Knowledge archiving methods, apparatus, equipment, storage media, and computer program products

Abstract

This application discloses a knowledge archiving method, apparatus, device, storage medium, and computer program product, relating to the field of data processing technology. The disclosed knowledge archiving method includes: classifying unstructured target requirement documents by business type using a vector machine algorithm to obtain a first classification result; generating a document summary of the target requirement document using a basic large model, and classifying the target requirement document by business type based on the document summary to obtain a second classification result, wherein the basic large model is a large language model trained using historical requirement documents and historical document summaries; determining the target classification result of the target requirement document based on the first and second classification results; and archiving the target requirement document, document summary, and target classification result in a structured form to a preset knowledge base. This application can efficiently archive unstructured business requirement documents into structured document knowledge.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to knowledge archiving methods, apparatus, devices, storage media, and computer program products. Background Technology

[0002] As businesses continue to grow and IT infrastructure development deepens, business requirements documents, as an important part of enterprise knowledge, record in detail the entire process and current status of business establishment and development, and are crucial for enhancing business expansion and support capabilities.

[0003] Currently, the main methods for knowledge archiving of business requirement documents include directly archiving the requirement baseline, converting the requirement document into a structured business baseline, and automatically converting the business baseline based on a structured requirement template. However, since the information filled in by business personnel in the requirement document is usually unstructured and has an inconsistent format, when knowledge organizers archive and manage knowledge based on the document, the document content is difficult to convert into structured document knowledge that can be flexibly applied. This often requires a lot of manpower and it is difficult to guarantee its accuracy.

[0004] In summary, how to efficiently archive unstructured business requirement documents into structured document knowledge has become a pressing technical problem that needs to be solved in this field. Summary of the Invention

[0005] The main purpose of this application is to provide a knowledge archiving method, apparatus, device, storage medium, and computer program product, which aims to efficiently archive unstructured business requirement documents into structured document knowledge.

[0006] To achieve the above objectives, this application proposes a knowledge archiving method, which includes:

[0007] The vector machine algorithm is used to classify unstructured target requirement documents by business type, and the first classification result is obtained;

[0008] A basic large model is used to generate a document summary of the target requirement document, and the target requirement document is classified by business type based on the document summary to obtain a second classification result. The basic large model is a large language model trained using historical requirement documents and historical document summaries.

[0009] The target classification result of the target requirement document is determined based on the first classification result and the second classification result.

[0010] The target requirement documents, document summaries, and target classification results are archived in a structured format to a pre-defined knowledge base.

[0011] In one embodiment, the knowledge base includes a vector library, and the step of using a vector machine algorithm to classify unstructured target requirement documents by business type to obtain a first classification result includes:

[0012] After performing word segmentation and document segmentation on the target requirement document, a vector machine algorithm is used to vectorize the target requirement document after word segmentation and document segmentation to obtain a structured document vector.

[0013] The structured document vectors are matched with the historical document vectors in the vector library to identify target historical requirement documents whose vector similarity is higher than a preset first similarity threshold.

[0014] Obtain the business type from the target historical requirements document and use the business type as the first classification result of the target requirements document.

[0015] In one embodiment, the knowledge base includes a large model, and the step of determining the target classification result of the target requirement document based on the first classification result and the second classification result further includes:

[0016] Based on the target classification results, search the knowledge base for historical target requirement documents with the same business type as the target requirement document, and analyze the differences between the target requirement document and the historical target requirement document to determine the business change content in the knowledge base.

[0017] Retrieve business questions and standard answers corresponding to the document summary from a pre-defined question list library;

[0018] Input the business change information into the previous version of the large model for model training to obtain the initial updated large model;

[0019] Ask business questions about the initial update of the large model and obtain the output answers;

[0020] Based on the comparison between the output answer and the standard answer, the target update model is obtained. Among them, the previous version model, the basic model, and the target update model are models of different update rounds. The target update model is used for the automatic archiving of subsequent requirement documents and knowledge Q&A services.

[0021] In one embodiment, the step of determining the target classification result of the target requirement document based on the first classification result and the second classification result includes:

[0022] The first classification result and the second classification result are cross-validated;

[0023] If the first classification result and the second classification result are consistent, the target classification result of the target requirement document is determined to be consistent with the first classification result and the second classification result.

[0024] If the first classification result and the second classification result are inconsistent, a verification window is generated to receive user operations, the target classification result is determined based on the user operations, and the first classification result and / or the second classification result is corrected based on the target classification result.

[0025] In one embodiment, the step of archiving the target requirement document, document summary, and target classification results in a structured form to a pre-defined knowledge base includes:

[0026] Construct a document index for the target requirement documents based on the target classification results;

[0027] The vectorized target requirement documents, document indexes, and target classification results are archived into a vector library.

[0028] Archive the target requirements document, document summary, and target classification results into the large model;

[0029] After the step of archiving the target requirements document, document summary, and target classification results in a structured format to a pre-defined knowledge base, the following steps are also included:

[0030] We continuously compare and verify the knowledge archived in the large model and the vector library.

[0031] In one embodiment, the method further includes:

[0032] Upon receiving an externally input content query command, identify the user intent corresponding to the content query command;

[0033] The similarity algorithm is used to select target content in the knowledge base and output the target content, where the target content is content whose similarity to the user's query intent is higher than a preset second similarity threshold.

[0034] Furthermore, to achieve the above objectives, this application also proposes a knowledge archiving device, which includes:

[0035] The first classification module is used to classify unstructured target requirement documents by business type using a vector machine algorithm to obtain the first classification result;

[0036] The second classification module is used to generate document summaries of target requirement documents using the basic large model, and classify the target requirement documents by business type based on the document summaries to obtain the second classification result. The basic large model is a large language model trained using historical requirement documents and historical document summaries.

[0037] The classification result determination module is used to determine the target classification result of the target requirement document based on the first classification result and the second classification result.

[0038] The archiving module is used to archive target requirement documents, document summaries, and target classification results in a structured format to a pre-defined knowledge base.

[0039] In addition, to achieve the above objectives, this application also proposes an electronic device, the device comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, the computer program being configured to implement the steps of the knowledge archiving method as described above.

[0040] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the knowledge archiving method described above.

[0041] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the knowledge archiving method described above.

[0042] This application proposes a knowledge archiving method. First, it uses a vector machine algorithm to classify unstructured target requirement documents by business type, obtaining a first classification result. Then, it uses a basic large model to generate a document summary of the target requirement document, and based on the generated document summary, it classifies the target requirement document by business type again, obtaining a second classification result. The basic large model is a large language model trained using historical requirement documents and historical document summaries. Then, it combines the first and second classification results to determine the target classification result of the target requirement document. After determining the target classification result, the target requirement document, document summary, and target classification result are archived together in a structured form into a pre-set knowledge base.

[0043] In summary, this application, by combining vector machine algorithms and large-model-based document summarization, can more accurately understand the content of unstructured target requirement documents, thus obtaining more accurate document classification results. Simultaneously, by storing the target requirement documents, document summaries, and classification results in a structured form in a knowledge base, it facilitates subsequent retrieval, querying, and utilization. Therefore, this application achieves the efficient archiving of unstructured business requirement documents into structured document knowledge, helping enterprises build a comprehensive knowledge management system and improve their overall knowledge management level and capabilities. Attached Figure Description

[0044] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0045] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 This is a flowchart illustrating an embodiment of the knowledge archiving method of this application.

[0047] Figure 2 This is a schematic diagram of the vector library construction process provided in Embodiment 2 of the knowledge archiving method of this application;

[0048] Figure 3 This is a schematic diagram of the word segmentation and document splitting process provided in Embodiment 2 of the knowledge archiving method of this application;

[0049] Figure 4 This is a schematic diagram of the indexing process provided in Embodiment 2 of the knowledge archiving method of this application;

[0050] Figure 5 This is a schematic diagram of the large model optimization process provided in Embodiment 2 of the knowledge archiving method of this application;

[0051] Figure 6 A schematic diagram of the large model and vector library comparison and verification process provided in Embodiment 2 of the knowledge archiving method of this application;

[0052] Figure 7 A schematic diagram of the large model construction framework provided in Embodiment 2 of the knowledge archiving method of this application;

[0053] Figure 8 A schematic diagram of the vector library construction framework provided in Embodiment 2 of the knowledge archiving method of this application;

[0054] Figure 9 This is a schematic diagram of the comparison and verification framework provided in Embodiment 2 of the knowledge archiving method of this application;

[0055] Figure 10 A schematic diagram of a vector similarity-based search framework provided for Embodiment 2 of the knowledge archiving method of this application;

[0056] Figure 11 A schematic diagram of the search framework for a large-model question-answering robot provided in Embodiment 2 of the knowledge archiving method of this application;

[0057] Figure 12 This is a schematic diagram of the knowledge archiving process provided in Embodiment 2 of the knowledge archiving method of this application;

[0058] Figure 13 This is a schematic diagram of the module structure of the knowledge archiving device according to an embodiment of this application;

[0059] Figure 14 This is a schematic diagram of the device structure of the hardware operating environment involved in the knowledge archiving method in the embodiments of this application.

[0060] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0061] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0062] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0063] The main solution of this application embodiment is as follows: A vector machine algorithm is used to classify unstructured target requirement documents by business type to obtain a first classification result; a basic large model is used to generate a document summary of the target requirement document, and the target requirement document is classified by business type based on the document summary to obtain a second classification result, wherein the basic large model is a large language model trained using historical requirement documents and historical document summaries; the target classification result of the target requirement document is determined based on the first and second classification results; and the target requirement document, document summary, and target classification result are archived in a structured form to a preset knowledge base.

[0064] Existing technologies for knowledge archiving primarily include three methods for archiving business requirement documents: directly archiving requirement baselines, converting requirement documents into structured business baselines, and automatically converting business baselines based on structured requirement templates. However, since the information filled in by business personnel in requirement documents is usually unstructured and has a variable format, when knowledge organizers archive and manage knowledge based on the documents, the document content is difficult to convert into structured document knowledge that can be flexibly applied. This often requires a lot of manpower and it is difficult to guarantee its accuracy.

[0065] This application provides a solution that, by combining vector machine algorithms and large-model-based document summarization, can more accurately understand the content of unstructured target requirement documents, thereby obtaining more accurate document classification results. Simultaneously, the target requirement document, document summary, and classification results are stored in a structured form in a knowledge base, facilitating subsequent retrieval, querying, and utilization. Thus, this application achieves the efficient archiving of unstructured business requirement documents into structured document knowledge, helping enterprises build a comprehensive knowledge management system and improve their overall knowledge management level and capabilities.

[0066] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or an electronic device capable of performing the above functions. The following description uses a knowledge archiving terminal as an example to illustrate this embodiment and the subsequent embodiments.

[0067] It should be noted that, for ease of understanding and explanation of this embodiment, the following section will first introduce traditional knowledge archiving solutions within enterprises, as detailed below:

[0068] For large IT-enabled enterprises, the largest component of knowledge management and knowledge archiving lies in the management and preservation of knowledge from requirement documents, business solution documents, and other related materials generated during the IT process. These documents contain the entire process and current status of the enterprise's business establishment and development. Extracting, organizing, and preserving knowledge from requirement and business documents is the most effective method for knowledge management and preservation.

[0069] Currently, there are two main ways to accumulate and archive knowledge from requirements documents or business documents.

[0070] The first approach: using the baseline of requirements as knowledge.

[0071] A requirements baseline refers to the version of the requirements specification that has been confirmed and approved during the project or product development process. It represents the set of requirements that is recognized as accurate, complete, and feasible at a specific point in time. The requirements baseline serves as the foundation and reference for ensuring the smooth progress of the project or product development.

[0072] This knowledge archiving method directly archives the baseline requirements in the knowledge base, meaning it directly stores confirmed and approved requirement documents. Simultaneously, when archiving requirement documents, they can be categorized and directory-based according to business characteristics or development methods, or business tags can be added to the documents to provide general document retrieval based on filename, tag, category, and directory. Some more advanced knowledge bases support word segmentation and word-segmentation-based indexing during document archiving, providing full-text content retrieval of the document library.

[0073] This method of knowledge archiving has the advantages of saving time and effort, but when directly archiving unstructured knowledge such as documents, it is difficult to guarantee the validity and uniqueness of the knowledge's interpretation. Furthermore, it is difficult to handle content conflicts between different documents.

[0074] The second approach is to convert requirement documents into structured business baselines.

[0075] The business baseline here refers to the set of business processes that have been summarized and abstracted from the start of IT infrastructure development to the present time, through the sorting out and summarizing of requirement documents from different versions and stages.

[0076] This knowledge archiving method adds an organization process to the knowledge accumulation process of the requirements baseline document. First, an initial version of the business baseline is established, containing some existing business knowledge. Subsequently, during the requirements development process, business knowledge is continuously summarized from new requirements documents. For the parts already existing in the business baseline, updates and version maintenance are performed; for business knowledge not present in the business baseline, a first version of the newly added business point is created.

[0077] The advantage of this knowledge archiving method is that it can effectively organize the structure of the knowledge base and allows for relatively precise control over its quality. The disadvantage is that the workload of organizing the knowledge is enormous, making it not worthwhile.

[0078] The third method: automatically converting business baselines based on structured requirement templates.

[0079] This knowledge archiving method is based on the idea of ​​converting requirement documents into structured business baselines. It pre-standardizes structured templates, allowing business personnel to use these templates to write and submit their requirements when they are submitted. This ensures that the requirement content has a certain degree of structure, saving the workload of subsequently organizing the business baselines from the requirement documents.

[0080] The advantage of this approach is that it significantly reduces the workload of knowledge organization while maintaining the knowledge structure and quality within the knowledge base. The disadvantages are that the workload of creating templates is substantial, and there are many restrictions on business personnel submitting requirements using templates; these technical limitations can easily hinder business development.

[0081] In the knowledge management systems of large enterprises with constantly evolving and frequently updated businesses, the current methods all have some shortcomings and technical problems.

[0082] The first approach: the shortcomings of using the baseline of requirements as a knowledge solution.

[0083] Directly archiving requirements documents as knowledge is a simple knowledge archiving process. However, it has the following three drawbacks:

[0084] First, it can only be used for reading and cannot be extended to other applications based on unstructured knowledge.

[0085] Second, the quality of knowledge is difficult to guarantee, and some knowledge is difficult to understand accurately.

[0086] Third, conflicts between different versions are difficult to identify.

[0087] The second approach: the drawbacks of converting requirement documents into structured business baselines.

[0088] Adding a processing step to the knowledge archiving process transforms unstructured documents into structured content for knowledge storage, significantly improving the accuracy of the knowledge. It also avoids the conflict handling issues associated with multiple versions and frequently changing business processes encountered with the first method. However, it has the following main drawbacks:

[0089] First, the mapping relationship between requirements documents and business knowledge is too complex.

[0090] Second, the quality of requirements documents affects the results of knowledge archiving.

[0091] Third, it requires a high level of business comprehension and breadth of knowledge from the personnel responsible for organizing the knowledge.

[0092] The third approach: The drawbacks of automatically converting business baselines based on structured requirement templates.

[0093] This method of standardizing the business requirements writing process through templates is a mainstream development trend in the industry. This approach essentially semi-structures and standardizes business requirements documents. It can, to some extent, address the issue of non-standardized requirements documents and reduce the cost of organizing business requirements documents into structured business knowledge documents. However, it has the following drawbacks:

[0094] First, templates impose certain limitations on business development.

[0095] Secondly, it is difficult to ensure through technical means that business personnel strictly follow the format when writing.

[0096] In summary, the shortcomings of the above solutions can be attributed to the fact that the information in the requirements documents is unstructured and its presentation is not fixed. When conducting knowledge management, it is difficult to convert this content into structured knowledge that can be flexibly applied, which often requires a lot of manpower and is difficult to guarantee its accuracy.

[0097] This embodiment primarily addresses the challenge of converting unstructured documents into knowledge and improving their accuracy with lower manpower costs, without mandating structured completion of business requirement documents.

[0098] Based on this, embodiments of this application provide a knowledge archiving method, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the knowledge archiving method of this application.

[0099] In this embodiment, the knowledge archiving method includes steps S10 to S40:

[0100] Step S10: Use the vector machine algorithm to classify the unstructured target requirement document by business type to obtain the first classification result;

[0101] It should be noted that the Support Vector Machine (SVM) algorithm is a supervised learning algorithm, mainly used for data classification problems. In this embodiment, through the trained vector machine, the terminal can automatically analyze the content of unstructured target requirement documents and perform preliminary classification based on the document's topic, domain, or other features to obtain the first classification result.

[0102] The knowledge archiving terminal uses a vector machine algorithm to perform a preliminary classification of unstructured target requirement documents, resulting in a preliminary document classification result, namely the first classification result. Among them, the target requirement documents are newly created business requirement documents. Unstructured requirement documents are difficult to be retrieved and indexed by automated tools. If the requirement documents are directly archived in an unstructured form, it is easy to cause information conflicts and inconsistencies.

[0103] Step S20: Generate a document summary of the target requirement document using the basic large model, and classify the target requirement document by business type according to the document summary to obtain the second classification result. The basic large model is a large language model trained using historical requirement documents and historical document summaries.

[0104] It should be noted that the basic large model is a private large language model trained using historical business requirement documents and already compiled historical document summaries, based on the open-source large language model.

[0105] The basic large model is used to generate document summaries of the target requirement documents. Based on the generated document summaries, the target requirement documents are classified by business type to obtain a second classification result. The document summary is a concise summary of the content of the target requirement documents, which can reflect the main content and core information of the target requirement documents.

[0106] Step S30: Determine the target classification result of the target requirement document based on the first classification result and the second classification result;

[0107] After obtaining the first and second classification results, the two results are compared and verified. By comprehensively considering the two classification results, a more accurate and reliable target classification result is obtained to ensure the accuracy of the classification.

[0108] Step S40: Archive the target requirement document, document summary, and target classification results into a pre-defined knowledge base in a structured format.

[0109] The target requirement document, its corresponding document summary, and the determined target classification results are archived in a structured form into a pre-set knowledge base. Structured archiving means that this information will be stored in the knowledge base in an orderly, easy-to-retrieve and manage manner, which facilitates subsequent use and query. At the same time, archiving the document, summary, and classification results together can provide users with comprehensive document information and convenient search services.

[0110] For example, the terminal receives a newly submitted requirement document. The document involves changes to employee benefits policies. A vector machine algorithm is used to classify the document by business type, resulting in a first classification: categorizing the document as Human Resources Management. Simultaneously, a document summary is generated using a basic large model, and this summary is then classified by business type, resulting in a second classification: categorizing the document as Employee Benefits Policy. The first and second classification results are then compared. If they match, the target classification for the requirement document is determined directly. If they do not match, further analysis is required.

[0111] In this example, since the first and second classification results are inconsistent, further analysis determines that the correct classification result should be employee welfare policy. Therefore, the target classification result of the requirement document is determined to be employee welfare policy, and the requirement document, document summary, and target classification result are archived to the knowledge base.

[0112] This embodiment provides a knowledge archiving method. By combining vector machine algorithms and document summarization based on large models, it can more accurately understand the content of unstructured target requirement documents, thereby obtaining more accurate document classification results. Simultaneously, the target requirement document, document summary, and classification results are stored in a structured form in a knowledge base, facilitating subsequent retrieval, querying, and utilization. Thus, this application achieves the efficient archiving of unstructured business requirement documents into structured document knowledge, helping enterprises build a comprehensive knowledge management system and improve their overall knowledge management level and capabilities.

[0113] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to that in the first embodiment described above can be referred to the above description and will not be repeated hereafter. Based on this, the knowledge base includes a vector library, and step S10 may include steps S101 to S103:

[0114] Step S101: After performing word segmentation and document segmentation on the target requirement document, the target requirement document after word segmentation and document segmentation is vectorized using a vector machine algorithm to obtain a structured document vector.

[0115] The knowledge archiving terminal first performs word segmentation and document splitting on the target requirement document to decompose the document content into smaller, meaningful units. Then, it uses a vector machine algorithm to vectorize these segmented and split document contents, transforming the target requirement document from text information into structured document vectors.

[0116] Step S102: Perform similarity matching between the structured document vector and the historical document vector in the vector library to determine the target historical requirement document whose vector similarity is higher than the preset first similarity threshold from the vector library.

[0117] The structured document vector obtained by transformation is matched with the historical document vectors in the vector library of the knowledge base to compare the similarity between the structured document vector of the target requirement document and the historical document vectors in the vector library. This determines the target historical requirement document whose vector similarity with the target requirement document is higher than the first similarity threshold. The first similarity threshold can be set and adjusted based on actual application requirements. This embodiment does not impose specific limitations on it.

[0118] Step S103: Obtain the business type of the target historical requirement document and use the business type as the first classification result of the target requirement document.

[0119] After finding a target historical requirement document with high similarity to the target requirement document through similarity matching, the business type of the target historical requirement document is obtained, and the business type is used as the first classification result of the target requirement document.

[0120] For example, to aid in understanding this embodiment, please refer to Figure 2 , Figure 2 A simplified flowchart illustrating the construction of a vector knowledge database (i.e., a vector library) is provided. The diagram depicts the initial construction process of the vector library on a terminal. Through word segmentation and vectorization techniques, the vector knowledge database is constructed to represent knowledge content in vector form. An efficient index is built for the vectorized knowledge database, enabling high-speed retrieval of document content corresponding to knowledge. The specific steps include:

[0121] Sub-step one: Batch read the text information of a batch of requirement documents.

[0122] The second sub-step involves document segmentation and word segmentation, which involves structuring, subdividing, and splitting unstructured knowledge information. Specifically, word segmentation involves processing the text content in historical requirement documents or large text fields in the database (such as the requirement description field). The mature Jieba tool (a Chinese word segmentation tool) is used, with some customization. Part-of-speech tagging is performed during segmentation, labeling words as adjectives, verbs, nouns, etc. After word segmentation, the resulting data is cleaned, removing noisy content and retaining only content relevant to the specific business or requirement. Removed elements include pronouns, dates, stop words, modal particles, punctuation, and stop words such as "we," "the reason," "furthermore," and "in summary." Data cleaning is primarily done in two ways: manual processing (mainly based on statically planned thesaurus to extract key features) and code scripting (mainly the program dynamically and automatically identifies and plans various thesauruses based on a rule base to extract key features).

[0123] like Figure 3 As shown, unstructured knowledge information is parsed, subdivided, and split through static programming and dynamic recognition. Specifically, this includes cleaning up invalid characters using a manual invalid character library (static programming) and a dynamic rule library for invalid characters (dynamic recognition); adding user dictionaries using a manual user dictionary (static programming) and a dynamic rule library for user dictionaries (dynamic recognition); processing synonyms using a manual thesaurus (static programming) and a dynamic rule library for synonyms (dynamic recognition); word segmentation; removing stop words using a manual stop word library (static programming) and a dynamic rule library for stop words (dynamic recognition); and removing non-interesting words using a manual reserved word library (static programming) and a dynamic rule library for reserved words (dynamic recognition). The resulting feature words are used for subsequent document vectorization representation.

[0124] Sub-step three involves vectorizing and classifying documents based on word segmentation and vector machine algorithms.

[0125] The vector similarity distance (representing the degree of similarity) of the vectorized document content is calculated using the vector machine method. Based on cluster analysis, historical documents are classified, and the business classification is manually verified to ensure the accuracy of the basic business classification. The accurate business classification is used as the basis (and should be consistent with the business classification of the large model).

[0126] In the absence of a basic business classification, cluster analysis can be performed on historical requirement categories, and initial business classifications can be constructed by manually classifying the data based on the clustering results. This eliminates the need for manually reading and classifying each file. If the project has accumulated basic business classifications, these historical categories can be used to label the business type of each file, resulting in more accurate classification results.

[0127] Based on the results of word segmentation and cleaning, SVM is used for document classification.

[0128] The classification algorithm employs a multi-class text classification support vector machine (SVM) algorithm, which has advantages in small-sample, non-linear, and high-dimensional pattern recognition. While SVM is inherently a binary classification algorithm, when handling multi-class problems, it requires constructing multi-class classifiers. During training, certain business descriptions are sequentially assigned to the positive set, while the remaining business descriptions are assigned to the negative set. For example, if 50 categories are defined to correspond to existing business types, 50 classifiers will be obtained as initial classification models. These classification models can then be used to classify newly added business requirement documents. The main function of classification is to compare and verify the business mapping results with the larger model, and to automatically correct errors.

[0129] Sub-step four involves matching the vectorized result of the new requirement document with the existing vectors in the vector library based on distance similarity. This identifies historical requirement documents with high similarity (e.g., similarity > 70%), and obtains the business type associated with this list of documents (i.e., historical requirement documents with high similarity). This business type is then used as the business type associated with the new requirement document, or as the business classification label after the new requirement document is vectorized.

[0130] Sub-step five involves comparing the vectorized matching business classification tags with the document business classification tags from the large model analysis to verify their accuracy.

[0131] Sub-step six: If the business classification label of the vectorized matching does not match the business classification label of the large model analysis, manual verification is required. The correct result is then synchronized to the label in the vector library and used as optimization training data for the large model.

[0132] Sub-step seven, the matching data from sub-step five, and the manually labeled data from sub-step six are correctly classified and entered into the vector database. The vector database can be implemented using the Milvus vector library (a vector database), which can store large-scale vectorized data and has powerful retrieval and configuration functions.

[0133] Sub-step eight: Document classification and construction of a document vectorized index. The document vectorized database is indexed using Faiss (Facebook AISimilarity Search, a library for large-scale vector retrieval) indexing technology.

[0134] Faiss is a vector-based similarity search technique. This application uses vectorized indexing of historical documents and business knowledge to construct a Faiss index.

[0135] For newly added business requirement documents after word segmentation and splitting, Faiss is also used to first perform similarity retrieval in the existing vector library to determine the category that best matches the new business requirement document in terms of similarity.

[0136] The document summaries output by the large model and the corresponding business information mapping relationships are also retrieved using the Faiss index to find the category with the highest similarity to the output results of the large model.

[0137] The two classification results are compared, replacing manual verification and saving labor costs.

[0138] Sub-step nine: Similarity matching retrieval based on vector distance. This provides vector index matching functionality, segmenting the user's query using the same rules and vectorizing it using the SVM algorithm. For example, if a user wants to search for "including all changes to a certain package," the retrieval process first segments the query "including all changes to a certain package" using the same method, removes noisy words, and then vectorizes it using the SVM algorithm. After vectorization, it performs similarity matching in a Faiss-indexed vector library to quickly obtain the desired content. This implements similarity matching retrieval based on vector distance.

[0139] like Figure 4 As shown, the terminal performs document content segmentation and vectorization representation on historical documents to construct a Faiss index for those historical documents. Then, when a new document is received, on the one hand, it can perform query vectorization and construct a Faiss index for the new document to achieve efficient Faiss indexing, and compare this index with the indexes of historical documents to obtain the desired content. On the other hand, it can also perform word segmentation, SVM vectorization, and construct a Faiss index for the new document, and compare this index with the indexes of historical documents to obtain the desired content.

[0140] In one feasible embodiment, the knowledge base includes a large model, and steps A10 to A50 may be included after step S30:

[0141] Step A10: Based on the target classification results, search the knowledge base for historical target requirement documents with the same business type as the target requirement document, and analyze the differences between the target requirement document and the historical target requirement document to determine the business change content in the knowledge base.

[0142] After determining the target classification result of the target requirement document, search the knowledge base for historical target requirement documents with the same business type as the target requirement document, analyze the differences in content between the target requirement document and the historical target requirement documents, and identify the business change content in the instruction library.

[0143] For example, if employee health insurance benefits are added to the target requirements document but not mentioned in the previous target requirements document, this will be marked as a business change. In addition, changes can involve business processes, data structures, policy adjustments, etc.

[0144] Step A20: Retrieve the business questions and standard answers corresponding to the document summary from the preset question list library;

[0145] The system retrieves business questions and their standard answers corresponding to the document summary from a pre-defined question list library. These questions and answers represent common problems and solutions in the business domain and are an important reference for evaluating the performance of large models.

[0146] Step A30: Input the business change content into the previous version of the large model for model training to obtain the initial updated large model;

[0147] The identified business changes are input into the previous version of the large model for model training, generating an initial updated large model, thereby integrating new business knowledge and changes into the large model.

[0148] Step A40: Ask business questions about the initial update of the large model and obtain the output answers;

[0149] After obtaining the initial updated large model, business questions are posed to the large model. These questions are generated from a question list library and are used to test the initial updated large model's ability to understand and answer business questions. The terminal can evaluate the model based on its output answers.

[0150] Step A50: Based on the comparison results between the output answer and the standard answer, the target update big model is obtained. Among them, the previous version big model, the basic big model and the target update big model are big models of different update rounds. The target update big model is used for the automatic archiving of subsequent requirement documents and knowledge question answering services.

[0151] It should be noted that the basic large model is a knowledge base model obtained by training the system using historical requirement documents and summaries of historical documents compiled by historical business personnel based on these documents. The previous version of the large model was a model version that was further optimized and trained based on the basic large model. In actual application scenarios, the basic large model can also be updated.

[0152] The output answer of the initial updated large model is compared with the standard answer to evaluate the accuracy of the initial updated large model in answering business questions. Based on the comparison results, the initial updated large model is adjusted and optimized as necessary to obtain the target updated large model. The target updated large model is obtained by merging business change content and testing model performance based on the previous version of the large model. It represents the optimal state of the large model in the current business environment and knowledge state.

[0153] Thus, based on the large model's processing of unstructured requirement documents, it can understand documents of various levels of detail, ideas, and formats, and can process complex mapping relationships as accurately as possible to obtain more accurate archived content.

[0154] In one feasible embodiment, step S30 may include steps S301 to S303:

[0155] Step S301: Verify the first classification result and the second classification result against each other;

[0156] After obtaining the first and second classification results, the knowledge archiving terminal first verifies the two classification results to ensure consistency and accuracy between them.

[0157] Step S302: If the first classification result and the second classification result are consistent, determine that the target classification result of the target requirement document is consistent with the first classification result and the second classification result.

[0158] If, after mutual verification, the first classification result and the second classification result are consistent, then the knowledge archiving terminal will directly determine that the target classification result of the target requirement document is the same as these two classification results.

[0159] Step S303: If the first classification result and the second classification result are inconsistent, a verification window is generated to receive user operations, the target classification result is determined according to the user operations, and the first classification result and / or the second classification result is corrected based on the target classification result.

[0160] If the first and second classification results are inconsistent, a manual verification mechanism is triggered. A verification window is generated, displaying the target requirement document and its two inconsistent classification results to a human verifier for verification. The human verifier, based on their professional knowledge and experience, reclassifies the target requirement document and performs corresponding user operations, such as clicking to select the correct document type classification option in the verification window, thereby determining the target classification result.

[0161] After determining the target classification results, the knowledge archiving terminal will also correct the first classification results and / or the second classification results based on the results of manual verification to ensure the consistency and accuracy of the classification results in the knowledge base.

[0162] For example, to aid in understanding this embodiment, please refer to Figure 5 , Figure 5 This document provides a simplified flowchart illustrating the process of accumulating and validating new business knowledge based on a large-scale model. It involves generating summaries and associated business types from the text data of requirement documents, acquiring a historical question database, performing question-and-answer exercises on a pre-trained large-scale model knowledge base containing new batches of requirement documents, and generating a new version of the large-scale model through automatic and manual verification. This process builds a continuously optimized large-scale model knowledge base, specifically including the following steps:

[0163] Sub-step one: Batch read the text information of a batch of requirement documents.

[0164] Sub-step two involves generating text summaries based on the large model, extracting key information from the input document content, obtaining summary information of these documents through the large model, and determining which business types in the knowledge base each requirement document changes (linking documents and business types).

[0165] When processing large volumes of documents, document summaries help users quickly grasp the main content and key information. A well-trained large model can understand the characteristics of a company's business and the different descriptions of vocabulary based on historical data, thereby identifying and summarizing business-related content in new business requirement documents.

[0166] The trained large model reads the file content, analyzes and models the document, determines which business types the document is related to in the knowledge base, and makes changes to those business types.

[0167] Based on the large model, documents are analyzed and modeled. Based on historical data and known business processes, relevant business content within the documents is identified and mapped. Furthermore, the large model can perform topic modeling and text classification to understand the document's theme and content. Through semantic analysis and classification of the text, entities within the document are identified and relationships between them are extracted, thus generating structured business knowledge, or a business baseline.

[0168] Sub-step three involves retrieving pre-summarized business questions and standard answers related to these summary information and business types from the historical question database, based on the document summary information and business types output by the large model.

[0169] Sub-step four involves using the files from this batch to optimize and train the large model, generating a new version of the large model. The new version of the large model theoretically already understands the relevant changes made to the business by the requirements documents in this batch. It also possesses the ability to answer questions related to these business-related issues.

[0170] Sub-step five: Use questions from the question bank to ask questions about the new version of the large model, and obtain the answers output by the new version of the large model.

[0171] Sub-step six involves comparing the semantic similarity of the answers output by the large model with the standard answers. For answers with high similarity to the standard answers in the question bank, the large model is considered to possess that knowledge. For answers with low similarity, manual review is required to confirm whether the differences are due to changes in these requirements affecting the business logic, and whether these differences are definitively caused by these changes. Specifically, when a difference is found between a question and a previously preset answer, manual evaluation is conducted to determine whether these differences are normal. If the business logic has indeed changed due to changes in the requirements of this batch, leading to changes in the question answers, and the new answers are found to be accurate during the verification process, then it is proven that the large model has grasped the content of these changes in requirements.

[0172] Sub-step seven: After manual review and confirmation that the optimized and trained new version of the large model has mastered the business knowledge in the new batch of requirement documents, the new version of the large model is merged into the main process to replace the previous version. This provides knowledge base question-and-answer services, as well as summary generation and business type matching functions for subsequently added documents.

[0173] Specifically, after manual judgment, when it is found that the answer of the large model differs from the previously preset answer, the manual judges whether these differences are normal. If the business has indeed changed due to the changes in the requirements of this batch, resulting in changes in the answer to the question, and the new answer is found to be accurate during the verification process, then it proves that the large model has grasped the changes in the requirements of this batch.

[0174] In one feasible embodiment, step S40 may include steps S401 to S403:

[0175] Step S401: Construct a document index for the target requirement documents based on the target classification results;

[0176] After determining the target classification result of the target requirement document, the knowledge archiving terminal will build a corresponding document index for the structured document vector based on the classification result. The document index is used to quickly retrieve and locate document data. By building the document index, the efficiency of subsequent document retrieval and management can be significantly improved.

[0177] Step S402: Archive the vectorized target requirement document, document index, and target classification results into the vector library;

[0178] The vectorized target requirement document, the corresponding document index, and the target classification results are archived together in the vector library. The vector library is specifically used to store vector data and document indexes. By archiving to the vector library, these data can be easily managed and queried.

[0179] Step S403: Archive the target requirement document, document summary, and target classification results into the large model;

[0180] In addition to archiving to the vector library, the knowledge archiving terminal will also archive the target requirement document, its document summary, and the target classification results into the large model. The large model is a trained and optimized large language model that can process and analyze large amounts of data and generate corresponding results.

[0181] By archiving the target requirements document and its related information into a large model, the large model can be fully utilized for in-depth analysis and processing of the document.

[0182] In one possible embodiment, step B10 may be included after step S40:

[0183] Step B10 involves continuously comparing and verifying the knowledge archived in the large model and the vector library.

[0184] To ensure the accuracy and consistency of archived knowledge, the knowledge archiving terminal continuously compares and verifies the knowledge in the large model and vector library to ensure the consistency and accuracy of the knowledge between the two. If inconsistencies or errors are found, they are corrected and updated in a timely manner to ensure the reliability and accuracy of the archived knowledge.

[0185] For example, to aid in understanding this embodiment, please refer to Figure 6 , Figure 6 A process for cross-verifying large models and vector libraries is provided, which specifically includes the following steps:

[0186] Sub-step 1: Construction of the large model knowledge base.

[0187] like Figure 7As shown, during the construction of the knowledge base large model, open-source large models such as GPT3.0 (General Pre-trained Transformer-3) or ChatGLM (Chinese-language Pre-training Model) can be used as the basic large model. The training data can be the enterprise's internal historical business requirement documents and the historical document summaries and business list information compiled by historical business personnel based on these documents. The knowledge base large model obtained after model training has the ability to extract summaries from document texts in a specified business domain.

[0188] Sub-step two: Building document vectorization indexing and classification capabilities.

[0189] like Figure 8 As shown, the core of building document vectorized indexing and classification capabilities lies in optimizing the word segmentation granularity, splitting granularity, and classification algorithm. In this embodiment, historical business requirement documents are used as input, and historical document summaries and business lists compiled by business personnel based on these documents, along with document and business category mappings, are used as the basic information for output verification. Optimization of parameters and granularity at each stage is performed, namely, optimizing word segmentation and document splitting, document vectorized indexing, and SVM vector machine classification algorithms in the vectorized indexing and classification tools, until the output files of the entire process are consistent with the content compiled by business personnel. Then, the construction of document vectorized indexing and classification capabilities is successful.

[0190] Sub-step 3: Mutual verification between the large model inference classification results and the vectorized index classification results.

[0191] like Figure 9 As shown in this embodiment, when a new business requirement document is input, inference of the large model and classification analysis of the similarity after document vectorization are performed simultaneously to generate two sets of document-business category mappings. That is, the document business change summary and business category are obtained through the knowledge base large model, and the document-business category similarity based on vector analysis is obtained through vectorization index and classification tools. Then, the two classification results are automatically compared and audited. For those that match, they are directly entered into the knowledge base. For documents with large differences, they are used as input for manual comparison and manual mapping. The mapping results are then used to optimize and train the knowledge base large model.

[0192] This continuous workflow ensures the accuracy of the knowledge base and enables the large model to be continuously optimized. Furthermore, the data collection process for optimizing training becomes a sustainable and low-cost endeavor.

[0193] In one feasible embodiment, the knowledge archiving method may further include steps B50 to B60:

[0194] Step B50: After receiving the externally input content query instruction, identify the user intent corresponding to the content query instruction;

[0195] When the knowledge archiving terminal receives a content query command from an external source, it identifies the user intent represented by the command. The user intent refers to the specific information or answer that the user wants to obtain from the knowledge base.

[0196] Step B60: Select target content from the knowledge base using a similarity algorithm and output the target content, wherein the target content is content whose similarity to the user's query intent is higher than a preset second similarity threshold.

[0197] After recognizing the user's intent, the knowledge archiving terminal uses a similarity algorithm to search and match within the knowledge base, and then outputs the found target content to the user to fulfill the user's query request. The target content is content whose similarity to the user's query intent is higher than a second similarity threshold. The second similarity threshold can be set and adjusted based on actual application needs; this embodiment does not impose specific limitations on it.

[0198] In addition, it should be noted that the terminal provides query content in three ways in response to external input query commands. The first is a knowledge base retrieval function similar to other knowledge bases, which will not be elaborated here. The second is knowledge and business document search based on vector similarity. The third is to build a question-answering robot based on a large model.

[0199] Among them, knowledge and business document search functions based on vector similarity, such as Figure 10 As shown, since this embodiment uses Faiss vectorized indexing based on similarity for documents and business knowledge, users can use similarity-based search when conducting knowledge searches. Their query content can be keywords or interactive language describing a query purpose. The query content is constructed into a Faiss index, and during the search process, a basic large model is used to perform natural language semantic analysis and inference of user intent on the input query. For example, inputting an interactive question like "Can employees do part-time work outside the office?" The search process doesn't simply use "Can employees do part-time work outside the office?" as a keyword for retrieval or full-text search. Instead, it understands that the user wants to ask about part-time work policies. After understanding the user intent, the similarity search algorithm finds content in the Faiss index of business needs and business knowledge, and calculates the similarity between each piece of content and the user intent. The program can then output the x pieces of content with the best similarity.

[0200] Among them, the function of building a question-answering robot based on a large model, such as Figure 11As shown in this embodiment, a large knowledge base model that encompasses all business knowledge in the knowledge base is constructed. This model provides interfaces to implement question-and-answer services based on the large model and knowledge base knowledge, and records usage records and user feedback. Specifically, a robot interface embedded in the client page is provided, and robot services and models are categorized. Questions and responses generated during the query process are stored in MySQL (a data management system). HTTP (Hypertext Transfer Protocol) requests are completed based on LangChain (Language Chain for building applications based on large language models) and the large language model to achieve personalized knowledge-based question-and-answer services.

[0201] Thus, in this embodiment, knowledge archiving does not require templates to standardize the process of filling in business requirements. Instead, it introduces large models and machine learning technology to assist in the process of extracting business knowledge from unstructured business requirement documents. Machines replace manual labor to realize the knowledge organization process, which has the highest human cost. Humans only need to verify the results.

[0202] In this embodiment, the final accumulated business knowledge is divided into three parts: business baseline (i.e., structured business knowledge base), document vectorization index, and large model trained using requirement documents.

[0203] The business baseline, which is highly abstracted business knowledge from the requirements document, is an overall view of the enterprise's business and is stored in a structured manner according to business categories. It also provides corresponding search and structured display applications.

[0204] A document vectorized index is a document index formed by natural language analysis and clustering methods for all required documents. It is a set of document classifications calculated based on natural language and vector machine algorithms. It stores the vectorized distance between documents, reflecting the similarity between documents.

[0205] The large-scale model trained using requirements documents is trained using historical requirements documents to understand the construction intent of the documents and thus comprehend the business meaning of the requirements. The trained model masters all the knowledge in the knowledge base and can organize this knowledge content in a human-like manner. This knowledge format enables the large-scale model to provide knowledge retrieval and question-and-answer capabilities through dialogue, question answering, and other methods.

[0206] All three types of knowledge originate from historical requirement documents and are generated using different technical methods. They can also be continuously generated through machine learning and large-scale model training. Furthermore, the three types of knowledge can mutually verify each other's accuracy, reducing the complexity of manually evaluating training results.

[0207] For example, to help understand the implementation process of the knowledge archiving method obtained by combining this embodiment with the above embodiments, please refer to... Figure 12 , Figure 12 A simplified flowchart of a knowledge archiving method is provided, specifically:

[0208] First, newly approved business requirement documents undergo word segmentation and document splitting, followed by document vectorization and classification using a vector machine algorithm (SVM algorithm), and a document vectorization index is constructed. Simultaneously, a basic large-scale model (i.e., a basic large-scale language model) is used to generate a summary of business changes in the requirement documents. By comparing the new business requirement document summary with existing business categories in the library through keyword frequency, the classification result of the new document change is output, serving as the classification label result of the large-scale model for the new document, thus obtaining the business and document classification mapping relationship. At the same time, the classification label result of the large-scale model is compared with the label result output by the vector machine algorithm. For parts where the two labeling methods match, the new document can be directly archived into the vector library, including the document classification and vector index library; only the parts that do not match require manual inspection and labeling.

[0209] As large models are continuously optimized and vector database data accumulates (including manual inspection and labeling), the labeling of these classification results will become increasingly accurate, gradually reducing the workload of manual inspection and labeling in this part of the work.

[0210] The foundational model for generating document summaries and structuring the knowledge information within documents is a proprietary model trained on top of an open-source model, using historical business requirement documents and pre-organized summary and structured knowledge information. This model is continuously optimized through training using summarization and structured knowledge results already determined to be accurate within the workflow. The determined accurate information includes: the consistent parts between the model's output and the vectorized output during the workflow, and the parts where inconsistencies arise are manually checked and labeled.

[0211] Meanwhile, the large model possesses the ability to answer questions based on knowledge base content. It queries a pre-maintained list of business and test questions related to new business needs and automatically answers questions based on the summary information of newly generated documents from the large model. The accuracy of the question-and-answer records can be manually verified to check if the merged large model meets expectations. Data that meets expectations is also used for optimizing and training the large model. The model's ability to generate answers is continuously optimized and trained.

[0212] The private large model exists in different versions. After optimization and training, the large model and the previous version of the large model are merged in terms of capabilities. Through continuous optimization and training, the accuracy of the large model will continue to improve with version iteration and capability merging, giving the process a certain degree of self-optimization.

[0213] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the knowledge archiving method of this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0214] This application also provides a knowledge archiving device; please refer to... Figure 13 The knowledge archiving device includes:

[0215] The first classification module 10 is used to classify unstructured target requirement documents by business type using a vector machine algorithm to obtain the first classification result;

[0216] The second classification module 20 is used to generate a document summary of the target requirement document using the basic large model, and classify the target requirement document by business type according to the document summary to obtain the second classification result. The basic large model is a large language model trained using historical requirement documents and historical document summaries.

[0217] The classification result determination module 30 is used to determine the target classification result of the target requirement document based on the first classification result and the second classification result.

[0218] The archiving module 40 is used to archive the target requirement documents, document summaries, and target classification results in a structured form to a pre-defined knowledge base.

[0219] Optionally, the knowledge base includes a vector library, and the first classification module 10 is also used for:

[0220] After performing word segmentation and document segmentation on the target requirement document, a vector machine algorithm is used to vectorize the target requirement document after word segmentation and document segmentation to obtain a structured document vector.

[0221] The structured document vectors are matched with the historical document vectors in the vector library to identify target historical requirement documents whose vector similarity is higher than a preset first similarity threshold.

[0222] Obtain the business type from the target historical requirements document and use the business type as the first classification result of the target requirements document.

[0223] Optionally, the knowledge base includes a large model, and the knowledge archiving device also includes a large model update module (not shown), which is used for:

[0224] Based on the target classification results, search the knowledge base for historical target requirement documents with the same business type as the target requirement document, and analyze the differences between the target requirement document and the historical target requirement document to determine the business change content in the knowledge base.

[0225] Retrieve business questions and standard answers corresponding to the document summary from a pre-defined question list library;

[0226] Input the business change information into the previous version of the large model for model training to obtain the initial updated large model;

[0227] Ask business questions about the initial update of the large model and obtain the output answers;

[0228] Based on the comparison between the output answer and the standard answer, the target update model is obtained. Among them, the previous version model, the basic model, and the target update model are models of different update rounds. The target update model is used for the automatic archiving of subsequent requirement documents and knowledge Q&A services.

[0229] Optionally, the classification result determination module 30 is also used for:

[0230] The first classification result and the second classification result are cross-validated;

[0231] If the first classification result and the second classification result are consistent, the target classification result of the target requirement document is determined to be consistent with the first classification result and the second classification result.

[0232] If the first classification result and the second classification result are inconsistent, a verification window is generated to receive user operations, the target classification result is determined based on the user operations, and the first classification result and / or the second classification result is corrected based on the target classification result.

[0233] Optionally, the knowledge archiving apparatus further includes a comparison and verification module (not shown), which is used for:

[0234] Construct a document index for the target requirement documents based on the target classification results;

[0235] The vectorized target requirement documents, document indexes, and target classification results are archived into a vector library.

[0236] Archive the target requirements document, document summary, and target classification results into the large model;

[0237] We continuously compare and verify the knowledge archived in the large model and the vector library.

[0238] Optionally, the knowledge archiving device also includes a content query module (not shown), which is used for:

[0239] Upon receiving an externally input content query command, identify the user intent corresponding to the content query command;

[0240] The similarity algorithm is used to select target content in the knowledge base and output the target content, where the target content is content whose similarity to the user's query intent is higher than a preset second similarity threshold.

[0241] The knowledge archiving device provided in this application, employing the knowledge archiving method described in the above embodiments, can efficiently archive unstructured business requirement documents into structured document knowledge. Compared with the prior art, the beneficial effects of the knowledge archiving device provided in this application are the same as those of the knowledge archiving method described in the above embodiments, and other technical features in the knowledge archiving device are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0242] This application provides an electronic device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform the knowledge archiving method in Embodiment 1 above.

[0243] The following is for reference. Figure 14 The diagram illustrates a structural schematic of an electronic device suitable for implementing embodiments of this application. The electronic devices in these embodiments may include, but are not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 14 The knowledge archiving device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0244] like Figure 14As shown, the electronic device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the electronic device. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. While electronic devices with various systems are shown in the figures, it should be understood that implementation or possession of all the systems shown is not required. More or fewer systems may be implemented alternatively.

[0245] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0246] The electronic device provided in this application embodiment employs the knowledge archiving method described above, which can efficiently archive unstructured business requirement documents into structured document knowledge. Compared with the prior art, the beneficial effects of the electronic device provided in this application embodiment are the same as those of the knowledge archiving method described above, and other technical features of this electronic device are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0247] It should be understood that the various parts disclosed in the embodiments of this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0248] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0249] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, which are used to execute the knowledge archiving method described above.

[0250] The computer-readable storage medium provided in this application embodiment may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections with one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0251] The aforementioned computer-readable storage medium may be included in an electronic device or may exist independently without being assembled into an electronic device.

[0252] The aforementioned computer-readable storage medium carries one or more programs. When the aforementioned one or more programs are executed by an electronic device, the electronic device causes the following to occur: It uses a vector machine algorithm to classify unstructured target requirement documents by business type, obtaining a first classification result; it uses a basic large model to generate a document summary of the target requirement document, and classifies the target requirement document by business type based on the document summary, obtaining a second classification result, wherein the basic large model is a large language model trained using historical requirement documents and historical document summaries; it determines the target classification result of the target requirement document based on the first and second classification results; and it archives the target requirement document, document summary, and target classification result in a structured form into a pre-defined knowledge base.

[0253] Computer program code for performing the operations of the embodiments of this application can be written in one or more programming languages ​​or a combination thereof. These programming languages ​​include object-oriented programming languages—such as Java, Smalltalk, and C++—and conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0254] 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 the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can 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.

[0255] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0256] The readable storage medium provided in this application embodiment is a computer-readable storage medium. This medium stores computer-readable program instructions (i.e., a computer program) for executing the aforementioned knowledge archiving method, enabling efficient archiving of unstructured business requirement documents into structured document knowledge. Compared to the prior art, the beneficial effects of the computer-readable storage medium provided in this application embodiment are the same as those of the knowledge archiving method provided in the above embodiments, and will not be repeated here.

[0257] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the knowledge archiving method described above.

[0258] The computer program product provided in this application can efficiently archive unstructured business requirement documents into structured document knowledge. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the knowledge archiving method provided in the above embodiments, and will not be repeated here.

[0259] The above are only some embodiments of this application and do not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A knowledge archiving method, characterized in that, The method includes: The vector machine algorithm is used to classify unstructured target requirement documents by business type, and the first classification result is obtained; A basic large model is used to generate a document summary of the target requirement document, and the target requirement document is classified by business type according to the document summary to obtain a second classification result. The basic large model is a large language model trained using historical requirement documents and historical document summaries. The target classification result of the target requirement document is determined based on the first classification result and the second classification result. The target requirement document, the document summary, and the target classification results are archived in a structured form to a pre-defined knowledge base; The knowledge base includes a large model, and after the step of determining the target classification result of the target requirement document based on the first classification result and the second classification result, the following is also included: Based on the target classification results, search the knowledge base for historical target requirement documents with the same business type as the target requirement document, and analyze the differences between the target requirement document and the historical target requirement document to determine the business change content in the knowledge base; Retrieve the business questions and standard answers corresponding to the document summary from a pre-set question list library; The business change content is input into the previous version of the large model for model training to obtain the initial updated large model; The business question is posed to the initial updated large model, and the output answer is obtained. Perform a semantic similarity comparison between the output answer and the standard answer; If the similarity is higher than a preset threshold, it is confirmed that the initial updated large model has grasped the content of the business change. If the similarity is lower than a preset threshold, manual review is initiated to determine the source of the difference. When the difference originates from a business change and the output answer is accurate, it is confirmed that the initial updated big model has grasped the content of the business change. The target updated large model is obtained based on the similarity comparison results and the manual review results. The previous version large model, the basic large model, and the target updated large model are large models from different update rounds. The target updated large model is used for the automatic archiving of subsequent requirement documents and knowledge Q&A services.

2. The method as described in claim 1, characterized in that, The knowledge base includes a vector library. The step of using a vector machine algorithm to classify unstructured target requirement documents by business type to obtain a first classification result includes: After performing word segmentation and document segmentation on the target requirement document, a vector machine algorithm is used to vectorize the target requirement document after word segmentation and document segmentation to obtain a structured document vector. The structured document vector is matched with the historical document vectors in the vector library to determine the target historical demand document whose vector similarity is higher than a preset first similarity threshold from the vector library; Obtain the business type of the target historical requirement document and use the business type as the first classification result of the target requirement document.

3. The method as described in claim 1, characterized in that, The step of determining the target classification result of the target requirement document based on the first classification result and the second classification result includes: The first classification result and the second classification result are mutually verified; If the first classification result and the second classification result are consistent, the target classification result of the target requirement document is determined to be consistent with the first classification result and the second classification result. If the first classification result and the second classification result are inconsistent, a verification window is generated to receive user operations, the target classification result is determined based on the user operations, and the first classification result and / or the second classification result is corrected based on the target classification result.

4. The method as described in claim 2, characterized in that, The step of archiving the target requirement document, the document summary, and the target classification result in a structured form to a preset knowledge base includes: A document index for the target requirement document is constructed based on the target classification results; The vectorized target requirement document, the document index, and the target classification result are archived into a vector library. The target requirement document, the document summary, and the target classification results are archived into the large model; After the step of archiving the target requirement document, the document summary, and the target classification result in a structured form to a preset knowledge base, the method further includes: The knowledge archived by the large model and the vector library is continuously compared and verified.

5. The method as described in claim 4, characterized in that, The method further includes: Upon receiving an externally input content query instruction, the system identifies the user intent corresponding to the content query instruction. The target content is selected from the knowledge base using a similarity algorithm and then output. The target content is content whose similarity to the user's intent is higher than a preset second similarity threshold.

6. A knowledge archiving device, characterized in that, The device includes: The first classification module is used to classify unstructured target requirement documents by business type using a vector machine algorithm to obtain the first classification result; The second classification module is used to generate a document summary of the target requirement document using the basic large model, and classify the target requirement document by business type according to the document summary to obtain the second classification result. The basic large model is a large language model trained using historical requirement documents and historical document summaries. The classification result determination module is used to determine the target classification result of the target requirement document based on the first classification result and the second classification result. The archiving module is used to archive the target requirement document, the document summary, and the target classification results in a structured form to a preset knowledge base; The knowledge base includes a large model, and the knowledge archiving device is further used for: Based on the target classification results, search the knowledge base for historical target requirement documents with the same business type as the target requirement document, and analyze the differences between the target requirement document and the historical target requirement document to determine the business change content in the knowledge base; Retrieve the business questions and standard answers corresponding to the document summary from a pre-set question list library; The business change content is input into the previous version of the large model for model training to obtain the initial updated large model; The business question is posed to the initial updated large model, and the output answer is obtained. Perform a semantic similarity comparison between the output answer and the standard answer; If the similarity is higher than a preset threshold, it is confirmed that the initial updated large model has grasped the content of the business change. If the similarity is lower than a preset threshold, manual review is initiated to determine the source of the difference. When the difference originates from a business change and the output answer is accurate, it is confirmed that the initial updated big model has grasped the content of the business change. The target updated large model is obtained based on the similarity comparison results and the manual review results. The previous version large model, the basic large model, and the target updated large model are large models from different update rounds. The target updated large model is used for the automatic archiving of subsequent requirement documents and knowledge Q&A services.

7. An electronic device, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the knowledge archiving method as described in any one of claims 1 to 5.

8. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the knowledge archiving method as described in any one of claims 1 to 5.

9. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps of the knowledge archiving method as described in any one of claims 1 to 5.

Images