Method for building enterprise wisdom knowledge management system based on artificial intelligence

Abstract

The invention discloses a method for constructing an enterprise intelligent knowledge management system based on artificial intelligence, and the method comprises the following steps: P1, constructing a knowledge vectorizing and storing business knowledge documents; p2, a knowledge base assistant system is constructed, IT operation and maintenance assistants oriented to the whole staff and professional assistants oriented to all business lines are constructed according to the enterprise informatization level and requirements, and the collective IT operation and maintenance, manpower, finance, internal control, law affairs, marketing and information key business fields are comprehensively covered; p3, knowledge base retrieval and construction: relevant retrieval is performed through a knowledge base assistant system and a localized knowledge base in a question and answer mode, a huge and rich knowledge treasure library can be constructed for an enterprise through the enterprise intelligent knowledge management system constructed by the invention, and the enterprise intelligent knowledge management system can be used for constructing a knowledge treasure library by closely tracking user feedback and implementing comprehensive evaluation, so that the enterprise intelligent knowledge management system is more intelligent and reliable. The accuracy of the knowledge assistants is stably kept at 85% or above, excellent service experience is provided for the user, and the improvement of the working efficiency is effectively promoted.

Description

Technical Field

[0001] The present invention relates to the technical field of enterprise intelligent knowledge management, and specifically to a method for building an enterprise intelligent knowledge management system based on artificial intelligence. Background Art

[0002] At present, with the rapid development of general artificial intelligence technology, generative large-scale models have shown unparalleled advantages in promoting profound changes in interaction methods and efficient processing and intelligent learning of unstructured documents. For large-scale and extensive group companies, with the continuous expansion and continuous operation of their businesses, as well as the increasingly accelerated pace of informatization construction, these companies will undoubtedly accumulate a large amount of unstructured document resources. These documents are usually stored in the enterprise network disk.

[0003] However, in the two key links of document retrieval and efficient use of knowledge, there are major technical bottlenecks and application barriers that cannot be ignored. These barriers not only seriously restrict enterprises from in-depth mining and effective use of existing document resources, but also make it impossible to fully release and utilize the valuable knowledge value contained in the company's stock documents. Therefore, building an enterprise-level knowledge system and providing convenient document retrieval applications have become important challenges for the digital and intelligent transformation of enterprises. To this end, we propose a method for building an enterprise intelligent knowledge management system based on artificial intelligence. Summary of the invention

[0004] The purpose of the present invention is to provide a method for building an enterprise intelligent knowledge management system based on artificial intelligence to solve the problems raised in the above background technology.

[0005] To achieve the above purpose, the present invention provides the following technical solution: a method for building an enterprise intelligent knowledge management system based on artificial intelligence, the method comprising constructing the following modules:

[0006] P1: Knowledge base construction, vectorizing and storing business knowledge documents;

[0007] P2: Construction of a knowledge base assistant system. Based on the enterprise's information level and needs, we build IT operation and maintenance assistants for all employees and professional assistants for each business line, covering the group's key business areas of IT operation and maintenance, human resources, finance, internal control, legal affairs, marketing, and information.

[0008] P3: Knowledge base retrieval construction, through the knowledge base assistant system and the localized knowledge base to conduct relevant retrieval through question and answer.

[0009] Preferably, the knowledge base construction comprises the following steps:

[0010] S01: Knowledge preprocessing, including reading, identifying, parsing, slicing and understanding business knowledge documents;

[0011] S02: Text vectorization, the Embedding model was deployed locally, and the model was adapted and fine-tuned;

[0012] S03: Vector storage: A weaviate vector database is built locally to store the text vectors calculated by the Embedding model.

[0013] Preferably, for the reading and recognition in S01, an OCR recognition component is deployed to support the recognition and analysis of common documents and pictures.

[0014] Preferably, the S01 supports following up the cloud document directory for layer-by-layer file search and automatic import, and sets the batch for automatic update. The user only needs to set a cloud document path and the frequency of automatic update, and the document can be automatically read and updated later, which is in line with the company's existing document cloud storage method.

[0015] Preferably, multiple segmentation rules are customized according to the size of custom blocks, special characters, and natural paragraphs. These rules can better segment the document and maintain the semantic integrity to a greater extent, thus providing a basis for the subsequent Embedding model to understand and vectorize the document.

[0016] Preferably, a management function is provided for the segmented text to support the administrator to adjust and modify the text blocks.

[0017] Preferably, a "1+N" intelligent assistant system is introduced in the construction of the knowledge base assistant system.

[0018] Preferably, the front end of the knowledge assistant system supports multiple entrances such as PC web pages, WeChat applets, and H5 applications, and accesses the application layer through RestAPI.

[0019] Preferably, the knowledge base retrieval construction comprises the following steps:

[0020] S04: The user raises a question. The user asks his or her own question in the knowledge assistant. At this time, the first permission check will be performed, that is, the permission check between the user and the knowledge assistant application. If the permission check passes, the question and the current user information will be passed to the Embedding model for the next step of processing;

[0021] S05: Question vectorization. After receiving the question from the knowledge assistant, the Embedding model first vectorizes the question and unifies the question and knowledge into the same dimension of "vector".

[0022] S06: Vector search: The Embedding model searches the vectorized question in the local vector library. After the search results are obtained, a second permission check is performed, that is, the permission check of the file associated with the search result. If the check passes, the search result fragment is retained. If the check fails, the fragment is discarded.

[0023] S07: Returns the TopK results. The Embedding model returns the search results after permission screening to the Rerank model. The model is locally deployed and has undergone parameter adjustment to achieve a certain balance between result hit rate and accuracy. The Rerank model sorts and organizes the results and returns them to the knowledge base assistant.

[0024] S08: Merge Prompt statements. A Prompt project is configured for each knowledge base assistant. The project sets the role, function, and restriction conditions of the assistant. When the result of the Rerank model is obtained, the Prompt statement and structure will be returned to the large language model.

[0025] S09: The large language model returns the result. The large language model can be deployed locally. At the same time, the present invention also supports the basic configuration and management switching functions of the large language model to meet the usage requirements of different security levels. After the large language model obtains the prompt sentence and the search results provided by the Rerank model, the information is integrated and returned to the knowledge assistant in the best interactive way;

[0026] S10: Answer the question. After double permission verification, Embedding model, optimized local vector library retrieval, Rerank model result processing and large language model integration optimization, the user is finally presented with an answer that meets the user's permissions and is relatively satisfactory.

[0027] Compared with the prior art, the present invention has the following beneficial effects:

[0028] 1. The enterprise intelligent knowledge management system constructed by the present invention can build a huge and rich knowledge treasure house for the enterprise. With their accurate response and efficient answering capabilities, they play an indispensable role in supporting the operation of business lines. By closely tracking user feedback and implementing comprehensive evaluations, the accuracy of these knowledge assistants is stably maintained at more than 85%, providing users with an excellent service experience and effectively promoting the improvement of work efficiency.

[0029] 2. In the field of operation and maintenance, the IT operation and maintenance assistant under the intelligent knowledge management system built by the present invention has achieved significant optimization of manpower input. In terms of IT operation and maintenance, the traditional "group-sector-company" multi-level hierarchical operation and maintenance model has led to a large scale of the operation support team, and the response and resolution efficiency is low. After the IT operation and maintenance assistant went online, it not only simplified the complex operation and maintenance hierarchy, but also realized the intelligent upgrade of operation and maintenance work. According to statistics, the assistant can directly solve and filter out at least 60% of daily operation and maintenance problems, greatly reducing the manual burden and improving overall operational efficiency.

[0030] 3. In the field of information security, the present invention avoids the risk of data leakage during cloud transmission and storage by storing data on local devices. At the same time, local storage also means that users do not need to rely on external cloud services, reducing the possibility of data loss and service interruption due to cloud service failures. In addition, users can customize data access and permission control according to their own needs and security policies, further enhancing data security. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 This is a schematic diagram of the enterprise intelligent knowledge management system and construction method architecture implemented by the present invention. DETAILED DESCRIPTION

[0032] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0033] See also Figure 1 , which is the enterprise intelligent knowledge management system and construction method architecture implemented by the present invention. It is based on the generative big model and the enterprise's knowledge documents, and can provide various knowledge assistants for all employees and professional lines. The method includes three parts: P1 knowledge base construction, P2 knowledge base assistant system and P3 knowledge base retrieval.

[0034] P1: Knowledge base construction. This part mainly vectorizes and stores business knowledge documents. The main steps are as follows:

[0035] S01: Knowledge preprocessing. This step mainly includes reading and identifying business knowledge documents, followed by parsing, slicing and understanding. For reading and identification, we deployed OCR recognition components to support the recognition and parsing of common documents and images. At the same time, we not only support the uploading and updating of local files, but also support following the cloud document directory for layer-by-layer file search and automatic import in order to match the company's existing document cloud storage methods, and set up automatic update batches. Users only need to set a cloud document path and the frequency of automatic updates, and then they can automatically read and update documents.

[0036] In terms of document segmentation, we have customized and implemented a variety of segmentation rules, such as by custom block size, by special characters (carriage return, @, etc.), by natural paragraphs, etc. These rules can better segment documents and maintain semantic integrity to a greater extent, providing a basis for the subsequent Embedding model to understand and vectorize documents. At the same time, for the segmented text, we provide management functions to support administrators to adjust and modify text blocks;

[0037] S02: Text vectorization, the Embedding model is deployed locally, and the model is adapted and fine-tuned. This step is to import the segmented text into the local Embedding model, and the model will vectorize the segmented text;

[0038] S03: Vector storage: We built a weaviate vector database locally to store the text vectors calculated by the Embedding model. For the local vector database, we optimized the index creation to ensure retrieval efficiency.

[0039] P2: Knowledge base assistant system. This part is mainly to build the enterprise's knowledge assistant system. According to the enterprise's information level and needs, we have built IT operation and maintenance assistants for all employees and professional assistants for various business lines, covering the group's IT operation and maintenance, human resources, finance, internal control, legal affairs, marketing, information and other key business areas, and ensuring the comprehensiveness and timeliness of knowledge, providing an efficient and accurate knowledge acquisition platform for internal users of the group.

[0040] On this basis, we have introduced the "1+N" intelligent assistant system, which has achieved a double leap in interactive experience and user needs. "1" represents a comprehensive intelligent assistant, which is like an omniscient guide, able to provide users with one-stop knowledge query services across different business fields. Users only need to enter their query requirements to quickly obtain relevant information from various knowledge bases, which greatly simplifies the information retrieval process and improves work efficiency.

[0041] "N" represents independent intelligent assistants in multiple professional fields. They each focus on a specific business field and are deeply integrated with the corresponding business systems. These professional assistants not only have a deep understanding of the knowledge in their respective fields, but can also provide more accurate and in-depth search services based on the specific needs of users. For professional line personnel, these assistants are undoubtedly their right-hand men in their work, helping them quickly solve business problems and promote work progress.

[0042] At the front end of the knowledge assistant system, it supports multiple entrances such as PC web pages, WeChat applets, H5 applications, etc., and accesses the application layer through RestAPI. Among them, H5 applications can better adapt to the internal office system and mobile applications of the enterprise.

[0043] P3: Knowledge base search. This part mainly involves users using the knowledge base assistant system and the localized knowledge base to conduct relevant searches in the form of questions and answers. The main steps are as follows:

[0044] S04: The user raises a question. The user raises his or her own question in the knowledge assistant. At this time, the first permission check will be performed, that is, the permission check between the user and the knowledge assistant application. If the permission check passes, the question and the current user information will be passed to the Embedding model for the next step of processing;

[0045] S05: Question vectorization: After receiving the question from the knowledge assistant, the Embedding model first vectorizes the question and unifies the question and knowledge into the same dimension of "vector".

[0046] S06: Vector search: The Embedding model searches the vectorized question in the local vector library. After the search results are obtained, a second permission check is performed, that is, the permission check of the file associated with the search result by the user. If the check passes, the search result fragment is retained. If the check fails, the fragment is discarded.

[0047] S07: Return the TopK results. The Embedding model returns the search results after permission screening to the Rerank model. The model is locally deployed and has been parameter adjusted to achieve a certain balance between result hit rate and accuracy. The Rerank model sorts and organizes the results and returns them to the knowledge base assistant.

[0048] S08: Merge Prompt statements. We configure a Prompt project for each knowledge base assistant. The project sets the assistant's role, function, and restrictions. When the result of the Rerank model is obtained, the Prompt statement and structure will be returned to the large language model.

[0049] S09: The large language model returns the result. Here, the large language model can be deployed locally. At the same time, the present invention also supports the basic configuration and management switching of the large language model to meet the usage requirements of different security levels. After the large language model obtains the prompt sentence and the search results provided by the Rerank model, it integrates this information and returns it to the knowledge assistant in the best interactive way;

[0050] S10: Answer the question. After double authority verification, Embedding model, optimized local vector library retrieval, Rerank model result processing and large language model integration optimization, the user is finally presented with an answer that meets the user's authority and is relatively satisfactory.

[0051] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0052] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for building an enterprise intelligent knowledge management system based on artificial intelligence, characterized by: The method involves constructing the following blocks: P1: Knowledge base construction, vectorizing and storing business knowledge documents; P2: Construction of a knowledge base assistant system. Based on the enterprise's information level and needs, we build IT operation and maintenance assistants for all employees and professional assistants for each business line, covering the group's key business areas of IT operation and maintenance, human resources, finance, internal control, legal affairs, marketing, and information. P3: Knowledge base retrieval construction, through the knowledge base assistant system and the localized knowledge base to conduct relevant retrieval through question and answer.

2. The method for building an enterprise intelligent knowledge management system based on artificial intelligence according to claim 1, characterized in that: The knowledge base construction includes the following steps: S01: Knowledge preprocessing, including reading, identifying, parsing, slicing and understanding business knowledge documents; S02: Text vectorization, the Embedding model was deployed locally, and the model was adapted and fine-tuned; S03: Vector storage: A weaviate vector database is built locally to store the text vectors calculated by the Embedding model.

3. The method for building an enterprise intelligent knowledge management system based on artificial intelligence according to claim 2, characterized in that: For the reading recognition in S01, an OCR recognition component is deployed to support the recognition and analysis of common documents and pictures.

4. The method for building an enterprise intelligent knowledge management system based on artificial intelligence according to claim 3, characterized in that: The S01 supports following up the cloud document directory for layer-by-layer file search and automatic import, and sets the batch for automatic update. The user only needs to set a cloud document path and the frequency of automatic update, and the document can be automatically read and updated later, which is in line with the company's existing document cloud storage method.

5. The method for building an enterprise intelligent knowledge management system based on artificial intelligence according to claim 4, characterized in that: Customize a variety of segmentation rules according to the size of custom blocks, special characters, and natural paragraphs. These rules better segment documents and maintain semantic integrity to a greater extent, providing a basis for the subsequent Embedding model to understand and vectorize documents.

6. The method for building an enterprise intelligent knowledge management system based on artificial intelligence according to claim 5, characterized in that: For the segmented text, management functions are provided to support administrators to adjust and modify the text blocks.

7. The method for building an enterprise intelligent knowledge management system based on artificial intelligence according to claim 6, characterized in that: A "1+N" intelligent assistant system is introduced in the construction of the knowledge base assistant system.

8. The method for building an enterprise intelligent knowledge management system based on artificial intelligence according to claim 7, characterized in that: At the front end of the knowledge assistant system, it supports multiple entrances such as PC web pages, WeChat applets, and H5 applications, and accesses the application layer through RestAPI.

9. The method for building an enterprise intelligent knowledge management system based on artificial intelligence according to claim 1, characterized in that: The knowledge base retrieval construction comprises the following steps: S04: The user raises a question. The user asks his or her own question in the knowledge assistant. At this time, the first permission check will be performed, that is, the permission check between the user and the knowledge assistant application. If the permission check passes, the question and the current user information will be passed to the Embedding model for the next step of processing; S05: Question vectorization. After receiving the question from the knowledge assistant, the Embedding model first vectorizes the question and unifies the question and knowledge into the same dimension of "vector". S06: Vector search: The Embedding model searches the vectorized question in the local vector library. After the search results are obtained, a second permission check is performed, that is, the permission check of the file associated with the search result. If the check passes, the search result fragment is retained. If the check fails, the fragment is discarded. S07: Returns the TopK results. The Embedding model returns the search results after permission screening to the Rerank model. The model is locally deployed and has undergone parameter adjustment to achieve a certain balance between result hit rate and accuracy. The Rerank model sorts and organizes the results and returns them to the knowledge base assistant. S08: Merge Prompt statements. A Prompt project is configured for each knowledge base assistant. The project sets the role, function, and restriction conditions of the assistant. When the result of the Rerank model is obtained, the Prompt statement and structure will be returned to the large language model. S09: The large language model returns the result. The large language model can be deployed locally. At the same time, the present invention also supports the basic configuration and management switching functions of the large language model to meet the usage requirements of different security levels. After the large language model obtains the prompt sentence and the search results provided by the Rerank model, the information is integrated and returned to the knowledge assistant in the best interactive way; S10: Answer the question. After double permission verification, Embedding model, optimized local vector library retrieval, Rerank model result processing and large language model integration optimization, the user is finally presented with an answer that meets the user's permissions and is relatively satisfactory.

Images