Knowledge management method and device, electronic equipment, medium and program product

Abstract

The invention provides a knowledge management method and device, electronic equipment, a medium and a program product, and can be applied to the technical field of big data and the technical field of artificial intelligence. The method comprises the following steps: acquiring authorized multi-source text interaction data, and performing semantic analysis on the multi-source text interaction data to form a semantic unit set; performing theme recognition on the semantic unit set to obtain a first target theme, and obtaining a first knowledge set corresponding to the first target theme; executing subject clustering or subject creation based on the first target subject and a historical subject set to obtain a second target subject, and determining a second knowledge set based on the first knowledge set and the second target subject; and updating a multi-layer memory structure in a knowledge management library based on the second target theme and the second knowledge set, wherein the multi-layer memory structure is used for storing knowledge coverage information of each knowledge participation entity under different themes.

Description

Technical Field

[0001] This application relates to the fields of big data technology and artificial intelligence technology, and more specifically to a knowledge management method, apparatus, device, medium and program product. Background Technology

[0002] As enterprises continue to advance digital transformation and intelligent office practices, they generate a large amount of unstructured knowledge assets in the process of business communication, project management, customer support, and knowledge accumulation. These assets contain rich business experience and tacit knowledge, which have high reuse value.

[0003] To promote knowledge sharing and innovation, enterprises generally build knowledge management systems to collect, organize, and apply knowledge. However, existing systems still have limitations in knowledge extraction, classification and updating, and semantic understanding, making it difficult to meet the needs of efficient collaboration and knowledge assetization. Some solutions include rule-based dialogue extraction and static knowledge base construction. The former relies on manually setting keywords and regular expressions to extract business information, requires manual maintenance of rules and classification systems, is difficult to adapt to diverse business semantics, and has high maintenance costs. The latter relies on employees manually uploading documents and storing knowledge in a fixed classification structure, resulting in slow update speed, limited coverage, and a lack of ability to identify tacit knowledge. In practical applications, both types of solutions suffer from poor topic adaptability, low update efficiency, and severe knowledge fragmentation. Knowledge on the same topic is stored in a scattered manner, making aggregation difficult, and content filtering is insufficient, with non-business information mixed into the knowledge base, leading to a decline in knowledge quality and affecting subsequent utilization. Summary of the Invention

[0004] In view of the above problems, this application provides knowledge management methods, apparatus, equipment, media and program products.

[0005] According to a first aspect of this application, a knowledge management method is provided, the method comprising: acquiring authorized multi-source text interaction data; performing semantic parsing on the multi-source text interaction data to form a semantic unit set; performing topic identification on the semantic unit set to obtain a first target topic, and acquiring a first knowledge set corresponding to the first target topic; performing topic clustering or topic creation based on the first target topic and a historical topic set to obtain a second target topic, and determining a second knowledge set based on the first knowledge set and the second target topic; and updating a multi-layer memory structure in a knowledge management base based on the second target topic and the second knowledge set, the multi-layer memory structure being used to store knowledge coverage information of each knowledge participating entity under different topics.

[0006] According to an embodiment of this application, updating the multi-layer memory structure in the knowledge management base based on the second target topic includes: using the second target topic as an index, extracting the topic-related knowledge set corresponding to the second target topic from the multi-layer memory structure; comparing the topic-related knowledge set with the second knowledge set to obtain knowledge change parameters; and updating the multi-layer memory structure based on the knowledge change parameters.

[0007] According to an embodiment of this application, the multi-layer memory structure includes: a short-term memory layer for caching semantic units and corresponding topics collected within a preset period; a topic memory layer for aggregating knowledge nodes under the same topic; and an individual memory layer for recording the participation frequency and knowledge contribution of each knowledge participating entity under different topics.

[0008] According to an embodiment of this application, updating the multi-layer memory structure based on the knowledge change parameters includes: parsing the knowledge change parameters to obtain short-term change parameters, topic change parameters, and individual change parameters; updating the time window weight or cache validity period of the corresponding semantic unit in the short-term memory layer according to the short-term change parameters; adjusting the weight of knowledge nodes corresponding to the second target topic in the topic memory layer according to the topic change parameters, and updating the semantic association information between knowledge nodes; and updating the participation frequency or knowledge contribution of each knowledge participating entity in the individual memory layer under the second target topic according to the individual change parameters.

[0009] According to an embodiment of this application, after updating the short-term memory layer, the topic memory layer, and the individual memory layer, the method further includes: extracting topic correspondence and semantic association information between different layers to generate an inter-layer association dataset; and based on the inter-layer association dataset, reconstructing the inter-layer association index of the multi-layer memory structure, outputting the updated inter-layer association index, and writing it into the knowledge management base.

[0010] According to an embodiment of this application, the step of semantically parsing the multi-source text interaction data to form a semantic unit set includes: performing sentence segmentation and dependency parsing on the multi-source text interaction data to extract entity, event, and attribute information; inputting the entity, event, and attribute information into a semantic embedding model to generate semantic unit vector representations; and aggregating the semantic unit vector representations based on vector similarity to form the semantic unit set.

[0011] According to an embodiment of this application, the step of performing topic clustering or topic creation based on the first target topic and the historical topic set to obtain a second target topic includes: calculating the semantic similarity between the first target topic and the historical topic set; when the semantic similarity is greater than or equal to a preset threshold, merging the first target topic into the corresponding historical topic cluster, and determining the cluster center node of the historical topic cluster as the second target topic; when the semantic similarity is less than the preset threshold, creating a target topic node, and determining the target topic node as the second target topic.

[0012] A second aspect of this application provides a knowledge management device, comprising: a data acquisition module, configured to: acquire authorized multi-source text interaction data, perform semantic parsing on the multi-source text interaction data, and form a semantic unit set; a topic identification module, configured to: perform topic identification on the semantic unit set, obtain a first target topic, and acquire a first knowledge set corresponding to the first target topic; a clustering or creation module, configured to: perform topic clustering or topic creation based on the first target topic and a historical topic set, obtain a second target topic, and determine a second knowledge set based on the first knowledge set and the second target topic; and a knowledge update module, configured to: update a multi-layer memory structure in a knowledge management base based on the second target topic and the second knowledge set, wherein the multi-layer memory structure is used to store knowledge coverage information of each knowledge participating entity under different topics.

[0013] A third aspect of this application provides an electronic device comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the method described above.

[0014] A fourth aspect of this application also provides a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method.

[0015] The fifth aspect of this application also provides a computer program product, including a computer program or instructions that, when executed by a processor, implement the steps of the above-described method.

[0016] According to embodiments of this application, by utilizing semantic parsing to structure multi-source text interaction data into a set of semantic units, the accuracy and universality of knowledge extraction are improved. Through topic identification and clustering based on semantic similarity, business topics can be automatically discovered and organized, achieving the aggregation and expansion of knowledge content at the semantic level. Furthermore, by combining a first knowledge set with a second target topic to generate a second knowledge set, the knowledge content is continuously updated as the topic evolves, avoiding the lag and fragmentation problems caused by manual maintenance. Finally, by updating the multi-layer memory structure based on the second target topic and the second knowledge set, the system can achieve orderly storage and continuous optimization of knowledge across multiple dimensions, including time, topic, and participating entities. Therefore, embodiments of this application can effectively improve the intelligence and response speed of enterprise knowledge management, enabling the knowledge system to self-learn, self-evolve, and maintain structural consistency. Attached Figure Description

[0017] The above-mentioned contents, other objects, features and advantages of this application will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0018] Figure 1 The illustrations depict application scenarios of knowledge management methods, apparatus, devices, media, and program products according to embodiments of this application.

[0019] Figure 2 A flowchart illustrating a knowledge management method according to an embodiment of this application is shown schematically.

[0020] Figure 3 The flowchart schematically illustrates a method for updating the multi-layer memory structure based on the knowledge change parameters according to some exemplary embodiments of this application;

[0021] Figure 4 The illustration shows a schematic diagram of a method for semantic parsing of multi-source text interaction data according to some exemplary embodiments of this application;

[0022] Figure 5 This schematically illustrates a structural block diagram of a knowledge management device according to an embodiment of this application; and

[0023] Figure 6 A block diagram schematically illustrates an electronic device suitable for implementing a knowledge management method according to an embodiment of this application. Detailed Implementation

[0024] The embodiments of this application will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of this application. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of this application for ease of explanation. However, it will be apparent that one or more embodiments may be implemented without these specific details. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of this application.

[0025] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0026] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0027] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0028] With the continuous advancement of enterprise digital transformation and intelligent office practices, enterprises generate a large amount of unstructured information in daily business communication, project management, customer support, and knowledge accumulation. This information contains rich business experience and tacit knowledge, possessing high reuse value. To promote knowledge sharing and business innovation, enterprises have generally built knowledge management systems to support the collection, organization, storage, and application of knowledge. However, existing knowledge management systems still have significant limitations in knowledge extraction, classification and updating, and semantic understanding, making it difficult to meet the needs of modern enterprises in terms of efficient collaboration and knowledge assetization.

[0029] Currently, enterprise knowledge management solutions mainly fall into two categories. One category is rule-based dialogue extraction solutions. These solutions typically rely on manually set keywords or regular expressions to extract business-related information from employee conversations. They also require manual maintenance of rule sets and knowledge classification systems, and regular import of dialogue records for review. While this solution achieves initial knowledge extraction and structured storage, its heavy reliance on manually set rules makes it inflexible when handling diverse business semantics, prone to missed or misidentified entries, and incurs high maintenance costs.

[0030] Another approach is the static knowledge base construction scheme, which primarily relies on manually uploaded and categorized work documents. Employees are required to proactively submit summaries, reports, or case files, and the system stores the knowledge according to a fixed classification structure, responding to user queries through pre-defined question-and-answer logic. This method can effectively organize existing knowledge, but the knowledge update speed is slow, the coverage is limited, and it is difficult to dynamically reflect employees' real-time work progress and business changes. Due to the lack of semantic recognition and contextual understanding capabilities, static knowledge bases often fail to capture tacit knowledge and struggle to support complex intelligent question-and-answer and knowledge reasoning tasks.

[0031] In practical applications, both solutions share common problems. First, they suffer from poor topic adaptability; the systems cannot automatically identify emerging business directions or adjust the knowledge topic structure, resulting in a lag in responding to rapidly changing business needs within the enterprise. Second, they are inefficient in updating; knowledge screening and entry rely on manual intervention, which is not only time-consuming and labor-intensive but also subject to subjective bias, easily leading to knowledge omissions or uneven quality. Third, knowledge fragmentation is severe; the same topic or related content is scattered across multiple subsystems, lacking effective aggregation and association mechanisms, resulting in problems such as duplicate entry, knowledge redundancy, and retrieval difficulties.

[0032] In addition, the existing system is also inadequate in terms of content filtering. Some irrelevant messages are often mixed with business-related conversations, and there is a lack of effective differentiation mechanism. This results in irrelevant content being mixed in with the knowledge base, affecting the overall data quality and subsequent knowledge utilization.

[0033] Based on this, embodiments of this application provide a knowledge management method, the method comprising: acquiring authorized multi-source text interaction data; performing semantic parsing on the multi-source text interaction data to form a semantic unit set; performing topic identification on the semantic unit set to obtain a first target topic, and acquiring a first knowledge set corresponding to the first target topic; performing topic clustering or topic creation based on the first target topic and a historical topic set to obtain a second target topic, and determining a second knowledge set based on the first knowledge set and the second target topic; and updating a multi-layer memory structure in a knowledge management base based on the second target topic and the second knowledge set, the multi-layer memory structure being used to store knowledge coverage information of each knowledge participating entity under different topics. According to embodiments of this application, by utilizing semantic parsing to structure multi-source text interaction data into a set of semantic units, the accuracy and universality of knowledge extraction are improved. Through topic identification and clustering based on semantic similarity, business topics can be automatically discovered and organized, achieving the aggregation and expansion of knowledge content at the semantic level. Furthermore, by combining a first knowledge set with a second target topic to generate a second knowledge set, the knowledge content is continuously updated as the topic evolves, avoiding the lag and fragmentation problems caused by manual maintenance. Finally, by updating the multi-layer memory structure based on the second target topic and the second knowledge set, the system can achieve orderly storage and continuous optimization of knowledge across multiple dimensions, including time, topic, and participating entities. Therefore, embodiments of this application can effectively improve the intelligence and response speed of enterprise knowledge management, enabling the knowledge system to self-learn, self-evolve, and maintain structural consistency.

[0034] It should be noted that the knowledge management methods, apparatus, devices, media, and program products defined in this application can be used in the fields of big data technology, artificial intelligence technology, and fintech, and can also be used in various other fields besides big data technology, artificial intelligence technology, and fintech. The application fields of the knowledge management methods, apparatus, devices, media, and program products provided in the embodiments of this application are not limited.

[0035] In the technical solution of this application, the user information (including but not limited to user personal information, user image information, user device information, such as location information) and data (including but not limited to data used for analysis, stored data, and displayed data) involved are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of related data all comply with relevant laws, regulations, and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entry points for users to choose to authorize or refuse.

[0036] In scenarios where personal information is used for automated decision-making, the methods, devices, and systems provided in this application all offer users corresponding operation entry points, allowing them to choose to agree to or reject the automated decision results; if the user chooses to reject, the process proceeds to expert decision-making. Here, "expert decision-making" refers to the decision-making activities of personnel who specialize in a particular field, possess specialized experience, knowledge, and skills, and have reached a certain level of professional expertise.

[0037] Figure 1 The illustrations depict application scenarios of knowledge management methods, apparatus, devices, media, and program products according to embodiments of this application.

[0038] like Figure 1 As shown, application scenario 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 serves as a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0039] Users can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 via the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).

[0040] In the embodiments of this application, the first terminal device 101 can be an example of the first device, and the second terminal device 102 and / or the third terminal device 103 can be an example of at least one second device. The first device and the second device can communicate collaboratively through an internal client mechanism to implement the data distribution and rendering logic described in the knowledge management method.

[0041] In some embodiments, the first device and at least one second device may be different display modules, windows or screens on the same computing terminal (such as a host), or multiple physical devices that work together through a network, such as different client instances deployed on a desktop computer, tablet terminal or mobile device respectively.

[0042] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with displays and support web browsing, including but not limited to smart mobile terminals, tablet computers, laptop computers, and desktop computers.

[0043] Server 105 can be a server that provides various services, such as a backend management server that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). The backend management server can analyze and process data such as received user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0044] It should be noted that the knowledge management method provided in this application embodiment can generally be executed by server 105. Correspondingly, the knowledge management device provided in this application embodiment can generally be located in server 105. The knowledge management method provided in this application embodiment can also be executed by a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Correspondingly, the knowledge management device provided in this application embodiment can also be located in a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105.

[0045] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0046] The following will be based on Figure 1 The described scene, through Figures 2-4 The knowledge management method of the disclosed embodiments is described in detail.

[0047] Figure 2 A flowchart illustrating a knowledge management method according to an embodiment of this application is shown schematically.

[0048] like Figure 2 As shown, the knowledge management method 200 of this embodiment includes operations S210 to S240.

[0049] In operation S210, authorized multi-source text interaction data is acquired, and semantic parsing is performed on the multi-source text interaction data to form a set of semantic units.

[0050] In the embodiments of this application, multi-source text interaction data within an enterprise can be uniformly collected and semantically processed to construct the semantic foundation layer of a knowledge management system. Specifically, the system can interface with authorized business systems within the enterprise to obtain text interaction data from multiple sources, including but not limited to customer relationship management systems, work order systems, project collaboration platforms, and meeting recording systems. Before being collected, this data undergoes authorization authentication and privacy anonymization to ensure that only authorized business corpora enter the subsequent analysis process, thereby guaranteeing data compliance and security. During the data collection process, the system can periodically retrieve the latest interaction records from different business sources and set differentiated collection strategies based on business importance and update frequency to achieve a dynamic balance of knowledge input.

[0051] In some embodiments, to improve data structuring efficiency, the acquired multi-source text interaction data can first be uniformly formatted and preprocessed. This includes steps such as text encoding conversion, irrelevant symbol removal, segmentation marker insertion, and language standardization, thereby eliminating differences in language style, format, and time stamps between different systems. Subsequently, the semantic parsing module performs sentence segmentation on the formatted text and, combined with dependency parsing and semantic role labeling, identifies entities, events, behavioral relationships, and attribute features within the sentences.

[0052] In some embodiments, to further enhance the semantic consistency of cross-domain corpora, the system can introduce a pre-trained semantic embedding model to vectorize the aforementioned semantic elements. The semantic embedding model can be based on a large language model and generate vector representations reflecting deep semantic relationships through a contextual attention mechanism. For data from different sources, such as customer service dialogues and project logs, the system can apply domain-adaptive semantic models to enhance the understanding of specific business domains through semantic encoding. Ultimately, all semantic units are mapped to a unified semantic vector space, thereby achieving semantic alignment and aggregation between different business data.

[0053] In some embodiments, the system can perform similarity calculation and aggregation operations based on the vector representations of semantic units to form a set of semantic units. This set of semantic units not only includes the original content of the text but also preserves its semantic relationships and contextual dependencies. The system can employ strategies based on vector cosine similarity or clustering algorithms to merge units with high semantic similarity into the same semantic group, thereby reducing redundant information and enhancing the aggregation degree of the knowledge structure. For example, in employee communication records, semantic units related to "customer complaint handling," "after-sales response," and "problem feedback" may be aggregated into sub-semantic groups under the same "customer service" theme for automatic classification in subsequent topic identification stages.

[0054] In some embodiments, to enhance the ability to parse complex contexts, semantic parsing can be further combined with a context window mechanism and a self-attention network to achieve hierarchical semantic understanding of long texts. The system can perform context modeling on multi-turn dialogues, meeting minutes, or work order history, so that the set of semantic units not only reflects local sentence meanings but also includes multi-dimensional information such as event time sequence and speaker roles. As a result, the system can more accurately capture semantic focus and business relevance in the subsequent topic recognition stage, providing high-quality input data for knowledge aggregation and topic evolution.

[0055] In some embodiments, semantic parsing can also support multimodal extended input. When business interaction data includes images, document attachments, or screenshots, the system can extract text content using optical character recognition (OCR) technology and perform semantic matching with corresponding text paragraphs. For structured table content, it can map table content into semantic entity nodes through fusion modeling of field label recognition and natural language description. Through multimodal parsing capabilities, the system can achieve cross-format and cross-carrier knowledge extraction and semantic unification, thereby expanding the coverage and depth of knowledge management.

[0056] In operation S220, topic identification is performed on the semantic unit set to obtain a first target topic, and a first knowledge set corresponding to the first target topic is obtained.

[0057] In the embodiments of this application, a topic structure with business significance can be identified from a set of semantic units, and knowledge content related to that topic can be extracted to form a first target topic and its corresponding first knowledge set. Specifically, based on the set of semantic units, a semantic aggregation and topic labeling process can be performed through a topic recognition model to determine the business semantic focus implied in the text content. The topic recognition model can employ a semantic classifier based on a large language model or a topic generation model to comprehensively analyze the semantic similarity, contextual association, and co-occurrence patterns of key entities between semantic units, thereby identifying the main topics that can represent the current business context. For example, in customer service communication data, the system can automatically identify topic tags such as "customer complaint," "refund application," and "product consultation"; in R&D project communication, business topics such as "version iteration," "defect repair," and "requirement change" can be identified.

[0058] In some embodiments, topic recognition can combine supervised and unsupervised learning methods to enhance accuracy. Specifically, the system can train a topic classifier using labeled historical business corpora, enabling it to learn the distribution characteristics of different topics in the semantic space. Simultaneously, for data in unknown or emerging business domains, the system can employ topic modeling algorithms for unsupervised topic discovery to identify potential topics that are not predefined. Through this combined strategy, the system maintains consistency in topic recognition while possessing adaptive recognition capabilities for new business semantics. Identified topics can be accompanied by confidence scores, which the system can use to select the most representative topics as the first target topics, ensuring the accuracy and relevance of subsequent knowledge extraction.

[0059] In some embodiments, to enhance the contextual understanding capability of topic recognition, the system can introduce a context window mechanism during the recognition process to perform local and global fusion analysis on continuous semantic units. This mechanism can capture semantic extension relationships across sentences and paragraphs, ensuring that topic recognition results not only rely on local keywords but also consider contextual evolution. For example, in lengthy meeting minutes or multi-round dialogues, although the explicit keywords of different statements vary significantly, the system can still identify implicit topics such as "project risk assessment" or "adjustment of cooperation plan" through contextual semantic aggregation. Furthermore, the system can also align the identified topics with the enterprise's internal knowledge graph through entity chaining and knowledge base mapping technology, thereby initially establishing a correspondence between topics and the existing knowledge system during the semantic recognition stage.

[0060] In some embodiments, while determining a first target topic, the system can simultaneously extract semantic units directly related to that topic to form a first knowledge set. The first knowledge set may include text fragments, entity nodes, event relationships, or external knowledge references related to the first target topic. To ensure the completeness of the knowledge set, the system can select content from the semantic unit set according to semantic relevance and perform hierarchical aggregation to construct a mapping relationship between the topic center and its semantic neighborhood. For example, after identifying the topic "customer complaint handling," the system can automatically aggregate related semantic units, such as "product quality issues," "customer service response time," and "compensation measures," to form the corresponding first knowledge set.

[0061] In some embodiments, to enhance the generalization ability of cross-domain topic recognition, the system can employ a multi-model collaborative judgment mechanism. Specifically, different large language models or semantic classifiers can independently predict topics for the set of semantic units, and the system then determines the final first target topic through weighted voting or confidence fusion algorithms. For text content with semantic ambiguity or multiple topics coexisting, the system can support multi-label recognition to reflect complex business contexts. For example, in a customer work order that involves both "invoice error" and "refund delay," the system can simultaneously label both topics and establish a corresponding first knowledge set for each topic, thereby achieving multi-dimensional knowledge extraction.

[0062] In operation S230, topic clustering or topic creation is performed based on the first target topic and the historical topic set to obtain the second target topic, and the second knowledge set is determined based on the first knowledge set and the second target topic.

[0063] In some embodiments, based on the topic identification results, cluster analysis can be performed on the current business topic and the topics in the historical knowledge system, or new topics can be created, thereby obtaining a higher-level topic expression, namely the second target topic, and constructing a corresponding second knowledge set accordingly.

[0064] Specifically, the system first takes the primary target topic as the center and, combined with the semantic vector representations in the historical topic set, calculates its semantic similarity with each historical topic. Semantic similarity can be calculated using a vector space model, a pre-trained language model, or a graph neural network model to reflect the semantic proximity between topics.

[0065] In some embodiments, the system determines whether a first target topic can be classified into an existing topic cluster based on a threshold: when the similarity is higher than the threshold, the system merges the topic into the corresponding topic cluster and updates the cluster center; when the similarity is lower than the threshold, the system automatically creates a new topic node to represent an emerging knowledge domain.

[0066] In some embodiments, to enhance the accuracy and robustness of clustering, the system can employ a hybrid algorithm based on hierarchical clustering and time-weighted clustering. Specifically, the system introduces a time decay factor when calculating topic similarity, assigning lower weights to earlier-formed topics to reflect the timeliness changes in the business context. This mechanism effectively prevents the semantic inertia of older topics from affecting the identification of new topics. For example, when the enterprise knowledge base already has the topic "customer complaint handling," and recently new semantic units related to "user experience improvement" are added, the system can identify that although its semantic similarity is high, its time weight is low, thus determining it as the emerging topic node "customer satisfaction optimization," achieving a sensitive response to new business directions. In addition, the system can dynamically adjust the similarity threshold during the clustering process and adaptively optimize the clustering boundary based on the stability of historical clustering, ensuring the rationality of topic division and semantic continuity.

[0067] In the embodiments of this application, after determining the second target topic, the system can generate a second knowledge set based on the semantic matching results between the first knowledge set and the second target topic. For example, this process may include: selecting knowledge units from the first knowledge set that are semantically highly relevant to the second target topic, and combining them with knowledge nodes related to that topic cluster from the historical topic set to form an expanded second knowledge set. In this way, the system not only inherits the knowledge content under the original topic, but also introduces cross-topic knowledge associations, realizing the semantic fusion of knowledge content. For example, when the topic "project delay management" is clustered with the historical topic "risk control," the system can automatically include knowledge units related to "project plan adjustment" and "risk level assessment" into the second knowledge set, forming a more hierarchical and comprehensive knowledge structure.

[0068] In some embodiments, to address complex contexts involving multiple overlapping topics, the system can introduce a multi-topic mapping mechanism during topic clustering. When a first target topic is simultaneously associated with multiple historical topics in terms of similarity, the system can assign weights to each topic based on the strength of the association and generate a composite second target topic. This second target topic can represent the intersection of multiple business dimensions, such as the fusion topic "user-driven product improvement" of "product performance optimization" and "user feedback analysis". In this case, the second knowledge set will contain semantic units from different topic sources, and the semantic fusion algorithm will eliminate duplicate and conflicting information, achieving comprehensive aggregation and unified expression of knowledge.

[0069] In some embodiments, the system can also incorporate enterprise knowledge graphs for topic clustering assistance. By mapping a first target topic to entity nodes in the knowledge graph, the system can obtain the hierarchical relationships and association paths of that topic within the knowledge network. During the clustering process, the system considers not only semantic similarity but also introduces structural similarity and relationship density as supplementary indicators, thereby achieving dual constraints of semantics and structure. For example, for the topic "supplier management," the system can determine from the knowledge graph structure that it has a strong relationship with the topic "procurement contract management." Although the semantic similarity is slightly lower, the two can still be aggregated into the same topic cluster, forming a set of knowledge units with semantics and business logic consistent.

[0070] In some embodiments, the system may also introduce a feedback learning mechanism to optimize the topic clustering results. After the second target topic is generated, the system can monitor the frequency of occurrence and contextual stability of the topic in subsequent business corpora; if the topic remains active and the associated knowledge continues to expand, the system confirms it as a stable topic and solidifies it into the knowledge base; if the topic appears frequently in the short term but lacks extensibility, the system can mark it as a transient topic or merge it back into the parent topic cluster.

[0071] In operation S240, the multi-layer memory structure in the knowledge management base is updated based on the second target topic and the second knowledge set. The multi-layer memory structure is used to store the knowledge coverage information of each knowledge participating entity under different topics.

[0072] In the embodiments of this application, after completing topic identification and topic clustering, the multi-layer memory structure in the knowledge management base can be dynamically updated based on the second target topic and its corresponding second knowledge set, so as to realize the hierarchical storage and continuous evolution of knowledge.

[0073] In some embodiments, when performing this operation, the system can extract topic-related features from the second knowledge set, including topic keywords, semantic vectors, and related entity information. Subsequently, the system retrieves the corresponding topic node in the multi-layered memory structure and calculates the similarity between the current topic node and the newly input knowledge. When the similarity is higher than a preset threshold, the new knowledge unit can be merged with the existing node, updating its weight and timestamp information; when the similarity is lower than the threshold, a new knowledge node is created in the topic layer and semantic links are established. Through this differentiated update strategy, the system can maintain the stability of the knowledge network while responding quickly to the emergence of new topics, achieving self-growth of the knowledge system. For example, when the system detects a new knowledge unit related to "feedback and follow-up" under the topic of "customer service quality improvement," it can automatically merge it into the existing node "customer satisfaction analysis" and adjust the semantic weight of the associated path to reflect the new knowledge contribution.

[0074] In some embodiments, to achieve interpretability and traceability of the knowledge update process, the system can introduce version management and time window mechanisms in a multi-layered memory structure. Each topic node generates a record entry containing a version number, update time, and update source upon update, supporting backtracking queries of historical knowledge. The time window mechanism controls the knowledge retention period in the short-term memory layer, enabling the system to dynamically adjust its caching strategy based on knowledge activity. For topic nodes that are frequently accessed or updated recently, the system can extend their retention time in the short-term memory layer; while knowledge nodes that have not been referenced for a long time are automatically downgraded to the archive layer or removed from the active index. Through this adaptive time adjustment mechanism, the system can reduce redundant data and improve knowledge access efficiency while ensuring knowledge integrity.

[0075] In some embodiments, the system can achieve the linkage update between different memory layers through a cross-layer information fusion algorithm. When a node is newly added or modified in the theme layer, the system can automatically trigger the synchronous update operations of the individual memory layer and the short-term memory layer. For example, when a knowledge node in a certain theme "Supply Chain Optimization" is updated, the system can simultaneously update the employee knowledge participation information related to this theme, adjust the knowledge coverage vector in the individual memory layer, so as to accurately reflect the change in the contribution degree of the knowledge participation entity under this theme. Correspondingly, the short-term memory layer will also record this knowledge update event and its semantic features for subsequent analysis modules to perform trend detection. Through the cross-layer linkage mechanism, the system realizes the consistent propagation of knowledge evolution in the global semantic network, ensuring the synchrony and consistency of the knowledge base in multiple dimensions.

[0076] In some embodiments, the update process of the multi-layer memory structure can also be combined with a knowledge quality assessment mechanism to ensure that the introduced knowledge content conforms to enterprise standards and business semantic specifications. Before writing the second knowledge set into the system, the system can use a pre-trained language model to automatically evaluate its semantic rationality, content integrity, and logical consistency, and decide whether to incorporate this knowledge into the theme memory layer or retain it in the short-term layer for manual review according to the scoring results. For example, for newly added knowledge from customer service chat records, if its semantic context is ambiguous or the theme deviates, the system will temporarily store it in the short-term memory layer and mark it as the "to be verified" state, and it will only be promoted to a formal knowledge node after passing semantic verification or manual confirmation.

[0077] In some embodiments, the system can also model and predict the knowledge evolution trend by statistically analyzing the update frequency and coverage of the second knowledge set. The system can input the update records in the multi-layer memory structure into a time series analysis model to identify trend signals of theme growth, decline, or reactivation, and present them graphically on the knowledge visualization interface. For example, for the "Product Performance Optimization" theme that has been continuously growing in the past few months, the system can automatically label it as a highly active theme and preferentially display its latest knowledge nodes; for the "Old System Maintenance" theme with gradually decreasing updates, it can prompt the knowledge administrator to perform archiving or merging operations. Through this trend prediction and visualization mechanism, the system can provide a decision-making basis for enterprise knowledge planning and resource allocation.

[0078] According to embodiments of this application, the dynamic evolution and self-learning optimization of the knowledge management system are achieved through semantic processing, topic extraction, and hierarchical updating of multi-source text interaction data of enterprises. First, the system obtains multi-source corpora from authorized business systems and, combined with semantic parsing and syntactic analysis, transforms unstructured text into a set of structured semantic units, fundamentally improving the computability of the data and cross-system semantic consistency. Second, by performing topic identification and knowledge association extraction on the semantic unit set, the system can automatically discover and organize business topics, generating knowledge sets closely corresponding to the topics, thus overcoming the limitations of manual screening and annotation. Furthermore, the system performs topic clustering and topic creation based on semantic similarity and time weighting, enabling the knowledge system to continuously expand with the evolution of business content and semantics, possessing the ability to automatically generate new topics and adaptively integrate old topics. Finally, by introducing a dynamic update mechanism with a multi-layered memory structure, the system performs hierarchical storage and semantic association reconstruction of knowledge across multiple dimensions, including time, topic, and participating entities, ensuring not only the timeliness and accuracy of knowledge updates but also achieving cross-layer information linkage and global consistency. In summary, the embodiments of this application effectively solve the problems of delayed topic identification, fragmented knowledge, and untimely updates in existing knowledge management systems, thereby improving the structuring level, update efficiency, and intelligent evolution capabilities of enterprise knowledge.

[0079] The knowledge management method of this application will be specifically described below by way of preferred embodiments.

[0080] In the embodiments of this application, the step of updating the multi-layer memory structure in the knowledge management base based on the second target topic can be used to achieve adaptive synchronization and differentiated adjustment of knowledge content in different memory systems, so as to ensure the consistency and timeliness of the knowledge base during topic evolution. Specifically, the system can use the second target topic as an index entry point to retrieve multi-layer memory data related to that topic from the knowledge management base.

[0081] Furthermore, the system performs a semantic comparison between the extracted topic-related knowledge set and the currently generated second knowledge set to identify differences in content, structure, and time dimensions. The comparison process can be implemented using techniques such as vector similarity calculation, knowledge node weight analysis, or semantic consistency verification, thereby obtaining knowledge change parameters that reflect the trend of knowledge change. These parameters can include multi-dimensional features such as the proportion of newly added knowledge, the decay coefficient of old knowledge, semantic offset indicators, and node update frequency, used to describe the dynamic changes in knowledge content within the current topic. For example, when the system detects a large number of new knowledge nodes related to "user experience improvement" under the topic of "product optimization solutions," the knowledge change parameters will reflect an increase in the activity of knowledge related to this topic, prompting the system to increase the memory weight of this topic in subsequent updates.

[0082] In some embodiments, to further improve the stability and interpretability of updates, the system may introduce a weight balancing and version control mechanism when performing update operations. The weight balancing mechanism can automatically allocate weight ratios when merging old and new knowledge to avoid semantic drift caused by sudden knowledge changes; the version control mechanism generates a version identifier and change record after each update, enabling the system to revert to the knowledge state at any point in time when needed, thereby improving the reliability and controllability of the system in complex business environments.

[0083] According to the embodiments of this application, by extracting the topic-related knowledge set with the second target topic as the index, generating knowledge change parameters based on semantic comparison with the second knowledge set, and then updating the multi-layer memory structure in layers, the intelligent evolution and global consistency maintenance of knowledge content are realized, thereby improving the dynamic response capability and semantic adaptive performance of the knowledge management system.

[0084] In the embodiments of this application, a multi-layered memory structure can be used to manage and dynamically update knowledge content at different time scales, semantic levels, and participating subject dimensions, thereby achieving persistent storage and structured evolution of the knowledge system. Specifically, the multi-layered memory structure can consist of a short-term memory layer, a topic memory layer, and an individual memory layer, with each layer working collaboratively to ensure the integrity and consistency of knowledge in the dimensions of capture, aggregation, and association. Through this multi-layered memory architecture, the system can achieve full-lifecycle management of knowledge when facing massive, multi-source, and dynamically changing business knowledge, enabling the enterprise knowledge base to possess the characteristics of real-time learning, semantic aggregation, individual association, and global evolution.

[0085] The short-term memory layer is used to cache semantic units and corresponding topic information collected within a preset period, and is the knowledge buffer layer with the fastest response speed and the highest update frequency in the system. This layer can record newly generated or changed business corpora, including short-cycle semantic content such as real-time dialogues, task communication, meeting minutes, and business logs.

[0086] The topic memory layer can perform knowledge aggregation and semantic structuring, which is used to integrate knowledge nodes belonging to the same topic at a higher level. Each topic node can contain multiple semantically related knowledge units, conceptual entities and their interrelationships, forming a semantically networked knowledge representation structure.

[0087] The individual memory layer records the knowledge activities of various knowledge-participating entities (such as employees, teams, or departments) under different topics, reflecting the flow and distribution of knowledge contributions within the organization. This layer establishes a knowledge participation profile for each participating entity, including its participation frequency, topic coverage, knowledge contribution, and collaborative relationships with others. By analyzing these parameters, the system can generate a vector representation reflecting the individual's knowledge growth trajectory, supporting personalized recommendations, knowledge assessment, and experience transfer management.

[0088] In the embodiments of this application, in order to enable the system to achieve multi-level linkage and semantic self-optimization based on the characteristics of knowledge change, the memory structure at different levels can be adaptively adjusted during the evolution of the knowledge system in order to maintain the semantic consistency and dynamic balance of the knowledge management base.

[0089] Figure 3 The flowchart illustrates a method for updating the multilayer memory structure based on the knowledge change parameters according to some exemplary embodiments of this application.

[0090] like Figure 3 As shown, the method for updating the multi-layer memory structure based on the knowledge change parameters includes operations S310 to S340.

[0091] In operation S310, the knowledge change parameters are analyzed to obtain short-term change parameters, topic change parameters, and individual change parameters.

[0092] Specifically, the system can obtain a comprehensive set of parameters reflecting knowledge dynamics during the preceding comparison process, including indicators such as the proportion of new knowledge, decay rate, node similarity, and changes in participation. These indicators can then be categorized and mapped to different levels of update dimensions: short-term change parameters reflect the intensity of changes in the latest semantic units, topic change parameters describe the evolutionary trend of topic-level semantic aggregation, and individual change parameters quantify the contribution and activity of knowledge-participating entities. Through parameter decomposition, the system can accurately locate the sources of knowledge changes and achieve hierarchical update decisions.

[0093] In operation S320, the time window weight or cache validity period of the corresponding semantic unit in the short-term memory layer is updated according to the short-term change parameters.

[0094] The short-term memory layer, acting as a fast response layer for knowledge input, is responsible for caching active semantic units within a preset period. When new or revised knowledge is detected, the time window weight or cache validity period of the corresponding semantic unit can be adjusted according to the changing parameters.

[0095] For example, semantic units marked as newly added and with high freshness in the comparison results can have their time window weight increased and their cache validity extended, so that they remain searchable and reusable in the short term. For semantic units that exist in the topic-related knowledge set but are missing in the second knowledge set and whose recent access frequency has decreased, their cache validity can be shortened or they can be marked as pending archiving. For semantic units that have only undergone minor text rewriting and whose semantic similarity changes are below the threshold, the original window is maintained but the version pointer is recorded to facilitate the reference of the latest content in subsequent topic layers.

[0096] In operation S330, the weights of knowledge nodes corresponding to the second target topic in the topic memory layer are adjusted according to the topic change parameters, and the semantic association information between knowledge nodes is updated.

[0097] In response to the topic change parameters, the weights of knowledge nodes corresponding to the second target topic in the topic memory layer can be adjusted, and the semantic association information between knowledge nodes can be updated synchronously.

[0098] For example, when the comparison results show that the frequency of a certain type of knowledge in the second knowledge set increases and it is closer to the topic center vector, the weight of that type of node is increased; when a knowledge item is weakened or removed in the second knowledge set, the weight of the corresponding node is reduced and its semantic edge with other nodes is weakened; when a new co-occurrence relationship or cross-topic reference relationship is identified, the corresponding semantic association information is added or strengthened within or across topics; for nodes that only undergo expression replacement but whose semantic coverage does not change, the weight can be kept unchanged, and only the latest version reference it points to can be updated.

[0099] In operation S340, based on the individual change parameters, the participation frequency or knowledge contribution of each knowledge participating entity in the individual memory layer under the second target topic is updated.

[0100] For individual changes in parameters, the participation frequency or knowledge contribution of each knowledge-participating entity in the individual's memory layer under the second objective topic can be updated.

[0101] For example, when a newly added knowledge item in the second knowledge set can be attributed to a participating entity, the participation frequency count of that entity under that topic can be increased, and its knowledge contribution can be updated by weighting according to item quality, number of citations, or review status; when the comparison shows that an item maintained by a participating entity is merged or downweighted, its contribution can be reduced accordingly or its participation frequency can be maintained but its weight adjusted; when an item is generated by multiple people in collaboration, the increment is distributed among the relevant entities according to the contribution slices formed in the comparison stage (such as submission ratio, revision ratio, and verification pass ratio).

[0102] For example, compared to the topic-related knowledge set, the second knowledge set adds two new categories of items: "Feedback and Follow-up Script Specifications" and "Abnormal Closed-Loop Time Limits," while weakening the "Single Compensation Standard" item. Based on this, the system generates knowledge change parameters: marking the two categories of items as newly added and highly recent, and marking the weakened items as having their weight reduced; the short-term memory layer increases the weight of the time window for newly added items and extends the cache validity period; the topic memory layer increases the weight of nodes related to "Follow-up Process" and "Closed-Loop Management," strengthening their semantic edge with "Satisfaction Assessment," while simultaneously applying weight decay to the "Compensation Policy" node; in the individual memory layer, entities contributing to the aforementioned newly added items have their participation frequency and contribution increased accordingly, while entities only maintaining weakened items have their contribution updated according to the decay coefficient and retain the version evolution pointer.

[0103] For example, the comparison results show that the second knowledge set deleted several outdated operation guidelines and merged two contents into a unified entry, and the number of visits has dropped significantly in the past week. The system generates "missing itemset", "merging flag" and "access decline rate" in the knowledge change parameters. The short-term memory layer shortens the cache validity period of the relevant semantic units and sets a higher window weight for the merged new entry accordingly; the topic memory layer performs redirection of the merged nodes to the unified node, reduces the weight of the replaced nodes and removes redundant semantic edges; the individual memory layer does not reduce the historical participation frequency but adjusts the contribution distribution of the merged entries, so that they are aggregated to the latest responsible subject of the unified entry, ensuring that the statistical caliber remains consistent after the change.

[0104] According to the embodiments of this application, a multi-layer memory structure update method based on knowledge change parameters enables the knowledge base to adaptively reflect semantic dynamics and business evolution, achieving high-precision, low-latency automated knowledge management and ensuring the consistency of data at each layer and the traceability of update basis.

[0105] In some embodiments, after updating the short-term memory layer, thematic memory layer, and individual memory layer, the system can also perform a reconstruction operation of the semantic associations between layers to ensure that the multi-layer memory structure maintains consistency and coherence between the semantic level and the entity level.

[0106] Specifically, after each layer is updated, the topic correspondence and semantic association information between different memory layers can be extracted first to form an inter-layer association dataset. This dataset can include the mapping relationship between newly generated or updated semantic units in the short-term layer and corresponding knowledge nodes in the topic layer, as well as the semantic link information between knowledge nodes in the topic layer and knowledge-participating entities in the individual layer. The system can identify the association strength between different layers through semantic embedding models, association path statistics, or co-occurrence probability calculations, thereby generating a multi-dimensional association matrix in the semantic space to reflect the topic consistency and participation dependency between layers.

[0107] After generating the inter-layer association dataset, the system can further reconstruct the inter-layer association index of the multi-layer memory structure. The inter-layer association index can be used to record semantic mappings and entity dependencies between different memory layers. During reconstruction, the system can clean up invalid semantic links in the old index based on the association weights in the inter-layer association dataset and write newly identified cross-layer association information into the index table. For example, when a new semantic unit "product defect feedback" is added to the short-term memory layer and has a high semantic relevance to the "quality improvement" node in the topic layer, the system will automatically establish a new cross-layer link in the index and supplement the corresponding knowledge association information of the participating employees in the individual memory layer. If there are redundant or duplicate relational paths in the original index, the system will merge them through weight fusion or path compression strategies to improve retrieval efficiency and the compactness of the data structure.

[0108] In another embodiment, to improve the dynamic availability of inter-layer indexes, an inter-layer consistency verification mechanism can be introduced during the reconstruction process. Inter-layer consistency verification checks the consistency of the number of topics, nodes, and semantic relevance coefficients between different layer indexes, ensuring that the reconstructed index structure is logically sound and free of data conflicts. When an update is detected for a topic in the short-term layer but no corresponding node has been formed in the topic layer, the system automatically triggers delayed updates or supplementary chain building operations to maintain the integrity of inter-layer semantics. Furthermore, the system can combine timestamps and version numbers to generate a unique version identifier for each index reconstruction, recording the update time and scope of changes, facilitating subsequent source tracing analysis and rollback operations.

[0109] In some embodiments, after index reconstruction is completed, the system can write the updated inter-layer association index to the persistent storage area of ​​the knowledge management repository. The write operation can be performed in batch synchronization or incremental update mode, and the index loading strategy can be set according to business priority. When a user or other system module initiates a knowledge retrieval request, the system can realize cross-layer query through the new index. For example, it can quickly locate relevant topic knowledge nodes from individual participation records, and then trace back to the latest semantic unit of the shortest layer, realizing full-link information tracking from people to knowledge and from events to topics.

[0110] Through the dynamic reconstruction and writing mechanism of inter-layer indexes, the embodiments of this application ensure the global consistency and retrieval of the multi-layer memory structure after the update, enabling the knowledge base to continuously reflect the semantic evolution state and support efficient knowledge retrieval and intelligent reasoning.

[0111] In the embodiments of this application, in order to transform diverse text data from various sources and formats within an enterprise into a structured set of semantic units, and to provide a unified and high-quality semantic input for subsequent topic recognition and knowledge aggregation, the following method is also provided.

[0112] Figure 4 The illustration shows a schematic diagram of a method for semantic parsing of multi-source text interaction data according to some exemplary embodiments of this application.

[0113] like Figure 4 As shown, the method for semantic parsing of multi-source text interaction data may include operations S410 to S430.

[0114] In operation S410, the multi-source text interaction data is segmented and subjected to dependency parsing to extract entity, event, and attribute information.

[0115] Specifically, the input raw text can be segmented and preprocessed syntactically. Sentence segmentation algorithms divide complex long sentences into logically complete semantic segments, and then dependency parsing models are used to identify subject-verb-object relationships, modification structures, and subordination relationships within the sentences. The generation of dependency trees enables the system to accurately extract core semantic units from the text, such as key entities and events like "customer," "complaint," "product," and "delayed delivery" in the phrase "customer complaint—delayed product delivery." Simultaneously, the system can combine named entity recognition technology to identify elements such as the organization, location, time, and business metrics involved, and extract their semantic attributes, such as quantitative features or timestamp information.

[0116] In operation S420, the entity, event, and attribute information are input into the semantic embedding model to generate a semantic unit vector representation.

[0117] The semantic embedding model can employ a large language model based on the Transformer architecture or an enterprise-customized semantic encoding model to map language units into a high-dimensional vector space, thereby capturing deep semantic relationships. Corpora from different sources (such as customer service records, project reports, work order comments, email summaries, etc.) can be uniformly represented as semantic vectors after semantic encoding, thus achieving cross-domain semantic alignment.

[0118] Furthermore, corresponding sub-models can be invoked based on the data type, such as using a dialogue embedding model for customer service corpora and a document embedding model for business reports, to enhance the context awareness capability of semantic understanding.

[0119] In operation S430, the semantic unit vector representations are aggregated based on vector similarity to form the semantic unit set.

[0120] Specifically, the semantic similarity of different semantic vectors can be determined by calculating the cosine similarity or Euclidean distance between semantic units. The system can use clustering algorithms to group vectors with high semantic similarity into the same group of semantic units, forming a cohesive semantic set structure. For corpora with ambiguous semantic boundaries or significant contextual changes, the system can use soft clustering or attention-weighted strategies to ensure that the aggregation results balance semantic accuracy and coverage. For example, when the system processes three types of statements simultaneously: "refund application delay," "order processing delay," and "delivery delay," the semantic aggregation module can aggregate them into the same semantic unit group, "process delay issues," so that they can be quickly assigned to relevant business themes during subsequent topic identification.

[0121] In some embodiments, semantic aggregation can also combine timestamps and source tag information to perform weighted aggregation of semantic units from different sources or periods, reflecting semantic evolution trends. The system can adjust the semantic aggregation weights based on time decay factors or source credibility parameters, giving higher weight to recently generated data from authoritative sources in semantic clustering. For example, the latest project reporting data from core business systems will preferentially influence the clustering results, thereby improving the system's responsiveness to emerging business semantics. Furthermore, the system can establish a preliminary semantic index through the aggregated semantic unit set for subsequent use by the topic recognition module, enabling rapid retrieval and incremental updates of semantic information.

[0122] According to embodiments of this application, by combining sentence segmentation analysis and dependency syntax trees, the system can accurately extract semantic hierarchical structures and improve the fine-grainedness of information parsing; the introduction of semantic embedding models enables the system to have cross-contextual semantic understanding capabilities, realizing the representation and fusion of multi-source corpora in a unified semantic space; through a semantic aggregation mechanism based on vector similarity, the system can automatically identify semantically similar content and generate a high-quality set of semantic units, reducing redundancy and semantic conflicts.

[0123] In some embodiments, topic clustering or topic creation can be performed based on a first target topic and a set of historical topics to obtain a second target topic, thereby enabling dynamic adaptive evolution of the topic hierarchy in the enterprise knowledge system. This allows the system to automatically determine whether newly identified topics should be included in existing topic clusters or whether new topic nodes should be created based on semantic similarity, thereby maintaining the continuous updating and semantic consistency of the topic structure in the knowledge management base.

[0124] In some embodiments, the system first calculates the semantic similarity between the first target topic and the set of historical topics. To this end, the system can utilize a semantic embedding model or a topic representation vector model to map each topic into a high-dimensional vector representation. Specifically, the system generates a semantic vector for the first target topic based on the keywords, core entities, and their contextual semantic relationships contained in the aforementioned set of semantic units, and extracts the vector representations of each historical topic from the set of historical topics. Subsequently, the system obtains the semantic similarity value between the first target topic and each historical topic by calculating cosine similarity or Euclidean distance.

[0125] When the semantic similarity is greater than or equal to a preset threshold, the system determines that the first target topic and the existing historical topic cluster belong to the same semantic domain, and thus merges the first target topic into the corresponding historical topic cluster. At this time, the system can update the cluster center node of the historical topic cluster based on a weighted average or center vector adjustment. For example, when the similarity of the topic "customer complaint handling process optimization" is higher than the threshold and it is merged into the topic cluster "customer service improvement", the center node of the topic cluster can be recalculated based on the weighted fusion result of the semantic vectors of the two, and the updated cluster center node can be determined as the second target topic.

[0126] When the semantic similarity is less than a preset threshold, the system assumes that there is no corresponding semantic cluster for the topic in the existing knowledge structure. Therefore, it creates a new target topic node and designates this node as the second target topic. The attributes of the newly created node can inherit the semantic features of the first target topic, including core keywords, typical sentences, and related corpus identifiers. At the same time, it exists as an independent node in the topic layer structure.

[0127] In some embodiments, the system can also record corresponding source identifiers and timestamp information for the second target topic based on the results of topic clustering or creation, for tracking the topic's evolution path and lifecycle management. Furthermore, to prevent topic clusters from overexpanding or cluster drifting, the system can perform topic integration and regrouping operations during periodic clustering. When the similarity of multiple topic clusters consistently exceeds a set threshold, they are merged into higher-level topic nodes; when the content divergence within a topic cluster is significant, it is automatically split into multiple sub-topic nodes, thereby maintaining the semantic clarity and hierarchical balance of the knowledge structure.

[0128] Through the aforementioned topic clustering and topic creation mechanisms, the embodiments of this application achieve adaptive dynamic evolution of the enterprise knowledge topic structure. By calculating semantic similarity and determining thresholds, the system can automatically distinguish semantic relationships between topics, avoiding subjective biases caused by manual classification. Based on the dynamic adjustment mechanism of the cluster center node, knowledge topic clusters can continuously evolve under the impetus of new knowledge, maintaining the accuracy and stability of topic boundaries. By automatically creating new topic nodes, the system can absorb emerging business semantics and concepts in real time, ensuring the timeliness and scalability of the knowledge management base. This provides a foundation for the long-term maintenance and intelligent evolution of the enterprise knowledge graph, enabling the knowledge system to possess self-organizing, self-updating, and sustainable learning capabilities.

[0129] According to embodiments of this application, by constructing a multi-layer semantic parsing and dynamic memory update framework, efficient transformation from multi-source unstructured data to structured semantic knowledge is achieved, significantly improving the automation level of enterprise knowledge acquisition, extraction, and parsing. The system introduces sentence segmentation parsing, syntactic dependency, and semantic embedding technologies at the semantic level, enabling accurate extraction of entity, event, and attribute information, and achieving unified expression of cross-source corpora in the semantic space. Based on semantic clustering and topic self-creation mechanisms, the system can adaptively identify the evolution trend of business semantics, dynamically aggregate newly emerging semantic topics, or add new nodes, achieving continuous optimization and expansion of the knowledge topic structure. Through a multi-layer memory structure and hierarchical update mechanism, the system supports real-time incremental updates in the time dimension, maintains clustering consistency in the topic dimension, and records knowledge contribution relationships in the participating entity dimension, thereby achieving multi-dimensional evolvable management of the knowledge system. Furthermore, through the reconstruction of inter-layer indexes and cross-layer semantic association mechanisms, the system can establish semantic mappings between different memory layers, achieving full-link retrieval and visual tracking from semantic units to topics and from topics to entities. Overall, the embodiments of this application effectively improve the semantic understanding capability, topic adaptability, and knowledge update efficiency of enterprise knowledge management systems, and construct an intelligent knowledge network that can learn sustainably and evolve through self-organization.

[0130] Corresponding to the above-described knowledge management method, embodiments of this application also provide a knowledge management device.

[0131] Figure 5 A schematic block diagram of a knowledge management device according to an embodiment of this application is shown.

[0132] like Figure 5 As shown, the knowledge management device 500 in this embodiment includes a data acquisition module 510, a topic identification module 520, a clustering or creation module 530, and a knowledge update module 540.

[0133] The data acquisition module 510 can be used to acquire authorized multi-source text interaction data, perform semantic parsing on the multi-source text interaction data, and form a set of semantic units. In one embodiment, the data acquisition module 510 can be used to perform the operation S210 described above, which will not be repeated here.

[0134] The topic identification module 520 can be used to identify topics in the semantic unit set, obtain a first target topic, and acquire a first knowledge set corresponding to the first target topic. In one embodiment, the topic identification module 520 can be used to perform the operation S220 described above, which will not be repeated here.

[0135] The clustering or creation module 530 can be used to perform topic clustering or topic creation based on the first target topic and the historical topic set to obtain a second target topic, and to determine a second knowledge set based on the first knowledge set and the second target topic. In one embodiment, the clustering or creation module 530 can be used to perform the operation S230 described above, which will not be repeated here.

[0136] The knowledge update module 540 can be used to update the multi-layer memory structure in the knowledge management base based on the second target topic and the second knowledge set. The multi-layer memory structure is used to store the knowledge coverage information of each knowledge participating entity under different topics. In one embodiment, the knowledge update module 540 can be used to perform the operation S240 described above, which will not be repeated here.

[0137] According to an embodiment of this application, the knowledge update module 540 can also be used to extract the topic-related knowledge set corresponding to the second target topic from the multi-layer memory structure using the second target topic as an index; compare the topic-related knowledge set with the second knowledge set to obtain knowledge change parameters; and update the multi-layer memory structure based on the knowledge change parameters. The multi-layer memory structure includes: a short-term memory layer for caching semantic units and corresponding topics collected within a preset period; a topic memory layer for aggregating knowledge nodes under the same topic; and an individual memory layer for recording the participation frequency and knowledge contribution of each knowledge participating entity under different topics.

[0138] According to an embodiment of this application, the knowledge update module 540 can also be used to parse the knowledge change parameters to obtain short-term change parameters, topic change parameters, and individual change parameters; update the time window weight or cache validity period of the corresponding semantic unit in the short-term memory layer according to the short-term change parameters; adjust the weight of the knowledge node corresponding to the second target topic in the topic memory layer according to the topic change parameters, and update the semantic association information between knowledge nodes; and update the participation frequency or knowledge contribution of each knowledge participating entity in the individual memory layer under the second target topic according to the individual change parameters.

[0139] According to an embodiment of this application, the knowledge update module 540 can also be used to extract the topic correspondence and semantic association information between different layers to generate an inter-layer association dataset; and based on the inter-layer association dataset, to reconstruct the inter-layer association index of the multi-layer memory structure, output the updated inter-layer association index, and write it into the knowledge management library.

[0140] According to an embodiment of this application, the data acquisition module 510 can also be used to perform sentence segmentation and dependency parsing on the multi-source text interaction data to extract entity, event and attribute information; input the entity, event and attribute information into a semantic embedding model to generate semantic unit vector representations; and aggregate the semantic unit vector representations based on vector similarity to form the semantic unit set.

[0141] According to an embodiment of this application, the clustering or creation module 530 can also be used to calculate the semantic similarity between the first target topic and the historical topic set; when the semantic similarity is greater than or equal to a preset threshold, the first target topic is merged into the corresponding historical topic cluster, and the cluster center node of the historical topic cluster is determined as the second target topic; when the semantic similarity is less than the preset threshold, a target topic node is created, and the target topic node is determined as the second target topic.

[0142] According to embodiments of this application, any multiple modules among the data acquisition module 510, topic identification module 520, clustering or creation module 530, and knowledge update module 540 can be merged into one module, or any one of these modules can be split into multiple modules. Alternatively, at least some of the functions of one or more of these modules can be combined with at least some of the functions of other modules and implemented in one module. According to embodiments of this application, at least one of the data acquisition module 510, topic identification module 520, clustering or creation module 530, and knowledge update module 540 can be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or any other reasonable means of integrating or packaging circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three implementation methods. Alternatively, at least one of the data acquisition module 510, topic identification module 520, clustering or creation module 530, and knowledge update module 540 may be implemented at least partially as a computer program module, which can perform corresponding functions when the computer program module is run.

[0143] Figure 6A block diagram schematically illustrates an electronic device suitable for implementing a knowledge management method according to an embodiment of this application.

[0144] like Figure 6 As shown, an electronic device 600 according to an embodiment of this application includes a processor 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage portion 608 into a random access memory (RAM) 603. The processor 601 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 601 may also include onboard memory for caching purposes. The processor 601 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of this application.

[0145] RAM 603 stores various programs and data required for the operation of electronic device 600. Processor 601, ROM 602, and RAM 603 are interconnected via bus 604. Processor 601 executes various operations of the method flow according to embodiments of this application by executing programs in ROM 602 and / or RAM 603. It should be noted that the programs may also be stored in one or more memories other than ROM 602 and RAM 603. Processor 601 may also execute various operations of the method flow according to embodiments of this application by executing programs stored in said one or more memories.

[0146] According to embodiments of this application, the electronic device 600 may further include an input / output (I / O) interface 605, which is also connected to a bus 604. The electronic device 600 may also include one or more of the following components connected to the input / output (I / O) interface 605: an input section 606 including a keyboard, mouse, etc.; an output section 607 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the input / output (I / O) interface 605 as needed. A removable medium 611, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 610 as needed so that computer programs read from it can be installed into the storage section 608 as needed.

[0147] This application also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0148] According to embodiments of this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium, such as including but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this application, the computer-readable storage medium may include ROM 602 and / or RAM 603 and / or one or more memories other than ROM 602 and RAM 603 described above.

[0149] Embodiments of this application also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code enables the computer system to implement the knowledge management method provided in the embodiments of this application.

[0150] When the computer program is executed by the processor 601, it performs the functions defined in the system / apparatus of this application embodiment. According to the embodiments of this application, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0151] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and downloaded and installed via the communication section 609, and / or installed from the removable medium 611. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0152] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 609, and / or installed from the removable medium 611. When the computer program is executed by the processor 601, it performs the functions defined in the system of this application embodiment. According to the embodiments of this application, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0153] According to embodiments of this application, program code for executing the computer programs provided in the embodiments of this application can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0154] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0155] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application.

[0156] The embodiments of this application have been described above. However, these embodiments are merely illustrative and not intended to limit the scope of this application. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. Without departing from the scope of this application, those skilled in the art can make various substitutions and modifications, all of which should fall within the scope of this application.

Claims

1. A knowledge management method characterized by, The method comprises: acquiring authorized multi-source text interaction data, performing semantic parsing on the multi-source text interaction data to form a semantic unit set; performing topic identification on the semantic unit set to obtain a first target topic, and acquiring a first knowledge set corresponding to the first target topic; performing topic clustering or topic creation based on the first target topic and a historical topic set to obtain a second target topic, and determining a second knowledge set based on the first knowledge set and the second target topic; and updating a multi-layer memory structure in a knowledge management library based on the second target topic and the second knowledge set, the multi-layer memory structure being used to store knowledge coverage information of each knowledge participating entity under different topics.

2. The method of claim 1, wherein, The updating of the multi-layer memory structure in the knowledge management library based on the second target topic comprises: extracting a topic-related knowledge set corresponding to the second target topic from the multi-layer memory structure with the second target topic as an index; comparing the topic-related knowledge set with the second knowledge set to obtain a knowledge change parameter; and updating the multi-layer memory structure based on the knowledge change parameter.

3. The method of claim 2, wherein, The multi-layer memory structure comprises: a short-term memory layer used to cache semantic units and corresponding topics collected within a preset period; a topic memory layer used to aggregate knowledge nodes under the same topic; an individual memory layer used to record the participation frequency and knowledge contribution degree of each knowledge participating entity under different topics.

4. The method according to claim 2 or 3, characterized in that, The updating of the multi-layer memory structure based on the knowledge change parameter comprises: parsing the knowledge change parameter to obtain a short-term change parameter, a topic change parameter and an individual change parameter; updating the time window weight or the cache validity period of the corresponding semantic unit in the short-term memory layer according to the short-term change parameter; adjusting the knowledge node weight corresponding to the second target topic in the topic memory layer according to the topic change parameter, and updating the semantic association information between knowledge nodes; and updating the participation frequency or the knowledge contribution degree of each knowledge participating entity under the second target topic in the individual memory layer according to the individual change parameter.

5. The method of claim 4, wherein, After the updating of the short-term memory layer, the topic memory layer and the individual memory layer is completed, the method further comprises: extracting the topic correspondence relationship and the semantic association information between different layers to generate an inter-layer association data set; and reconstructing the inter-layer association index of the multi-layer memory structure based on the inter-layer association data set, outputting the updated inter-layer association index and writing it into the knowledge management library.

6. The method of claim 1, wherein, The semantic parsing on the multi-source text interaction data to form a semantic unit set comprises: performing sentence segmentation and dependency syntax analysis on the multi-source text interaction data to extract entity, event and attribute information; inputting the entity, event and attribute information into a semantic embedding model to generate semantic unit vector representation; and aggregating the semantic unit vector representation based on vector similarity to form the semantic unit set.

7. The method of claim 1, wherein, The obtaining of the second target topic based on the first target topic and the historical topic set comprises: calculating semantic similarity of the first target topic and the set of historical topics; when the semantic similarity is greater than or equal to a preset threshold, incorporating the first target topic into a corresponding historical topic cluster, and determining a cluster center node of the historical topic cluster as the second target topic; when the semantic similarity is less than the preset threshold, creating a target topic node, and determining the target topic node as the second target topic.

8. A knowledge management apparatus characterized by comprising: The apparatus comprises: a data acquisition module configured to acquire authorized multi-source text interaction data, and perform semantic analysis on the multi-source text interaction data to form a set of semantic units; a topic identification module configured to perform topic identification on the set of semantic units to obtain a first target topic, and acquire a first knowledge set corresponding to the first target topic; a clustering or creating module configured to perform topic clustering or topic creation based on the first target topic and a set of historical topics to obtain a second target topic, and determine a second knowledge set based on the first knowledge set and the second target topic; and a knowledge updating module configured to update a multi-layer memory structure in a knowledge management library based on the second target topic and the second knowledge set, the multi-layer memory structure being configured to store knowledge coverage information of each knowledge participating entity under different topics. 9.An electronic device comprising: one or more processors; a memory for storing one or more computer programs, characterized in that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1-7.

10. A computer readable storage medium having stored thereon a computer program or instructions, characterized in that, The computer program or instructions, when executed by the processor, implement the steps of the method according to any one of claims 1-7.

11. A computer program product comprising computer programs or instructions, characterized in that, The computer program or instructions, when executed by the processor, implement the steps of the method according to any one of claims 1-7.

Images