Knowledge base management method, device, and storage medium

Abstract

The application discloses a knowledge base management method and device and a storage medium, and relates to the technical field of data management. The method comprises the following steps: acquiring external knowledge original text data, sequentially performing classification and semantic embedding processing, acquiring knowledge field classification label data and knowledge original high-dimensional semantic vector data; performing wavelet decomposition compression processing on the knowledge original high-dimensional semantic vector data, and acquiring standardized low-dimensional compressed semantic vector data; encapsulating the external knowledge original text data, the knowledge field classification label data, the knowledge original high-dimensional semantic vector data and the standardized low-dimensional compressed semantic vector data into knowledge nodes; storing the knowledge nodes into a knowledge base after allocating unique identifications to the knowledge nodes; and acquiring target style vector data according to target style reference text data, and taking the target style vector data as an output constraint of the knowledge nodes. The application can guarantee the accuracy of knowledge input of the knowledge base and personalized knowledge output.

Description

Technical Field

[0001] This application relates to the field of data management technology, specifically to a knowledge base management method, device, and storage medium. Background Technology

[0002] With the development of AI knowledge base technology, private knowledge bases have become the core carrier for knowledge accumulation, content retrieval, and intelligent generation by the controlling entity. Current mainstream knowledge base management solutions mostly involve simply classifying and storing raw knowledge text data, and then outputting the text using pre-defined language and a generic style. This approach to knowledge base management has the following drawbacks in practical applications: The content of knowledge output is highly homogenized. As the original knowledge text data is continuously accumulated and stored, there is a risk of semantic and logical conflicts between new input data and existing data, making it impossible to guarantee the accuracy of knowledge. Summary of the Invention

[0003] In view of the shortcomings of the existing technology, the technical problem to be solved by this application is: how to avoid homogeneous output of knowledge and ensure the accuracy of the knowledge input into the knowledge base.

[0004] To achieve the above objectives, in a first aspect, embodiments of this application provide a knowledge base management method, the method comprising the following steps: We acquire raw text data of external knowledge, classify and semantically embed it sequentially, and obtain knowledge domain classification label data and raw high-dimensional semantic vector data of knowledge. Wavelet decomposition and compression are performed on the original high-dimensional semantic vector data of knowledge to obtain standardized low-dimensional compressed semantic vector data. External knowledge raw text data, knowledge domain classification label data, knowledge raw high-dimensional semantic vector data, and standardized low-dimensional compressed semantic vector data are encapsulated into knowledge nodes; After assigning a unique identifier to a knowledge node, it is stored in the knowledge base; Based on the target style reference text data, obtain the target style vector data, and use the target style vector data as the output constraint for the knowledge nodes.

[0005] In conjunction with the first aspect, in one implementation, the process for obtaining the knowledge domain classification label data includes: Based on external knowledge and raw text data, extract word frequency features, text source features, and text structure features; Input the original text data of external knowledge, along with the corresponding word frequency features, text source features, and text structure features, into an offline knowledge classifier to obtain knowledge domain classification label data.

[0006] In conjunction with the first aspect, in one implementation, the process of performing wavelet decomposition and compression processing on the original high-dimensional semantic vector data of knowledge to obtain standardized low-dimensional compressed semantic vector data includes: The numerical distribution characteristics of each dimension of the original high-dimensional semantic vector data of knowledge are obtained. The absolute value of the value of each dimension is counted one by one. When the absolute value is greater than a preset threshold, the dimension is determined to be a strong signal dimension; when the frequency of the dimension value is less than a preset threshold, the dimension is determined to be a weak signal dimension. Determine whether the overall number of dimensions of the original high-dimensional semantic vector data is less than the preset number of dimensions. If so, strengthen the strong signal dimension according to the weight coefficient of the first preset ratio, weaken the weak signal dimension according to the weight coefficient of the second preset ratio, and then compress it through wavelet decomposition at a compression ratio of 2:1 to obtain standardized low-dimensional compressed semantic vector data. Otherwise, after segmenting the original high-dimensional semantic vector data of knowledge with a fixed number of dimensions, the data is compressed by wavelet decomposition at a compression ratio of 4:1, and then weighted and fused to obtain standardized low-dimensional compressed semantic vector data.

[0007] In conjunction with the first aspect, in one implementation, after acquiring a new knowledge node, it is determined whether the new knowledge node can be stored in the knowledge base. This determination process includes: Based on the knowledge domain classification label data, existing knowledge nodes in the same domain as the new knowledge node are selected from the knowledge base; Calculate the semantic similarity between the new knowledge node and the corresponding standardized low-dimensional compressed semantic vector data of the existing knowledge node. When the semantic similarity is greater than a preset threshold, the new knowledge node and the corresponding existing knowledge node are determined to be similar knowledge nodes. Determine whether there are semantic and logical conflicts between similar knowledge nodes. If so, proceed to manual review; otherwise, assign a unique identifier to the new knowledge node and store it in the knowledge base.

[0008] In conjunction with the first aspect, in one implementation, the manual review process includes: The content of new knowledge nodes and existing knowledge nodes is manually reviewed. When a new knowledge node is accepted, it is directly assigned a unique identifier and stored in the knowledge base. When adopting existing knowledge nodes, the original text data of external knowledge corresponding to the new knowledge node is modified before the new knowledge node is re-acquired.

[0009] In conjunction with the first aspect, in one implementation, the process of obtaining target style vector data based on target style reference text data includes: Collect original compliant text data from the controlling entity, preprocess it, and then obtain target style reference text data; Input the target style reference text data into the pre-trained style vector extraction model to obtain the target style vector data.

[0010] In conjunction with the first aspect, in one implementation, the style vector extraction model is used to extract vocabulary usage features, syntactic structure features, rhetorical expression features, and logical thinking features from target style reference text data, and to perform feature concatenation and dimensionality reduction mapping on the vocabulary usage features, syntactic structure features, rhetorical expression features, and logical thinking features to obtain target style vector data.

[0011] In conjunction with the first aspect, in one implementation, the style vector extraction model is constrained by a joint loss value; this joint loss value is obtained by weighted summation of the contrastive learning loss of vocabulary usage features, the contrastive learning loss of syntactic structure features, the classification cross-entropy loss of rhetorical expression features, and the contrastive learning loss of logical thinking features.

[0012] Secondly, embodiments of this application provide a knowledge base management device, the knowledge base management device including a processor, a memory, and a knowledge base management program stored in the memory and executable by the processor, wherein when the knowledge base management program is executed by the processor, it implements the method provided in the first aspect.

[0013] Thirdly, embodiments of this application provide a computer-readable storage medium storing a knowledge base management program, wherein when the knowledge base management program is executed, it implements the method provided in the first aspect.

[0014] Compared with the prior art, the advantages of this application are: (1) Knowledge nodes are constructed using external knowledge original text data, and style vectors are trained by controlling the subject's original compliant text data. The style vectors are used as output constraints for all knowledge nodes. This allows all output content to conform to the controlling subject's personal language style and thinking logic without changing the authenticity of the knowledge base's objective knowledge, thus avoiding the defects of homogenization in traditional knowledge base output.

[0015] (2) By encapsulating the original text data of external knowledge, the classification label data of knowledge domain, the original high-dimensional semantic vector data of knowledge, and the standardized low-dimensional compressed semantic vector data into knowledge nodes, it is possible not only to preserve the integrity and traceability of the original knowledge, but also to provide structured data support for semantic comparison and conflict detection. In this way, during the accumulation and storage of the original knowledge text data, the accuracy of the content in the knowledge base can be effectively guaranteed by identifying semantic logical conflict content. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of the knowledge base management method in the embodiments of this application; Figure 2 This is a schematic diagram of the hardware structure of the knowledge base management device involved in the embodiments of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the order described. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0020] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0021] Firstly, embodiments of this application provide a knowledge base management method, referring to... Figure 1 The method includes the following steps: S1. Obtain raw text data of external knowledge (such as documents, notes, data texts, work documents, etc. actively input by the control subject), classify and semantically embed them in sequence, and obtain knowledge domain classification label data and raw high-dimensional semantic vector data of knowledge. S2. Perform wavelet decomposition and compression on the original high-dimensional semantic vector data of knowledge to obtain standardized low-dimensional compressed semantic vector data. S3. Encapsulate the original text data of external knowledge, the classification label data of knowledge domain, the original high-dimensional semantic vector data of knowledge, and the standardized low-dimensional compressed semantic vector data into knowledge nodes; S4. After assigning a unique identifier to the knowledge node, store it in the knowledge base; S5. Based on the original compliant text data of the controlling entity (such as the controlling entity's original notes, insights, and creative texts), obtain the target style vector data, and use the target style vector data as the output constraint of the knowledge node.

[0022] To further explain, encapsulation as a knowledge node refers to organizing external knowledge raw text data, knowledge domain classification label data, knowledge raw high-dimensional semantic vector data, and standardized low-dimensional compressed semantic vector data into an independently identifiable knowledge unit in a logically related manner. This logical relationship includes, but is not limited to: encapsulation within the same data structure, distributed storage linked by a unique identifier, and heterogeneous storage linked by an index.

[0023] By constructing knowledge nodes using original text data from external sources and training style vectors using original, compliant text data from the controlling entity, and using these style vectors as output constraints for all knowledge nodes, it is possible to ensure that all output content aligns with the controlling entity's personal language style and thought process without altering the objective authenticity of the knowledge base. This avoids the homogenization of traditional knowledge base outputs.

[0024] In one embodiment, the process of obtaining target style vector data based on target style reference text data includes: Collect original compliant text data from the control entity, preprocess it (word segmentation, syntactic parsing, and semantic embedding), and then obtain target style reference text data; Input the target style reference text data into the pre-trained style vector extraction model to obtain the target style vector data.

[0025] By using the target style vector data as the output constraint for knowledge nodes, the output of data can be achieved by controlling the style of the subject when outputting knowledge base content.

[0026] The aforementioned target style vector data is a preferred output constraint method used to control the expressive style of the knowledge base's output content. Aligning the target style vector data with the compliant text style features of the controlling entity ensures that the knowledge base's output content conforms to the controlling entity's language style and thought process while maintaining the objectivity of knowledge. After the target style vector data is bound to knowledge nodes, all outputs of the knowledge base are constrained by this target style vector data, thereby achieving personalized adaptation of the knowledge base's output.

[0027] The above-mentioned style vector extraction model can be constructed in the following ways: We adopt a lightweight Transformer Encoder architecture, remove the general semantic classification head, and build a four-channel style feature extraction branch to extract style features of different dimensions (lexical usage features, syntactic structure features, rhetorical expression features, and logical thinking features). After each channel's style feature extraction branch independently extracts style features of the corresponding dimension, the feature fusion layer performs feature splicing and dimensionality reduction mapping to obtain the target style vector data.

[0028] Based on this, the training process of the above style vector extraction model includes: The vocabulary usage features and syntactic structure features of the original compliant text data of the control subject are extracted by self-supervised method. After adding rhetorical tags and logical chain tags to the original compliant text data of the control subject, a sample data is generated. After accumulating at least 50 sample data, a dataset is generated. The dataset is divided into a training set and a validation set according to a set ratio (8:2); The style vector extraction model is trained using the training set. During the process, the Adam optimizer is used for iterative training with a maximum of 50 iterations. The contrastive learning loss of vocabulary usage features, contrastive learning loss of syntactic structure features, classification cross-entropy loss of rhetorical expression features, and contrastive learning loss of logical thinking features are weighted (the weight coefficients of each loss term are determined by grid search) and summed to obtain the joint loss value. After the joint loss value converges or reaches the maximum number of iterations, the style vector extraction model is validated using a validation set. The training of the style vector extraction model is completed when the style recognition accuracy output by the style vector extraction model is greater than a preset threshold (≥93%).

[0029] In one embodiment, the process of acquiring and classifying raw text data of external knowledge includes: Based on external knowledge and raw text data, extract word frequency features, text source features, and text structure features; The raw text data of external knowledge, along with its corresponding word frequency features, text source features, and text structure features, are input into an offline knowledge classifier to obtain knowledge domain classification label data. This enables domain classification of the raw text data of external knowledge, thereby narrowing the scope of subsequent semantic comparison and improving data processing efficiency.

[0030] In one embodiment, the process of performing wavelet decomposition and compression on the original high-dimensional semantic vector data of knowledge to obtain standardized low-dimensional compressed semantic vector data includes: S21. Obtain the numerical distribution characteristics of each dimension of the original high-dimensional semantic vector data of knowledge, and count the absolute value of each dimension's value. When the absolute value is greater than a preset threshold (which can be manually defined), the dimension is determined to be a strong signal dimension; when the frequency of the dimension's value is less than a preset threshold, the dimension is determined to be a weak signal dimension. S22. Determine whether the overall number of dimensions of the original high-dimensional semantic vector data of knowledge is less than the preset number of dimensions (≤512 dimensions). If yes, proceed to S23; otherwise, proceed to S24. S23. After strengthening the strong signal dimension with a weight coefficient of the first preset ratio (e.g., 1.2 to 1.5 times) and weakening the weak signal dimension with a weight coefficient of the second preset ratio (e.g., 0.1 times), the data is compressed by wavelet decomposition at a compression ratio of 2:1 to obtain standardized low-dimensional compressed semantic vector data. S24. After segmenting the original high-dimensional semantic vector data of knowledge with a fixed number of dimensions (e.g., 512 dimensions), the data is compressed using wavelet decomposition at a compression ratio of 4:1. After segment-by-segment compression, weighted fusion is performed to obtain standardized low-dimensional compressed semantic vector data. The weight coefficients for weighted fusion are obtained using the following methods: Obtain the number of strong signal dimensions in the original high-dimensional semantic vector data of each knowledge segment, and obtain the ratio of the number of strong signal dimensions to the total number of strong signal dimensions. Use this ratio as the weight coefficient of the original high-dimensional semantic vector data of that knowledge segment.

[0031] Furthermore, for knowledge original high-dimensional semantic vector data segments that do not have strong signal dimensions, fixed weight coefficients are assigned according to the number of knowledge original high-dimensional semantic vector data segments, and the weight coefficients of the remaining knowledge original high-dimensional semantic vector data segments are adjusted proportionally to ensure that the sum of the weight coefficients of all knowledge original high-dimensional semantic vector data segments is one.

[0032] This leads to the construction of a differentiated compression mechanism for original high-dimensional semantic vector data. By statistically analyzing the absolute values ​​of each dimension of the original high-dimensional semantic vector data, strong and weak signal dimensions are distinguished, and targeted dimension enhancement and noise reduction are performed. Simultaneously, based on the number of dimensions in the original high-dimensional semantic vector data, high- and low-dimensional vector branches are divided, and different compression ratios and segmented fusion strategies are matched. Overall optimization compression of low-dimensional vectors and segmented weighted fusion compression of high-dimensional vectors are performed. This effectively eliminates redundant noise in the original high-dimensional semantic vector data, enhances core semantic features, and avoids semantic loss, feature ambiguity, and wasted computing power caused by uniform compression, thus achieving a balance between semantic integrity and device computing efficiency.

[0033] As an extension, when acquiring standardized low-dimensional compressed semantic vector data, wavelet decomposition compression can be replaced by performing discrete cosine transform, wavelet packet transform, or other frequency domain analysis methods on the original high-dimensional semantic vector data.

[0034] In one embodiment, when acquiring new external knowledge original text data, the above steps S1-S3 are repeated to acquire new knowledge nodes.

[0035] Based on this, the process for determining whether a new knowledge node can be stored in the knowledge base includes: Define the knowledge nodes in the knowledge base as existing knowledge nodes, and filter existing knowledge nodes in the same domain as the new knowledge node based on the knowledge domain classification label data. Calculate the semantic similarity between the new knowledge node and the corresponding standardized low-dimensional compressed semantic vector data of the existing knowledge node (cosine similarity can be used for calculation). Determine if the semantic similarity is greater than a preset threshold (>0.8). If so, determine if the new knowledge node and the corresponding existing knowledge node are similar knowledge nodes; otherwise, directly assign a unique identifier to the new knowledge node and store it in the knowledge base. Determine whether similar knowledge nodes have semantic and logical conflicts (such as conflicting definitions, conflicting parameters, conflicting conditions for establishment, conflicting conclusions, etc.). If so, proceed to manual review; otherwise, directly assign a unique identifier to the new knowledge node and store it in the knowledge base.

[0036] In this way, during the accumulation and storage of original knowledge text data, semantic similarity and semantic logical conflicts are processed layer by layer to identify content with semantic logical conflicts. Combined with manual review, this effectively ensures the accuracy of the content in the knowledge base.

[0037] The above manual review process includes: The content of new knowledge nodes and existing knowledge nodes is manually reviewed. When a new knowledge node is accepted, it is directly assigned a unique identifier and stored in the knowledge base. When adopting existing knowledge nodes, after correcting the content of the original text data of external knowledge corresponding to the new knowledge node, repeat steps S1-S3 to obtain the new knowledge node, and re-select existing knowledge nodes in the same field as the new knowledge node, and make a judgment on whether there are semantic and logical conflicts between similar knowledge nodes.

[0038] Secondly, embodiments of this application provide a knowledge base management device, which can be a personal computer (PC), laptop computer, server, or other device with data processing capabilities.

[0039] Reference Figure 2 , Figure 2 This is a schematic diagram of the hardware structure of the knowledge base management device involved in the embodiments of this application. In the embodiments of this application, the knowledge base management device may include a processor, a memory, a communication interface, and a communication bus.

[0040] The communication bus can be of any type and is used to interconnect the processor, memory, and communication interface.

[0041] The communication interface includes input / output (I / O) interfaces, physical interfaces, and logical interfaces used for interconnecting devices within the knowledge base management device, as well as interfaces used for interconnecting the knowledge base management device with other devices (such as other computing devices or control units). Physical interfaces can be Ethernet interfaces, fiber optic interfaces, ATM interfaces, etc.; control units can be displays, keyboards, etc.

[0042] Memory can be various types of storage media, such as random access memory (RAM), read-only memory (ROM), non-volatile RAM (NVRAM), flash memory, optical storage, hard disk, programmable ROM (PROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), etc.

[0043] The processor can be a general-purpose processor, which can call the knowledge base management program stored in the memory and execute the knowledge base management method provided in the embodiments of this application. For example, the general-purpose processor can be a central processing unit (CPU). The method executed when the knowledge base management program is called can be referred to in the various embodiments of the knowledge base management method of this application, and will not be repeated here.

[0044] Those skilled in the art will understand that Figure 2 The hardware structure shown does not constitute a limitation of this application and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0045] Thirdly, embodiments of this application also provide a computer-readable storage medium.

[0046] The computer-readable storage medium of this application stores a knowledge base management program, wherein when the knowledge base management program is executed by a processor, it implements the steps of the knowledge base management method described above.

[0047] The method implemented when the knowledge base management program is executed can be referred to in various embodiments of the knowledge base management method of this application, and will not be repeated here.

[0048] It should be noted that the sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0049] The terms "comprising" and "having," and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus. The terms "first," "second," and "third," etc., are used to distinguish different objects, etc., and do not indicate a sequence, nor do they limit "first," "second," and "third" to different types.

[0050] In the description of the embodiments of this application, terms such as "exemplary," "for example," or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "exemplary," "for example," or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary," "for example," or "for instance" is intended to present the relevant concepts in a concrete manner.

[0051] In the description of the embodiments of this application, unless otherwise stated, " / " means "or". For example, A / B can mean A or B. The "and / or" in the text is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist simultaneously, and B exists alone. In addition, in the description of the embodiments of this application, "multiple" means two or more.

[0052] In some processes described in the embodiments of this application, multiple operations or steps are included in a specific order. However, it should be understood that these operations or steps may not be executed in the order they appear in the embodiments of this application, or they may be executed in parallel. The sequence number of the operation is only used to distinguish different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed sequentially or in parallel, and these operations or steps may be combined.

[0053] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device to execute the methods described in the various embodiments of this application.

[0054] The above are merely specific embodiments of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application. Therefore, the protection scope of this application should be determined by the scope of the claims.

Claims

1. A knowledge base management method, characterized in that, The method includes the following steps: We acquire raw text data of external knowledge, classify and semantically embed it sequentially, and obtain knowledge domain classification label data and raw high-dimensional semantic vector data of knowledge. Wavelet decomposition and compression are performed on the original high-dimensional semantic vector data of knowledge to obtain standardized low-dimensional compressed semantic vector data. External knowledge raw text data, knowledge domain classification label data, knowledge raw high-dimensional semantic vector data, and standardized low-dimensional compressed semantic vector data are encapsulated into knowledge nodes; After assigning a unique identifier to a knowledge node, it is stored in the knowledge base; Based on the target style reference text data, obtain the target style vector data, and use the target style vector data as the output constraint for the knowledge nodes.

2. The knowledge base management method as described in claim 1, characterized in that, The process for obtaining the knowledge domain classification label data includes: Based on external knowledge and raw text data, extract word frequency features, text source features, and text structure features; Input the original text data of external knowledge, along with the corresponding word frequency features, text source features, and text structure features, into an offline knowledge classifier to obtain knowledge domain classification label data.

3. The knowledge base management method as described in claim 1, characterized in that, The process of performing wavelet decomposition and compression on the original high-dimensional semantic vector data of knowledge to obtain standardized low-dimensional compressed semantic vector data includes: The numerical distribution characteristics of each dimension of the original high-dimensional semantic vector data of knowledge are obtained. The absolute value of the value of each dimension is counted one by one. When the absolute value is greater than a preset threshold, the dimension is determined to be a strong signal dimension; when the frequency of the dimension value is less than a preset threshold, the dimension is determined to be a weak signal dimension. Determine whether the overall number of dimensions of the original high-dimensional semantic vector data is less than the preset number of dimensions. If so, strengthen the strong signal dimension according to the weight coefficient of the first preset ratio, weaken the weak signal dimension according to the weight coefficient of the second preset ratio, and then compress it through wavelet decomposition at a compression ratio of 2:1 to obtain standardized low-dimensional compressed semantic vector data. Otherwise, after segmenting the original high-dimensional semantic vector data of knowledge with a fixed number of dimensions, the data is compressed by wavelet decomposition at a compression ratio of 4:1, and then weighted and fused to obtain standardized low-dimensional compressed semantic vector data.

4. The knowledge base management method as described in claim 1, characterized in that, After acquiring a new knowledge node, the process of determining whether the new knowledge node can be stored in the knowledge base includes: Based on the knowledge domain classification label data, existing knowledge nodes in the same domain as the new knowledge node are selected from the knowledge base; Calculate the semantic similarity between the new knowledge node and the corresponding standardized low-dimensional compressed semantic vector data of the existing knowledge node. When the semantic similarity is greater than a preset threshold, the new knowledge node and the corresponding existing knowledge node are determined to be similar knowledge nodes. Determine whether there are semantic and logical conflicts between similar knowledge nodes. If so, proceed to manual review; otherwise, assign a unique identifier to the new knowledge node and store it in the knowledge base.

5. The knowledge base management method as described in claim 4, characterized in that, The manual review process includes: The content of new knowledge nodes and existing knowledge nodes is manually reviewed. When a new knowledge node is accepted, it is directly assigned a unique identifier and stored in the knowledge base. When adopting existing knowledge nodes, the original text data of external knowledge corresponding to the new knowledge node is modified before the new knowledge node is re-acquired.

6. The knowledge base management method as described in claim 1, characterized in that, The process of obtaining target style vector data based on target style reference text data includes: Collect original compliant text data from the controlling entity, preprocess it, and then obtain target style reference text data; Input the target style reference text data into the pre-trained style vector extraction model to obtain the target style vector data.

7. The knowledge base management method as described in claim 6, characterized in that, The style vector extraction model is used to extract vocabulary usage features, syntactic structure features, rhetorical expression features, and logical thinking features from target style reference text data. It then performs feature concatenation and dimensionality reduction mapping on the vocabulary usage features, syntactic structure features, rhetorical expression features, and logical thinking features to obtain target style vector data.

8. The knowledge base management method as described in claim 7, characterized in that, The style vector extraction model is constrained by a joint loss value, which is obtained by weighted summation of the contrastive learning loss of vocabulary usage features, the contrastive learning loss of syntactic structure features, the classification cross-entropy loss of rhetorical expression features, and the contrastive learning loss of logical thinking features.

9. A knowledge base management device, characterized in that, The knowledge base management device includes a processor, a memory, and a knowledge base management program stored in the memory and executable by the processor, wherein when the knowledge base management program is executed by the processor, it implements the steps of the knowledge base management method as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a knowledge base management program, wherein when the knowledge base management program is executed, it implements the steps of the knowledge base management method as described in any one of claims 1 to 8.

Images