AI-based intelligent knowledge base management method and system

Abstract

The application relates to the technical field of artificial intelligence, and discloses an AI-based intelligent knowledge base management method and system, which comprises the following steps: acquiring user interaction data and multi-source heterogeneous data, performing word segmentation and vectorization processing, and generating a semantic feature vector set; performing information contribution degree analysis according to the semantic feature vector set, and generating a key semantic feature set; performing time sequence correlation analysis according to the key semantic feature set and the user interaction data, and obtaining a time feature vector; performing correlation analysis according to the key semantic feature set and the time feature vector, and obtaining an optimized classification logic; performing knowledge graph construction according to the optimized classification logic and the key semantic feature set, and generating a dynamic knowledge item set; and performing vector retrieval matching according to the dynamic knowledge item set and the user interaction data, and obtaining a final output result. The method can realize dynamic knowledge item division based on user behavior.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to an AI-based intelligent knowledge base management method and system. Background Technology

[0002] Currently, with the rapid development of the digital age, the scale of data is exploding, encompassing various forms such as text, images, and user interaction records. Intelligent knowledge base management, as one of the core technologies in modern information processing, aims to efficiently organize, store, and utilize massive amounts of data through advanced technological means, providing accurate and efficient support for scenarios such as enterprise decision-making, customer service, and personalized recommendations. Intelligent knowledge base management not only needs to extract valuable information from complex and diverse data but also needs to dynamically adjust knowledge content based on changes in user behavior and needs to improve information retrieval efficiency and user experience satisfaction. However, current intelligent knowledge base management technologies still face significant challenges in handling multi-source heterogeneous data and dynamic user needs, particularly in accurately capturing deep semantic relationships within content and performing real-time dynamic classification based on user behavior, making it difficult to meet the actual needs of diverse scenarios.

[0003] In existing technologies, intelligent knowledge base management primarily relies on preset rules or a single analytical dimension to process and organize data. For example, existing methods identify key information from text data and categorize it into knowledge entries through keyword matching, word frequency statistics, or extraction techniques based on fixed templates. This approach can achieve rapid information classification and storage to a certain extent when processing structured or semi-structured data. However, when faced with diverse and complex multi-source heterogeneous data, it is difficult to effectively capture the deep semantic relationships within the text content. Furthermore, when analyzing user behavior data, existing technologies are typically limited to recording simple interaction frequencies or behavior types, such as click counts, browsing durations, or query records, lacking in-depth analysis of the semantic intent behind user behavior. For instance, in customer service scenarios, users may exhibit specific behavioral patterns due to repeated queries about a particular issue, but existing systems often fail to dynamically correlate these behavioral data with the semantic features of the content. This results in potentially duplicated, redundant, or irrelevant knowledge entries, thus affecting the practicality and accuracy of the knowledge base. Furthermore, existing technologies lack a flexible mechanism for adjusting classification logic when dealing with dynamically changing user needs, making it difficult to dynamically update the organizational structure of knowledge items based on real-time changes in user interaction. Therefore, existing technologies suffer from the problem of failing to achieve dynamic knowledge item segmentation based on user behavior.

[0004] In summary, existing technologies have the problem of making it difficult to achieve dynamic knowledge item segmentation based on user behavior. Summary of the Invention

[0005] This invention provides an AI-based intelligent knowledge base management method and system to achieve dynamic knowledge item segmentation based on user behavior.

[0006] Firstly, in order to solve the above-mentioned technical problems, the present invention provides an AI-based intelligent knowledge base management method, comprising:

[0007] Acquire user interaction data and multi-source heterogeneous data, extract text content, perform word segmentation and vectorization processing, and generate a set of semantic feature vectors;

[0008] Based on the set of semantic feature vectors, information entropy regularization and information contribution analysis are performed to generate a set of key semantic features;

[0009] Based on the set of key semantic features and the user interaction data, temporal correlation analysis and contextual association analysis are performed to obtain a time feature vector;

[0010] Based on the set of key semantic features and the time feature vector, correlation analysis and matching optimization are performed to obtain optimized classification logic;

[0011] Based on the optimized classification logic and the key semantic feature set, a knowledge graph is constructed and the item priority is updated to generate a dynamic knowledge item set;

[0012] Based on the dynamic knowledge item set and the user interaction data, vector retrieval matching is performed to obtain the final output result.

[0013] In one optional implementation, the step of acquiring user interaction data and multi-source heterogeneous data, extracting text content, performing word segmentation and vectorization processing, and generating a semantic feature vector set includes:

[0014] Acquire user interaction data and heterogeneous data from multiple sources;

[0015] Based on the multi-source heterogeneous data, text content is extracted, and standardized processing is performed to obtain a cleaned text set;

[0016] Based on the cleaned text set, unit word segmentation and labeling are performed using a hidden Markov model to obtain a set of labeled unit words;

[0017] Based on the labeled unit word set, low-frequency word embedding is performed, the semantic similarity between unit words is calculated, and the word embedding vector set is integrated.

[0018] Based on the set of word embedding vectors, the context relevance weights are calculated, and the word embedding vectors are weighted and adjusted to generate a set of semantic feature vectors.

[0019] In one optional implementation, the step of performing information entropy regularization and information contribution analysis based on the semantic feature vector set to generate a key semantic feature set includes:

[0020] Based on the set of semantic feature vectors, calculate the contribution weight of unit words to the grammatical framework, and construct the first layer of attention distribution;

[0021] Based on the first layer of attention distribution, information entropy regularization is performed, and the contribution weights are dynamically adjusted to obtain a set of unit word contribution weights;

[0022] Based on the first layer attention distribution and the unit word contribution weight set, semantic aggregation is performed through a cross-layer attention mechanism to generate a key semantic feature set.

[0023] In one optional implementation, the step of performing temporal correlation analysis and contextual association analysis based on the key semantic feature set and the user interaction data to obtain a time feature vector includes:

[0024] Based on the user interaction data, timestamp-behavior matching is performed to obtain behavior sequence data;

[0025] Based on the behavioral sequence data, a time-series correlation analysis is performed to obtain a time-series correlation vector;

[0026] Based on the temporal correlation vector and the set of key semantic features, a context association analysis is performed to obtain a context vector;

[0027] The time-series correlation vector and the context vector are concatenated to obtain the time feature vector.

[0028] In one optional implementation, the step of performing association analysis and matching optimization based on the key semantic feature set and the time feature vector to obtain optimized classification logic includes:

[0029] Based on the set of key semantic features and the time feature vector, the semantic concentration is calculated, user-related semantic feature analysis is performed, and user-related semantic feature vector is obtained.

[0030] Based on the user-related semantic feature vector and the time feature vector, cluster analysis is performed to obtain cluster grouping results;

[0031] Based on the clustering results, feature integration is performed to obtain the initial classification logic;

[0032] Based on the initial classification logic, and combined with the previously acquired historical classification results, matching optimization is performed to obtain the optimized classification logic.

[0033] In one optional implementation, the step of constructing a knowledge graph and updating item priorities based on the optimized classification logic and the key semantic feature set to generate a dynamic knowledge item set includes:

[0034] Based on the optimized classification logic and the key semantic feature set, semantic sentiment analysis is performed, and semantic units are supplemented by a preset domain dictionary to obtain an enhanced semantic set;

[0035] Based on the enhanced semantic set, entity node mapping and node relationship analysis are performed to generate a structured knowledge graph;

[0036] Based on the structured knowledge graph and the optimized classification logic, item relevance analysis and item priority updates are performed to generate a dynamic set of knowledge items.

[0037] In one optional implementation, the step of performing vector retrieval matching based on the dynamic knowledge item set and the user interaction data to obtain the final output result includes:

[0038] Based on the user interaction data, extract the user query content, perform word segmentation and semantic parsing, and generate a user query vector;

[0039] Based on the user query vector and the dynamic knowledge item set, vector retrieval matching is performed to generate a matching result list;

[0040] Based on the matching result list, the corresponding knowledge items are extracted, and a comprehensive analysis of sentiment tendency and knowledge domain coverage is performed to obtain the final output result.

[0041] Secondly, the present invention provides an AI-based intelligent knowledge base management system, comprising:

[0042] The data acquisition module is used to acquire user interaction data and multi-source heterogeneous data, extract text content, perform word segmentation and vectorization processing, and generate a set of semantic feature vectors.

[0043] The semantic extraction module is used to perform information entropy regularization and information contribution analysis based on the set of semantic feature vectors to generate a set of key semantic features.

[0044] The time series analysis module is used to perform time series correlation analysis and context association analysis based on the key semantic feature set and the user interaction data to obtain a time feature vector;

[0045] The logic optimization module is used to perform correlation analysis and matching optimization based on the key semantic feature set and the time feature vector to obtain optimized classification logic;

[0046] The knowledge graph construction module is used to construct a knowledge graph and update the item priority based on the optimized classification logic and the key semantic feature set, and generate a dynamic knowledge item set.

[0047] The result output module is used to perform vector retrieval matching based on the dynamic knowledge item set and the user interaction data to obtain the final output result.

[0048] Thirdly, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the AI-based intelligent knowledge base management method described in any one of the above.

[0049] Fourthly, the present invention also provides a computer-readable storage medium comprising a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute the AI-based intelligent knowledge base management method described in any one of the above.

[0050] Compared with the prior art, the present invention has the following beneficial effects:

[0051] (1) This invention obtains user interaction data and multi-source heterogeneous data, extracts text content, performs word segmentation and vectorization processing, and generates a set of semantic feature vectors, which can effectively improve the structuring degree of data processing. In the process, irrelevant characters are removed through standardization and unit word splitting and labeling is performed based on the Hidden Markov Model, which can initially screen out meaningful text units and avoid information redundancy. Then, by embedding low-frequency words and calculating the semantic similarity between unit words, a set of word embedding vectors is integrated, and then weighted and adjusted by combining context relevance weights. The generated set of semantic feature vectors not only retains the semantic core of the text, but also incorporates context relevance, providing clear and accurate basic data for subsequent analysis, thereby improving the overall efficiency of knowledge management.

[0052] (2) Based on the semantic feature vector set, this invention performs information entropy regularization and information contribution analysis to generate a key semantic feature set, which can significantly enhance the targeting of semantic extraction. In the process, the first-layer attention distribution is constructed by calculating the contribution weight of unit words to the grammatical framework, which can highlight the core content in the text. Then, the contribution weight is dynamically adjusted through information entropy regularization to reduce the interference of non-critical information. Combined with the cross-layer attention mechanism, semantic aggregation is performed, and the generated key semantic feature set focuses on the most important semantic information, avoiding information dispersion and effectively improving the accuracy of subsequent classification and knowledge organization.

[0053] (3) Based on the key semantic feature set and the user interaction data, this invention performs temporal correlation analysis and contextual association analysis to obtain a time feature vector, which can comprehensively reflect the dynamic patterns of user behavior. In the process, behavior sequence data is generated through timestamp-behavior matching, and then temporal correlation analysis is performed to obtain a temporal correlation vector. Combined with the context vector generated by contextual association analysis of the key semantic feature set, this captures the correlation between user behavior and content. Finally, by concatenating and integrating this information through vector splicing, the resulting time feature vector accurately reflects the user's temporal patterns and semantic preferences, laying a solid foundation for subsequent dynamic classification.

[0054] (4) Based on the key semantic feature set and the time feature vector, this invention performs correlation analysis and matching optimization to obtain optimized classification logic, which can improve the adaptability of classification results. In the process, user-related semantic feature analysis is performed by calculating semantic concentration, and the generated user-related semantic feature vector accurately reflects the correlation between users and content. Then, clustering analysis is combined with the time feature vector to obtain clustering grouping results, and then the initial classification logic is generated by feature integration. The optimized classification logic is then combined with historical classification results for matching optimization. The obtained optimized classification logic can dynamically adapt to user needs and ensure the accuracy and consistency of classification.

[0055] (5) Based on the optimized classification logic and the key semantic feature set, this invention constructs a knowledge graph and updates item priorities to generate a dynamic knowledge item set, which enhances the flexibility of knowledge organization. During the process, semantic sentiment analysis is performed, and semantic units are supplemented using a pre-set domain dictionary, resulting in an enhanced semantic set covering a wider range of knowledge domains. Next, a structured knowledge graph is generated through entity node mapping and node relationship analysis. Then, combined with the optimized classification logic, item relevance analysis and priority updates are performed. The resulting dynamic knowledge item set can adjust its structure according to user needs, improving the adaptability of knowledge management.

[0056] (6) This invention performs vector retrieval matching based on the dynamic knowledge item set and the user interaction data to obtain the final output result, which can significantly improve the accuracy of retrieval. In the process, the query vector generated by extracting the user query content and performing word segmentation and semantic parsing accurately reflects the user's needs. Then, by performing vector retrieval matching with the dynamic knowledge item set to generate a matching result list, and then by performing sentiment-knowledge domain coverage comprehensive analysis, the final output result is not only highly relevant to the user query, but also takes into account the quality and applicability of the content, thereby improving the overall effect of knowledge retrieval. Attached Figure Description

[0057] Figure 1This is a schematic diagram of the AI-based intelligent knowledge base management method provided in the first embodiment of the present invention;

[0058] Figure 2 This is a schematic diagram of the structure of the AI-based intelligent knowledge base management system provided in the second embodiment of the present invention. Detailed Implementation

[0059] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0060] Reference Figure 1 The first embodiment of the present invention provides an AI-based intelligent knowledge base management method, including the following steps:

[0061] S11: Acquire user interaction data and multi-source heterogeneous data, extract text content, perform word segmentation and vectorization processing, and generate a set of semantic feature vectors;

[0062] S12, Based on the set of semantic feature vectors, perform information entropy regularization and information contribution analysis to generate a set of key semantic features;

[0063] S13, Based on the set of key semantic features and the user interaction data, perform temporal correlation analysis and contextual association analysis to obtain a time feature vector;

[0064] S14. Based on the set of key semantic features and the time feature vector, perform association analysis and matching optimization to obtain optimized classification logic;

[0065] S15, Based on the optimized classification logic and the key semantic feature set, construct a knowledge graph and update the item priority to generate a dynamic knowledge item set;

[0066] S16, Perform vector retrieval matching based on the dynamic knowledge item set and the user interaction data to obtain the final output result.

[0067] In step S11, it is necessary to acquire user interaction data and multi-source heterogeneous data, extract text content, perform word segmentation and vectorization processing, and generate a set of semantic feature vectors.

[0068] In one implementation, user interaction data and multi-source heterogeneous data are acquired, text content is extracted, and word segmentation and vectorization processing are performed to generate a set of semantic feature vectors, including:

[0069] Acquire user interaction data and multi-source heterogeneous data; extract text content from the multi-source heterogeneous data, perform standardization processing, and obtain a cleaned text set; based on the cleaned text set, perform unit word splitting and labeling based on a Hidden Markov Model to obtain a labeled unit word set; based on the labeled unit word set, perform low-frequency word embedding, calculate the semantic similarity between unit words, and integrate them to obtain a word embedding vector set; based on the word embedding vector set, calculate the context relevance weight, perform weighted adjustment of the word embedding vectors, and generate a semantic feature vector set.

[0070] It should be noted that the step of acquiring user interaction data and multi-source heterogeneous data is achieved by collecting user interaction records on the platform and multi-source heterogeneous data from different sources, such as social media posts or enterprise documents. User operation data, such as browsing and liking behaviors, and text data, such as comments or document content, are collected in real time. After data collection, the data is grouped by timestamp and initially processed to obtain user interaction data and multi-source heterogeneous data. The timestamp grouping rule is to divide the data into segments by hour to ensure that the amount of data in each time period is balanced and facilitates subsequent analysis. After grouping, the data is initially processed by removing duplicate records through deduplication and converting the format to unify the data from different sources into JSON format, eliminating format differences and providing structured data for subsequent processing. Based on the aforementioned multi-source heterogeneous data, the text content is extracted and standardized to obtain a cleaned text set. This step involves extracting text data from the multi-source heterogeneous data and processing it with a unified encoding format. Regular expressions are used to remove non-text characters such as emojis or HTML tags. Combined with preset text cleaning rules, redundant information is eliminated to ensure data cleanliness and consistency. These preset text cleaning rules include removing duplicate words, filtering sentences exceeding a preset character count threshold, and removing phrases irrelevant to the topic, ensuring concise and relevant text content. For example, a character count threshold of 500 characters is set. This threshold is obtained by analyzing the average length distribution of sentences in the training data, and is based on the median plus one standard deviation, ensuring that excessively long sentences containing redundant information are removed. Based on the cleaned text set, a unit word segmentation and annotation step is performed using a Hidden Markov Model (HMM) to obtain an annotated unit word set. This step uses an HMM to identify word boundaries through a probability transition matrix and annotates each unit word with a part-of-speech tag, such as noun or verb. When processing Chinese text, the word segmentation results are optimized based on dictionary matching and contextual information to generate a structured annotated unit word set. Based on the labeled unit word set, the following steps are performed: Low-frequency word embedding is performed, semantic similarity between unit words is calculated, and a word embedding vector set is obtained. This step utilizes a pre-trained language model such as BERT to convert low-frequency words into high-dimensional vectors. The semantic similarity between two word vectors is determined by calculating the cosine of the angle between them. K-means clustering is used to group semantically similar words, aggregating them into a unified word embedding vector set based on semantic similarity. Based on this word embedding vector set, context relevance weights are calculated, and the word embedding vectors are weighted and adjusted to generate a semantic feature vector set. This step analyzes the contextual role of words in a sentence using an attention mechanism, assigns appropriate weights, and then weights and adjusts the word embedding vectors to generate a semantic feature vector set that integrates word embedding dimensions and contextual information. The semantic feature vector set is a mathematical representation of text content, containing semantic information of words and contextual relevance, and can be used for subsequent tasks such as semantic analysis, classification, or knowledge graph construction.

[0071] In step S12, information entropy regularization and information contribution analysis are performed based on the set of semantic feature vectors to generate a set of key semantic features.

[0072] In one implementation, based on the semantic feature vector set, information entropy regularization and information contribution analysis are performed to generate a key semantic feature set, including:

[0073] Based on the set of semantic feature vectors, the contribution weight of unit words to the grammatical framework is calculated, and a first-layer attention distribution is constructed. Based on the first-layer attention distribution, information entropy regularization is performed, and the contribution weights are dynamically adjusted to obtain a set of unit word contribution weights. Based on the first-layer attention distribution and the set of unit word contribution weights, semantic aggregation is performed through a cross-layer attention mechanism to generate a set of key semantic features.

[0074] It should be noted that the step of calculating the contribution weight of unit words to the grammatical structure based on the semantic feature vector set and constructing the first layer of attention distribution is implemented through the attention mechanism of the Transformer model. First, the semantic feature vector set is input into the model. The model calculates the semantic relevance of each unit word to the grammatical structure, such as the subject or predicate, based on the position and context of each unit word in the sentence. A preliminary weight distribution is generated through a weighted allocation process, highlighting the core parts of the sentence structure, and finally forming the first layer of attention distribution that reflects the semantic focus of the text.

[0075] Based on the first layer of attention distribution, information entropy regularization is performed, and contribution weights are dynamically adjusted to obtain a set of unit word contribution weights. This step introduces information entropy calculation to evaluate the semantic uncertainty of each unit word. By analyzing the degree of information dispersion in the sentence, if the unit word's information entropy value is higher than the preset information entropy threshold, it indicates semantic ambiguity, and its weight is dynamically reduced. A linear decay method is adopted, gradually reducing the weight according to the proportion of information entropy value exceeding the threshold, with a reduction of 10% to 50% linearly decreasing from the excess portion, ensuring smooth adjustment and not excessively affecting the semantic structure. An optimized set of unit word contribution weights is generated to enhance semantic focus. The preset information entropy threshold is obtained by analyzing the average information entropy distribution of unit words in the training corpus. The average information entropy distribution of unit words in the training corpus is obtained by collecting a corpus of 100,000 documents covering the fields of science, medicine, and education. A random stratified sampling method is used to extract from multi-source heterogeneous data, covering user interaction records and public texts, to ensure sample diversity and representativeness. The specific value is based on the median or upper quartile value of information entropy in the statistical sample to ensure that the threshold can effectively distinguish high-uncertainty unit words and maintain the stability of semantic analysis. Based on the first-layer attention distribution and the unit word contribution weight set, a key semantic feature set is generated through semantic aggregation via a cross-layer attention mechanism. This step utilizes a multi-layer attention structure, employing a three-layer Transformer structure, with each layer containing 512-dimensional hidden states and using the ReLU activation function. Residual connections between layers pass information to avoid gradient vanishing. The first-layer attention distribution and the optimized unit word contribution weight set are input into the multi-layer network, integrating cross-layer semantic information layer by layer. By aligning the attention weights of different layer outputs, a multi-head attention mechanism is used to calculate the semantic relevance between layers. Residual connections add the output of the previous layer to the output of the current layer, ensuring smooth information transmission. The outputs of each layer are weighted and summarized through the attention mechanism. Initial weights are extracted from the unit word contribution weight set, and dynamic weights between layer outputs are calculated using the multi-head attention mechanism. A weighted average is used to fuse the outputs of different layers, extracting the most representative unit word semantic information to generate the key semantic feature set. The key semantic feature set is a refined representation of the core semantic content in the text and can be used for subsequent tasks such as temporal analysis, classification logic optimization, or knowledge graph construction.

[0076] In step S13, time-series correlation analysis and contextual association analysis are performed based on the key semantic feature set and the user interaction data to obtain a time feature vector.

[0077] In one implementation, based on the set of key semantic features and the user interaction data, temporal correlation analysis and contextual association analysis are performed to obtain a time feature vector, including:

[0078] Based on the user interaction data, timestamp-behavior matching is performed to obtain behavior sequence data; based on the behavior sequence data, temporal correlation analysis is performed to obtain a temporal correlation vector; based on the temporal correlation vector and the key semantic feature set, context association analysis is performed to obtain a context vector; based on the temporal correlation vector and the context vector, vector concatenation is performed to obtain a time feature vector.

[0079] It should be noted that the step of performing timestamp-behavior matching to obtain behavior sequence data based on the user interaction data analyzes the user interaction records on the platform, such as browsing, liking, or commenting. Based on the timestamp records, it matches the behavior type with the corresponding time point, generating behavior sequence data containing time sequence and behavior frequency. The data is arranged in ascending order of timestamps to ensure a clear temporal relationship between behaviors. The step of performing temporal correlation analysis based on the behavior sequence data to obtain a temporal correlation vector uses a recurrent neural network (RNN) model. The behavior sequence data is input, and the model analyzes the sequential relationship and time dependency between behaviors. By processing each time point in the sequence one by one, the temporal patterns of the behavior are extracted, generating a temporal correlation vector reflecting the dynamics of user operations. Based on the temporal correlation vector and the key semantic feature set, context association analysis is performed to obtain a context vector. This step projects the temporal correlation vector and key semantic feature vector into a unified low-dimensional space through feature alignment. Principal Component Analysis (PCA) is used to extract common feature dimensions, and the cosine of the angle between the projected vectors is calculated as the similarity, measuring the correlation between the temporal correlation vector and each semantic feature vector. If the similarity exceeds a preset correlation threshold, context association is considered to exist. The preset correlation threshold is obtained by analyzing the average similarity distribution between user behavior and semantic features in the training data, specifically taking the median plus one standard deviation to ensure that the threshold can effectively distinguish between relevant and irrelevant associations. The context vector is then integrated with the contextual information to generate the context vector. The next step, vector concatenation, uses vector concatenation technology to connect the temporal correlation vector and context vector in dimensional order, generating a comprehensive vector containing user behavior time patterns and semantic associations. The time feature vector is a multi-dimensional data representation that integrates user behavior time patterns and semantic preferences, and can be used for subsequent tasks such as dynamic classification, optimizing classification logic, or generating knowledge entries.

[0080] In step S14, association analysis and matching optimization are performed based on the set of key semantic features and the time feature vector to obtain optimized classification logic.

[0081] In one implementation, association analysis and matching optimization are performed based on the key semantic feature set and the time feature vector to obtain optimized classification logic, including:

[0082] Based on the set of key semantic features and the time feature vector, the semantic concentration is calculated, and user-related semantic feature analysis is performed to obtain user-related semantic feature vectors. Based on the user-related semantic feature vectors and the time feature vectors, cluster analysis is performed to obtain clustering results. Based on the clustering results, feature integration is performed to obtain initial classification logic. Based on the initial classification logic and combined with the pre-acquired historical classification results, matching optimization is performed to obtain optimized classification logic.

[0083] It should be noted that the step of calculating semantic concentration based on the key semantic feature set and the time feature vector, and performing user-related semantic feature analysis to obtain user-related semantic feature vectors, involves statistically analyzing the frequency of occurrence of each feature in the key semantic feature set, calculating semantic concentration, and determining the key areas of user attention by calculating the proportion of feature frequency to total frequency. Combined with behavioral patterns in the time feature vector, cosine similarity is used to filter out semantic features highly related to user behavior. The similarity threshold is set to 0.7 by calculating the cosine of the angle between the time feature vector and each key semantic feature vector. This similarity threshold is obtained by analyzing the average similarity distribution between user behavior and semantic features in the training data, and is set to the median plus one standard deviation. Features with similarity higher than the threshold are filtered out, and the top 20% of the highest similarity features are retained as relevant semantic features. This ensures that the filtering results focus on the semantic content most relevant to user behavior, generating user-related semantic feature vectors that reflect user preferences. The step of performing cluster analysis based on the user-related semantic feature vector and the time feature vector to obtain the clustering grouping results adopts the K-means clustering algorithm. The user-related semantic feature vector and the time feature vector are used as inputs. The similarity is calculated based on Euclidean distance, and the data points are grouped into several clusters. The number of clusters is determined according to the diversity of user behavior patterns in the training data. For example, the elbow rule is used to analyze the stable point of the sum of squared errors within the cluster, and the clustering grouping results that reflect user behavior and semantic patterns are generated. Based on the clustering results, feature integration is performed to obtain the initial classification logic. This step involves weighted averaging of the central features of each cluster, assigning weights based on the number of samples within each cluster (higher weights for larger samples). The central features are then weighted and summed to generate an integrated vector. This vector integrates representative patterns of user-related semantic and temporal feature vectors. Preliminary classification rules are generated by combining behavior frequency and semantic relevance within each cluster. The average frequency of user behavior within each cluster is calculated as the behavior frequency. Cosine similarity is used to quantify semantic relevance. A scoring function is designed to sum the two and normalize the results, generating a comprehensive score reflecting both behavior frequency and semantic relevance. Preliminary classification rules are defined based on these scores, forming the initial classification logic.Based on the initial classification logic and combined with the pre-acquired historical classification results, the matching optimization step is performed to obtain the optimized classification logic. This step compares the initial classification logic with the historical classification results and calculates the matching degree. By representing the initial classification logic and the historical classification results as vectors, the cosine value of the angle between the two is calculated. Specifically, the weight distribution or feature values ​​of the two are standardized into unit vectors, and then the cosine value is obtained by dividing the dot product by the product of the two vector magnitudes. The value ranges from -1 to 1. The closer the value is to 1, the more consistent the direction is, i.e., the higher the matching degree. If the matching degree is lower than the preset matching degree threshold, the weights and category rules are adjusted to optimize the classification logic. The preset matching degree threshold is obtained by analyzing the average matching degree distribution of the classification results in the historical data. The value is the median value plus one standard deviation to ensure that the optimized logic is consistent with the historical pattern, thus generating the optimized classification logic.

[0084] Optimized classification logic is a dynamically adjusted classification rule that reflects a comprehensive classification pattern of users' current behavior and historical preferences. It can be used for subsequent tasks such as knowledge graph construction, item priority updating, or dynamic knowledge item generation.

[0085] In step S15, a knowledge graph needs to be constructed and the item priority updated based on the optimized classification logic and the key semantic feature set to generate a dynamic knowledge item set.

[0086] In one implementation, based on the optimized classification logic and the set of key semantic features, a knowledge graph is constructed and the item priority is updated to generate a dynamic knowledge item set, including:

[0087] Based on the optimized classification logic and the key semantic feature set, semantic sentiment analysis is performed, and semantic units are supplemented by a preset domain dictionary to obtain an enhanced semantic set. Based on the enhanced semantic set, entity node mapping and node relationship analysis are performed to generate a structured knowledge graph. Based on the structured knowledge graph and the optimized classification logic, item relevance analysis and item priority updates are performed to generate a dynamic knowledge item set.

[0088] It should be noted that, based on the optimized classification logic and the key semantic feature set, semantic sentiment analysis is performed. This step, supplementing semantic units with a pre-defined domain dictionary to obtain an enhanced semantic set, utilizes a sentiment analysis model such as VADER to analyze the sentiment tendency in the key semantic feature set. This is achieved by detecting the positive, negative, or neutral sentiment scores of words, combined with user preference weights in the optimized classification logic. These user preference weights are calculated based on historical user interaction data during the optimized classification logic generation process, assigning weight values ​​accordingly (e.g., adding 0.3 to high-frequency categories and subtracting 0.2 from low-frequency categories). This ensures that sentiment analysis prioritizes reflecting user interests, generating preliminary sentiment labels. The VADER model then analyzes word sentiment scores, categorizing them into three types: positive (e.g., "excellent" score above 0.5), negative (e.g., "failure" score below -0.5), or neutral (e.g., "data" score close to 0). The overall tendency is determined based on the weighted average of word scores in a sentence. Subsequently, a pre-defined domain dictionary, such as a technical or medical terminology list, is used to supplement missing or ambiguous semantic units, enhancing the domain coverage and completeness of the semantic set, thus generating an enhanced semantic set. Based on the enhanced semantic set, the step of generating a structured knowledge graph by mapping entity nodes and analyzing node relationships employs Named Entity Recognition (NER) technology. Entities such as names, locations, or concepts are extracted from the enhanced semantic set and mapped to nodes in the knowledge graph. Relationships between entities are identified and dependency analysis is performed, such as "belongs to" or "influences". The extracted entities are stored as nodes using the graph database Neo4j. Edge relationships between nodes are defined using the dependency analysis results, such as "influences" or "associations". By traversing all entities and relationships in the enhanced semantic set, a directed acyclic graph is constructed to ensure a clear graph structure without redundant connections. Based on the structured knowledge graph and the optimized classification logic, the step of performing item relevance analysis and item priority update to generate a dynamic knowledge item set involves analyzing the relevance between each node in the structured knowledge graph and the user classification logic using a graph traversal algorithm. A depth-first search (DFS) algorithm is employed, traversing the graph from the root node and calculating the cosine similarity between each node and the user preference features in the optimized classification logic. This similarity value is defined as the relevance value. If the relevance value is lower than a preset relevance threshold, the node weight is adjusted. This preset relevance threshold is obtained by analyzing the historical matching degree distribution between user queries and knowledge nodes in the training data, and is set to the median plus one standard deviation to ensure high-relevance items are prioritized. Combined with the optimized classification logic, the item priorities are updated to generate the dynamic knowledge item set. The dynamic knowledge item set is a collection of knowledge units that is dynamically adjusted according to user needs, containing high-priority items from the structured knowledge graph, and can be used for subsequent tasks such as vector retrieval matching or knowledge recommendation.

[0089] In step S16, vector retrieval matching is performed based on the dynamic knowledge item set and the user interaction data to obtain the final output result.

[0090] In one implementation, vector retrieval matching is performed based on the dynamic knowledge item set and the user interaction data to obtain the final output result, including:

[0091] Based on the user interaction data, user query content is extracted, and word segmentation and semantic parsing are performed to generate user query vectors. Based on the user query vectors and the dynamic knowledge item set, vector retrieval matching is performed to generate a matching result list. Based on the matching result list, corresponding knowledge items are extracted, and sentiment-knowledge domain coverage comprehensive analysis is performed to obtain the final output result.

[0092] It should be noted that the step of extracting user query content based on the user interaction data, performing word segmentation and semantic parsing, and generating user query vectors involves extracting real-time query input, such as search keywords or question descriptions, from the user interaction data. This is done using rule-based word segmentation tools like Jieba to segment the Chinese text, combined with a pre-trained BERT model for semantic parsing. The analysis of contextual relationships between words is used to generate a vector representation containing semantic information and syntactic structure, i.e., the user query vector. The step of performing vector retrieval matching based on the user query vector and the dynamic knowledge item set to generate a matching result list employs a nearest neighbor search algorithm such as Annoy. The cosine similarity is calculated between the user query vector and each item vector in the dynamic knowledge item set to obtain the item matching similarity. Items with matching similarity higher than a preset similarity threshold are selected. This preset threshold is obtained by analyzing the average similarity distribution of historical queries and knowledge item matches, and is set to the median plus one standard deviation to ensure highly relevant results are included. The matching result list is then generated by sorting the results in descending order of similarity. Based on the matching result list, the following steps are performed: First, extracting relevant knowledge items from the matching result list and conducting a comprehensive analysis of sentiment and knowledge domain coverage to obtain the final output. This involves extracting the top N knowledge items by similarity ranking from the matching result list, evaluating their sentiment scores using a sentiment analysis model such as TextBlob, and calculating the coverage of the knowledge items with the user's query domain using a pre-defined domain dictionary. The percentage of hit words in the knowledge items relative to the corresponding domain words in the pre-defined domain dictionary is used as the coverage score. Then, a weighted average is used to integrate the sentiment and domain coverage scores to generate a comprehensive score. By setting the sentiment score weight to 0.4 and the domain coverage score weight to 0.6, a weighted sum is generated based on the TextBlob output sentiment score range of -1 to 1 and the coverage percentage, resulting in a comprehensive score of 0 to 1. The highest-scoring item is then selected as the final output. The final output is knowledge content that is highly relevant to the user's query and has optimized sentiment and domain coverage, which can be used to provide users with accurate knowledge recommendations or answer support.

[0093] In summary, this invention discloses an AI-based intelligent knowledge base management method. This invention extracts semantic features from user interaction data and multi-source heterogeneous data, performs temporal and contextual analysis, combines optimized classification logic to construct a dynamic knowledge graph, and performs vector retrieval matching, thereby realizing dynamic knowledge item division based on user behavior.

[0094] Reference Figure 2 The second embodiment of the present invention provides an AI-based intelligent knowledge base management system, comprising:

[0095] The data acquisition module is used to acquire user interaction data and multi-source heterogeneous data, extract text content, perform word segmentation and vectorization processing, and generate a set of semantic feature vectors.

[0096] The semantic extraction module is used to perform information entropy regularization and information contribution analysis based on the set of semantic feature vectors to generate a set of key semantic features.

[0097] The time series analysis module is used to perform time series correlation analysis and context association analysis based on the key semantic feature set and the user interaction data to obtain a time feature vector;

[0098] The logic optimization module is used to perform correlation analysis and matching optimization based on the key semantic feature set and the time feature vector to obtain optimized classification logic;

[0099] The knowledge graph construction module is used to construct a knowledge graph and update the item priority based on the optimized classification logic and the key semantic feature set, and generate a dynamic knowledge item set.

[0100] The result output module is used to perform vector retrieval matching based on the dynamic knowledge item set and the user interaction data to obtain the final output result.

[0101] It should be noted that the AI-based intelligent knowledge base management device provided in this embodiment of the invention is used to execute all the process steps of the AI-based intelligent knowledge base management method in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0102] This invention also provides an electronic device. The electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, such as an AI-based intelligent knowledge base management program. When the processor executes the computer program, it implements the steps described in the various AI-based intelligent knowledge base management method embodiments above, for example... Figure 1 The step S11 shown. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the above-described device embodiments, such as the map construction module.

[0103] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0104] The electronic device may be a desktop computer, laptop, handheld computer, or smart tablet, etc. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0105] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting all parts of the electronic device via various interfaces and lines.

[0106] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0107] Wherein, if the modules / units integrated in the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0108] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0109] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. An AI-based intelligent knowledge base management method, characterized in that, include: Acquire user interaction data and multi-source heterogeneous data, extract text content, perform word segmentation and vectorization processing, and generate a set of semantic feature vectors; Based on the set of semantic feature vectors, information entropy regularization and information contribution analysis are performed to generate a set of key semantic features; Based on the set of key semantic features and the user interaction data, temporal correlation analysis and contextual association analysis are performed to obtain a time feature vector; Based on the set of key semantic features and the time feature vector, correlation analysis and matching optimization are performed to obtain optimized classification logic; Based on the optimized classification logic and the key semantic feature set, a knowledge graph is constructed and the item priority is updated to generate a dynamic knowledge item set; Based on the dynamic knowledge item set and the user interaction data, vector retrieval matching is performed to obtain the final output result.

2. The AI-based intelligent knowledge base management method according to claim 1, characterized in that, The process of acquiring user interaction data and multi-source heterogeneous data, extracting text content, performing word segmentation and vectorization processing, and generating a set of semantic feature vectors includes: Acquire user interaction data and heterogeneous data from multiple sources; Based on the multi-source heterogeneous data, text content is extracted, and standardized processing is performed to obtain a cleaned text set; Based on the cleaned text set, unit word segmentation and labeling are performed using a hidden Markov model to obtain a set of labeled unit words; Based on the labeled unit word set, low-frequency word embedding is performed, the semantic similarity between unit words is calculated, and the word embedding vector set is integrated. Based on the set of word embedding vectors, the context relevance weights are calculated, and the word embedding vectors are weighted and adjusted to generate a set of semantic feature vectors.

3. The AI-based intelligent knowledge base management method according to claim 1, characterized in that, The step of generating a key semantic feature set by performing information entropy regularization and information contribution analysis based on the semantic feature vector set includes: Based on the set of semantic feature vectors, calculate the contribution weight of unit words to the grammatical framework, and construct the first layer of attention distribution; Based on the first layer of attention distribution, information entropy regularization is performed, and the contribution weights are dynamically adjusted to obtain a set of unit word contribution weights; Based on the first layer attention distribution and the unit word contribution weight set, semantic aggregation is performed through a cross-layer attention mechanism to generate a key semantic feature set.

4. The AI-based intelligent knowledge base management method according to claim 1, characterized in that, The step of performing temporal correlation analysis and contextual association analysis based on the key semantic feature set and the user interaction data to obtain a time feature vector includes: Based on the user interaction data, timestamp-behavior matching is performed to obtain behavior sequence data; Based on the behavioral sequence data, a time-series correlation analysis is performed to obtain a time-series correlation vector; Based on the temporal correlation vector and the set of key semantic features, a context association analysis is performed to obtain a context vector; The time-series correlation vector and the context vector are concatenated to obtain the time feature vector.

5. The AI-based intelligent knowledge base management method according to claim 1, characterized in that, The step of performing association analysis and matching optimization based on the key semantic feature set and the time feature vector to obtain optimized classification logic includes: Based on the set of key semantic features and the time feature vector, the semantic concentration is calculated, user-related semantic feature analysis is performed, and user-related semantic feature vector is obtained. Cluster analysis is performed based on the user-related semantic feature vector and the time feature vector to obtain clustering results; Based on the clustering results, feature integration is performed to obtain the initial classification logic; Based on the initial classification logic, and combined with the previously acquired historical classification results, matching optimization is performed to obtain the optimized classification logic.

6. The AI-based intelligent knowledge base management method according to claim 1, characterized in that, The process of constructing a knowledge graph and updating item priorities based on the optimized classification logic and the key semantic feature set to generate a dynamic knowledge item set includes: Based on the optimized classification logic and the key semantic feature set, semantic sentiment analysis is performed, and semantic units are supplemented by a preset domain dictionary to obtain an enhanced semantic set; Based on the enhanced semantic set, entity node mapping and node relationship analysis are performed to generate a structured knowledge graph; Based on the structured knowledge graph and the optimized classification logic, item relevance analysis and item priority updates are performed to generate a dynamic set of knowledge items.

7. The AI-based intelligent knowledge base management method according to claim 1, characterized in that, The step of performing vector retrieval matching based on the dynamic knowledge item set and the user interaction data to obtain the final output result includes: Based on the user interaction data, extract the user query content, perform word segmentation and semantic parsing, and generate a user query vector; Based on the user query vector and the dynamic knowledge item set, vector retrieval matching is performed to generate a matching result list; Based on the matching result list, the corresponding knowledge items are extracted, and a comprehensive analysis of sentiment tendency and knowledge domain coverage is performed to obtain the final output result.

8. An AI-based intelligent knowledge base management system, characterized in that, include: The data acquisition module is used to acquire user interaction data and multi-source heterogeneous data, extract text content, perform word segmentation and vectorization processing, and generate a set of semantic feature vectors. The semantic extraction module is used to perform information entropy regularization and information contribution analysis based on the set of semantic feature vectors to generate a set of key semantic features. The time series analysis module is used to perform time series correlation analysis and context association analysis based on the key semantic feature set and the user interaction data to obtain a time feature vector; The logic optimization module is used to perform correlation analysis and matching optimization based on the key semantic feature set and the time feature vector to obtain optimized classification logic; The knowledge graph construction module is used to construct a knowledge graph and update the item priority based on the optimized classification logic and the key semantic feature set, and generate a dynamic knowledge item set. The result output module is used to perform vector retrieval matching based on the dynamic knowledge item set and the user interaction data to obtain the final output result.

9. An electronic device, characterized in that, The system includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements the AI-based intelligent knowledge base management method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the AI-based intelligent knowledge base management method as described in any one of claims 1 to 7.

Images