Method and system for knowledge management based on intelligent information retrieval

Abstract

The application discloses a knowledge management method and system based on information intelligent retrieval, relates to the field of data processing, and comprises the following steps: importing historical input modal data of a permission user within a current period for retrieval; adopting a modal recognition algorithm to count the frequency of various types of modal data in the historical input modal data, and determining the dominant input modal of the permission user; constructing a knowledge conversion rule model according to the retrieval expression of the dominant input modal; connecting an initial knowledge base of an information intelligent retrieval system, adaptively converting the knowledge content in the initial knowledge base by using the knowledge conversion rule model, outputting a converted knowledge base, and searching the real-time input modal of the permission user in the converted knowledge base to return a search result. The application solves the technical problems of low search efficiency and insufficient matching degree of the existing knowledge management based on information intelligent retrieval, and achieves the technical effect of improving the search matching degree and efficiency.

Description

Technical Field

[0001] This application relates to the field of data processing, and in particular to knowledge management methods and systems based on intelligent information retrieval. Background Technology

[0002] In the era of information overload, efficiently and accurately meeting users' personalized knowledge retrieval needs has become a key issue in improving the practicality and user satisfaction of knowledge management systems. Currently, the main approach to solving this problem is to build a unified knowledge base and design a universal retrieval interface to adapt to different user input modalities. However, because current methods do not fully consider the individual differences and dynamic changes in user input modalities, the retrieval results have a low degree of matching with users' actual needs, especially in multimodal interaction scenarios where retrieval efficiency and accuracy are insufficient.

[0003] At present, knowledge management based on intelligent information retrieval suffers from technical problems such as low retrieval efficiency and insufficient matching degree. Summary of the Invention

[0004] This application provides a knowledge management method and system based on intelligent information retrieval. By collecting and analyzing historical input modality data of authorized users within a specific time period, identifying their dominant input modality characteristics, constructing a knowledge conversion rule model adapted to this modality, and reconstructing the initial knowledge base in a targeted manner, the application achieves accurate retrieval in the converted knowledge base that highly matches the user's real-time input modality. This solves the technical problems of low retrieval efficiency and insufficient matching degree in existing knowledge management based on intelligent information retrieval, and achieves the technical effect of improving retrieval matching degree and efficiency.

[0005] This application provides a knowledge management method based on intelligent information retrieval, comprising: importing historical input modality data used by authorized users for retrieval within the current period; using a modality recognition algorithm to statistically analyze the frequency of various modalities in the historical input modality data to determine the dominant input modality of the authorized users; constructing a knowledge conversion rule model according to the retrieval expression of the dominant input modality; connecting to the initial knowledge base of the intelligent information retrieval system, using the knowledge conversion rule model to adaptively convert the knowledge content in the initial knowledge base, outputting a converted knowledge base, retrieving the real-time input modality of the authorized users in the converted knowledge base, and returning the retrieval results.

[0006] In a possible implementation, the following processing is performed: A modality recognition algorithm is used to count the frequency of each modality in the historical input modality data, wherein the modality categories include at least text-based dialogue modality, image modality, and file modality; the occurrence count of each input modality in the historical input modality data is counted according to the modality category; a weighted frequency factor is defined, and the occurrence count of each input modality is calculated according to the weighted frequency factor, outputting a modality frequency vector for each input modality; based on the modality frequency vector of each input modality, the dominant input modality of the authorized user is determined.

[0007] In a possible implementation, the retrieval expression of the dominant input modality is extracted, and the following processing is performed: multiple representative retrieval samples of the historical input of the dominant input modality are obtained; the expression content structure of the multiple representative retrieval samples is analyzed, wherein the expression content structure includes at least one corresponding retrieval expression element; a first mapping rule is established based on the expression content structure, and the retrieval expression of the dominant input modality is output, wherein the first mapping rule is a mapping rule between the modal data input by the authorized user under the dominant input modality and the knowledge expression used for retrieval.

[0008] In a possible implementation, a knowledge conversion rule model is constructed according to the retrieval expression of the dominant input modality, and the following processing is performed: a second mapping rule is constructed, which is a mapping rule between the retrieval expression of the dominant input modality and the knowledge content expression of the knowledge base; the first mapping rule and the second mapping rule are combined using the IF-THEN rule set to construct the knowledge conversion rule model.

[0009] In a possible implementation, after constructing a knowledge conversion rule model by combining the first mapping rule and the second mapping rule using the IF-THEN rule set, the following processing is performed: collecting the retrieval result feedback data of the authorized user; scoring the knowledge conversion rule model according to the retrieval result feedback data to obtain the model scoring result, including retrieval efficiency score and retrieval accuracy score; when the model scoring result is less than a preset scoring threshold, adjusting the parameter weights of the first mapping rule and the second mapping rule, and reconstructing the knowledge conversion rule model.

[0010] In a possible implementation, the knowledge content in the initial knowledge base is adaptively transformed using the knowledge transformation rule model to output a transformed knowledge base, and the following processing is performed: parsing the retrieval expression of the dominant input modality in the second mapping rule of the knowledge transformation rule model; adaptively transforming the knowledge content in the initial knowledge base according to the retrieval expression of the dominant input modality in the second mapping rule to output the transformed knowledge base; and establishing a modality adaptation index for the transformed knowledge base to output the transformed knowledge base.

[0011] In a possible implementation, based on the modality frequency vector of each input modality, the dominant input modality of the authorized user is determined, and the following processing is also performed: determining whether the dominant input modality of the authorized user is a unimodal input mode; if the dominant input modality of the authorized user is a unimodal input mode, determining the dominant input modality of the authorized user, and constructing a knowledge conversion rule model according to the retrieval expression of the dominant input modality.

[0012] In a possible implementation, the following processing is performed: if the dominant input modality of the authorized user is a multimodal input mode, determine multiple dominant input modalities of the authorized user, wherein the modal frequency vectors of the multiple dominant input modalities are all greater than a preset frequency vector threshold; construct multiple knowledge conversion rule models according to multiple retrieval expressions of the multiple dominant input modalities; based on the real-time input modality of the authorized user, switch to the corresponding knowledge conversion rule model among the multiple knowledge conversion rule models for retrieval, and return the retrieval results.

[0013] In a possible implementation, the following processing is performed: the search result feedback data includes search dwell time, search result click frequency, search content download rate, and search rollback times.

[0014] This application also provides a knowledge management system based on intelligent information retrieval, comprising: a historical input modality data import module, used to import historical input modality data used by authorized users for retrieval within the current period; a dominant input modality determination module, used to use a modality recognition algorithm to count the frequency of various modalities in the historical input modality data and determine the dominant input modality of the authorized user; a knowledge conversion rule model construction module, used to construct a knowledge conversion rule model according to the retrieval expression of the dominant input modality; and a conversion retrieval module, used to connect to the initial knowledge base of the intelligent information retrieval system, use the knowledge conversion rule model to adaptively convert the knowledge content in the initial knowledge base, output a converted knowledge base, retrieve the real-time input modality of the authorized user in the converted knowledge base, and return the retrieval results.

[0015] The proposed knowledge management method and system based on intelligent information retrieval first imports historical input modality data used by authorized users for retrieval within the current timeframe. Then, a modality recognition algorithm is used to statistically analyze the frequency of various modalities in the historical input modality data to determine the dominant input modality of the authorized user. Next, a knowledge conversion rule model is constructed according to the retrieval expression of the dominant input modality. Finally, this model is connected to the initial knowledge base of the intelligent information retrieval system. The knowledge content in the initial knowledge base is adaptively converted using the knowledge conversion rule model, resulting in a converted knowledge base. The real-time input modality of the authorized user is then retrieved from this converted knowledge base, and the retrieval results are returned. This achieves the technical effect of improving retrieval matching accuracy and efficiency. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings of the embodiments of the present invention will be briefly described below. Flowcharts are used in this application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from these processes.

[0017] Figure 1 This is a flowchart illustrating the knowledge management method based on intelligent information retrieval provided in an embodiment of this application.

[0018] Figure 2 This is a schematic diagram of the structure of a knowledge management system based on intelligent information retrieval provided in an embodiment of this application.

[0019] Figure labeling: Historical input modality data import module 10, dominant input modality determination module 20, knowledge conversion rule model construction module 30, conversion retrieval module 40. Detailed Implementation

[0020] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application.

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description of this application will be provided in conjunction with the accompanying drawings. The described embodiments should not be considered as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] In the following description, references to "some embodiments" describe a subset of all possible embodiments. However, it is understood that "some embodiments" can be the same or different subsets of all possible embodiments and can be combined with each other without conflict. The terms "first" and "second" are used merely to distinguish similar objects and do not represent a specific ordering of objects. The terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules not explicitly listed or inherent to these processes, methods, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only.

[0023] This application provides a knowledge management method based on intelligent information retrieval, such as... Figure 1 As shown, the method includes:

[0024] Step S100: Import the historical input modality data used by the authorized user for retrieval within the current period.

[0025] Specifically, a unique identifier for the authorized user is determined, such as the user ID assigned by the system upon login. The current time range is set, for example, in days, weeks, or months, specifying the start and end times. Using the query functionality of a relational database management system (such as MySQL or Oracle) or a non-relational database (such as MongoDB), a database query statement or data interface request parameter is constructed, using the user ID and time range as key conditions. The query operation is executed to retrieve the corresponding historical input modality data from a dedicated database table storing user retrieval history. If the historical input modality data is stored in other systems or services, a request with user authorization verification information and time range parameters is sent through a predefined data interface (such as a RESTful API) to obtain the required data. This historical input modality data includes the user-inputted text content, characteristic parameters of the speech signal (such as speech frequency and rate), and image feature vectors used in image retrieval.

[0026] For example, in an enterprise's internal knowledge management system, when a user searches for knowledge within the system, their search behavior data is stored in the "user_search_history" table in a MySQL database. The table structure is shown in Table 1. To import the historical input modality data used by a user (assuming user ID is 123) for searches within the past month, the SQL query is: `SELECT * FROM user_search_history WHERE user_id = 123 AND search_time BETWEENDATE_SUB(NOW(), INTERVAL 1 MONTH) AND NOW()`. After executing this statement, the system retrieves all search records for the user within that time range, including input data from different modalities.

[0027] Table 1: Example of a database storing user retrieval history data

[0028] field name Data types illustrate user_id INT User unique identifier search_time DATETIME Search time input_modality VARCHAR Input modality type, such as "text", "voice", or "image". input_data TEXT Input modal data: text content is stored directly, while speech and image data store the corresponding feature parameters.

[0029] Step S200: Use a modality recognition algorithm to count the frequency of each modality in the historical input modality data to determine the dominant input modality of the authorized user.

[0030] Specifically, the imported historical input modal data is preprocessed to ensure accuracy and consistency. For example, text data can be cleaned (stop words, punctuation, etc.); speech data can be denoised; and image data can be normalized. The preprocessed data is then categorized according to different modal types. For example, text data, speech data, and image data are categorized separately. Data statistics tools are used to count the frequency of each modality. For example, the frequency of text modality, speech modality, and image modality is counted. The frequencies of different modalities are compared, and the modality with the highest frequency is identified as the dominant input modality.

[0031] In one possible implementation, step S200 further includes step S210, which uses a modality recognition algorithm to statistically analyze the frequency of various modalities in the historical input modality data. The modality categories include at least text-based dialogue, image-based, and file-based modalities. Specifically, the modality recognition algorithm is an algorithmic framework capable of handling multiple modalities (such as text-based dialogue, images, and files), automatically identifying the modality category of the input data based on its characteristics. This algorithm can be a multi-classification model or a system containing multiple sub-modules, each specializing in one modality, but all sub-modules working together to complete the overall modality recognition task. The historical input modality data is preprocessed to ensure consistent data format. For example, for text data, word segmentation and cleaning are performed; for image data, size normalization is performed; for file data, file header information and extensions are extracted. The modality recognition algorithm is used to classify each piece of historical input modality data, labeling it as a text-based dialogue modality, image modality, or file modality.

[0032] Step S220: Count the occurrence frequency of each input modality in the historical input modality data according to the modality category. Specifically, based on the modality recognition results, the data is classified into the corresponding modality category, and a data processing library (such as Pandas) is used to count the occurrence frequency of each modality category.

[0033] Step S230: Define a weighted frequency factor, calculate the frequency of occurrence of each input modality according to the weighted frequency factor, and output the modality frequency vector of each input modality. Specifically, according to business requirements, assign a weight to each modality category to reflect the importance of different modalities in retrieval. For example, the weight of text dialogue modality is 0.5, the weight of image modality is 0.3, and the weight of file modality is 0.2. Multiply the frequency of each modality by the corresponding weight to obtain the weighted frequency, and combine the weighted frequencies into a vector as the modality frequency vector.

[0034] Step S240: Determine the dominant input modality of the authorized user based on the modality frequency vector of each input modality. Specifically, a frequency vector threshold is set according to business requirements. For example, the threshold is 0.4. Each component of the modality frequency vector is compared with the frequency vector threshold, and the modality greater than the frequency vector threshold is selected as the dominant input modality. This implementation method, through weighted frequency factors and frequency vector threshold discrimination, can more accurately identify the user's dominant input modality and improve the personalized service capability of the retrieval system.

[0035] In one possible implementation, the dominant input modality of the authorized user is determined based on the modality frequency vector of each input modality. Step S240 further includes step S241, determining whether the dominant input modality of the authorized user is a unimodal input mode; step S242, if the dominant input modality of the authorized user is a unimodal input mode, determining the dominant input modality of the authorized user, and constructing a knowledge conversion rule model according to the retrieval expression of the dominant input modality.

[0036] Specifically, if only one modality exceeds the frequency vector threshold, it is classified as a unimodal input method; if multiple modalities exceed the frequency vector threshold, it is classified as a multimodal input method. If the dominant input modality of a user with access privileges is a unimodal input method, a retrieval expression suitable for that modality is constructed based on the characteristics of the dominant modality. Based on the retrieval expression and the business rules of the knowledge management system, a knowledge transformation rule model is constructed. For example, if the dominant modality is text dialogue, natural language processing technology is used to extract keywords and semantic information from the text, construct a retrieval expression, and transform the format of the corresponding category of knowledge content in the knowledge base according to the rules to make it more in line with the user's retrieval habits.

[0037] In one possible implementation, step S240 further includes step S243: if the dominant input modality of the authorized user is a multimodal input mode, determine multiple dominant input modalities of the authorized user, wherein the modal frequency vectors of the multiple dominant input modalities are all greater than a preset frequency vector threshold. Specifically, if the weighted frequency of multiple modalities is greater than the frequency vector threshold, then these modalities are all dominant input modalities.

[0038] Step S244: Construct multiple knowledge conversion rule models according to the multiple retrieval expressions of the multiple dominant input modalities. Specifically, for each dominant input modality, a retrieval expression is constructed based on its features. For example, for the text dialogue modality, natural language processing techniques are used to extract keywords and semantic information from the text; for the image modality, image features are extracted. Based on the retrieval expression and combined with the business rules of the knowledge management system, a knowledge conversion rule model is constructed. For example, if the retrieval expression contains specific keywords, the knowledge content of the corresponding category in the knowledge base is formatted according to the rules to better suit user retrieval habits.

[0039] Step S245: Based on the real-time input modality of the authorized user, switch to the corresponding knowledge conversion rule model among the multiple knowledge conversion rule models for retrieval, and return the retrieval result. Specifically, a modality recognition algorithm is used to identify the user's real-time input modality; based on the real-time input modality, the corresponding knowledge conversion rule model is selected; the selected knowledge conversion rule model is used to retrieve the knowledge base; and the retrieval result is returned.

[0040] By determining whether the input method is unimodal, the system can flexibly adapt to the input habits of different users and provide appropriate search services for both unimodal and multimodal users.

[0041] Step S300: Construct a knowledge conversion rule model according to the retrieval expression of the dominant input modality.

[0042] Specifically, a retrieval expression is a query statement or condition used to retrieve information from a knowledge base. It is constructed based on the user's input modality and retrieval needs, accurately expressing the characteristics and requirements of the knowledge content the user wants to retrieve. The knowledge transformation rule model is a model built based on the retrieval expression of the dominant input modality and the business rules of the knowledge management system. It guides the transformation of knowledge content in the initial knowledge base into a format suitable for retrieval and display. It contains a series of rule logics that specify the transformation methods for the knowledge's format, structure, etc.

[0043] Based on the determined dominant input modality, appropriate techniques are selected to extract key information and construct retrieval expressions. For example, if the dominant input modality is text, NLP techniques such as text parsing and semantic understanding are used to construct retrieval expressions. For instance, techniques like part-of-speech tagging and dependency parsing are used to extract key information from the text to form retrieval expressions. If the dominant input modality is speech, speech recognition technology is first used to convert speech into text, and then the aforementioned NLP techniques are used to construct retrieval expressions. If the dominant input modality is image, image recognition technology is used to extract image features and construct image feature-based retrieval expressions. For instance, the Locality Sensitive Hash (LSH) algorithm is used to process image feature vectors to form expressions usable for retrieval.

[0044] Based on the constructed search expression and combined with the business rules of the knowledge management system, such as knowledge classification and relationships, a rule engine (such as Drools) is used to build a knowledge transformation rule model. The rule engine can match the search expression with the knowledge content in the knowledge base and formulate transformation rules according to predefined rule logic. For example, if the search expression contains specific keywords, the corresponding category of knowledge content in the knowledge base will be formatted according to the rules to better suit user search habits.

[0045] In one possible implementation, the retrieval expression of the dominant input modality is extracted, and step S300 further includes step S310, obtaining multiple representative retrieval samples from the historical inputs of the dominant input modality. Specifically, all input records of the dominant input modality are queried from the historical input data, and multiple representative retrieval samples are selected according to certain criteria (such as frequency, diversity, etc.).

[0046] Step S320: Analyze the expression content structure of the multiple representative retrieval samples, whereby the expression content structure includes at least one corresponding retrieval expression element. Specifically, preprocess the selected representative retrieval samples, such as text cleaning and image normalization. Analyze the expression content structure of each representative retrieval sample and extract the corresponding retrieval expression element. For example, for the text modality, NLP techniques are used to analyze the text structure and extract elements such as keywords and phrases; for the image modality, image processing techniques are used to extract image features such as color and texture; for the file modality, the file content is analyzed to extract key information.

[0047] Step S330: Establish a first mapping rule based on the expression content structure, and output the retrieval expression for the dominant input modality. The first mapping rule is the mapping rule between the modal data input by the authorized user under the dominant input modality and the knowledge representation used for retrieval. Specifically, the rule engine defines the first mapping rule between input modal data and knowledge representation based on the analyzed expression content structure. According to the first mapping rule, the input modal data is converted into a retrieval expression.

[0048] For example, assuming the dominant input modality is text-based dialogue, the following representative search samples are selected: "2023 Q3 New Energy Vehicle Sales Ranking", "Battery Range Parameters of Car A", and "Comparison of Intelligent Driving Functions of Car X and Car Y". NLP processing is performed on each representative search sample, and the search expression elements are extracted as shown in Table 2.

[0049] Table 2: Examples of the content structure of analytical expressions

[0050] Search Samples Word segmentation results Keyword extraction Semantic role analysis "Top 3 New Energy Vehicle Sales Rankings for Q3 2023" [2023 Q3 New Energy Vehicle Sales Ranking] New energy vehicle sales ranking (numerical indicator) (sorting requirement) Objective: To obtain new energy vehicle data sorted by sales volume; Condition: Time = Q3 2023 Battery range parameters for car A [Car A's Battery Range Specifications] Battery range (technical parameters) for Car A (product name) Objective: To query the technical parameters of a specific product. "Comparing the intelligent driving functions of Car X and Car Y" [Comparison of "Car X" and "Car Y" "Intelligent Driving" Functions] Car X (compare entity) Car Y (compare entity) Intelligent Driving (functional module) Objective: Comparison of features across multiple products

[0051] The expression's content structure can be summarized as follows: 1. Entity elements: Product name (Car A), technical parameters (battery range), comparison subject (Car X vs. Car Y); 2. Condition elements: Time range (Q3 2023), sorting requirements (sales ranking); 3. Operation elements: Comparison, parameter query, sorting.

[0052] Based on this, the first mapping rule can be: When the system analyzes the user's input search text: if the text contains comparison keywords such as "compare" or "vs," it is automatically identified as a comparison query, and the product names and specific items to be compared (such as battery life, price, etc.) are extracted. If the text contains keywords such as "parameters" or "specifications," it is automatically identified as a parameter query, and the specific product names and technical parameters to be searched are extracted. For other ordinary queries, the core nouns in the query are extracted as keywords, and possible time range requirements and sorting requirements (such as "newest," "best-selling," etc.) are automatically identified.

[0053] This implementation method, by analyzing the expression structure of representative search samples, can accurately extract search expressions, thus improving search accuracy. After establishing the first mapping rule, the system can quickly convert user input modal data into search expressions, improving search efficiency.

[0054] In one possible implementation, a knowledge transformation rule model is constructed according to the retrieval expression of the dominant input modality. Step S300 further includes step S340, constructing a second mapping rule. The second mapping rule is a mapping rule between the retrieval expression of the dominant input modality and the knowledge content expression of the knowledge base. Specifically, the core of the second mapping rule is to establish an adaptation relationship between the retrieval expression and the original content of the knowledge base, which includes the following three-dimensional transformations: The first dimension is knowledge presentation modality matching, that is, determining the best output form of the knowledge base content based on the dominant input modality (e.g., text-dominated user preference for question-and-answer fragments, image-dominated user need for visualization charts). The second dimension is knowledge metadata enhancement, that is, dynamically inserting structured tags (e.g., timestamps, keyword highlighting), summary generation, and other auxiliary information for the features of the retrieval expression. The third dimension is knowledge preprocessing optimization, that is, preprocessing the knowledge content as needed (e.g., text vectorization, image feature extraction) to adapt to subsequent retrieval efficiency.

[0055] Step S350: The first mapping rule and the second mapping rule are combined using the IF-THEN rule set to construct a knowledge conversion rule model. Specifically, the first mapping rule (user input → retrieval expression) and the second mapping rule (retrieval expression → knowledge conversion) are linked together and combined into a complete and executable knowledge conversion rule model through the "IF-THEN" condition judgment logic.

[0056] For example, assuming the dominant input modality is text, and "comparison" type queries frequently appear in historical inputs, the first mapping rule could be: IF input contains "comparison M and N" → THEN generate the search expression [COMPARE(M,N)]. The second mapping rule could be: IF search expression is [COMPARE(M,N)] → THEN extract the parameters of M and N from the knowledge base and generate a comparison table.

[0057] This implementation method, by constructing a second mapping rule and combining the first mapping rule, can accurately convert the user's search expression into a knowledge content expression in the knowledge base, thereby improving the efficiency and accuracy of knowledge retrieval and reducing retrieval time.

[0058] In one possible implementation, after constructing a knowledge conversion rule model by combining the first mapping rule and the second mapping rule using an IF-THEN rule set, step S300 further includes step S360, collecting the retrieval result feedback data from the authorized user. Specifically, a feedback mechanism is designed in the user interface to allow authorized users to evaluate the retrieval results, such as satisfaction ratings and relevance evaluations. User feedback data on the retrieval results is collected through the user interface and a logging system. The logging system records the authorized user's actions related to the retrieval results, such as clicks and dwell time.

[0059] Step S370: The knowledge conversion rule model is scored according to the search result feedback data to obtain model scoring results, including a search efficiency score and a search accuracy score. Specifically, the calculation methods for the search efficiency score and the search accuracy score are defined. For example, the search efficiency score can be based on search time, and the search accuracy score can be based on user satisfaction and relevance evaluation. For example, the search efficiency score can be defined as 100 minus 10 times the search time, and the search accuracy score can be defined as the user satisfaction score multiplied by 20. The scoring result of the knowledge conversion rule model is calculated based on the collected feedback data.

[0060] Step S380: When the model score is less than a preset score threshold, adjust the parameter weights of the first mapping rule and the second mapping rule, and reconstruct the knowledge conversion rule model. Specifically, a preset score threshold is set according to business needs. The model score is compared with the preset score threshold to determine whether the model needs adjustment. If the model score is less than the preset score threshold, a parameter adjustment algorithm (such as gradient descent, genetic algorithm, etc.) is used to adjust the parameter weights of the first and second mapping rules. Based on the adjusted parameter weights, the knowledge conversion rule model is reconstructed. This implementation method, by collecting user feedback data and adjusting the model according to the score results, can continuously optimize the knowledge conversion rule model and improve retrieval efficiency and accuracy.

[0061] In one possible implementation, step S360 further includes: the search result feedback data includes search dwell time, search result click frequency, search content download rate, and search rollback times.

[0062] Specifically, the behavior of authorized users on the search results page is monitored through the user interface and backend logging system. This includes search dwell time, search result click frequency, search content download rate, and search rollback frequency. Search dwell time refers to the length of time authorized users spend on the search results page; a longer dwell time indicates user interest, while a shorter dwell time may indicate the user did not find the desired information. The dwell time is calculated by recording the time authorized users enter and leave the search results page through frontend code. Search result click frequency refers to the number of times authorized users click on the search results page, indicating their level of interest. A high click frequency indicates that authorized users are searching for information among multiple results, potentially suggesting the search results are not precise enough. Click events by authorized users on the search results page are recorded through frontend code, and the number of clicks is counted. Search content download rate refers to the proportion of search content downloaded by authorized users; a higher download rate indicates higher user satisfaction with the search content, while a lower download rate may indicate user dissatisfaction with the search results. Download behavior by authorized users is recorded through backend logs, and the ratio of downloads to total searches is calculated. The number of times a user returns to the search input page from the search results page indicates their level of dissatisfaction with the search results. A higher number of returns suggests that the user has adjusted their search criteria multiple times, potentially indicating that the search results are not precise enough. This method records the user's actions of returning to the search input page through front-end code and counts the number of returns. By collecting various behavioral data from authorized users on the search results page, this approach can comprehensively evaluate user satisfaction and usage of the search results. Combining multiple user behavior data to calculate a model score allows for a more accurate evaluation of the performance of the knowledge conversion rule model.

[0063] Step S400: Connect to the initial knowledge base of the intelligent information retrieval system, use the knowledge conversion rule model to adaptively convert the knowledge content in the initial knowledge base, output a converted knowledge base, and retrieve the real-time input modality of the authorized user in the converted knowledge base, returning the retrieval results.

[0064] Specifically, the initial knowledge base refers to the storage structure such as a database or file system that stores the original knowledge content of the system. It contains knowledge that has not been transformed or optimized and is the original data source of the knowledge management system. The transformed knowledge base is the knowledge base that has been processed by the knowledge transformation rule model. It stores the transformed knowledge content, and the format and structure of this knowledge content are more suitable for search engine retrieval operations, which can improve retrieval efficiency and accuracy.

[0065] The system connects to the initial knowledge base of the intelligent information retrieval system using database connectivity technologies (such as JDBC and ODBC) or through a predefined knowledge base service interface (such as RESTful API) to retrieve knowledge content. Based on the constructed knowledge transformation rule model, data transformation tools (such as ETL tools, such as Talend) or data manipulation libraries in programming languages ​​(such as Pandas in Python) are used to transform the knowledge content in the initial knowledge base. The transformation process adjusts the format and structure of the knowledge according to the rule logic in the knowledge transformation rule model to adapt it to retrieval requirements. For example, knowledge may be converted from one format (such as long text) to another format (such as structured tables) according to rules, or it may be categorized or tagged. The transformed knowledge is stored in a new knowledge base (the transformed knowledge base), and an index is built using a retrieval engine (such as Elasticsearch or Solr). The system also receives real-time input modal data from authorized users and converts this data into retrieval expressions according to the retrieval expression construction method corresponding to the dominant input modality. The search engine performs a search operation in the transformed knowledge base, and returns the search results to authorized users based on rules such as relevance ranking.

[0066] In one possible implementation, the knowledge content in the initial knowledge base is adaptively transformed using the knowledge transformation rule model to output a transformed knowledge base. Step S400 further includes step S410, parsing the retrieval expression of the dominant input modality in the second mapping rule of the knowledge transformation rule model. Specifically, the constructed knowledge transformation rule model is loaded, the second mapping rule in the model is parsed using a rule parser, the retrieval expression of the dominant input modality is extracted, the structure of the retrieval expression is analyzed, and key information such as keywords, phrases, and semantic information is extracted.

[0067] Step S420: Adaptively transform the knowledge content in the initial knowledge base according to the retrieval expression of the dominant input modality described in the second mapping rule, and output the transformed knowledge base. Specifically, load the knowledge content in the initial knowledge base, and according to the retrieval expression in the second mapping rule, use a content transformation tool (such as a text processing tool, image processing tool, etc.) to adaptively transform the knowledge content in the initial knowledge base, and store the transformed knowledge content in the new knowledge base.

[0068] Step S430 involves building a modality-adaptive index for the transformed knowledge base to output the transformed knowledge base. Specifically, the knowledge content in the transformed knowledge base is loaded, and an indexing tool (such as Elasticsearch, Solr, etc.) is used to build a modality-adaptive index based on the characteristics of the dominant input modality. The indexed knowledge base is then output for subsequent retrieval. This implementation, by parsing and applying the second mapping rule in the knowledge transformation rule model, accurately converts the knowledge content in the initial knowledge base into a retrieval-suitable format, improving the accuracy of knowledge retrieval. By building a modality-adaptive index, retrieval efficiency and accuracy are ensured, improving the overall performance of the system.

[0069] This application's embodiments solve the technical problems of low retrieval efficiency and insufficient matching degree in existing knowledge management based on information intelligent retrieval, by collecting and analyzing the historical input modality data of authorized users within a specific time period, identifying their dominant input modality features, constructing a knowledge conversion rule model adapted to the modality, reconstructing the initial knowledge base in a targeted manner, and finally achieving accurate retrieval in the converted knowledge base that highly matches the user's real-time input modality.

[0070] In the above text, refer to Figure 1 A knowledge management method based on intelligent information retrieval according to embodiments of the present invention is described in detail. Next, reference will be made to... Figure 2 This invention describes a knowledge management system based on intelligent information retrieval according to an embodiment of the present invention.

[0071] The knowledge management system based on intelligent information retrieval according to embodiments of the present invention addresses the technical problems of low retrieval efficiency and insufficient matching degree in existing knowledge management systems based on intelligent information retrieval, thereby achieving the technical effect of improving retrieval matching degree and efficiency. The knowledge management system based on intelligent information retrieval includes: a historical input modality data import module 10, a dominant input modality determination module 20, a knowledge conversion rule model construction module 30, and a conversion retrieval module 40.

[0072] The historical input modality data import module 10 is used to import historical input modality data used by authorized users for retrieval within the current period; the dominant input modality determination module 20 is used to use a modality recognition algorithm to count the frequency of various modalities in the historical input modality data and determine the dominant input modality of the authorized user; the knowledge conversion rule model construction module 30 is used to construct a knowledge conversion rule model according to the retrieval expression of the dominant input modality; the conversion retrieval module 40 is used to connect to the initial knowledge base of the information intelligent retrieval system, use the knowledge conversion rule model to adaptively convert the knowledge content in the initial knowledge base, output a converted knowledge base, retrieve the real-time input modality of the authorized user in the converted knowledge base, and return the retrieval results.

[0073] The specific configuration of the dominant input modality determination module 20 will be described in detail below. As mentioned above, the dominant input modality determination module 20 may further include: a frequency statistics unit for using a modality recognition algorithm to count the frequency of various modalities in the historical input modality data, wherein the modality categories include at least text dialogue modality, image modality, and file modality, and the number of occurrences of each input modality in the historical input modality data is counted according to the modality category; a modality frequency vector output unit for defining a weighted frequency factor, calculating the number of occurrences of each input modality according to the weighted frequency factor, and outputting the modality frequency vector of each input modality; and a dominant input modality determination unit for determining the dominant input modality of the authorized user based on the modality frequency vector of each input modality.

[0074] The specific configuration of the knowledge conversion rule model construction module 30 will be described in detail below. As mentioned above, to extract the retrieval expression of the dominant input modality, the knowledge conversion rule model construction module 30 may further include: a representative retrieval sample acquisition unit for acquiring multiple representative retrieval samples from the historical input of the dominant input modality; an expression content structure analysis unit for analyzing the expression content structure of the multiple representative retrieval samples, wherein the expression content structure includes at least one corresponding retrieval expression element; and a first mapping rule establishment unit for establishing a first mapping rule based on the expression content structure and outputting the retrieval expression of the dominant input modality, wherein the first mapping rule is a mapping rule between the modal data input by the authorized user under the dominant input modality and the knowledge expression used for retrieval.

[0075] The knowledge conversion rule model construction module 30, which constructs a knowledge conversion rule model according to the retrieval expression of the dominant input modality, may further include: a second mapping rule construction unit for constructing a second mapping rule, wherein the second mapping rule is a mapping rule between the retrieval expression of the dominant input modality and the knowledge content expression of the knowledge base; and a knowledge conversion rule model construction unit for combining the first mapping rule and the second mapping rule using the IF-THEN rule set to construct a knowledge conversion rule model.

[0076] The knowledge conversion rule model construction module 30, after combining the first mapping rule and the second mapping rule using the IF-THEN rule set to construct the knowledge conversion rule model, may further include: a retrieval result feedback data collection unit for collecting retrieval result feedback data from the authorized user; a scoring unit for scoring the knowledge conversion rule model according to the retrieval result feedback data and obtaining model scoring results, including retrieval efficiency score and retrieval accuracy score; and a parameter weight adjustment unit for adjusting the parameter weights of the first mapping rule and the second mapping rule and reconstructing the knowledge conversion rule model when the model scoring result is less than a preset scoring threshold.

[0077] The specific configuration of the conversion retrieval module 40 will be described in detail below. As mentioned above, the knowledge content in the initial knowledge base is adaptively converted using the knowledge conversion rule model to output a converted knowledge base. The conversion retrieval module 40 may further include: a retrieval expression parsing unit for parsing the retrieval expression of the dominant input modality in the second mapping rule of the knowledge conversion rule model; an adaptive conversion unit for adaptively converting the knowledge content in the initial knowledge base according to the retrieval expression of the dominant input modality in the second mapping rule to output the converted knowledge base; and a modality adaptation index building unit for building a modality adaptation index on the converted knowledge base to output the converted knowledge base.

[0078] Specifically, based on the modality frequency vector of each input modality, the dominant input modality of the authorized user is determined. The dominant input modality determination unit may further include: a judgment subunit for judging whether the dominant input modality of the authorized user is a unimodal input mode; and a knowledge conversion rule model construction subunit for determining the dominant input modality of the authorized user if the dominant input modality of the authorized user is a unimodal input mode, and constructing a knowledge conversion rule model according to the retrieval expression of the dominant input modality.

[0079] The dominant input modality determination unit may further include: multiple dominant input modality determination subunits for determining multiple dominant input modalities of the authorized user if the dominant input modality of the authorized user is a multimodal input mode, wherein the modality frequency vectors of the multiple dominant input modalities are all greater than a preset frequency vector threshold; multiple knowledge conversion rule model construction subunits for constructing multiple knowledge conversion rule models according to multiple retrieval expressions of the multiple dominant input modalities; and a switching retrieval subunit for switching to the corresponding knowledge conversion rule model among the multiple knowledge conversion rule models for retrieval based on the real-time input modality of the authorized user, and returning the retrieval results.

[0080] The search result feedback data collection unit may further include: the search result feedback data includes search dwell time, search result click frequency, search content download rate, and search rollback times.

[0081] The knowledge management system based on intelligent information retrieval provided in the embodiments of the present invention can execute the knowledge management method based on intelligent information retrieval provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0082] Although this application makes various references to certain modules in the system according to the embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of this invention.

[0083] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application. In some cases, the actions or steps described in this application can be performed in a different order than that shown in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A knowledge management method based on intelligent retrieval of information, characterized in that, The method comprises: Importing historical input modal data of a permission user for retrieval within a current period; Using a modal recognition algorithm to count the frequency of each type of modal in the historical input modal data, and determining the dominant input modal of the permission user; Building a knowledge conversion rule model according to the retrieval expression of the dominant input modal; Connecting an initial knowledge base of an intelligent retrieval system, using the knowledge conversion rule model to adaptively convert the knowledge content in the initial knowledge base, outputting a converted knowledge base, and retrieving the real-time input modal of the permission user in the converted knowledge base to return the retrieval result; The method for building a knowledge conversion rule model according to the retrieval expression of the dominant input modal comprises: Building a second mapping rule, which is a mapping rule between the retrieval expression of the dominant input modal and the expression of the knowledge content in the knowledge base; Combining the first mapping rule and the second mapping rule using an IF-THEN rule set to build a knowledge conversion rule model, wherein the first mapping rule is a mapping rule between the modal data input by the permission user under the dominant input modal and the knowledge expression used for retrieval; The method for collecting retrieval result feedback data of the permission user comprises: Scoring the knowledge conversion rule model according to the retrieval result feedback data to obtain a model score result, including a retrieval efficiency score and a retrieval accuracy score; When the model score result is less than a preset score threshold, adjusting the parameter weight of the first mapping rule and the second mapping rule to rebuild the knowledge conversion rule model.

2. The knowledge management method based on intelligent information retrieval according to claim 1, wherein, The method for using a modal recognition algorithm to count the frequency of each type of modal in the historical input modal data comprises: Counting the number of occurrences of each input modal in the historical input modal data according to the modal categories; Defining a weighted frequency factor to calculate the number of occurrences of each input modal according to the weighted frequency factor, and outputting a modal frequency vector of each input modal; Determining the dominant input modal of the permission user according to the modal frequency vector of each input modal.

3. The knowledge management method based on intelligent information retrieval according to claim 1, wherein, The method for extracting the retrieval expression of the dominant input modal comprises: Obtaining multiple representative retrieval samples of the historical input of the dominant input modal; Analyzing the expression content structure of the multiple representative retrieval samples, which includes at least one corresponding retrieval expression element; Building a first mapping rule according to the expression content structure to output the retrieval expression of the dominant input modal.

4. The knowledge management method based on intelligent information retrieval according to claim 1, wherein, The method for using the knowledge conversion rule model to adaptively convert the knowledge content in the initial knowledge base to output a converted knowledge base comprises: Parsing the retrieval expression of the dominant input modal in the second mapping rule in the knowledge conversion rule model; Adaptively converting the knowledge content in the initial knowledge base according to the retrieval expression of the dominant input modal in the second mapping rule to output the converted knowledge base; Building a modal adaptation index for the converted knowledge base to output the converted knowledge base.

5. The knowledge management method based on intelligent information retrieval according to claim 2, wherein, According to a modal frequency vector of each input modality, a dominant input modality of the authorized user is determined, and the method further comprises: judging whether the dominant input modality of the authorized user is a single-modality input mode; if the dominant input modality of the authorized user is a single-modality input mode, determining the dominant input modality of the authorized user, and constructing a knowledge conversion rule model according to a retrieval expression of the dominant input modality.

6. The knowledge management method based on intelligent information retrieval according to claim 5, wherein, if the dominant input modality of the authorized user is a multi-modality input mode, determining a plurality of dominant input modalities of the authorized user, wherein modal frequency vectors of the plurality of dominant input modalities are all greater than a preset frequency vector threshold; constructing a plurality of knowledge conversion rule models according to a plurality of retrieval expressions of the plurality of dominant input modalities; switching to a corresponding knowledge conversion rule model for retrieval according to a real-time input modality of the authorized user among the plurality of knowledge conversion rule models, and returning a retrieval result.

7. The knowledge management method based on intelligent information retrieval according to claim 1, wherein, The retrieval result feedback data comprises a retrieval stay duration, a retrieval result click frequency, a retrieval content download rate, and a retrieval backtracking number.

8. A knowledge management system based on intelligent retrieval of information, characterized in that, The system is used to implement the knowledge management method based on information intelligent retrieval according to any one of claims 1-7, and the system comprises: a historical input modality data import module, used for importing historical input modality data of an authorized user used for retrieval within a current period; a dominant input modality determination module, used for determining a dominant input modality of the authorized user by using a modal recognition algorithm to count frequencies of various modalities in the historical input modality data; a knowledge conversion rule model construction module, used for constructing a knowledge conversion rule model according to a retrieval expression of the dominant input modality; a conversion retrieval module, used for connecting an initial knowledge base of an information intelligent retrieval system, adaptively converting knowledge content in the initial knowledge base by using the knowledge conversion rule model, outputting a converted knowledge base, retrieving a real-time input modality of the authorized user in the converted knowledge base, and returning a retrieval result.

Images