising: integrating multimodal data in an enterprise, constructing a private knowledge base, and combining the private knowledge base with a public knowledge base; performing intention analysis on multimodal retrieval requirements submitted by a user based on a combination result, performing conditional retrieval on the private knowledge base and the public knowledge base based on the intention analysis result, and outputting multimodal query data; performing quality inspection on the conditional retrieval process, and performing dynamic optimization on the combination result based on the quality inspection result to obtain a final multimodal enterprise knowledge base, and performing deployment management on the multimodal enterprise knowledge base. The management effect of multimodal knowledge in the enterprise is greatly improved, and the retrieval accuracy of the knowledge is also improved.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a management method for a multimodal enterprise knowledge base system. Background Art

[0002] With the development of enterprises, a large amount of knowledge resources have been accumulated, including data in multiple modalities such as text, images, audio, and video. Due to the rapid rise of the DeepSeek model, in order to respond to the market and improve product competitiveness, it is necessary to apply DeepSeek to enable call centers to create new scenarios and new value.

[0003] However, traditional enterprise knowledge bases often focus on data management and knowledge mining in a single modality (such as text), and struggle to effectively integrate and utilize multimodal information. For example, internal training videos contain lecturers' explanations (audio), demonstration images (video), and accompanying handouts (text). Existing knowledge base systems struggle to link these different modalities for comprehensive knowledge representation and retrieval. Furthermore, during the retrieval process, users' query requirements cannot be accurately analyzed, resulting in a mismatch between retrieval results and query requirements, significantly reducing the effectiveness of enterprise knowledge management.

[0004] Therefore, in order to overcome the above-mentioned defects, the present invention provides a management method for a multimodal enterprise knowledge base system. Summary of the Invention

[0005] The present invention provides a management method for a multimodal enterprise knowledge base system, which is used to integrate the multimodal data within the enterprise, build a private knowledge base, and combine the private knowledge base with the public knowledge base, so as to facilitate the accurate and effective determination of the user's retrieval needs, and then facilitate the accurate retrieval of knowledge in the knowledge base. Secondly, the multimodal retrieval needs submitted by the user are analyzed for intent, and effective knowledge traversal is achieved based on the intent analysis results, ensuring that the final retrieval results meet the user's retrieval needs and ensure the effective integration and utilization of the enterprise's multimodal information. Finally, the retrieval process is quality inspected to optimize the combined results of the private knowledge base and the public knowledge base based on the results, further ensuring the operational reliability of the multimodal enterprise knowledge base. At the same time, the multimodal enterprise knowledge base is deployed and managed, which facilitates the multimodal enterprise knowledge base to be applied in multiple scenarios, greatly improving the management effect of multimodal knowledge within the enterprise, and also improving the accuracy of knowledge retrieval.

[0006] The present invention provides a management method for a multimodal enterprise knowledge base system, comprising:

[0007] Step 1: Integrate the enterprise's internal multimodal data, build a private knowledge base, and combine the private knowledge base with the public knowledge base;

[0008] Step 2: Based on the joint results, the user's submitted multimodal search requirements are analyzed for intent, and based on the intent analysis results, conditional searches are performed on private and public knowledge bases to output multimodal query data.

[0009] Step 3: Perform quality inspection on the conditional retrieval process, and dynamically optimize the joint results based on the quality inspection results to obtain the final multimodal enterprise knowledge base, and deploy and manage the multimodal enterprise knowledge base.

[0010] Preferably, a method for managing a multimodal enterprise knowledge base system includes, in step 1, integrating the multimodal data within the enterprise to construct a private knowledge base, including:

[0011] Acquire multimodal data within the enterprise and classify the multimodal data into business categories to obtain multimodal data sets under different business categories;

[0012] Build databases for different business categories and configure multimodal data interfaces for the databases;

[0013] Based on the configured multimodal data interface, the multimodal data sets under different business categories are uploaded to the corresponding database, and based on the uploaded results, the multimodal data sets under each business category are configured and managed to obtain the corresponding private knowledge base.

[0014] Preferably, a management method of a multimodal enterprise knowledge base system performs configuration management on a multimodal data set under each business category based on the uploaded results, including:

[0015] Traverse the multimodal data set in each database based on the uploaded results, and determine the data composition and the number of knowledge base files in each database based on the traversal results;

[0016] Generate a knowledge base description for each database based on the data structure, and generate a unique knowledge base label and knowledge base name for each database based on the business category;

[0017] Based on the upload results of the multimodal data set, the update time information of each knowledge base is determined, and the number of knowledge base files, knowledge base description, knowledge base label, knowledge base name and update time information are summarized to obtain the knowledge base card of each database;

[0018] The knowledge base card is bound to the corresponding database, and based on the binding result, the knowledge base card of each database is displayed on the knowledge base management page to obtain the corresponding private knowledge base.

[0019] In this embodiment, the private knowledge base includes a sensitive word library, a business question and answer library, and a polite language library.

[0020] Preferably, a method for managing a multimodal enterprise knowledge base system includes, in step 1, integrating the multimodal data within the enterprise to construct a private knowledge base, including:

[0021] Receive updated data uploaded from within the enterprise in real time, analyze the updated data, and determine the target private knowledge base and knowledge validity period corresponding to the updated data;

[0022] Automatically store updated data into the corresponding target private knowledge base. At the same time, set automatic reminders and cleanup notifications for updated data based on the knowledge validity period.

[0023] When the automatic reminder cleanup notification is started, a secondary confirmation pop-up window will pop up on the knowledge base management page, and after receiving the cleanup instruction, the data cleanup operation will be executed to complete the dynamic update of the private knowledge base.

[0024] Preferably, a method for managing a multimodal enterprise knowledge base system, in step 1, combining the private knowledge base and the public knowledge base, includes:

[0025] Performing a first association between the private knowledge base and the knowledge base of the public knowledge base, and obtaining a business service target of the enterprise knowledge base according to the management terminal based on the first association result;

[0026] Based on the business service goals, the work engine of the private knowledge base is trained on the private training platform using the company's proprietary corpus, and the first business processing flow of the private knowledge base is obtained based on the training results;

[0027] At the same time, the business execution categories of the private knowledge base and the public knowledge base are obtained based on the management terminal, and the business collaboration logic of the private knowledge base and the public knowledge base is determined based on the business execution categories;

[0028] A second business processing flow of the public knowledge base is obtained, and based on the business collaboration logic, a second association is performed between the business processing flow of the private knowledge base and the public knowledge base, and the private knowledge base and the public knowledge base are combined based on the second association result.

[0029] Preferably, a method for managing a multimodal enterprise knowledge base system, step 2: performing intent analysis on a multimodal search requirement submitted by a user based on the joint result, and performing conditional search on a private knowledge base and a public knowledge base based on the intent analysis result, and outputting multimodal query data, including:

[0030] Receive multimodal search requirements submitted by users, parse the submitted multimodal search requirements, and determine the current search data category;

[0031] The preset algorithm is called based on the current search data category to translate the multi-modal search requirement into text, and the text is semantically analyzed to obtain the target semantics corresponding to the text;

[0032] The text is filtered based on the target semantics, and search keywords are obtained based on the multi-condition filtering result;

[0033] The search dimension is determined based on the search keywords, and the target query intention of the user is determined in combination with the target semantics of the text, and the query type submitted by the user is determined based on the target query intention, wherein the query type includes professional knowledge query and comprehensive query;

[0034] When the query type is professional knowledge query, the enterprise internal knowledge in the private knowledge base is traversed based on the search keywords, and the matching degree between the enterprise internal knowledge and the search keywords is determined based on the traversal result;

[0035] The target knowledge is sorted in descending order of relevance based on the value of the matching degree, and the multi-modal query data corresponding to the multi-modal search requirement is obtained based on the sorting result, and the multi-modal query data is fed back to the user terminal;

[0036] The feedback notification of the user terminal is received in real time based on the feedback result, and when the user makes a satisfactory feedback, the professional knowledge query is completed, and when the user makes an unsatisfactory feedback, the comprehensive query is linked;

[0037] When the query type is comprehensive query, the linkage query request sent is received in real time, and the search engine of the public knowledge base is started based on the received result;

[0038] The target semantics and the search keywords are analyzed in multiple rounds based on the starting result, and the request intention of the user is determined based on the multi-round intention analysis result;

[0039] The query strategy is generated based on the request intention, and the private knowledge base is conditionally searched based on the query strategy to obtain the target knowledge corresponding to the linkage query request or the multi-modal query data corresponding to the comprehensive query;

[0040] The obtained multi-modal query data is fed back to the user terminal.

[0041] Preferably, a management method of a multi-modal enterprise knowledge base system, the obtained multi-modal query data is fed back to the user terminal, comprising:

[0042] The obtained multi-modal query data is converted into speech based on a preset algorithm to obtain response speech;

[0043] The response speech is broadcasted to the user through the user terminal based on a media server.

[0044] Preferably, a management method of a multi-modal enterprise knowledge base system, in step 3, the conditional retrieval process is quality inspected, and the joint result is dynamically optimized based on the quality inspection result to obtain a final multi-modal enterprise knowledge base, comprising:

[0045] The conditional retrieval process is monitored, and the call recording between the user and the multi-modal enterprise knowledge base system is offline parsed or real-time parsed according to the monitoring requirement to obtain the corresponding dialogue record text, wherein the monitoring requirement includes offline quality inspection and real-time quality inspection;

[0046] The corresponding target private knowledge base is locked based on the dialogue record text, and the dialogue record text is globally traversed based on the business attribute of the target private knowledge base to obtain a scoring index in the conditional retrieval process;

[0047] The scoring index is scored based on the scoring rule to obtain a quality inspection score corresponding to the conditional retrieval process, and a quality inspection report is generated based on the quality inspection score and the scoring index;

[0048] At the same time, the dialogue record text is keyword and sentence extracted based on the work engine in the public knowledge base, and the automatic filling of the work order is performed according to the keyword and sentence extraction result;

[0049] The target defect existing in the joint result of the private knowledge base and the public knowledge base is determined based on the quality inspection report and the automatically filled work order, and the target defect is dynamically optimized to obtain the final multi-modal enterprise knowledge base.

[0050] Preferably, a management method of a multi-modal enterprise knowledge base system, the target defect is dynamically optimized to obtain a final multi-modal enterprise knowledge base, comprising:

[0051] Obtain historical dialogue record text, and perform data enhancement expansion training set on the historical dialogue record text;

[0052] Determine the target defect existing in the joint result of the private knowledge base and the public knowledge base based on the quality inspection report and the automatically filled work order, and construct a multi-dimensional virtual user query scene;

[0053] Perform user query training based on the training set and the multi-dimensional virtual user query scene, and determine the output accuracy and recall rate of the conditional retrieval process based on the training result;

[0054] Determine the training score and optimization features based on the output accuracy and recall rate;

[0055] The target defect is dynamically optimized based on the training score and the optimization features to obtain the final multi-modal enterprise knowledge base.

[0056] Preferably, in a method for managing a multimodal enterprise knowledge base system, in step 3, deploying and managing the multimodal enterprise knowledge base includes:

[0057] Acquire the obtained multimodal enterprise knowledge base and configure multi-dimensional interfaces for the multimodal enterprise knowledge base;

[0058] At the same time, based on the multi-dimensional interface configuration results, a microservice architecture with hybrid local and cloud deployment is constructed for the multimodal enterprise knowledge base, and standardized interfaces are added to the constructed microservice architecture;

[0059] Based on the addition results, the permission parameter adaptive strategy configuration and emergency response adaptive strategy configuration are performed on the multimodal enterprise knowledge base to complete the deployment management of the multimodal enterprise knowledge base.

[0060] Compared with the prior art, the present invention has the following beneficial effects:

[0061] By integrating the multimodal data within the enterprise, building a private knowledge base, and combining the private knowledge base with the public knowledge base, it is convenient to accurately and effectively determine the user's search needs, and then facilitate accurate retrieval of knowledge in the knowledge base. Secondly, the multimodal retrieval needs submitted by the user are analyzed for intent, and effective knowledge traversal is achieved based on the intent analysis results, ensuring that the final retrieval results meet the user's retrieval needs and ensure the effective integration and utilization of the enterprise's multimodal information. Finally, the retrieval process is quality inspected to optimize the combined results of the private knowledge base and the public knowledge base based on the results, further ensuring the operational reliability of the multimodal enterprise knowledge base. At the same time, the multimodal enterprise knowledge base is deployed and managed to facilitate the application of the multimodal enterprise knowledge base in multiple scenarios, greatly improving the management effect of multimodal knowledge within the enterprise and improving the accuracy of knowledge retrieval.

[0062] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures specifically pointed out in this application document.

[0063] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0065] Figure 1 This is a flow chart of a method for managing a multimodal enterprise knowledge base system in an embodiment of the present invention;

[0066] Figure 2 This is a schematic diagram of offline quality inspection in a management method for a multimodal enterprise knowledge base system according to an embodiment of the present invention;

[0067] Figure 3 This is a schematic diagram of the principle of real-time quality inspection in a management method of a multimodal enterprise knowledge base system in an embodiment of the present invention. DETAILED DESCRIPTION

[0068] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0069] Example 1:

[0070] This embodiment provides a management method for a multimodal enterprise knowledge base system. Figure 1 As shown, including:

[0071] Step 1: Integrate the enterprise's internal multimodal data, build a private knowledge base, and combine the private knowledge base with the public knowledge base;

[0072] Step 2: Based on the joint results, the user's submitted multimodal search requirements are analyzed for intent, and based on the intent analysis results, conditional searches are performed on private and public knowledge bases to output multimodal query data.

[0073] Step 3: Perform quality inspection on the conditional retrieval process, and dynamically optimize the joint results based on the quality inspection results to obtain the final multimodal enterprise knowledge base, and deploy and manage the multimodal enterprise knowledge base.

[0074] In this embodiment, the multimodal data includes text data, image data, video data, etc.

[0075] In this embodiment, the private knowledge base refers to data information related to the internal business of the enterprise.

[0076] In this embodiment, the public knowledge base refers to data information that is accessible to all users, such as the DeepSeek large model.

[0077] In this embodiment, combining the private knowledge base and the public knowledge base means organically associating the private knowledge base and the public knowledge base. When the user needs a comprehensive query, the engine of the public knowledge base can be called first to perform a detailed analysis of the user's needs, generate a corresponding query strategy, and then further perform knowledge query on the private knowledge base according to the query strategy. When the user directly queries the private knowledge base, the private knowledge base can also be directly accessed for knowledge query. At the same time, when combining, the public knowledge base can be used to analyze the enterprise's private data (historical query data), give scores and suggestions, and then optimize the combination. After the combination, the content of the private knowledge base can be dynamically updated.

[0078] In this embodiment, the multimodal retrieval requirement refers to the content input by the user, which can be in the form of voice, text, or pictures.

[0079] In this embodiment, intent analysis refers to analyzing the user's search needs in order to determine whether the user directly queries the private knowledge base or performs a comprehensive query (which requires combining the public knowledge base with the private knowledge base).

[0080] In this embodiment, conditional retrieval refers to using a corresponding query mechanism according to the determined intention, that is, directly querying the private knowledge base or querying the private knowledge base in combination with the public knowledge base.

[0081] In this embodiment, the multimodal query data refers to a plurality of different types of data that meet the user's needs, obtained after searching the knowledge base according to the user's multimodal search needs.

[0082] In this embodiment, quality inspection refers to a quality inspection of the system's processing capability during the user's search process, in order to determine the system's operating status.

[0083] In this embodiment, deployment management refers to lightweight deployment of the resulting multimodal enterprise knowledge base and setting up API interface development.

[0084] The beneficial effects of the above technical solution are: by integrating the multimodal data within the enterprise, building a private knowledge base, and combining the private knowledge base with the public knowledge base, it is convenient to accurately and effectively determine the user's search needs, and then facilitate accurate retrieval of knowledge in the knowledge base; secondly, the multimodal retrieval needs submitted by the user are analyzed for intent, and effective knowledge traversal is achieved based on the intent analysis results, ensuring that the final retrieval results meet the user's retrieval needs and ensure the effective integration and utilization of the enterprise's multimodal information; finally, the retrieval process is quality inspected to optimize the combined results of the private knowledge base and the public knowledge base based on the results, further ensuring the operational reliability of the multimodal enterprise knowledge base; at the same time, the multimodal enterprise knowledge base is deployed and managed, which facilitates the application of the multimodal enterprise knowledge base in multiple scenarios, greatly improving the management effect of multimodal knowledge within the enterprise, and also improving the accuracy of knowledge retrieval.

[0085] Example 2:

[0086] Based on Example 1, this embodiment provides a method for managing a multimodal enterprise knowledge base system. In step 1, the multimodal data within the enterprise is integrated to build a private knowledge base, including:

[0087] Acquire multimodal data within the enterprise and classify the multimodal data into business categories to obtain multimodal data sets under different business categories;

[0088] Build databases for different business categories and configure multimodal data interfaces for the databases;

[0089] Based on the configured multimodal data interface, the multimodal data sets under different business categories are uploaded to the corresponding database, and based on the uploaded results, the multimodal data sets under each business category are configured and managed to obtain the corresponding private knowledge base.

[0090] In this embodiment, business category classification refers to classifying the obtained multimodal data into business types, for example, classification into polite expressions and professional terms.

[0091] In this embodiment, the database is a library for storing multimodal data sets corresponding to different business categories.

[0092] In this embodiment, multimodal data interface configuration refers to configuring different modal data interfaces on the constructed database, with the purpose of being able to upload different modal data through the interface.

[0093] The beneficial effects of the above technical solution are: by integrating the multimodal data within the enterprise and storing the integration results in the constructed database, at the same time, configuring the multimodal data interface of the database, a comprehensive and reliable construction of the private knowledge base is achieved, providing reliable data support for the management of the multimodal enterprise knowledge base system.

[0094] Example 3:

[0095] Based on Example 2, this embodiment provides a management method for a multimodal enterprise knowledge base system, which performs configuration management on a multimodal data set under each business category based on the uploaded results, including:

[0096] Traverse the multimodal data set in each database based on the uploaded results, and determine the data composition and the number of knowledge base files in each database based on the traversal results;

[0097] Generate a knowledge base description for each database based on the data structure, and generate a unique knowledge base label and knowledge base name for each database based on the business category;

[0098] Based on the upload results of the multimodal data set, the update time information of each knowledge base is determined, and the number of knowledge base files, knowledge base description, knowledge base label, knowledge base name and update time information are summarized to obtain the knowledge base card of each database;

[0099] The knowledge base card is bound to the corresponding database, and based on the binding result, the knowledge base card of each database is displayed on the knowledge base management page to obtain the corresponding private knowledge base.

[0100] In this embodiment, the private knowledge base includes a sensitive word library, a business question and answer library, and a polite language library.

[0101] In this embodiment, data composition refers to the data categories contained in each database.

[0102] In this embodiment, the knowledge base description is text information that explains the data categories contained in the database and the corresponding business types.

[0103] In this embodiment, the knowledge base tag refers to a description or label that can be used to mark the database for business purposes, for example, it can be a "polite language" tag.

[0104] In this embodiment, the knowledge base management page is set in advance and is used to display and view the knowledge base.

[0105] The beneficial effect of the above technical solution is: by traversing the multimodal data set in the database, the knowledge base description, knowledge base label, knowledge base name and update time information of each database are determined, and finally the knowledge base cards are generated and displayed, which is convenient for users and staff to quickly and effectively understand the situations of different private knowledge bases, and also convenient for quick reading and locking of the corresponding private knowledge base according to needs, thereby meeting the user's query needs.

[0106] Example 4:

[0107] Based on Example 1, this embodiment provides a method for managing a multimodal enterprise knowledge base system. In step 1, the multimodal data within the enterprise is integrated to build a private knowledge base, including:

[0108] Receive updated data uploaded from within the enterprise in real time, analyze the updated data, and determine the target private knowledge base and knowledge validity period corresponding to the updated data;

[0109] Automatically store updated data into the corresponding target private knowledge base. At the same time, set automatic reminders and cleanup notifications for updated data based on the knowledge validity period.

[0110] When the automatic reminder cleanup notification is started, a secondary confirmation pop-up window will pop up on the knowledge base management page, and after receiving the cleanup instruction, the data cleanup operation will be executed to complete the dynamic update of the private knowledge base.

[0111] In this embodiment, the target private knowledge base refers to the private knowledge base corresponding to the current update data.

[0112] In this embodiment, the knowledge validity period refers to the period during which the updated data is allowed to be used normally, for example, one month or one year.

[0113] The beneficial effects of the above technical solution are: through real-time structuring of updated data uploaded within the enterprise, and updating the updated data in the corresponding private knowledge base, at the same time, setting the knowledge validity period of the updated data, it is convenient to perform cleanup operations in time after the deadline is reached, and realize secondary confirmation during cleanup, thereby ensuring the security and reliability of the data while realizing dynamic updates of the private knowledge base.

[0114] Example 5:

[0115] Based on Example 1, this embodiment provides a method for managing a multimodal enterprise knowledge base system. In step 1, the private knowledge base and the public knowledge base are combined, including:

[0116] Performing a first association between the private knowledge base and the knowledge base of the public knowledge base, and obtaining a business service target of the enterprise knowledge base according to the management terminal based on the first association result;

[0117] Based on the business service goals, the work engine of the private knowledge base is trained on the private training platform using the company's proprietary corpus, and the first business processing flow of the private knowledge base is obtained based on the training results;

[0118] At the same time, the business execution categories of the private knowledge base and the public knowledge base are obtained based on the management terminal, and the business collaboration logic of the private knowledge base and the public knowledge base is determined based on the business execution categories;

[0119] A second business processing flow of the public knowledge base is obtained, and based on the business collaboration logic, a second association is performed between the business processing flow of the private knowledge base and the public knowledge base, and the private knowledge base and the public knowledge base are combined based on the second association result.

[0120] In this embodiment, the first association refers to associating the private knowledge base with the public knowledge base, that is, enabling collaborative access to knowledge in the private knowledge base and the public knowledge base.

[0121] In this embodiment, the business service objectives refer to the types of business that need to be executed through the enterprise knowledge base and the goals or standards that each type of business needs to achieve during operation.

[0122] In this embodiment, the private training platform is built in advance and is a platform for training and optimizing the knowledge base.

[0123] In this embodiment, the enterprise-specific corpus is known in advance, including information such as professional speech.

[0124] In this embodiment, the first business processing flow refers to the business processing flow corresponding to the private knowledge base.

[0125] In this embodiment, the service execution category refers to the service execution type corresponding to the private knowledge base and the public knowledge base respectively.

[0126] In this embodiment, the business collaboration logic refers to a collaborative processing solution between a private knowledge base and a public knowledge base when processing a business.

[0127] In this embodiment, the second business processing flow refers to the specific processing steps of the public knowledge base when processing the business.

[0128] In this embodiment, the second association refers to associating the business process flow of the private knowledge base with that of the public knowledge base.

[0129] The beneficial effect of the above technical solution is that by combining the private knowledge base with the public knowledge base, it is convenient to jointly solve the query requests submitted by the user according to the combined results, thereby ensuring the accuracy and reliability of the processing of the user query requests.

[0130] Example 6:

[0131] Based on Example 1, this embodiment provides a method for managing a multimodal enterprise knowledge base system. Step 2: Based on the joint results, the multimodal search requirements submitted by the user are analyzed for intent, and based on the intent analysis results, conditional searches are performed on the private knowledge base and the public knowledge base to output multimodal query data, including:

[0132] Receive multimodal search requirements submitted by users, parse the submitted multimodal search requirements, and determine the current search data category;

[0133] Based on the current search data category, the preset algorithm is called to translate the multimodal search requirements into text, and the text is semantically parsed to obtain the target semantics corresponding to the text;

[0134] Perform multi-condition filtering on the text based on the target semantics, and obtain search keywords based on the multi-condition filtering results;

[0135] Determine the search dimension based on the search keyword, and determine the user's target query intent based on the target semantics of the text, and determine the query type submitted by the user based on the target query intent, where the query type includes professional knowledge query and comprehensive query;

[0136] When the query type is professional knowledge query, the internal knowledge of the enterprise in the private knowledge base is traversed based on the search keywords, and the matching degree between the internal knowledge of the enterprise and the search keywords is determined based on the traversal results;

[0137] Based on the matching degree values, the target knowledge is sorted in descending order of relevance, and based on the sorting results, multimodal query data corresponding to the multimodal retrieval requirements is obtained, and the multimodal query data is fed back to the user terminal;

[0138] Receive feedback notifications from user terminals in real time based on feedback results, and complete professional knowledge query when users provide satisfactory feedback. At the same time, link with comprehensive query when users provide unsatisfactory feedback;

[0139] When the query type is a comprehensive query, the linked query request is received in real time, and the search engine of the public knowledge base is activated based on the received result;

[0140] Based on the launch results, multiple rounds of intent analysis are performed on the target semantics and search keywords, and the user's request intent is determined based on the results of multiple rounds of intent analysis;

[0141] Generate a query strategy based on the request intent, and perform conditional search on the private knowledge base based on the query strategy to obtain the target knowledge corresponding to the linkage query request or the multimodal query data corresponding to the comprehensive query;

[0142] The obtained multimodal query data is fed back to the user terminal.

[0143] In this embodiment, the retrieval data category refers to the data category corresponding to the multimodal retrieval requirement currently submitted by the user, for example, it can be any one of text, picture and voice.

[0144] In this embodiment, the preset algorithm is pre-trained and is used to perform text conversion on data in different forms.

[0145] In this embodiment, the target semantics refers to the specific meaning corresponding to the text.

[0146] In this embodiment, multi-condition filtering refers to filtering the data in the text according to the target semantics, that is, filtering out valid text data related to the retrieval.

[0147] In this embodiment, the search keyword refers to the final text data obtained after performing multi-condition filtering on the text.

[0148] In this embodiment, the search dimension refers to the business category corresponding to the search keyword, such as polite terms and professional terms.

[0149] In this embodiment, the target query intent refers to the data content that the user ultimately needs to query.

[0150] In this embodiment, the target knowledge refers to data that meets the user's search requirements and is obtained by traversing the internal knowledge of the enterprise in the private knowledge base.

[0151] In this embodiment, the linkage query request refers to linking the private knowledge base with the public knowledge base to execute the user's search request.

[0152] In this embodiment, generating a query strategy based on request intent refers to determining a strategy or specific method for querying data in a private knowledge base based on a determined user's request intent.

[0153] The beneficial effects of the above technical solution are: by parsing the multimodal retrieval requirements submitted by the user, the corresponding algorithm is called according to the retrieval data category to convert the multimodal retrieval requirements into text, and then the search keywords are determined through the text. Secondly, the user's query type is determined through the search keywords, and then the user's multimodal data query requirements are responded to according to the query type. Finally, the multimodal query data is determined and the multimodal query data is fed back to the user terminal for viewing, thereby improving the accuracy and reliability of the user's multimodal data query requirements.

[0154] Example 7:

[0155] Based on Example 6, this embodiment provides a method for managing a multimodal enterprise knowledge base system, which feeds back the obtained multimodal query data to a user terminal, including:

[0156] Performing voice conversion on the obtained multimodal query data based on a preset algorithm to obtain a response voice;

[0157] The media server will broadcast the response voice to the user through the user terminal.

[0158] In this embodiment, the preset algorithm is set in advance.

[0159] The beneficial effect of the above technical solution is: by converting the obtained multimodal query data into voice, the response voice is determined, and finally, the response voice is voice broadcast through the user terminal, so that the user can understand and view the query request in time.

[0160] Example 8:

[0161] Based on Example 1, this embodiment provides a management method for a multimodal enterprise knowledge base system. In step 3, the conditional search process is quality-checked, and the combined results are dynamically optimized based on the quality-check results to obtain the final multimodal enterprise knowledge base, including:

[0162] Monitor the conditional retrieval process and, based on monitoring requirements, perform offline or real-time analysis of recorded conversations between users and the multimodal enterprise knowledge base system to obtain the corresponding conversation transcripts. Monitoring requirements include offline and real-time quality checks.

[0163] Based on the conversation record text, the corresponding target private knowledge base is locked, and the conversation record text is globally traversed based on the business attributes of the target private knowledge base to obtain the scoring index in the conditional retrieval process;

[0164] Scoring the scoring indicators based on the scoring rules, obtaining the quality inspection score corresponding to the conditional retrieval process, and generating a quality inspection report based on the quality inspection score and scoring indicators;

[0165] At the same time, the work engine based on the public knowledge base extracts key words and sentences from the conversation record text, and automatically fills in the work order based on the key word and sentence extraction results;

[0166] Based on the quality inspection report and automatically filled work orders, the target defects in the joint results of the private knowledge base and the public knowledge base are determined, and the target defects are dynamically optimized to obtain the final multimodal enterprise knowledge base.

[0167] In this embodiment, the target private knowledge base refers to the private knowledge base corresponding to the current conversation record text.

[0168] In this embodiment, the business attribute refers to the business category corresponding to the target private knowledge base and the specific requirements at runtime, etc.

[0169] In this embodiment, the scoring indicator refers to the parameter that can be evaluated in the conditional retrieval process, i.e., the corresponding value, for example, the number of polite expressions.

[0170] In this embodiment, the target defect refers to the drawbacks existing in the joint result of the private knowledge base and the public knowledge base, i.e., the vulnerabilities that cause the abnormal working effect of the joint result.

[0171] In this embodiment, the scoring rules include: conversation duration score (70 points): less than 2 minutes (0-50 points), 2-8 minutes (50-60 points), 8-15 minutes (60-70 points), and more than 15 minutes (50-60 points);

[0172] Polite expression score (15 points): no polite expression (+0 point), use 1-3 polite expressions (+5 points), use 3-6 polite expressions (+10 points), and use more than 6 polite expressions (+15 points);

[0173] Professional degree of conversation score (15 points): conversation contains business words less than 10% (+0 points), conversation contains business words 11%-50% (+10 points), and conversation contains business words more than 50% (+15 points);

[0174] Sensitive word deduction (deduction): deduct 5 points for each sensitive word in the conversation, and the upper limit is not capped.

[0175] In this embodiment, the offline quality inspection schematic diagram is as shown in Figure 2 .

[0176] In this embodiment, the quality inspection score = conversation duration score + polite expression score + professional degree of conversation score - sensitive word deduction.

[0177] In this embodiment, real-time quality inspection detects sensitive words in the agent conversation in quality inspection, and gives a pop-up window warning. At the same time, the key words in the customer conversation are matched in the knowledge base, and the real-time reply content is recommended to the agent;

[0178] At the same time, click the call state as

in call

[0179] If the agent is detected to use sensitive words during the call, a prompt "Sensitive words detected: XX" will be popped up in the dialogue window. At the same time, the agent can give recommended responses by searching the knowledge base or intelligent answering based on the customer's questions. The real-time quality inspection principle diagram is as follows Figure 3 shown.

[0180] The beneficial effects of the above technical solution are: by performing quality inspection on the conditional retrieval process, generating quality inspection reports and work orders based on the quality inspection results, and determining the existing target defects based on the quality inspection reports and work orders, and then dynamically optimizing the target defects, and realizing the construction of the final multimodal enterprise knowledge base.

[0181] Example 9:

[0182] Based on Example 8, this embodiment provides a management method for a multimodal enterprise knowledge base system, which dynamically optimizes target defects to obtain a final multimodal enterprise knowledge base, including:

[0183] Obtain historical conversation record texts and perform data augmentation on the historical conversation record texts to expand the training set;

[0184] Based on quality inspection reports and automatically filled work orders, target defects in the joint results of private and public knowledge bases are identified, and multi-dimensional virtual user query scenarios are constructed;

[0185] Conduct user query training based on the training set and multi-dimensional virtual user query scenarios, and determine the output accuracy and recall rate of the conditional retrieval process based on the training results;

[0186] Determine training scores and optimize features based on output accuracy and recall;

[0187] The target defects are dynamically optimized based on the training scores and optimized features to obtain the final multimodal enterprise knowledge base.

[0188] In this embodiment, data enhancement includes back-translation and entity replacement.

[0189] In this embodiment, the multi-dimensional virtual user query scenarios include customer complaints and business consultations.

[0190] In this embodiment, the target defect refers to a weak link in the conditional retrieval process, such as insufficient professional knowledge.

[0191] In this embodiment, the optimization feature refers to a parameter or item that needs to be specifically optimized.

[0192] The beneficial effects of the above technical solution are: by determining the text of historical conversation records and expanding the text of historical conversation records through data enhancement, the training set can be effectively determined. Secondly, the target defects are trained based on the training set and the constructed multi-dimensional virtual user query scenario, and finally a multimodal enterprise knowledge base is constructed.

[0193] Example 10:

[0194] Based on Example 1, this embodiment provides a method for managing a multimodal enterprise knowledge base system. In step 3, the multimodal enterprise knowledge base is deployed and managed, including:

[0195] Acquire the obtained multimodal enterprise knowledge base and configure multi-dimensional interfaces for the multimodal enterprise knowledge base;

[0196] At the same time, based on the multi-dimensional interface configuration results, a microservice architecture with hybrid local and cloud deployment is constructed for the multimodal enterprise knowledge base, and standardized interfaces are added to the constructed microservice architecture;

[0197] Based on the addition results, the permission parameter adaptive strategy configuration and emergency response adaptive strategy configuration are performed on the multimodal enterprise knowledge base to complete the deployment management of the multimodal enterprise knowledge base.

[0198] In this embodiment, the multi-dimensional interface configuration includes:

[0199] Knowledge retrieval API: supports keywords, semantic similarity, and multimodal hybrid queries;

[0200] Ticket creation API: automatically populates customer information, issue classification, and priority tags;

[0201] Third-party system integration: Pre-installed integration templates for CRM, work order systems, and voice platforms, support for Webhook callbacks, and real-time synchronization of work order status and customer information.

[0202] In this embodiment, deploying and managing the multimodal enterprise knowledge base further includes:

[0203] (1) Performance requirements:

[0204] Concurrency capability: supports ≥1000 concurrent requests, interface response time ≤500ms;

[0205] Model inference latency: single request ≤ 300ms (in GPU acceleration environment);

[0206] (2) Security requirements:

[0207] Data transmission: full-link HTTPS + SM4 national secret algorithm encryption;

[0208] Permission control: RBAC hierarchical permissions (e.g., customer service can only view public knowledge, while administrators can edit core data), i.e., permission parameter adaptive policy configuration;

[0209] Audit log: records all data changes, API calls, and user login behaviors, with a retention period of ≥ 6 months;

[0210] Data desensitization: Customer sensitive information (such as mobile phone numbers and email addresses) is encrypted using AES when stored and partially masked when displayed (such as 138****1234);

[0211] Compliance requirements: Comply with GDPR, China's Personal Information Protection Law and other regulations, and provide data deletion interfaces and compliance audit reports;

[0212] (3) Reliability requirements:

[0213] Disaster recovery capability: supports remote multi-active deployment, with fault recovery time ≤ 15 minutes;

[0214] Data backup: Daily incremental backup + weekly full backup, supporting one-click recovery; i.e., emergency response adaptive policy configuration.

[0215] The beneficial effect of the above technical solution is that by deploying and managing the obtained multimodal enterprise knowledge base, the multimodal enterprise knowledge base is easily deployed in different enterprise systems, thereby facilitating the application of the multimodal enterprise knowledge base.

[0216] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A management method for a multimodal enterprise knowledge base system, characterized in that: include: Step 1: Integrate the enterprise's internal multimodal data, build a private knowledge base, and combine the private knowledge base with the public knowledge base; Step 2: Based on the joint results, the user's submitted multimodal search requirements are analyzed for intent, and based on the intent analysis results, conditional searches are performed on private and public knowledge bases to output multimodal query data. Step 3: Perform quality inspection on the conditional retrieval process and dynamically optimize the joint results based on the quality inspection results to obtain the final multimodal enterprise knowledge base, and deploy and manage the multimodal enterprise knowledge base; Among them, step 2: based on the joint results, perform intent analysis on the multimodal search requirements submitted by the user, and perform conditional search on the private knowledge base and the public knowledge base based on the intent analysis results, and output multimodal query data, including: Receive multimodal search requirements submitted by users, parse the submitted multimodal search requirements, and determine the current search data category; Based on the current search data category, the preset algorithm is called to translate the multimodal search requirements into text, and the text is semantically parsed to obtain the target semantics corresponding to the text; Perform multi-condition filtering on the text based on the target semantics, and obtain search keywords based on the multi-condition filtering results; Determine the search dimension based on the search keyword, and determine the user's target query intent based on the target semantics of the text, and determine the query type submitted by the user based on the target query intent, where the query type includes professional knowledge query and comprehensive query; When the query type is professional knowledge query, the internal knowledge of the enterprise in the private knowledge base is traversed based on the search keywords, and the matching degree between the internal knowledge of the enterprise and the search keywords is determined based on the traversal results; Based on the matching degree, the target knowledge is sorted in descending order of relevance, and based on the sorting result, multimodal query data corresponding to the multimodal retrieval requirement is obtained, and the multimodal query data is fed back to the user terminal; Receive feedback notifications from user terminals in real time based on feedback results, and complete professional knowledge query when users provide satisfactory feedback. At the same time, link with comprehensive query when users provide unsatisfactory feedback; When the query type is a comprehensive query, the linked query request is received in real time, and the search engine of the public knowledge base is activated based on the received result; Based on the launch results, multiple rounds of intent analysis are performed on the target semantics and search keywords, and the user's request intent is determined based on the results of multiple rounds of intent analysis; Generate a query strategy based on the request intent, and perform conditional search on the private knowledge base based on the query strategy to obtain the target knowledge corresponding to the linkage query request or the multimodal query data corresponding to the comprehensive query; The obtained multimodal query data is fed back to the user terminal.

2. A management method for a multimodal enterprise knowledge base system according to claim 1, characterized in that: In step 1, the enterprise's internal multimodal data is integrated to build a private knowledge base, including: Acquire multimodal data within the enterprise and classify the multimodal data into business categories to obtain multimodal data sets under different business categories; Build databases for different business categories and configure multimodal data interfaces for the databases; Based on the configured multimodal data interface, the multimodal data sets under different business categories are uploaded to the corresponding database, and based on the uploaded results, the multimodal data sets under each business category are configured and managed to obtain the corresponding private knowledge base.

3. A method for managing a multimodal enterprise knowledge base system according to claim 2, characterized in that: Based on the uploaded results, configure and manage the multimodal data sets under each business category, including: Traverse the multimodal data set in each database based on the uploaded results, and determine the data composition and the number of knowledge base files in each database based on the traversal results; Generate a knowledge base description for each database based on the data structure, and generate a unique knowledge base label and knowledge base name for each database based on the business category; Based on the upload results of the multimodal data set, the update time information of each knowledge base is determined, and the number of knowledge base files, knowledge base description, knowledge base label, knowledge base name and update time information are summarized to obtain the knowledge base card of each database; The knowledge base card is bound to the corresponding database, and based on the binding result, the knowledge base card of each database is displayed on the knowledge base management page to obtain the corresponding private knowledge base.

4. A management method for a multimodal enterprise knowledge base system according to claim 1, characterized in that: In step 1, the enterprise's internal multimodal data is integrated to build a private knowledge base, including: Receive updated data uploaded from within the enterprise in real time, analyze the updated data, and determine the target private knowledge base and knowledge validity period corresponding to the updated data; Automatically store updated data into the corresponding target private knowledge base. At the same time, set automatic reminders and cleanup notifications for updated data based on the knowledge validity period. When the automatic reminder cleanup notification is started, a secondary confirmation pop-up window will pop up on the knowledge base management page, and after receiving the cleanup instruction, the data cleanup operation will be executed to complete the dynamic update of the private knowledge base.

5. The method for managing a multimodal enterprise knowledge base system according to claim 1, characterized in that: In step 1, the private knowledge base and the public knowledge base are combined, including: Performing a first association between the private knowledge base and the knowledge base of the public knowledge base, and obtaining a business service target of the enterprise knowledge base according to the management terminal based on the first association result; Based on the business service goals, the work engine of the private knowledge base is trained on the private training platform using the company's proprietary corpus, and the first business processing flow of the private knowledge base is obtained based on the training results; At the same time, the business execution categories of the private knowledge base and the public knowledge base are obtained based on the management terminal, and the business collaboration logic of the private knowledge base and the public knowledge base is determined based on the business execution categories; A second business processing flow of the public knowledge base is obtained, and based on the business collaboration logic, a second association is performed between the business processing flow of the private knowledge base and the public knowledge base, and the private knowledge base and the public knowledge base are combined based on the second association result.

6. A management method for a multimodal enterprise knowledge base system according to claim 1, characterized in that: Feedback of the obtained multimodal query data to the user terminal includes: Performing voice conversion on the obtained multimodal query data based on a preset algorithm to obtain a response voice; The media server will broadcast the response voice to the user through the user terminal.

7. A method for managing a multimodal enterprise knowledge base system according to claim 1, characterized in that: In step 3, the conditional retrieval process is quality-checked, and the joint results are dynamically optimized based on the quality-check results to obtain the final multimodal enterprise knowledge base, including: Monitor the conditional retrieval process and, based on monitoring requirements, perform offline or real-time analysis of recorded conversations between users and the multimodal enterprise knowledge base system to obtain the corresponding conversation transcripts. Monitoring requirements include offline and real-time quality checks. Based on the conversation record text, the corresponding target private knowledge base is locked, and the conversation record text is globally traversed based on the business attributes of the target private knowledge base to obtain the scoring index in the conditional retrieval process; Scoring the scoring indicators based on the scoring rules, obtaining the quality inspection score corresponding to the conditional retrieval process, and generating a quality inspection report based on the quality inspection score and scoring indicators; At the same time, the work engine based on the public knowledge base extracts key words and sentences from the conversation record text, and automatically fills in the work order based on the key word and sentence extraction results; Based on the quality inspection report and automatically filled work orders, the target defects in the joint results of the private knowledge base and the public knowledge base are determined, and the target defects are dynamically optimized to obtain the final multimodal enterprise knowledge base.

8. A method for managing a multimodal enterprise knowledge base system according to claim 7, characterized in that: Dynamically optimize the target defects to obtain the final multimodal enterprise knowledge base, including: Obtain historical conversation record texts and perform data augmentation on the historical conversation record texts to expand the training set; Based on quality inspection reports and automatically filled work orders, target defects in the joint results of private and public knowledge bases are identified, and multi-dimensional virtual user query scenarios are constructed; Conduct user query training based on the training set and multi-dimensional virtual user query scenarios, and determine the output accuracy and recall rate of the conditional retrieval process based on the training results; Determine training scores and optimize features based on output accuracy and recall; The target defects are dynamically optimized based on the training scores and optimized features to obtain the final multimodal enterprise knowledge base.

9. The method for managing a multimodal enterprise knowledge base system according to claim 1, characterized in that: In step 3, the multimodal enterprise knowledge base is deployed and managed, including: Acquire the obtained multimodal enterprise knowledge base and configure multi-dimensional interfaces for the multimodal enterprise knowledge base; At the same time, based on the multi-dimensional interface configuration results, a microservice architecture with hybrid local and cloud deployment is constructed for the multimodal enterprise knowledge base, and standardized interfaces are added to the constructed microservice architecture; Based on the addition results, the permission parameter adaptive strategy configuration and emergency response adaptive strategy configuration are performed on the multimodal enterprise knowledge base to complete the deployment management of the multimodal enterprise knowledge base.

Images