An AI-based enterprise knowledge management and intelligent question answering system

Abstract

This invention relates to the field of artificial intelligence technology, and specifically to an AI-based enterprise knowledge management and intelligent question-answering system. The system includes a data acquisition module, an intent acquisition module, an intelligent question-answering module, an employee analysis module, and a learning planning module. The data acquisition module collects multi-source heterogeneous data from various office systems within the enterprise, performs entity recognition and relationship extraction on the heterogeneous data, and constructs an enterprise knowledge graph including enterprise business entities and their relationships. The intent acquisition module acquires natural language questions input by employees, performs semantic recognition on the natural language questions to extract semantic features, obtains the question content based on the semantic features, and determines the query path in the enterprise knowledge graph based on the question content. The intelligent question-answering module retrieves the corresponding answer from the enterprise knowledge graph based on the query path and pushes it to the employee. The employee analysis module records the interaction data of employees during the questioning process, including the semantic features of natural language questions, question frequency, and feedback behavior to the pushed answers, and assesses the employee's cognitive level of related business entities based on the interaction data. The learning planning module determines the knowledge gap between the employee's current cognitive level and the preset target level for the corresponding position, retrieves and plans the optimal learning path including several business nodes and learning materials in the enterprise knowledge base based on the knowledge gap, and pushes the optimal learning path to the employee.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and specifically to an AI-based enterprise knowledge management and intelligent question-answering system. Background Technology

[0002] In the course of daily business operations, various business systems generate a large and continuously growing amount of knowledge assets, including experience summaries accumulated during project execution, standardized business process documents, code libraries and design specifications developed through technical research and development, and work insights gained by employees in their practical work. These knowledge resources are of great value to the company's sustainable operation and innovative development, and how to effectively manage and utilize these knowledge assets has become an important issue in the company's information technology construction.

[0003] Currently, many companies have recognized the importance of knowledge management and have begun building internal knowledge base systems to centrally store and categorize documents scattered across various departments. Through the construction of a document management system, companies can digitally preserve knowledge assets such as project experience, technical documents, and process specifications, preventing the loss of critical knowledge due to employee turnover or job changes. At the same time, the knowledge base also provides a learning platform for new employees, enabling them to understand the company's business and culture by consulting documents, thus shortening their adaptation period. In their daily work, when employees encounter problems, they can use keyword searches to find relevant information and solutions in the knowledge base, reducing the time cost of cross-departmental communication.

[0004] However, existing enterprise knowledge management solutions still have significant limitations in practical applications. First, at the knowledge acquisition and organization level, existing solutions mainly rely on manual sorting and classification, requiring a significant investment of manpower for document annotation and maintenance. With business development, the total amount of knowledge grows rapidly, making it difficult for manual methods to keep up, resulting in lagging knowledge base updates and the inability to incorporate a large amount of newly generated project experience and process changes into the system in a timely manner. Simultaneously, heterogeneous data scattered across different systems lacks effective integration methods. Structured data such as database records, semi-structured data such as process logs, and unstructured data such as technical documents are isolated from each other, making it difficult to form a unified knowledge system.

[0005] Secondly, in terms of knowledge retrieval and acquisition, most existing systems rely on keyword matching for search. Employees need to extract precise keywords to find the information they need, which can be difficult for new employees unfamiliar with business terminology to accurately express their needs. When employees encounter complex problems, they may need to adjust keywords multiple times and search repeatedly, or even browse multiple documents to piece together a complete answer, resulting in low retrieval efficiency and failing to meet the need for immediate answers. Thirdly, regarding employee learning and development, existing knowledge bases lack consideration for individual employee differences. All new employees receive the same training materials, failing to provide differentiated learning content based on their job characteristics and existing knowledge base. When experienced employees use the knowledge base, the system cannot identify their proficiency level, resulting in content that does not match actual needs and wasting time. Furthermore, existing systems cannot perceive employees' knowledge mastery, making it difficult for managers to identify the team's overall knowledge gaps in a timely manner, and training resources lack targeting. Overall, existing enterprise knowledge management solutions are essentially passive document storage tools, lacking a deep understanding of knowledge and intelligent processing capabilities. They cannot achieve proactive knowledge services and personalized adaptation, limiting the full realization of the value of knowledge assets. Summary of the Invention

[0006] The technical problem solved by this invention is to provide an AI-based enterprise knowledge management and intelligent question-answering system that can automatically integrate multi-source enterprise knowledge, achieve accurate and intelligent question-answering, and dynamically assess employees' cognitive levels to recommend personalized learning paths.

[0007] The basic solution provided by this invention is an AI-based enterprise knowledge management and intelligent question-answering system, characterized by including a data acquisition module, an intent acquisition module, an intelligent question-answering module, an employee analysis module, and a learning planning module. The data acquisition module is used to collect multi-source heterogeneous data from various office systems within the enterprise, perform entity recognition and relationship extraction on the multi-source heterogeneous data, and construct an enterprise knowledge graph that includes enterprise business entities and their related relationships. The intent acquisition module is used to acquire natural language questions input by employees, perform semantic recognition on the natural language questions to extract semantic features, obtain the question content based on the semantic features, and determine the query path in the enterprise knowledge graph based on the question content. The intelligent question-and-answer module is used to retrieve the corresponding answer to the question from the enterprise knowledge graph based on the query path and push it to the employee. The employee analysis module is used to record the interaction data of employees during the questioning process. The interaction data includes the semantic features of natural language questions, the frequency of questions, and the feedback behavior to the pushed answers. The interaction data is used to assess the employee's cognitive level of the related business entities. The learning planning module is used to determine the knowledge gap between an employee's current cognitive level and the preset target level for the corresponding position. Based on the knowledge gap, it retrieves and plans the optimal learning path from the enterprise knowledge base, which includes several business nodes and learning materials, and pushes the optimal learning path to the employee.

[0008] The principle and advantages of this invention are as follows: The data acquisition module automatically captures heterogeneous data from various internal office systems, including project documents, process files, and technical materials. It then performs entity recognition and relationship extraction on this data, ultimately constructing a structured enterprise knowledge graph. The intent acquisition module processes natural language questions input by employees, extracts the core content of the questions through semantic recognition technology, and determines the query path in the knowledge graph accordingly. The intelligent question-answering module retrieves the corresponding answer from the knowledge graph according to this path and pushes it to the employee. Simultaneously, the employee analysis module records all employee interactions during the questioning process, including the questioning method, frequency, and feedback to answers. Based on this data, it assesses the employee's mastery of relevant business knowledge, forming a cognitive level. The learning planning module, based on the gap between the employee's current cognitive level and the target level required for their position, plans an optimal learning path in the knowledge base, including learning content and sequence, and pushes it to the employee. Take a newly hired software engineer as an example. When he asks how to conduct code reviews, the system can not only provide a standard process, but also analyze his understanding of the process. If it finds that he is not familiar with the use of review tools, the system will subsequently push relevant teaching materials for the tools.

[0009] Compared with existing technologies, the advantage of this system is that it is no longer a simple document storage and keyword search, but rather constructs a dynamic knowledge network that can understand the semantics of questions and provide differentiated answers and learning suggestions based on employees' individual cognitive levels. It truly realizes proactive knowledge services and personalized employee training, avoiding the problems of traditional one-size-fits-all training and difficulty in knowledge retrieval.

[0010] Meanwhile, identifying employees' proficiency in business knowledge by analyzing their natural language questions is a core component of the system's personalized learning and precise knowledge services. When employees encounter problems in their daily work, they ask the system questions in natural language, and these questions themselves carry a wealth of information reflecting their cognitive level. The intent acquisition module performs semantic recognition on the questions, extracting the business entities involved and the complexity of the problems. The employee analysis module then further records and analyzes the semantic characteristics of the questions, such as the number of technical terms used, the sentence structure of the questions, the depth of the knowledge graph the questions point to, the frequency of questions, and the feedback behavior to the pushed answers. This data collectively constitutes the basis for assessing employee proficiency. For example, a newly hired finance employee might initially ask, "What is the reimbursement process?" This type of question involves few technical terms, has a simple sentence structure, and points to the surface level of the knowledge graph, leading the system to initially determine that the employee is a novice. As their work progresses, when they begin asking, "How are budget items adjusted when reimbursing across projects?" the questions contain more technical terms, have a more complex sentence structure, and point to deeper rule nodes in the knowledge graph, allowing the system to recognize that their proficiency in that process has significantly improved. This dynamic recognition method based on questioning behavior allows the system to grasp the knowledge status of employees in real time without relying on additional testing or human evaluation.

[0011] Furthermore, the data acquisition module includes a preprocessing module, an entity recognition module, a relationship recognition module, and a knowledge graph module; The preprocessing module is used to clean the format and parse the content of data from different sources, including extracting table fields from structured data, identifying document level tags from semi-structured data, and performing natural language processing on unstructured text data. The entity recognition module is used to identify the enterprise's business entities from the preprocessed data. The business entities include project entities, employee entities, process entities, document entities, and professional and technical entities, and each business entity is assigned a type label and a unique identifier. The relationship identification module is used to identify business relationships between entities and link business entities with related relationships. The knowledge graph module is used to obtain an enterprise knowledge graph based on the business relationships between business entities, with business entities as nodes and business relationships as edges.

[0012] The preprocessing module first cleans and parses the data from different sources, such as extracting fields from tables exported from project management systems, identifying chapter titles from workflow documents in office automation systems, and performing natural language processing such as word segmentation and stop word removal on the main text of technical documents to ensure that subsequent steps process clean and standardized data. The entity recognition module is responsible for identifying the company's core business elements from the preprocessed data, such as specific project names, employee names, approval processes, technical manuals, and technical terms like cloud computing, and assigning a type label and unique identifier to each identified element. The relationship recognition module further analyzes the business relationships between these entities, such as which projects an employee is responsible for, which approval nodes a process includes, and which technical terms a document references, and links related entities together. Finally, the knowledge graph module uses these entities as nodes and the relationships between them as edges to organize all the information into a huge network structure—the enterprise knowledge graph. Compared to the simple file classification and storage methods in existing technologies, this solution transforms scattered information into an interconnected and logically clear knowledge network through refined preprocessing, entity recognition, and relation extraction. This lays a solid data foundation for subsequent intelligent question answering and in-depth analysis, enabling knowledge retrieval to no longer rely on filename or keyword matching, but rather on the inherent connections between entities.

[0013] Furthermore, the intent acquisition module includes a natural language processing module, an intent classification module, an entity linking module, and a query path generation module; The natural language processing module is used to perform word segmentation, part-of-speech tagging, and syntactic analysis on natural language queries input by employees, and generate structured language feature vectors. The intent classification module is used to identify the intent category of the question based on the language feature vector using a pre-trained intent classification model. The intent categories include process query, personnel query, document query, and terminology explanation. The entity linking module is used to identify the mentioned enterprise business entity references from the natural language question and match the entity references with the corresponding entity nodes in the enterprise knowledge graph. The query path generation module is used to generate a traversal path for searching in the enterprise knowledge graph based on the intent category and the matched entity nodes, and output the query path to the intelligent question answering module.

[0014] The Natural Language Processing (NLP) module receives natural language sentences input by employees, performs word segmentation to break the sentence down into individual words, parts-of-speech tagging to distinguish between nouns, verbs, and adjectives, and syntactic analysis to clarify the modification relationships and structure between words. Finally, it transforms this information into a structured language feature vector that a computer can process. The Intent Classification module uses a pre-trained intent classification model to determine the type of intent behind the employee's question based on this feature vector. For example, is the employee asking how to operate a process, finding someone's contact information, retrieving a specific document, or explaining the meaning of a technical term? The Entity Linking module identifies words in the question that point to specific business entities, such as project codes or employee names, and precisely matches these words with existing entity nodes in the enterprise knowledge graph to determine the specific object the user is asking about. The Query Path Generation module combines the outputs of the first two modules, generating a traversal path in the knowledge graph based on the identified intent type and the linked specific entities. For example, starting from a process node and following the inclusion relationship to find the approval step node, this path directly guides the intelligent question answering module in finding the answer. When an employee asks Manager Zhang what the acceptance process is, the natural language processing module analyzes the sentence structure, the intent classification module determines it to be a process query, the entity linking module matches Manager Zhang to a manager node named Zhang San in the knowledge graph, matches the acceptance process to the acceptance process node, and the query path generation module generates a path from the Zhang San node along the responsibility relationship to find the acceptance process node and then along the inclusion relationship to find the specific steps.

[0015] Furthermore, the employee analysis module includes an interactive data acquisition module, a semantic analysis module, a knowledge mastery assessment module, a cognitive level classification module, and a dynamic profile generation module; The interactive data acquisition module is used to record the interaction behavior data between employees and the system in real time. The interaction behavior data includes the natural language question text of the business entity entered by the employee, the question timestamp, the question frequency, the click to view the pushed answer, and the reading time. The semantic analysis module is used to perform semantic evaluation on the natural language questions entered by employees. By analyzing the number of professional terms involved in the questions, the complexity of the question sentence, and the depth of the knowledge graph level to which the questions point, it generates a cognitive depth index that represents the employee's understanding of the current business entity. The knowledge mastery assessment module is used to calculate the employee's knowledge mastery score based on the trend of the frequency of questions asked by the employee within a preset time period, the adoption rate of the pushed answers, and the accuracy rate of correctly answering verification questions. The cognitive level classification module is used to input cognitive depth indicators and knowledge mastery scores into a pre-trained cognitive level assessment model and output the employee's cognitive level for business entities. The cognitive levels include novice, beginner, proficient and expert levels. The dynamic profile generation module is used to generate a dynamic employee profile that includes the employee's knowledge structure, cognitive shortcomings, and learning preferences, based on the cognitive level and in combination with the employee's job information, department information, and historical learning records.

[0016] The interaction data acquisition module collects employee interactions with the system, including the questions asked, the time of the questions, the frequency of questions about specific knowledge points, whether the pushed answers were viewed, and the reading time. The semantic analysis module mines these recorded questions, analyzing the professional terminology used, the simplicity or complexity of the question phrasing, and the depth of knowledge within the knowledge graph the question points to, thereby generating an indicator to quantify the employee's understanding of the current business knowledge. The knowledge mastery assessment module evaluates from another perspective, determining whether the number of times an employee asks questions in a certain area increases or decreases over a period of time, the proportion of times they adopt system-push answers, and their accuracy rate when answering system-pushed test questions. This data is combined to calculate the employee's knowledge mastery score in that area. The cognitive level classification module inputs the previously obtained understanding depth indicator and mastery score into a trained model, ultimately outputting the employee's cognitive stage for different business knowledge areas, such as novice, beginner with basic understanding, proficient, or expert capable of guiding others. The dynamic profile generation module aggregates all these assessment results, combining them with the employee's position, department, and past learning records to ultimately generate a dynamic profile that includes the employee's knowledge structure, existing cognitive gaps, and personal learning preferences. For example, a sales employee frequently asks about product pricing and discount policies. Semantic analysis shows that his questions are relatively superficial, and a mastery assessment reveals that he has a low probability of answering related test questions correctly. The system might classify him as a beginner and mark pricing policies as a knowledge gap in his profile. Compared to existing technologies that can only record which documents an employee has viewed, this solution constructs a comprehensive employee knowledge and ability model through multi-dimensional and multi-level analysis, providing detailed data support for subsequent precise learning and knowledge recommendation.

[0017] Furthermore, the semantic analysis module calculates the employee's cognitive depth index of business entities in the following manner. : Employee data is obtained from the interactive data acquisition module. Record the N natural language queries for business entity e; For the i-th question, extract the question text and input it into the pre-trained language model to identify the number of technical terms contained in the question. Calculate the syntactic complexity score And query the enterprise knowledge graph to determine the depth level at which the question is hit within the knowledge graph. ; The depth contribution value of the i-th question is calculated using the following formula. :

[0018] in The maximum total number of technical terms associated with entity e. The maximum syntactic complexity score is preset. Calculate the arithmetic mean of the depth contribution values ​​of N questions to obtain the cognitive depth index: ...

[0019] Furthermore, the Apple module for assessing knowledge of business entities calculates employees' knowledge assessment scores using the following methods. : Employee data is obtained from the interactive data acquisition module. Design the historical interaction data of business entity e and count its total number of questions. and the number of repeated questions The repeated question refers to a question whose semantic similarity to a historical question exceeds a preset threshold; Statistical analysis of employee adoption rates for answers related to the business entity e pushed to them. :

[0020] in The total number of answers related to business entity e pushed to employees. The number of answers that employees click to view and read for a duration exceeding a preset threshold; Obtain the accuracy rate of employees' responses to verification questions proactively pushed by the system related to business entity e. :

[0021] in The number of times the verification question was pushed to business entity e. The number of times an employee answers correctly; Calculation knowledge mastery score :

[0022] in , , As presupposed rights, their sum is 1.

[0023] Furthermore, the cognitive level classification module calculates the quantitative value of an employee's cognitive level towards a business entity using the following method. : Employee data is obtained from the interactive data collection module. The timestamp of the last interaction of business entity e. Get the current system time Calculate the time interval The time decay factor is obtained. ; The cognitive level quantification value is calculated using the following formula:

[0024] in , , Preset weighting coefficients; The quantized value of the cognitive level is input into a pre-trained mapping function, which outputs the corresponding cognitive level. The mapping function divides the quantized value range into multiple continuous intervals, each interval corresponding to a preset cognitive level. The cognitive levels include novice, beginner, proficient, and expert levels.

[0025] Furthermore, the dynamic profile generation module generates dynamic employee profiles in the following ways: Get the collection of all business entities that the employee has interacted with. ; For each business Determine the level of gas cognition respectively And extract the employee's behavioral feature vector for this business entity from the interactive data acquisition module. Generate employee profiles: .

[0026] The semantic analysis module collects multiple natural language questions from employees regarding specific business entities. For each question, it identifies the number of technical terms, calculates the syntactic complexity score, and determines the depth level of the hit in the knowledge graph. A formula incorporating terminology density, syntactic complexity, and knowledge depth is used to calculate the depth contribution value of a single question. The average of multiple questions is then used as the employee's cognitive depth index for that business entity. The knowledge mastery assessment module calculates the total number of questions and repeated questions asked by employees regarding the business entity, calculates their acceptance rate of pushed answers and the accuracy rate of answering verification questions proactively pushed by the system. The complement of the repeated question percentage, acceptance rate, and accuracy rate are weighted and summed to obtain a knowledge mastery score. The cognitive level classification module introduces a time decay factor, calculating the forgetting effect based on the time interval between the employee's last interaction. The cognitive depth index, knowledge mastery score, and time decay factor are weighted and summed to obtain a comprehensive cognitive level quantification value. Finally, this quantification value is mapped to a preset cognitive level range. The dynamic profile generation module collects all business entities that employees have interacted with, determines the cognitive level of each entity, and extracts behavioral feature vectors from the interaction data. Finally, it generates a dynamic employee profile in the form of a set, containing the employee's cognitive level and behavioral characteristics for each entity. Compared to traditional methods that rely on periodic assessments or subjective evaluations, this solution provides more objective and realistic evaluation results based on daily work behavior. It can accurately depict employees' knowledge structure and learning habits, providing a precise and reliable data foundation for subsequent personalized learning path planning. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of an embodiment of the present invention. Detailed Implementation

[0028] The following detailed description illustrates the specific implementation method: The basic implementation examples are as follows: Figure 1 As shown: An AI-based enterprise knowledge management and intelligent question-answering system is characterized by comprising a data acquisition module, an intent acquisition module, an intelligent question-answering module, an employee analysis module, and a learning planning module. The data acquisition module is used to collect multi-source heterogeneous data from various office systems within the enterprise, perform entity recognition and relationship extraction on the multi-source heterogeneous data, and construct an enterprise knowledge graph that includes enterprise business entities and their related relationships.

[0029] The data acquisition module includes a preprocessing module, an entity recognition module, a relationship recognition module, and a knowledge graph module; The preprocessing module is used to clean the format and parse the content of data from different sources, including extracting table fields from structured data, identifying document level tags from semi-structured data, and performing natural language processing on unstructured text data. The entity recognition module is used to identify the enterprise's business entities from the preprocessed data. The business entities include project entities, employee entities, process entities, document entities, and professional and technical entities, and each business entity is assigned a type label and a unique identifier. The relationship identification module is used to identify business relationships between entities and link business entities with related relationships. The knowledge graph module is used to obtain an enterprise knowledge graph based on the business relationships between business entities, with business entities as nodes and business relationships as edges.

[0030] Specifically, taking a software development company as an example, the company has multiple data sources, including project documents from a project management system, workflow documents from an office automation system, technical documents from a code repository, and employee files from a human resources system. The data acquisition module connects to these systems by configuring data source connectors. Specifically, it collected 120 project summary reports from the past three years from the project management system, 15 workflow documents such as financial reimbursement processes and leave approval processes from the office automation system, 80 technical documents such as technical architecture specifications and interface specifications from the code repository, and basic information data of all employees from the human resources system. These data come in various formats, including Excel spreadsheets, Word documents with chapter tags, and plain text log files. The preprocessing module first cleans the data, removing blank lines and duplicate records from tables, identifying first-level and second-level headings as document hierarchy tags from Word documents, and performing word segmentation and stop word removal on the main text of technical documents. The entity recognition module then identifies business entities from the cleaned data. For example, it identifies 15 project entities, such as the smart government affairs platform project and the smart healthcare project, from project summary reports; 8 process entities, such as expense reimbursement processes and leave application processes, from process documents; 45 employee entities, such as Zhang Wei and Li Na, from employee files; and 22 professional technical entities, such as database connection pools and Redis caches, from technical documents. The relationship recognition module further analyzes the relationships between entities. For example, it establishes the relationship that Zhang Wei is responsible for the smart government affairs platform project; it establishes the relationship that the expense reimbursement process includes three nodes: filling out an application form, departmental approval, and financial review; and it establishes the relationship that the database connection pool configuration guide in the technical documents references the database connection pool of professional technical entities. The knowledge graph module uses these entities as nodes and relationships as edges, storing the above information in a graph database, ultimately constructing an enterprise knowledge graph containing 180 nodes and 320 edges.

[0031] The intent acquisition module is used to acquire natural language questions input by employees, perform semantic recognition on the natural language questions to extract semantic features, obtain the question content based on the semantic features, and determine the query path in the enterprise knowledge graph based on the question content.

[0032] The intent acquisition module includes a natural language processing module, an intent classification module, an entity linking module, and a query path generation module; The natural language processing module is used to perform word segmentation, part-of-speech tagging, and syntactic analysis on natural language queries input by employees, and generate structured language feature vectors. The intent classification module is used to identify the intent category of the question based on the language feature vector using a pre-trained intent classification model. The intent categories include process query, personnel query, document query, and terminology explanation. The entity linking module is used to identify the mentioned enterprise business entity references from the natural language question and match the entity references with the corresponding entity nodes in the enterprise knowledge graph. The query path generation module is used to generate a traversal path for searching in the enterprise knowledge graph based on the intent category and the matched entity nodes, and output the query path to the intelligent question answering module.

[0033] Specifically, taking Wang Li, a new employee joining the R&D department, as an example, she encounters a problem at work and asks a question in natural language through the system interface. The intent acquisition module receives the question text input by Wang Li and processes it. Taking Wang Li's first question, "What is the project acceptance process that Engineer Zhang is responsible for?" as an example, the natural language processing module first performs word segmentation on the sentence, obtaining words such as "Engineer Zhang," "responsible for," "of," "project," "acceptance," "process," and "what." Part-of-speech tagging identifies "Engineer Zhang" as a noun, "responsible for" as a verb, and "acceptance process" as a noun phrase. Syntactic analysis clarifies that "Engineer Zhang" is the subject, "responsible for" is the predicate, and "project acceptance process" is the object. Finally, a language feature vector containing this information is generated. The intent classification module inputs this feature vector into a pre-trained intent classification model, which determines that the intent category of the question is process query. The entity linking module identifies two entity references from the question: "Engineer Zhang" and "project acceptance process." By querying the enterprise knowledge graph, "Engineer Zhang" is matched to the "Zhang Wei" node in the employee entity, and "project acceptance process" is matched to the "project acceptance process" node in the process entity. The query path generation module generates a traversal path based on the intent of the process query class and the two matched entity nodes: starting from the Zhang Wei node, it finds all the project nodes under his responsibility along the responsibility relationship, then finds the project acceptance process node along the inclusion relationship from these project nodes, and finally finds the specific step node under this process along the inclusion relationship. This query path is then output to the intelligent question answering module.

[0034] The intelligent question-and-answer module is used to retrieve the corresponding answers to questions from the enterprise knowledge graph based on the query path and push them to employees.

[0035] The employee analytics module records employee interaction data during the questioning process. This interaction data includes the semantic features of natural language questions, question frequency, and feedback behavior to pushed answers. The module assesses employees' cognitive level of related business entities based on the interaction data.

[0036] The employee analysis module includes an interactive data collection module, a semantic analysis module, a knowledge mastery assessment module, a cognitive level classification module, and a dynamic profile generation module. The interactive data acquisition module is used to record the interaction behavior data between employees and the system in real time. The interaction behavior data includes the natural language question text of the business entity entered by the employee, the question timestamp, the question frequency, the click to view the pushed answer, and the reading time. The semantic analysis module is used to perform semantic evaluation on the natural language questions entered by employees. By analyzing the number of professional terms involved in the questions, the complexity of the question sentence, and the depth of the knowledge graph level to which the questions point, it generates a cognitive depth index that represents the employee's understanding of the current business entity. The knowledge mastery assessment module is used to calculate the employee's knowledge mastery score based on the trend of the frequency of questions asked by the employee within a preset time period, the adoption rate of the pushed answers, and the accuracy rate of correctly answering verification questions. The cognitive level classification module is used to input cognitive depth indicators and knowledge mastery scores into a pre-trained cognitive level assessment model and output the employee's cognitive level for business entities. The cognitive levels include novice, beginner, proficient and expert levels. The dynamic profile generation module is used to generate a dynamic employee profile that includes the employee's knowledge structure, cognitive shortcomings, and learning preferences, based on the cognitive level and in combination with the employee's job information, department information, and historical learning records.

[0037] The semantic analysis module calculates the employee's cognitive depth index of business entities in the following way. : Employee data is obtained from the interactive data acquisition module. Record the N natural language queries for business entity e; For the i-th question, extract the question text and input it into the pre-trained language model to identify the number of technical terms contained in the question. Calculate the syntactic complexity score And query the enterprise knowledge graph to determine the depth level at which the question is hit within the knowledge graph. ; The depth contribution value of the i-th question is calculated using the following formula. :

[0038] in The maximum total number of technical terms associated with entity e. The maximum syntactic complexity score is preset. Calculate the arithmetic mean of the depth contribution values ​​of N questions to obtain the cognitive depth index: ...

[0039] The Apple module for assessing knowledge of business entities calculates employee knowledge scores using the following methods. : Employee data is obtained from the interactive data acquisition module. Design the historical interaction data of business entity e and count its total number of questions. and the number of repeated questions The repeated question refers to a question whose semantic similarity to a historical question exceeds a preset threshold; Statistical analysis of employee adoption rates for answers related to the business entity e pushed to them. :

[0040] in The total number of answers related to business entity e pushed to employees. The number of answers that employees click to view and read for a duration exceeding a preset threshold; Obtain the accuracy rate of employees' responses to verification questions proactively pushed by the system related to business entity e. :

[0041] in The number of times the verification question was pushed to business entity e. The number of times an employee answers correctly; Calculation knowledge mastery score :

[0042] in , , As presupposed rights, their sum is 1.

[0043] The cognitive level classification module calculates the quantitative value of an employee's cognitive level towards a business entity using the following method. : Employee data is obtained from the interactive data collection module. The timestamp of the last interaction of business entity e. Get the current system time Calculate the time interval The time decay factor is obtained. ; The cognitive level quantification value is calculated using the following formula:

[0044] in , , Preset weighting coefficients; The quantized value of the cognitive level is input into a pre-trained mapping function, which outputs the corresponding cognitive level. The mapping function divides the quantized value range into multiple continuous intervals, each interval corresponding to a preset cognitive level. The cognitive levels include novice, beginner, proficient, and expert levels.

[0045] Specifically, Table 1 records some of Wang Li's interaction data with the system during her first month at the company: Table 1

[0046] The learning planning module is used to determine the knowledge gap between an employee's current cognitive level and the preset target level for the corresponding position. Based on the knowledge gap, it retrieves and plans the optimal learning path from the enterprise knowledge base, which includes several business nodes and learning materials, and pushes the optimal learning path to the employee.

[0047] The employee analysis module's interaction data acquisition module records these interactions in real time, generating the aforementioned table data for use by other sub-modules. The semantic analysis module analyzes Wang Li's questions regarding the "project acceptance process" entity. As shown in the table, Wang Li asked two questions related to this entity: one on March 1st and another on March 15th. Regarding the March 1st question about the steps involved in the project acceptance process, the language model identified the number of technical terms contained within. ("Project Acceptance", "Process"), the maximum total number of technical terms associated with this entity. Syntactic complexity score Preset maximum syntactic complexity score The question hits a depth level K1=2 in the knowledge graph. Therefore, the depth contribution value d1 of this question is:

[0048]

[0049]

[0050] Regarding the question from 03-15 about who needs to sign off on project acceptance, the number of technical terms identified is T2=1 (project acceptance), the syntactic complexity score is H2=0.4, and the hit depth level is K2=3. Therefore, the depth contribution value d2 is:

[0051] Therefore, Wang Li's cognitive depth index of the project acceptance process entity is the arithmetic mean of the depth contribution values ​​of the two questions:

[0052] The knowledge mastery assessment module calculates Wang Li's knowledge mastery of the "database connection pool" entity. It retrieves Wang Li's historical interaction data related to this entity from the interaction data acquisition module, including the total number of questions asked. The two questions, from March 10th, are "How to configure a database connection pool?" and "What is the maximum number of connections to set in a database connection pool?". Semantic similarity comparison showed that these two questions did not overlap with previous questions; therefore, the number of duplicate questions is not counted. .

[0053] The system pushed answers related to this entity to Wang Li three times, including the configuration steps answer pushed on March 10th, the maximum connection number setting answer pushed on March 10th, and the connection pool principle explanation answer pushed on March 20th. Wang Li clicked to view and read the answers for more than 30 seconds twice: the configuration steps answer (320 seconds) and the maximum connection number setting answer (150 seconds). Therefore, the adoption rate is:

[0054] The system proactively sent Wang Li two verification questions about "database connection pools," one on the function of connection pools and the other on common parameters. Wang Li answered correctly once, therefore the accuracy rate is A:

[0055] Preset weighting coefficients , ,

[0056] Knowledge mastery score :

[0057]

[0058]

[0059] The cognitive level classification module calculates Wang Li's cognitive level quantification value for the "database connection pool entity". First, it obtains from the interaction data acquisition module that Wang Li's last interaction with this entity was at 16:38 on March 10th, and the current system time is 10:00 on March 25th, with a time interval Δt = 14.72 days. Assuming an attenuation coefficient λ = 0.05, the time attenuation factor is:

[0060] Wang Li calculated the semantic depth index of the "database connection pool" entity based on her two queries (the calculation process is similar to that described above, and the result is given directly here): Assuming weighting coefficients α=0.3, β=0.5, and γ=0.2, the quantified value of the cognitive level is:

[0061]

[0062] The quantified value is input into a mapping function, which sets the range as follows: 0-0.3 for beginner level, 0.3-0.6 for intermediate level, 0.6-0.8 for intermediate level, and 0.8-1.0 for expert level. 0.5188 falls within the 0.3-0.6 range; therefore, Wang Li's understanding level of the "database connection pool" entity is intermediate. Similarly, her understanding level of the "project acceptance process" entity is calculated to be beginner level, and her understanding level of the "code review" entity is also intermediate.

[0063] The dynamic profile generation module generates dynamic employee profiles in the following ways: Get the collection of all business entities that the employee has interacted with. ; For each business Determine the level of gas cognition respectively And extract the employee's behavioral feature vector for this business entity from the interactive data acquisition module. Generate employee profiles: .

[0064] The dynamic profile generation module obtains the set E_u of all business entities that Wang Li has interacted with, including three entities: "Project Acceptance Process," "Code Review," and "Database Connection Pool." A cognitive level is determined for each entity: "Project Acceptance Process" is novice level, "Code Review" is beginner level, and "Database Connection Pool" is beginner level. Behavioral feature vectors are extracted from the interaction data collection module. For the "Project Acceptance Process" entity, the question frequency is 3 times, the average reading time is 195 seconds, and the feedback tendency is 1 positive feedback and no negative feedback. The final dynamic employee profile of Wang Li is shown in Table 2.

[0065] The profile also identified Wang Li's knowledge gap as "project acceptance process" and her learning preference as a tendency to read technical configuration documents. The profile was stored in a database for use by the learning planning module.

[0066] Specifically, the learning planning module identifies Wang Li's knowledge gaps based on her dynamic employee profile. The pre-set target competency requirements for the software development position in the R&D department are: beginner level for "Project Acceptance Process," proficient level for "Code Review," and proficient level for "Database Connection Pool." Comparing Wang Li's current understanding level with the target levels, there are gaps in her understanding of "Project Acceptance Process" (from novice to beginner), "Code Review" (from beginner to proficient), and "Database Connection Pool" (from beginner to proficient). The module retrieves learning materials related to these three business entities from the enterprise knowledge base, including documents such as "Detailed Explanation of Project Acceptance Process," "Best Practice Guide for Code Review," and video tutorials on "Database Connection Pool Configuration." Based on the size of the knowledge gaps and the prerequisite dependencies between learning materials, an optimal learning path is planned: first, learn the "Detailed Explanation of Project Acceptance Process" document to fill the most basic process understanding gap; then learn the "Database Connection Pool Configuration" video tutorial; and finally, learn the "Best Practice Guide for Code Review" document. The learning planning module pushes this path to Wang Li and dynamically adjusts the path based on her learning progress and new questions during her subsequent interactions with the system. For example, if Wang Li's questions about "project acceptance" become more complex after she completes the process documentation, the system will update her profile accordingly and adjust the subsequent recommended content.

[0067] The above are merely embodiments of the present invention. Commonly known structures and characteristics are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are aware of all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, under the guidance of this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. An AI-based enterprise knowledge management and intelligent question-answering system, characterized in that: It includes a data acquisition module, an intent acquisition module, an intelligent question answering module, an employee analysis module, and a learning planning module; The data acquisition module is used to collect multi-source heterogeneous data from various office systems within the enterprise, perform entity recognition and relationship extraction on the multi-source heterogeneous data, and construct an enterprise knowledge graph that includes enterprise business entities and their related relationships. The intent acquisition module is used to acquire natural language questions input by employees, perform semantic recognition on the natural language questions to extract semantic features, obtain the question content based on the semantic features, and determine the query path in the enterprise knowledge graph based on the question content. The intelligent question-and-answer module is used to retrieve the corresponding answer to the question from the enterprise knowledge graph based on the query path and push it to the employee. The employee analysis module is used to record the interaction data of employees during the questioning process. The interaction data includes the semantic features of natural language questions, the frequency of questions, and the feedback behavior to the pushed answers. The interaction data is used to assess the employee's cognitive level of the related business entities. The learning planning module is used to determine the knowledge gap between an employee's current cognitive level and the preset target level for the corresponding position. Based on the knowledge gap, it retrieves and plans the optimal learning path from the enterprise knowledge base, which includes several business nodes and learning materials, and pushes the optimal learning path to the employee.

2. The AI-based enterprise knowledge management and intelligent question-answering system according to claim 1, characterized in that: The data acquisition module includes a preprocessing module, an entity recognition module, a relationship recognition module, and a knowledge graph module; The preprocessing module is used to clean the format and parse the content of data from different sources, including extracting table fields from structured data, identifying document level tags from semi-structured data, and performing natural language processing on unstructured text data. The entity recognition module is used to identify the enterprise's business entities from the preprocessed data. The business entities include project entities, employee entities, process entities, document entities, and professional and technical entities, and each business entity is assigned a type label and a unique identifier. The relationship identification module is used to identify business relationships between entities and link business entities with related relationships. The knowledge graph module is used to obtain an enterprise knowledge graph based on the business relationships between business entities, with business entities as nodes and business relationships as edges.

3. The AI-based enterprise knowledge management and intelligent question-answering system according to claim 2, characterized in that: The intent acquisition module includes a natural language processing module, an intent classification module, an entity linking module, and a query path generation module; The natural language processing module is used to perform word segmentation, part-of-speech tagging, and syntactic analysis on natural language queries input by employees, and generate structured language feature vectors. The intent classification module is used to identify the intent category of the question based on the language feature vector using a pre-trained intent classification model. The intent categories include process query, personnel query, document query, and terminology explanation. The entity linking module is used to identify the mentioned enterprise business entity references from the natural language question and match the entity references with the corresponding entity nodes in the enterprise knowledge graph. The query path generation module is used to generate a traversal path for searching in the enterprise knowledge graph based on the intent category and the matched entity nodes, and output the query path to the intelligent question answering module.

4. The AI-based enterprise knowledge management and intelligent question-answering system according to claim 3, characterized in that: The employee analysis module includes an interactive data collection module, a semantic analysis module, a knowledge mastery assessment module, a cognitive level classification module, and a dynamic profile generation module. The interactive data acquisition module is used to record the interaction behavior data between employees and the system in real time. The interaction behavior data includes the natural language question text of the business entity entered by the employee, the question timestamp, the question frequency, the click to view the pushed answer, and the reading time. The semantic analysis module is used to perform semantic evaluation on the natural language questions entered by employees. By analyzing the number of professional terms involved in the questions, the complexity of the question sentence, and the depth of the knowledge graph level to which the questions point, it generates a cognitive depth index that represents the employee's understanding of the current business entity. The knowledge mastery assessment module is used to calculate the employee's knowledge mastery score based on the trend of the frequency of questions asked by the employee within a preset time period, the adoption rate of the pushed answers, and the accuracy rate of correctly answering verification questions. The cognitive level classification module is used to input cognitive depth indicators and knowledge mastery scores into a pre-trained cognitive level assessment model and output the employee's cognitive level for business entities. The cognitive levels include novice, beginner, proficient and expert levels. The dynamic profile generation module is used to generate a dynamic employee profile that includes the employee's knowledge structure, cognitive shortcomings, and learning preferences, based on the cognitive level and in combination with the employee's job information, department information, and historical learning records.

5. The AI-based enterprise knowledge management and intelligent question-answering system according to claim 4, characterized in that: The semantic analysis module calculates the employee's cognitive depth index of business entities in the following way. : Employee data is obtained from the interactive data acquisition module. Record the N natural language queries for business entity e; For the i-th question, extract the question text and input it into the pre-trained language model to identify the number of technical terms contained in the question. Calculate the syntactic complexity score And query the enterprise knowledge graph to determine the depth level at which the question is hit within the knowledge graph. ; The depth contribution value of the i-th question is calculated using the following formula. : in The maximum total number of technical terms associated with entity e. The maximum syntactic complexity score is preset. Calculate the arithmetic mean of the depth contribution values ​​of N questions to obtain the cognitive depth index: 。 6. The AI-based enterprise knowledge management and intelligent question-answering system according to claim 5, characterized in that: The Apple module for assessing knowledge of business entities calculates employee knowledge scores using the following methods. : Employee data is obtained from the interactive data acquisition module. Design the historical interaction data of business entity e and count its total number of questions. and the number of repeated questions The repeated question refers to a question whose semantic similarity to a historical question exceeds a preset threshold; Statistical analysis of employee adoption rates for answers related to the business entity e pushed to them. : in The total number of answers related to business entity e pushed to employees. The number of answers that employees click to view and read for a duration exceeding a preset threshold; Obtain the accuracy rate of employees' responses to verification questions proactively pushed by the system related to business entity e. : in The number of times the verification question was pushed to business entity e. The number of times an employee answers correctly; Calculation knowledge mastery score : in , , As presupposed rights, their sum is 1.

7. The AI-based enterprise knowledge management and intelligent question-answering system according to claim 6, characterized in that: The cognitive level classification module calculates the quantitative value of an employee's cognitive level towards a business entity using the following method. : Employee data is obtained from the interactive data collection module. The timestamp of the last interaction of business entity e. Get the current system time Calculate the time interval The time decay factor is obtained. ; The cognitive level quantification value is calculated using the following formula: in , , Preset weighting coefficients; The quantized value of the cognitive level is input into a pre-trained mapping function, which outputs the corresponding cognitive level. The mapping function divides the quantized value range into multiple continuous intervals, each interval corresponding to a preset cognitive level. The cognitive levels include novice, beginner, proficient, and expert levels.

8. The AI-based enterprise knowledge management and intelligent question-answering system according to claim 7, characterized in that: The dynamic profile generation module generates dynamic employee profiles in the following ways: Get the collection of all business entities that the employee has interacted with. ; For each business Determine the level of gas cognition respectively And extract the employee's behavioral feature vector for this business entity from the interactive data acquisition module. Generate employee profiles: 。

Images