Intelligent Swarm-style Quality Optimization Management Method and System for Multi-source Label Data

Abstract

The present invention provides an intelligent swarm-style quality optimization management method and system for multi-source label data, which relates to the technical field of data labeling. It includes extracting the comprehensive feature vector of the label through a feature extraction module, using a deep feature enhancement network to process and generate a similarity and dependence feature matrix, and constructing a label quality evaluation model; mapping the label data to swarm network nodes and performing three-level partitioning to construct a node optimization association network; implementing optimization based on a collaborative optimization rule set until the quality improvement value is less than a preset threshold. The present invention can effectively improve the quality of label data, reduce data processing costs, and improve optimization efficiency.

Description

Technical Field

[0001] The present invention relates to data tagging technology, and particularly to an intelligent swarm-based quality optimization management method and system for multi-source tag data. Background Art

[0002] With the advent of the big data era, tag data, as an important data organization and management method, has been widely used in various fields. Tag data can not only effectively describe and classify information, but also help users quickly retrieve and understand relevant content. In practical applications, tag data often comes from multiple different data sources, which may include various channels such as user annotation, automatic generation, and expert annotation. The quality of multi-source tag data directly affects the usability and application effect of the data. Therefore, it is of great significance to optimize and manage the quality of multi-source tag data.

[0003] Currently, in the quality management of multi-source tag data, mainly methods such as manual review and simple rule filtering are adopted. These traditional methods have some obvious deficiencies in dealing with large-scale multi-source tag data:

[0004] Existing tag quality assessment methods often only focus on features in a single dimension, ignoring the semantic relevance and timeliness features between tags, resulting in incomplete and inaccurate quality assessment results.

[0005] Traditional tag optimization methods lack systematicness and collaboration. The optimization processes of each tag are relatively independent, and the similarity and dependency relationships between tags are not fully utilized to improve the optimization effect.

[0006] Existing optimization methods usually adopt static optimization strategies and cannot adaptively adjust the optimization scheme according to the dynamic changes of tag quality, resulting in an unsatisfactory optimization effect and being prone to falling into local optima.

[0007] Therefore, there is an urgent need for a tag data quality optimization management method that can comprehensively consider multi-dimensional features, support collaborative optimization, and has self-adaptive capabilities to improve the overall quality level of multi-source tag data. Summary of the Invention

[0008] Embodiments of the present invention provide an intelligent swarm-based quality optimization management method and system for multi-source tag data, which can solve the problems in the prior art.

[0009] In the first aspect of the embodiments of the present invention, [[ID= thirty-two ]]

[0010] An intelligent swarm-based quality optimization management method for multi-source tag data is provided, including:

[0011] Obtain multi-source label data input, and respectively extract the comprehensive feature vectors of each label through the feature extraction module. The comprehensive feature vectors include semantic feature vectors, association feature vectors, and timeliness feature vectors. Input the comprehensive feature vectors into the feature fusion layer to generate a label comprehensive feature matrix; process the label comprehensive feature matrix through a deep feature enhancement network, and calculate the similarity feature matrix and dependence feature matrix between labels; construct a label quality evaluation model based on each feature matrix, calculate the initial quality score of each label, and generate a label quality grading sequence;

[0012] According to the label quality grading sequence, map the label data to the bee colony network nodes, and perform a three-level division of the nodes based on the initial quality score: set the nodes with quality scores higher than the preset threshold as instruction bees, responsible for formulating optimization strategies; set the nodes with quality scores equal to the preset threshold as follower bees, responsible for transmitting optimization instructions; set the nodes with quality scores lower than the preset threshold as execution bees, responsible for implementing optimization operations; construct a node optimization association network using the similarity feature matrix and dependence feature matrix, and design a collaborative optimization rule set based on the node optimization association network;

[0013] Start the optimization process based on the collaborative optimization rule set. The instruction bee analyzes the feature pattern of the first-quality label to generate an optimization instruction set. The follower bee plans the optimization path through the node optimization association network to generate a task allocation plan. The execution bee performs feature optimization on the second-quality label according to the task allocation plan; after the optimization is completed, input the optimization result back into the feature extraction module for quality evaluation and update the label quality grading sequence; when the improvement value of the quality score after continuous multi-round optimization is less than the preset improvement threshold, end the optimization process and output the final label data set.

[0014] Process the label comprehensive feature matrix through a deep feature enhancement network, and calculate the similarity feature matrix and dependence feature matrix between labels; construct a label quality evaluation model based on each feature matrix, calculate the initial quality score of each label, and the generated label quality grading sequence includes:

[0015] Construct a deep feature enhancement network, which includes a feature mapping layer, a feature enhancement layer, and a feature reconstruction layer. The feature mapping layer uses a non-linear activation function to map the label comprehensive feature matrix to a high-dimensional latent space. The feature enhancement layer calculates the attention weights between feature vectors and performs weighted fusion based on the attention weights to achieve feature enhancement. The feature reconstruction layer uses a deconvolution operation to remap the enhanced features back to the original feature space;

[0016] The cosine similarity method is used to calculate the semantic similarity between tags to obtain a similarity feature matrix. At the same time, a temporal correlation analysis method is introduced to construct a directed weighted graph through the temporal pattern and co-occurrence frequency of tag appearances to obtain a dependency feature matrix;

[0017] The similarity feature matrix and the dependency feature matrix are fused with the original tag comprehensive feature matrix to construct a tag quality evaluation model. The tag quality evaluation model calculates the quality of the tag's own features, the similarity relationship between tags, and the dependency relationship between tags respectively, and obtains an initial quality score through weighted summation;

[0018] The initial quality score is normalized, and the density clustering method is used to automatically determine the quality level division threshold to generate a tag quality grading sequence containing quality level information.

[0019] Using the cosine similarity method to calculate the semantic similarity between tags to obtain a similarity feature matrix, and at the same time introducing a temporal correlation analysis method to construct a directed weighted graph through the temporal pattern and co-occurrence frequency of tag appearances to obtain a dependency feature matrix includes:

[0020] Obtain the tag text set, input the tag text set into a pre-trained word vector model to obtain the initial word vector corresponding to each tag text; for tags containing multiple words in the tag text set, calculate the term frequency-inverse document frequency value of each word in the tag as the weight coefficient, and perform weighted averaging on the initial word vectors of each word according to the weight coefficient to obtain the overall semantic feature vector of the tag;

[0021] Based on the overall semantic feature vector, calculate the cosine similarity between any two tags in the tag text set, and fill the cosine similarity into the corresponding row and column positions of the similarity feature matrix to obtain the semantic similarity feature matrix of the tag text set. The dimension of the semantic similarity feature matrix is the product of the total number of tags and the total number of tags;

[0022] Obtain the temporal data of the tag text set, set a time window, analyze the co-occurrence pattern of the tag text set within the time window, count the co-occurrence frequency of tag pairs, record the relative appearance order of tags, and calculate the time interval distribution of tag pairs; based on the co-occurrence frequency, the relative appearance order, and the time interval distribution, use a preset weight coefficient for weighted combination to obtain the temporal dependency strength of tag pairs;

[0023] Construct a directed weighted graph according to the temporal dependence intensity, map the labels in the label text set to the nodes of the directed weighted graph, determine the connection direction between nodes based on the relative occurrence order of label pairs, and use the temporal dependence intensity as the weight value of the corresponding connection to obtain a directed temporal weighted graph describing the temporal association of labels; construct a dependence feature matrix based on the connection relationship and weight value of the directed temporal weighted graph.

[0024] Fuse the similarity feature matrix and the dependence feature matrix with the original label comprehensive feature matrix to construct a label quality evaluation model. The label quality evaluation model calculates the quality of the label's own features, the similarity relationship between labels, and the dependence relationship between labels respectively, and obtains an initial quality score through weighted summation, including:

[0025] Perform weighted averaging on the semantic similarity feature and the structural similarity feature in the similarity feature matrix to obtain a comprehensive similarity feature between labels, and perform weighted averaging on the hierarchical dependence feature and the association dependence feature in the dependence feature matrix to obtain a comprehensive dependence feature between labels;

[0026] Normalize the language normality feature, the professional term normality feature, and the semantic integrity feature in the original label comprehensive feature matrix to obtain a normalized feature, and perform logarithmic transformation on the word frequency distribution feature and the query coverage feature to obtain a statistical feature;

[0027] Concatenate and combine the comprehensive similarity feature between labels, the comprehensive dependence feature between labels, the normalized feature, and the statistical feature to construct a feature fusion matrix, and construct a label quality evaluation model based on the feature fusion matrix;

[0028] Calculate the quality score of the label's own features based on the label quality evaluation model, where the quality score of the label's own features is obtained by weighted summation of the normalized feature and the statistical feature;

[0029] Calculate the similarity relationship score between labels based on the label quality evaluation model, where the similarity relationship score between labels is obtained by threshold screening of the comprehensive similarity feature between labels;

[0030] Calculate the dependence relationship score between labels based on the label quality evaluation model, where the dependence relationship score between labels is obtained by weighted calculation of the comprehensive dependence feature between labels;

[0031] Perform weighted summation on the quality score of the label's own features, the similarity relationship score between labels, and the dependence relationship score between labels to obtain an initial label quality score.

[0032] Start the optimization process based on the collaborative optimization rule set. The instruction bee analyzes the feature patterns of the first quality label to generate an optimization instruction set. The follower bee generates a task assignment plan through optimizing the path planning of the node optimization association network. The executor bee performs feature optimization on the second quality label according to the task assignment plan, including:

[0033] Construct a node optimization association network and establish a collaborative optimization rule set. The collaborative optimization rule set includes a feature extraction rule, an optimization instruction generation rule, and a quality optimization rule. Start the label optimization process based on the collaborative optimization rule set;

[0034] The instruction bee analyzes the first quality label according to the feature extraction rule, obtains semantic features through word vector representation, obtains structural features through hierarchical relationships, obtains usage features through application frequencies, and performs weighted combination on the semantic features, the structural features, and the usage features to generate label feature data;

[0035] The instruction bee processes the label feature data based on the optimization instruction generation rule, inputs the label feature data into the feature vector space for clustering analysis, extracts feature patterns to establish a feature pattern library, and generates an optimization instruction set according to the feature pattern library;

[0036] The follower bee receives the optimization instruction set, calculates the optimization distance and optimization cost between nodes through the node optimization association network, selects the optimal optimization path using a heuristic algorithm, and generates a task assignment plan based on the optimal optimization path and node optimization capabilities;

[0037] The executor bee analyzes the semantic content and word vector features of the second quality label based on the quality optimization rule and the task assignment plan, selects words with similar semantic content from a preset synonym library for supplementation, and at the same time expands the dimensional information of the word vector features to generate an optimized label.

[0038] The follower bee receives the optimization instruction set, calculates the optimization distance and optimization cost between nodes through the node optimization association network, selects the optimal optimization path using a heuristic algorithm, and generates a task assignment plan based on the optimal optimization path and node optimization capabilities, including:

[0039] Receive the optimization instruction set, extract the priority parameters in the optimization instruction set, and determine the semantic distance weight, structural distance weight, and functional distance weight based on the priority parameters;

[0040] Calculate the cosine similarity of the node label word vectors to obtain the semantic distance, calculate the hierarchical span and the depth of the common ancestor node between nodes in the label hierarchical tree to obtain the structural distance, and calculate the co-occurrence frequency and complementarity degree between nodes to obtain the functional distance;

[0041] Multiplying the semantic distance, the structural distance, and the functional distance by corresponding weights respectively and adding them together to generate a comprehensive optimized distance matrix of the node optimized association network;

[0042] Obtaining the task queue length and resource occupancy of the node to obtain a node load value, analyzing the time resources of the node to obtain a time constraint value, and calculating the optimization cost based on the comprehensive optimization distance matrix, the node load value and the time constraint value;

[0043] Setting an initial pheromone distribution according to the priority information of the optimization instruction set, generating a state transition probability based on the optimization cost and the node optimization capability, and performing a path search based on the state transition probability;

[0044] During the path search process, local pheromones are updated in real time, and historical optimal path information is recorded. When the number of search iterations reaches a preset threshold, the optimal optimized path is selected;

[0045] Decomposing the optimization task based on the optimal optimization path and node processing capabilities, generating a set of subtasks, determining dependencies between subtasks, and calculating resource requirements for each subtask;

[0046] A task scheduling strategy is formulated according to the subtask set, the dependency relationship and the resource requirements, and a task allocation plan including execution order and parallelism is generated.

[0047] Calculate the cosine similarity of the node label word vector to obtain the semantic distance, calculate the hierarchical span and common ancestor node depth between nodes in the label hierarchy tree to obtain the structural distance, and calculate the co-occurrence frequency and complementarity between nodes to obtain the functional distance, including:

[0048] Obtain node label text, input the node label text into a pre-trained word vector model, and generate a word vector corresponding to the node label text; calculate the cosine similarity of the word vectors between the node labels to obtain the semantic distance between the node labels;

[0049] The semantic distance calculation formula is as follows:

[0050] ;

[0051] in, is the semantic similarity, t i , t j For two label nodes, is the corresponding word vector;

[0052] Obtain a pre - constructed tag hierarchy tree, locate the node positions in the tag hierarchy tree based on the semantic distance; calculate the hierarchical span between nodes, where the hierarchical span is determined by the path length between nodes; identify the common ancestor nodes between nodes and calculate the depth of the common ancestor nodes; perform a weighted combination of the hierarchical span and the depth to obtain the structural distance between node tags;

[0053] The structural distance calculation formula is as follows:

[0054] ;

[0055] where α is the weight coefficient of the structural distance, path(t i ,t j ) is the hierarchical span between nodes, maxPath is the maximum path length in the tree, depth(LCA) is the depth of the nearest common ancestor node, and maxDepth is the maximum depth of the tree;

[0056] Determine associated node pairs based on the semantic distance and the structural distance, count the co - occurrence frequencies of the associated node pairs; extract the usage scenario information of the associated node pairs, calculate the complementary degree of node tags in the usage scenario; perform a weighted combination of the co - occurrence frequency and the complementary degree to obtain the functional distance between node tags;

[0057] The functional distance calculation formula is as follows:

[0058] ;

[0059] where t i is the i - th tag node to be calculated, t j is the j - th tag node to be calculated, β is the weight coefficient of the functional distance, coot(t i ,t j ) is the co - occurrence times of the tag pair, is the total co - occurrence times of all tag pairs, t k is the k - th tag combination in the tag set, is the distribution vector of tag t i in all scenarios, is the distribution vector of tag t j in all scenarios, is the scenario vector of the tag, is the Jensen - Shannon divergence of the scenario vector.

[0060] In the second aspect of the embodiments of the present invention,

[0061] Provide an intelligent swarm - type quality optimization management system for multi - source tag data, including:

[0062] The first unit is used to obtain multi-source tag data input, and respectively extract the comprehensive feature vectors of each tag through a feature extraction module. The comprehensive feature vectors include semantic feature vectors, association feature vectors, and timeliness feature vectors. The comprehensive feature vectors are input into a feature fusion layer to generate a tag comprehensive feature matrix. The tag comprehensive feature matrix is processed by a deep feature enhancement network to calculate a similarity feature matrix and a dependency feature matrix between tags. A tag quality evaluation model is constructed based on each feature matrix to calculate the initial quality score of each tag and generate a tag quality grading sequence.

[0063] The second unit is used to map the tag data into bee colony network nodes according to the tag quality grading sequence, and perform a three-level division of the nodes based on the initial quality score: the nodes with quality scores higher than a preset threshold are set as command bees, responsible for formulating optimization strategies; the nodes with quality scores equal to the preset threshold are set as follower bees, responsible for transmitting optimization instructions; the nodes with quality scores lower than the preset threshold are set as execution bees, responsible for implementing optimization operations. A node optimization association network is constructed using the similarity feature matrix and the dependency feature matrix, and a collaborative optimization rule set is designed based on the node optimization association network.

[0064] The third unit is used to start an optimization process based on the collaborative optimization rule set. The command bee analyzes the feature pattern of the first-quality tag to generate an optimization instruction set. The follower bee performs an optimization path planning through the node optimization association network to generate a task allocation plan. The execution bee performs feature optimization on the second-quality tag according to the task allocation plan. After optimization, the optimization result is re-input into the feature extraction module for quality evaluation and the tag quality grading sequence is updated. When the improvement value of the quality score after continuous multi-round optimization is less than a preset improvement threshold, the optimization process ends and the final tag data set is output.

[0065] In the third aspect of the embodiments of the present invention,

[0066] An electronic device is provided, including:

[0067] A processor;

[0068] A memory for storing instructions executable by the processor;

[0069] Wherein, the processor is configured to call the instructions stored in the memory to execute the method described above.

[0070] In the fourth aspect of the embodiments of the present invention,

[0071] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0072] The beneficial effects of this application are as follows:

[0073] 1. By constructing a deep feature enhancement network to process the label comprehensive feature matrix, it can effectively capture the similarity and dependency relationships between labels, achieve accurate evaluation of the label data quality, and improve the accuracy and reliability of label quality evaluation.

[0074] 2. Adopting the hierarchical structure design of the bee colony network, dividing label nodes into three levels: instruction bees, follower bees, and executor bees, establishing a clear optimization division mechanism, and realizing the efficient optimization of multi-source label data through the formulation of collaborative optimization rule sets, significantly improving the execution efficiency of label optimization.

[0075] 3. Based on the node optimization association network for task allocation and path planning, combined with the dynamic feedback mechanism of quality evaluation, it realizes the adaptive adjustment of the label optimization process, ensures the continuous improvement of the optimization results, and finally outputs a high-quality label data set, providing a reliable data basis for subsequent data applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 It is a schematic flow chart of the intelligent bee colony-based quality optimization management method for multi-source label data in an embodiment of the present invention;

[0077] Figure 2 It is the overall architecture diagram of the data label platform in an embodiment of the present invention;

[0078] Figure 3 It is a schematic structural diagram of the intelligent bee colony-based quality optimization management system for multi-source label data in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0079] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0080] The technical solutions of the present invention will be described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments.

[0081] Figure 1 It is a schematic flow chart of the intelligent bee colony-based quality optimization management method for multi-source label data in an embodiment of the present invention, as Figure 1 shown, and the method includes:

[0082] S11. Obtain multi-source label data input, and respectively extract the comprehensive feature vectors of each label through the feature extraction module. The comprehensive feature vectors include semantic feature vectors, association feature vectors, and timeliness feature vectors. Input the comprehensive feature vectors into the feature fusion layer to generate a label comprehensive feature matrix; process the label comprehensive feature matrix through a deep feature enhancement network, and calculate and obtain a similarity feature matrix and a dependence feature matrix between labels; construct a label quality evaluation model based on each feature matrix, calculate the initial quality score of each label, and generate a label quality grading sequence;

[0083] S12. According to the label quality grading sequence, map the label data to the bee colony network nodes, and perform a three-level division of the nodes based on the initial quality score: set the nodes with quality scores higher than the preset threshold as command bees, responsible for formulating optimization strategies; set the nodes with quality scores equal to the preset threshold as follower bees, responsible for transmitting optimization instructions; set the nodes with quality scores lower than the preset threshold as execution bees, responsible for implementing optimization operations; construct a node optimization association network using the similarity feature matrix and the dependence feature matrix, and design a collaborative optimization rule set based on the node optimization association network;

[0084] S13. Start the optimization process based on the collaborative optimization rule set. The command bee analyzes the feature pattern of the first-quality label to generate an optimization instruction set. The follower bee performs an optimization path planning through the node optimization association network to generate a task assignment plan. The execution bee performs feature optimization on the second-quality label according to the task assignment plan; after the optimization is completed, input the optimization result back into the feature extraction module for quality evaluation and update the label quality grading sequence; when the improvement value of the quality score after continuous multi-round optimization is less than the preset improvement threshold, end the optimization process and output the final label data set.

[0085] In an optional implementation manner, processing the label comprehensive feature matrix through a deep feature enhancement network, calculating and obtaining a similarity feature matrix and a dependence feature matrix between labels; constructing a label quality evaluation model based on each feature matrix, calculating the initial quality score of each label, and generating a label quality grading sequence includes:

[0086] Construct a deep feature enhancement network. The deep feature enhancement network includes a feature mapping layer, a feature enhancement layer, and a feature reconstruction layer. The feature mapping layer uses a non-linear activation function to map the label comprehensive feature matrix to a high-dimensional latent space. The feature enhancement layer calculates the attention weights between feature vectors and performs weighted fusion based on the attention weights to achieve feature enhancement. The feature reconstruction layer uses a deconvolution operation to remap the enhanced features back to the original feature space;

[0087] The cosine similarity method is used to calculate the semantic similarity between tags to obtain a similarity feature matrix. At the same time, a temporal correlation analysis method is introduced to construct a directed weighted graph through the temporal pattern and co-occurrence frequency of tag appearances to obtain a dependency feature matrix;

[0088] The similarity feature matrix and the dependency feature matrix are fused with the original tag comprehensive feature matrix to construct a tag quality evaluation model. The tag quality evaluation model calculates the quality of the tag's own features, the similarity relationship between tags, and the dependency relationship between tags respectively, and obtains an initial quality score through weighted summation;

[0089] The initial quality score is normalized, and the density clustering method is used to automatically determine the quality level division threshold, generating a tag quality grading sequence containing quality level information.

[0090] The present invention proposes a tag quality evaluation method based on a deep feature enhancement network. This method processes the tag comprehensive feature matrix, calculates the similarity feature matrix and the dependency feature matrix between tags, constructs a tag quality evaluation model based on these feature matrices, and finally generates a tag quality grading sequence.

[0091] First, a deep feature enhancement network is constructed. This network includes a feature mapping layer, a feature enhancement layer, and a feature reconstruction layer. The feature mapping layer uses a non-linear activation function such as ReLU to map the tag comprehensive feature matrix to a high-dimensional latent space. For example, assuming the original feature matrix has a dimension of 100x50, it may be mapped to a high-dimensional space of 100x200 through the feature mapping layer. The feature enhancement layer realizes feature enhancement by calculating the attention weights between feature vectors and performing weighted fusion based on these weights. Specifically, the dot product attention mechanism can be used to calculate the attention scores between each pair of feature vectors, and then the softmax function is used to normalize these scores into weights. Then, these weights are used to perform weighted summation on the feature vectors to obtain the enhanced features. The feature reconstruction layer uses a transposed convolution operation to remap the enhanced features back to the original feature space. This step can use a transposed convolution operation to remap the high-dimensional features (such as 100x200) back to the original dimension (100x50).

[0092] Next, calculate the similarity feature matrix and dependence feature matrix between tags. For the similarity feature matrix, the cosine similarity method is used to calculate the semantic similarity between tags. Specifically, for the feature vectors of any two tags, calculate their dot product divided by their respective L2 norms to obtain a similarity value between -1 and 1. In this way, for n tags, an n×n similarity matrix is finally obtained. For the dependence feature matrix, a time series correlation analysis method is introduced. By analyzing the occurrence patterns and co-occurrence frequencies of tags in the time series, a directed weighted graph is constructed. For example, the frequency of tag B following tag A within a certain time window can be counted, and this frequency is used as the weight of the edge from A to B. Finally, an n×n dependence matrix is obtained, where each element represents the degree of dependence of one tag on another tag.

[0093] Then, fuse the similarity feature matrix and dependence feature matrix with the original tag comprehensive feature matrix to construct a tag quality evaluation model. This model calculates the quality of the features of the tags themselves, the similarity relationship between tags, and the dependence relationship between tags respectively. For the quality of the features of the tags themselves, factors such as the usage frequency and coverage range of the tags can be considered. The similarity relationship between tags can directly use the values in the similarity feature matrix. The dependence relationship between tags can be calculated based on the dependence feature matrix. Finally, the initial quality score of each tag is obtained by weighted summation. For example, weights of 0.4, 0.3, and 0.3 can be set, and the scores of the three aspects are weighted and summed.

[0094] Finally, normalize the initial quality scores, use the density clustering method to automatically determine the quality level division threshold, and generate a tag quality grading sequence containing quality level information. The normalization process can use the min-max scaling method to map all scores to between 0 and 1. Then, use density clustering algorithms such as DBSCAN to automatically determine several clustering centers according to the distribution characteristics of the scores, and the midpoints between these centers can be used as the quality level division thresholds. For example, if three clustering centers are determined, the tag quality can be divided into four levels: excellent, good, average, and to be improved. Finally, according to these thresholds, each tag is assigned to the corresponding quality level to form a tag quality grading sequence.

[0095] The solution of this application can:

[0096] This method enhances the expressive ability of label features through a deep feature enhancement network, enabling better capture of the complex relationships between labels, thereby improving the accuracy and reliability of label quality assessment. By introducing temporal correlation analysis and constructing a directed weighted graph, this method not only considers the static similarity between labels but also dynamic dependency relationships, making the quality assessment results more comprehensive and objective. The density clustering method is used to automatically determine the threshold for quality level division, avoiding the subjectivity and inaccuracy that may be brought about by manually setting the threshold, making the label quality grading more scientific and reasonable, and facilitating subsequent label optimization and management work.

[0097] In an alternative implementation, the cosine similarity method is used to calculate the semantic similarity between labels to obtain a similarity feature matrix. At the same time, a temporal correlation analysis method is introduced to construct a directed weighted graph through the temporal pattern and co-occurrence frequency of label appearances to obtain a dependency feature matrix, including:

[0098] Obtain a label text set, input the label text set into a pre-trained word vector model to obtain an initial word vector corresponding to each label text; for a label containing multiple words in the label text set, calculate the term frequency-inverse document frequency value of each word in the label as a weight coefficient, and perform weighted averaging on the initial word vectors of each word according to the weight coefficient to obtain an overall semantic feature vector of the label.

[0099] Based on the overall semantic feature vector, calculate the cosine similarity between any two labels in the label text set, and fill the cosine similarity into the corresponding row and column positions of the similarity feature matrix to obtain the semantic similarity feature matrix of the label text set. The dimension of the semantic similarity feature matrix is the product of the total number of labels and the total number of labels.

[0100] Obtain the temporal data of the label text set, set a time window, analyze the co-occurrence pattern of the label text set within the time window, count the co-occurrence frequency of label pairs, record the relative occurrence order of labels, and calculate the time interval distribution of label pairs; based on the co-occurrence frequency, the relative occurrence order, and the time interval distribution, use a preset weight coefficient for weighted combination to obtain the temporal dependency strength of label pairs.

[0101] Construct a directed weighted graph according to the temporal dependency strength, map the labels in the label text set to the nodes of the directed weighted graph, determine the connection direction between nodes based on the relative occurrence order of label pairs, and use the temporal dependency strength as the weight value of the corresponding connection to obtain a directed temporal weighted graph describing label temporal correlation; construct a dependency feature matrix based on the connection relationship and weight value of the directed temporal weighted graph.

[0102] The present invention proposes a method for constructing a label feature matrix based on semantic similarity and temporal correlation, and the method includes the following steps:

[0103] First, obtain a set of label texts. This set contains all the labels to be analyzed. For example, suppose there is a set containing 5 labels: {"Machine Learning", "Deep Learning", "Artificial Intelligence", "Data Mining", "Neural Network"}.

[0104] Next, input the obtained set of label texts into a pre-trained word vector model. This model can be Word2Vec, GloVe, FastText, etc. Suppose a 300-dimensional word vector model is used, then each label will obtain an initial 300-dimensional word vector.

[0105] For labels containing multiple words, it is necessary to calculate the term frequency-inverse document frequency (TF-IDF) values of each word in the label as weight coefficients. For example, for the label "Deep Learning", calculate the TF-IDF values of "deep" and "learning" respectively. Suppose the TF-IDF value of "deep" is 0.6 and the TF-IDF value of "learning" is 0.4.

[0106] Then, perform weighted averaging on the initial word vectors of each word according to the calculated weight coefficients to obtain the overall semantic feature vector of the label. Continuing with the example of "Deep Learning", multiply the word vector of "deep" by 0.6, multiply the word vector of "learning" by 0.4, and then add the two results to obtain the overall semantic feature vector of the label "Deep Learning".

[0107] Based on the obtained overall semantic feature vector, calculate the cosine similarity between any two labels in the set of label texts. For example, calculate the cosine similarity between "Machine Learning" and "Deep Learning", and suppose the obtained result is 0.85. Fill the calculated cosine similarity into the corresponding row and column positions of the similarity feature matrix. For the above example of 5 labels, the finally obtained semantic similarity feature matrix is a 5x5 matrix.

[0108] Next, obtain the temporal data of the set of label texts. These data include the timestamp information of the label appearance.

[0109] Set a suitable time window, such as 1 hour. Analyze the co-occurrence pattern of the set of label texts within this time window. Count the co-occurrence frequency of label pairs. For example, within 1 hour, "Machine Learning" and "Deep Learning" co-occur 10 times. At the same time, record the relative appearance order of the labels, such as "Machine Learning" appears before "Deep Learning" in 80% of the cases.

[0110] Calculate the time interval distribution of tag pairs. For example, by counting the time intervals when the tag pair "machine learning" and "deep learning" co-occur, the possible results could be: in 50% of the cases, the interval is within 5 minutes; in 30% of the cases, the interval is between 5 - 15 minutes; and in 20% of the cases, the interval is between 15 - 30 minutes.

[0111] Based on the co-occurrence frequency, relative occurrence order, and time interval distribution, a preset weight coefficient is used for weighted combination to obtain the temporal dependence strength of the tag pair. Assume that the weight of the co-occurrence frequency is 0.5, the weight of the relative occurrence order is 0.3, and the weight of the time interval distribution is 0.2. Then, the temporal dependence strength of "machine learning" with respect to "deep learning" can be calculated according to these weights.

[0112] According to the calculated temporal dependence strength, a directed weighted graph is constructed. The tags in the tag text set are mapped to the nodes of the directed weighted graph. Based on the relative occurrence order of the tag pair, the connection direction between the nodes is determined. For example, a directed edge is established from the "machine learning" node to the "deep learning" node. The temporal dependence strength is used as the weight value of the corresponding connection.

[0113] Finally, based on the connection relationship and weight values of the constructed directed temporal weighted graph, a dependence feature matrix is constructed. The dimension of this matrix is also the product of the total number of tags and the total number of tags, i.e., 5x5. Each element in the matrix represents the degree of dependence of the row tag on the column tag.

[0114] Through the above steps, two feature matrices are obtained: the semantic similarity feature matrix and the dependence feature matrix. These two matrices jointly describe the semantic relationship and temporal correlation between tags, providing rich feature information for subsequent tag analysis and recommendation tasks.

[0115] The solution of this application can:

[0116] By combining semantic similarity and temporal correlation analysis, this method comprehensively depicts the relationship between tags, improving the expression ability and accuracy of tag features. Using the pre-trained word vector model and TF-IDF weighted average, it effectively captures the semantic information of tags, overcoming the limitations of traditional methods that rely only on literal matching. Introducing temporal correlation analysis, it considers the time pattern and co-occurrence characteristics of tag appearances, revealing the dynamic dependence relationship between tags and providing more valuable information support for time-series related application scenarios.

[0117] In an alternative embodiment, the similarity feature matrix and the dependence feature matrix are fused with the original tag comprehensive feature matrix to construct a tag quality evaluation model. The tag quality evaluation model calculates the quality of the tag's own features, the similarity relationship between tags, and the dependence relationship between tags respectively, and obtains an initial quality score through weighted summation, including:

[0118] Perform weighted averaging on the semantic similarity features and the structural similarity features in the similarity feature matrix to obtain the comprehensive similarity features between labels, and perform weighted averaging on the hierarchical dependence features and the associative dependence features in the dependence feature matrix to obtain the comprehensive dependence features between labels;

[0119] Normalize the language normality features, professional term normality features, and semantic integrity features in the original label comprehensive feature matrix to obtain the normalized features, and perform logarithmic transformation on the word frequency distribution features and the query coverage features to obtain the statistical features;

[0120] Concatenate and combine the comprehensive similarity features between labels, the comprehensive dependence features between labels, the normalized features, and the statistical features to construct a feature fusion matrix, and construct a label quality evaluation model based on the feature fusion matrix;

[0121] Calculate the quality score of the label's own features based on the label quality evaluation model, where the quality score of the label's own features is obtained by performing weighted summation on the normalized features and the statistical features;

[0122] Calculate the similarity relationship score between labels based on the label quality evaluation model, where the similarity relationship score between labels is obtained by performing threshold screening on the comprehensive similarity features between labels;

[0123] Calculate the dependence relationship score between labels based on the label quality evaluation model, where the dependence relationship score between labels is obtained by performing weighted calculation on the comprehensive dependence features between labels;

[0124] Perform weighted summation on the quality score of the label's own features, the similarity relationship score between labels, and the dependence relationship score between labels to obtain the initial label quality score.

[0125] This embodiment provides a method for constructing a label quality evaluation model, which realizes a comprehensive evaluation of label quality through feature fusion and multi-dimensional evaluation.

[0126] First, perform weighted averaging on the semantic similarity features and the structural similarity features in the similarity feature matrix to obtain the comprehensive similarity features between labels. Specifically, different weights can be assigned to the semantic similarity features and the structural similarity features. For example, the semantic similarity weight is 0.6, and the structural similarity weight is 0.4. In this way, the semantic and structural similarities between labels can be comprehensively considered.

[0127] Meanwhile, perform a weighted average on the hierarchical dependence features and association dependence features in the dependence feature matrix to obtain the comprehensive dependence features between labels. Different weights can be assigned to the hierarchical dependence features and association dependence features respectively. For example, the weight of the hierarchical dependence is 0.7, and the weight of the association dependence is 0.3. This can comprehensively consider the dependence relationship between labels.

[0128] Next, normalize the language normality features, professional term normality features, and semantic integrity features in the original label comprehensive feature matrix to obtain the normalized features. The min-max normalization method can be used for normalization, mapping the feature values to the range of 0 to 1. For the word frequency distribution feature and the query coverage feature, perform logarithmic transformation to obtain the statistical features. Logarithmic transformation can effectively compress the data range and make the feature distribution more uniform.

[0129] Concatenate and combine the above-obtained comprehensive similarity features between labels, comprehensive dependence features between labels, normalized features, and statistical features to construct a feature fusion matrix. This feature fusion matrix contains all the feature information required for label quality assessment. Based on this feature fusion matrix, construct a label quality assessment model.

[0130] In the label quality assessment model, first calculate the quality score of the label's own features. This score is obtained by performing a weighted sum on the normalized features and statistical features. Different weights can be assigned to different features. For example, the weight of the language normality feature is 0.3, the weight of the professional term normality feature is 0.2, the weight of the semantic integrity feature is 0.2, the weight of the word frequency distribution feature is 0.15, and the weight of the query coverage feature is 0.15.

[0131] Then, calculate the similarity relationship score between labels. This score is obtained by performing threshold screening on the comprehensive similarity features between labels. A similarity threshold can be set. For example, when the comprehensive similarity between two labels is greater than or equal to this threshold, it is considered that they have a significant similarity relationship and a higher score is assigned; otherwise, a lower score is assigned.

[0132] Next, calculate the label dependence relationship score. This score is obtained by performing a weighted calculation on the comprehensive dependence features between labels. Different weights can be set according to the strength of the dependence relationship. For example, the weight of the strong dependence relationship is 0.8, the weight of the medium dependence relationship is 0.5, and the weight of the weak dependence relationship is 0.2.

[0133] Finally, perform a weighted sum on the quality score of the label's own features, the similarity relationship score between labels, and the label dependence relationship score to obtain the initial label quality score. Different weights can be assigned to these three scores respectively. For example, the weight of the quality score of the label's own features is 0.5, the weight of the similarity relationship score is 0.3, and the weight of the dependence relationship score is 0.2.

[0134] To better understand this process, a specific data case can be used to illustrate. Suppose there is a label "Artificial Intelligence" with the following characteristic values:

[0135] Language normativity characteristic: 0.95; Professional term normativity characteristic: 0.88; Semantic integrity characteristic: 0.92; Word frequency distribution characteristic: 1000 (6.91 after logarithmic transformation); Query coverage characteristic: 0.85; Comprehensive similarity characteristic with other labels: 0.82 (with "Machine Learning"), 0.75 (with "Deep Learning"); Comprehensive dependence characteristic with other labels: 0.9 (with "Algorithm"), 0.85 (with "Big Data").

[0136] First, perform a weighted sum of the normalization characteristics and statistical characteristics to obtain the label's own characteristic quality score:

[0137] 0.95 * 0.3 + 0.88 * 0.2 + 0.92 * 0.2 + 6.91 * 0.15 + 0.85 * 0.15 = 1.8965;

[0138] Then, perform threshold screening on the similarity relationships between labels. Assume the threshold is 0.8. Then the similarity relationship score between "Artificial Intelligence" and "Machine Learning" is 1, and the similarity relationship score with "Deep Learning" is 0.

[0139] Next, calculate the label dependence relationship score. Assume that "Artificial Intelligence" has a strong dependence relationship with "Algorithm" and a medium dependence relationship with "Big Data":

[0140] 0.9 * 0.8 + 0.85 * 0.5 = 1.145;

[0141] Finally, perform a weighted sum of these three scores to obtain the initial label quality score:

[0142] 1.8965 * 0.5 + 1 * 0.3 + 1.145 * 0.2 = 1.4233;

[0143] In this way, an evaluation result of label quality that comprehensively considers multiple factors can be obtained.

[0144] The solution of this application can:

[0145] Through feature fusion and multi-dimensional evaluation, this technical solution realizes a comprehensive evaluation of label quality, improving the accuracy and reliability of the evaluation. By performing weighted averaging on similarity features and dependency features, and normalizing and logarithmically transforming the original features, different types of features can be effectively fused together, making full use of various feature information. The final label quality score is calculated by weighted summation, making the evaluation result more flexible and adjustable. The weights of different features can be adjusted according to actual needs to adapt to different application scenarios.

[0146] In an optional implementation manner, an optimization process is started based on the collaborative optimization rule set. The instruction bee analyzes the feature pattern of the first quality label to generate an optimization instruction set. The follower bee generates a task assignment plan through optimizing the path planning of the node optimization association network. The executor bee performs feature optimization on the second quality label according to the task assignment plan, including:

[0147] Construct a node optimization association network and establish a collaborative optimization rule set. The collaborative optimization rule set includes a feature extraction rule, an optimization instruction generation rule, and a quality optimization rule. Start the label optimization process based on the collaborative optimization rule set;

[0148] The instruction bee analyzes the first quality label according to the feature extraction rule, obtains semantic features through word vector representation, obtains structural features through hierarchical relationships, obtains usage features through application frequencies, and performs weighted combination on the semantic features, the structural features, and the usage features to generate label feature data;

[0149] The instruction bee processes the label feature data based on the optimization instruction generation rule, inputs the label feature data into the feature vector space for clustering analysis, extracts feature patterns to establish a feature pattern library, and generates an optimization instruction set according to the feature pattern library;

[0150] The follower bee receives the optimization instruction set, calculates the optimization distance and optimization cost between nodes through the node optimization association network, selects the optimal optimization path using a heuristic algorithm, and generates a task assignment plan based on the optimal optimization path and node optimization capabilities;

[0151] The executor bee analyzes the semantic content and word vector features of the second quality label based on the quality optimization rule and the task assignment plan, selects words with similar semantic content from a preset synonym library for supplementation, and at the same time expands the dimensional information of the word vector features to generate an optimized label.

[0152] The label quality optimization system based on the node - optimized association network first constructs an optimized association network. By establishing the connection relationships between nodes, a network topology structure is formed. Each node has independent optimization ability indicators, including parameters such as computing resources and storage capacity. At the same time, a collaborative optimization rule set is formulated, which includes three parts: feature extraction rules, optimization instruction generation rules, and quality optimization rules.

[0153] In the feature extraction stage, the instruction bee conducts multi - dimensional analysis on the first quality label. Through word vector models such as Word2Vec, the semantic features of the label are obtained, and each word is mapped to a 300 - dimensional vector space. Analyze the hierarchical position of the label in the classification system, and extract the hypernym - hyponym relationship as a structural feature. Statistically analyze the usage features such as the usage frequency and coverage of the label in historical data. Combine the three types of features in a weighted manner to generate a label feature vector.

[0154] In the optimization instruction generation stage, the instruction bee inputs the label feature vector into the feature space. The K - means clustering algorithm is used to perform clustering analysis on the feature vector to extract feature patterns. According to the clustering results, a feature pattern library is established, which contains the typical feature patterns of various types of labels. Based on the pattern library, a targeted optimization instruction set is generated.

[0155] After receiving the optimization instruction set, the follower bee conducts path planning through the node - optimized association network. Calculate the optimized distance between nodes, considering factors such as network latency and bandwidth. The ant colony algorithm is used to select the optimal path, weighing between the path length and the optimization effect. Combine the optimization ability of the nodes to generate a task allocation scheme.

[0156] The executor bee optimizes the second quality label according to the allocation scheme. Analyze the semantic content of the label, and select similar words from the thesaurus for supplementation. At the same time, expand the dimension of the word vector features to enhance the semantic expression ability. Finally, an optimized high - quality label is generated.

[0157] After analysis, the original label "mobile phone" obtains a semantic feature vector [0.82, 0.56,...], the structural feature is {"hypernym": "electronic product", "hyponym": "smartphone"}, and the usage frequency is 2,850 times. After optimization, it is expanded to "mobile phone, smartphone, communication device, mobile terminal", and the dimension of the word vector is expanded from 300 to 500.

[0158] The solution of this application can:

[0159] Improve the semantic expression ability of the label through multi - dimensional feature analysis and collaborative optimization, making the label more accurately reflect the content features. Adopt a distributed node collaborative optimization method, improve the processing efficiency of label optimization, and is suitable for large - scale data processing scenarios. The optimization scheme based on feature patterns and heuristic algorithms has strong adaptability and can perform targeted optimization according to the characteristics of different types of labels.

[0160] In an alternative embodiment, the follower bees receive the optimization instruction set, calculate the optimized distances and optimized costs between nodes through the node optimization association network, select the optimal optimization path using a heuristic algorithm, and generate a task allocation scheme based on the optimal optimization path and node optimization capabilities, including:

[0161] Receive the optimization instruction set, extract the priority parameters in the optimization instruction set, and determine the semantic distance weight, structural distance weight, and functional distance weight based on the priority parameters;

[0162] Calculate the cosine similarity of the node label word vectors to obtain the semantic distance, calculate the hierarchical span between nodes and the depth of the common ancestor node in the label hierarchical tree to obtain the structural distance, and calculate the co-occurrence frequency and complementarity degree between nodes to obtain the functional distance;

[0163] Multiply and add the semantic distance, the structural distance, and the functional distance with their corresponding weights respectively to generate a comprehensive optimization distance matrix of the node optimization association network;

[0164] Obtain the node load value by getting the task queue length and resource occupancy of the node, analyze the time resources of the node to obtain the time constraint value, and calculate the optimization cost based on the comprehensive optimization distance matrix, the node load value, and the time constraint value;

[0165] Set the initial pheromone distribution according to the priority information of the optimization instruction set, generate the state transition probability by combining the optimization cost and node optimization capabilities, and perform path search based on the state transition probability;

[0166] During the path search process, update the local pheromone in real time and record the historical optimal path information. When the search iteration times reach the preset threshold, select the optimal optimization path;

[0167] Decompose the optimization task based on the optimal optimization path and node processing capabilities, generate a sub-task set and determine the dependency relationship between sub-tasks, and calculate the resource requirements for each sub-task;

[0168] Formulate a task scheduling strategy according to the sub-task set, the dependency relationship, and the resource requirements, and generate a task allocation scheme including the execution order and parallelism.

[0169] After the follower bee system receives the optimized instruction set, it first parses the priority parameter information contained in the instruction set. The priority parameters usually include weight values in three dimensions: the semantic distance weight ranges from 0.2 to 0.4, the structural distance weight ranges from 0.3 to 0.5, and the functional distance weight ranges from 0.2 to 0.4. For example, when the system emphasizes semantic similarity, the semantic distance weight can be set to 0.4, the structural distance weight to 0.3, and the functional distance weight to 0.3.

[0170] After determining the weights, the system starts to calculate the multi-dimensional distances between nodes. For the semantic distance, the system uses a pre-trained word vector model to convert the node labels into vector representations, and obtains the semantic distance value by calculating the cosine similarity between the vectors. Taking nodes A and B as an example, if the similarity of their label word vectors is 0.85, then the semantic distance is 0.15.

[0171] The calculation of the structural distance is based on the label hierarchy tree. Suppose node A is on the third layer and node B is on the fifth layer, and their nearest common ancestor node is on the second layer, then the hierarchical span is 3 and the depth of the common ancestor is 2. By combining these two factors, the structural distance value of 0.4 can be obtained.

[0172] The functional distance is obtained by analyzing historical task data. The system counts the task collaboration frequency and functional complementarity degree between nodes. For example, the historical collaboration times between nodes A and B are 50 times, and the functional complementarity degree is 0.7. From this, the functional distance value of 0.3 is calculated.

[0173] Multiply the distance values in the three dimensions by the corresponding weights and sum them to obtain the comprehensive optimization distance. Continuing with nodes A and B as an example, their comprehensive optimization distance is 0.28. The system performs this calculation for all node pairs to generate a complete optimization distance matrix.

[0174] Next, the system collects node status information to calculate the optimization cost. For each node, its current task queue length and resource utilization rate are obtained. For example, the task queue length of node A is 10, the CPU utilization rate is 75%, and the memory utilization rate is 60%. From this, the node load value of 0.8 is obtained. At the same time, considering the time resource constraints of the node, such as the available time of node A within the next 2 hours is 1.5 hours, the time constraint value of 0.75 is obtained.

[0175] Based on the optimization distance matrix, node load values, and time constraint values, the system calculates a complete optimization cost network. On this basis, an improved ant colony algorithm is used for path search. Initially, the pheromone distribution is set according to the priority information, and the paths with higher priorities receive more initial pheromones.

[0176] During the iterative search process, the system updates the local pheromone concentration in real time. After each path search is completed, the pheromone value is adjusted according to the optimization cost of the path. At the same time, a global optimal path record is maintained. When the number of iterations reaches 1000, the optimal path found during this period is selected as the final optimized path.

[0177] After obtaining the optimal path, the system decomposes the optimization task according to the processing capabilities of the nodes. For example, a data processing task can be decomposed into subtasks such as data cleaning, feature extraction, and model training. The system analyzes the dependencies between subtasks, constructs a task dependency graph, and estimates the resource requirements for each subtask.

[0178] Finally, the system generates a detailed task allocation plan based on the subtask set, dependencies, and resource requirements. This plan includes specific information such as the execution order of tasks, parallelism settings, and resource allocation.

[0179] The solution of this application can:

[0180] Through the multi-dimensional distance calculation and comprehensive weight allocation mechanism, the correlation degree between nodes can be comprehensively evaluated, improving the accuracy and rationality of task allocation. By using an improved ant colony algorithm for path optimization, combined with real-time pheromone update and historical optimal path record, the optimization efficiency is significantly improved, and the system resource consumption is reduced. Based on fine-grained task decomposition and dependency analysis, the optimal scheduling of tasks and resource allocation are realized, improving the overall execution efficiency and resource utilization rate of the system.

[0181] In an alternative embodiment, the cosine similarity of the node label word vectors is calculated to obtain the semantic distance, the hierarchical span and the depth of the common ancestor node between nodes in the label hierarchy tree are calculated to obtain the structural distance, and the co-occurrence frequency and complementarity degree between nodes are calculated to obtain the functional distance, including:

[0182] Obtain the node label text, input the node label text into a pre-trained word vector model to generate the word vector corresponding to the node label text; calculate the cosine similarity between the word vectors of the node labels to obtain the semantic distance between the node labels;

[0183] The semantic distance calculation formula is as follows:

[0184] ;

[0185] Where is the semantic similarity, t i , t j are two label nodes, is the corresponding word vector;

[0186] Obtain a pre-built tag hierarchy tree, locate the node positions in the tag hierarchy tree based on the semantic distance; calculate the hierarchical span between nodes, where the hierarchical span is determined by the path length between nodes; identify the common ancestor nodes between nodes and calculate the depth of the common ancestor nodes; perform a weighted combination of the hierarchical span and the depth to obtain the structural distance between node tags;

[0187] The structural distance calculation formula is as follows:

[0188] ;

[0189] where α is the weight coefficient of the structural distance, path(t i ,t j ) is the hierarchical span between nodes, maxPath is the maximum path length in the tree, depth(LCA) is the depth of the nearest common ancestor node, and maxDepth is the maximum depth of the tree;

[0190] Determine associated node pairs based on the semantic distance and the structural distance, count the co-occurrence frequencies of the associated node pairs; extract the usage scenario information of the associated node pairs, and calculate the complementary degree of node tags in the usage scenario; perform a weighted combination of the co-occurrence frequencies and the complementary degree to obtain the functional distance between node tags;

[0191] The functional distance calculation formula is as follows:

[0192] ;

[0193] where t i is the i-th tag node to be calculated, t j is the j-th tag node to be calculated, β is the weight coefficient of the functional distance, coot(t i ,t j ) is the number of co-occurrences of the tag pair, is the total number of co-occurrences of all tag pairs, t k is the k-th pair of tag combinations in the tag set, is the distribution vector of tag t i in all scenarios, is the distribution vector of tag t j in all scenarios, is the scenario vector of the tag, is the Jensen-Shannon divergence of the scenario vector.

[0194] When calculating label node similarity, the text content of the node label is first obtained. This label text is then vectorized using a pre-trained word embedding model. Mainstream word embedding models such as Word2Vec or BERT can be used. For example, the labels "data mining" and "machine learning" are input into the pre-trained model to generate corresponding high-dimensional vector representations. By calculating the cosine similarity between these two vectors, the semantic distance between the labels can be determined. The smaller the semantic distance, the closer the two labels are semantically.

[0195] After obtaining the semantic distance, the node position needs to be located within the pre-built label hierarchy tree. Taking the technology field label tree as an example, "artificial intelligence" is the parent node, with child nodes such as "machine learning" and "deep learning." The hierarchical span between any two nodes is calculated, which is the length of the shortest path between them. The nearest common ancestor node of these two nodes is also identified, and the depth value of this ancestor node is calculated. Combining hierarchical span and depth information yields a structural distance that reflects the structural relationships between nodes.

[0196] After determining semantic and structural distances, we can identify highly correlated node pairs. We then count the co-occurrence frequencies of these node pairs in real-world application scenarios. For example, in the case of technical document annotation, we analyze the co-occurrence of the tags "deep learning" and "neural network" within documents. We also extract the usage of these tags in different application scenarios and construct scenario distribution vectors. By calculating the differences between scenario distribution vectors, we can assess the complementarity between the tags. Finally, we perform a weighted combination of the co-occurrence frequency and complementarity information to derive a functional distance that reflects the strength of the association between nodes.

[0197] In practical applications, the weights of semantic distance, structural distance, and functional distance can be adjusted according to specific needs. For example, in the case of technical document classification, the weight of semantic distance can be appropriately increased; in the case of knowledge graph construction, the weight of structural distance can be increased; and in the case of recommendation systems, the influence of functional distance can be strengthened.

[0198] The solution of this application can:

[0199] By calculating the semantic distance, the semantic knowledge contained in the pre-trained model is fully utilized, which can accurately capture the semantic correlation between labels and effectively improve the accuracy of label association analysis. Combining the structural information of the label hierarchy tree, both the hierarchical span between nodes and the depth of the common ancestor nodes are considered, comprehensively reflecting the hierarchical organization relationship between labels and improving the rationality of similarity calculation. By integrating co-occurrence frequency and scenario complementarity analysis, both the statistical laws of label usage and the complementary role of labels in different scenarios are considered, characterizing the association features between labels from multiple dimensions and enhancing the comprehensiveness and reliability of label relationship analysis.

[0200] Figure 2 This is the overall architecture diagram of the data label platform according to the embodiments of the present invention. As Figure 2 shown, the method further includes:

[0201] The platform supports the access of multi-source heterogeneous data, covering structured data (such as SQL databases, data warehouses), semi-structured data (such as JSON, XML files), and unstructured data (such as logs, text files). Through the data source management module, users can easily configure the data source connection parameters to ensure the real-time access and efficient processing of multi-source data. The addition, deletion, and maintenance of data sources are all completed through the built-in visual interface of the platform, and the system will strictly verify the connectivity and access rights of the data sources.

[0202] The feature extraction module uses advanced feature engineering techniques to automatically extract high-value features from the accessed data. The system supports rule-based feature selection (such as correlation analysis based on statistical indicators, information gain, etc.), and provides model applications such as automatic clustering and classification to generate preliminary user labels. This process not only reduces the dependence on manual operations but also improves the accuracy and timeliness of feature extraction. Users can further adjust the feature selection rules through custom SQL scripts or drag-and-drop operation interfaces.

[0203] The automated configuration engine integrated in the platform is based on predefined business rules and can automatically complete the matching of feature data, the generation and configuration of labels. The configuration process includes entering basic label information, defining label values, associating data source tables, setting filtering conditions, and update methods. The system will dynamically adjust the label configuration according to the preset rules to ensure that corresponding update strategies are automatically generated when the data changes, and users can review and fine-tune the configuration results on the platform interface.

[0204] The feedback optimization module adopts a closed-loop management mechanism. Through user behavior analysis, historical data monitoring, and the input of real-time data streams, it automatically adjusts the key parameters in the label system. This module integrates an incremental learning model and an adaptive algorithm, which can quickly adjust the weights of labels, screening conditions, and data update frequencies according to business changes, ensuring that the label system can always maintain the best state. Through this mechanism, the platform realizes the continuous optimization and adaptive adjustment of labels, greatly improving the accuracy and dynamic response ability of labels.

[0205] The platform adopts a containerized deployment solution based on a microservices architecture. The core technology stack includes the SpringBoot framework, Docker containers, and the Kubernetes orchestration and management tool. The platform supports horizontal scaling and elastic scaling, and can automatically expand the processing capacity according to business needs. The service modules in the system perform service discovery and registration through Zookeeper. The data storage adopts a combined solution of MySQL and Elasticsearch to ensure fast read and write capabilities in a high-concurrency environment. The system also supports distributed log management, monitoring and alerting, and automated operation and maintenance to ensure high availability and stability.

[0206] The present invention proposes an intelligent label management and optimization platform for a big data environment, aiming to solve the problems of poor flexibility and insufficient automation ability in traditional data label management systems. This platform integrates functions such as multi-source data access, feature extraction, automated label generation and configuration, and dynamic feedback optimization. Through a microservices architecture and a distributed computing framework, it provides efficient data processing and intelligent label management. Its innovation points include an automated configuration engine, a closed-loop feedback optimization mechanism, multi-level permission management, and scalability, which are suitable for the big data management needs of enterprise users in scenarios such as user profiling and precision marketing.

[0207] Figure 3 It is a schematic structural diagram of the intelligent swarm-based quality optimization management system for multi-source label data in an embodiment of the present invention. As Figure 3 shown, the system includes:

[0208] A first unit, which is used to obtain the input of multi-source label data, and respectively extract the comprehensive feature vectors of each label through a feature extraction module. The comprehensive feature vectors include semantic feature vectors, association feature vectors, and timeliness feature vectors. Input the comprehensive feature vectors into a feature fusion layer to generate a label comprehensive feature matrix; process the label comprehensive feature matrix through a deep feature enhancement network to calculate the similarity feature matrix and dependence feature matrix between labels; construct a label quality evaluation model based on each feature matrix, calculate the initial quality score of each label, and generate a label quality grading sequence;

[0209] A second unit, configured to map the tag data into the nodes of the swarm network according to the tag quality grading sequence, and perform a three-level division on the nodes based on the initial quality score: set the nodes with quality scores higher than a preset threshold as command bees responsible for formulating optimization strategies; set the nodes with quality scores equal to the preset threshold as follower bees responsible for transmitting optimization instructions; set the nodes with quality scores lower than the preset threshold as executing bees responsible for implementing optimization operations; construct a node optimization association network by using the similarity feature matrix and the dependence feature matrix, and design a collaborative optimization rule set based on the node optimization association network;

[0210] A third unit, configured to initiate an optimization process based on the collaborative optimization rule set. The command bees analyze the feature patterns of the first quality tags to generate an optimization instruction set. The follower bees perform optimization path planning through the node optimization association network to generate a task allocation plan. The executing bees perform feature optimization on the second quality tags according to the task allocation plan. After the optimization is completed, the optimization results are re-input into the feature extraction module for quality evaluation and update the tag quality grading sequence. When the improvement value of the quality score after continuous multiple rounds of optimization is less than a preset improvement threshold, the optimization process is ended and the final tag data set is output.

[0211] In a third aspect of the embodiments of the present invention,

[0212] Provided is an electronic device, including:

[0213] A processor;

[0214] A memory for storing instructions executable by the processor;

[0215] Wherein, the processor is configured to call the instructions stored in the memory to execute the method described above.

[0216] In a fourth aspect of the embodiments of the present invention,

[0217] Provided is a computer-readable storage medium, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0218] The present invention may be a method, an apparatus, a system, and / or a computer program product. The computer program product may include a computer-readable storage medium, on which computer-readable program instructions for executing various aspects of the present invention are uploaded.

[0219] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the various embodiments of the present invention.

Claims

1. An intelligent swarm-style quality optimization management method for multi-source tag data, characterized in that Including: Obtain multi-source label data input, and respectively extract the comprehensive feature vectors of each label through a feature extraction module. The comprehensive feature vectors include semantic feature vectors, association feature vectors, and timeliness feature vectors. Input the comprehensive feature vectors into a feature fusion layer to generate a label comprehensive feature matrix; process the label comprehensive feature matrix through a deep feature enhancement network, and calculate and obtain a similarity feature matrix and a dependence feature matrix between labels; Construct a label quality evaluation model based on each feature matrix, calculate the initial quality score of each label, and generate a label quality grading sequence; According to the label quality grading sequence, map the label data to the nodes of the bee colony network, and perform a three-level division of the nodes based on the initial quality score: set the nodes with quality scores higher than the preset threshold as command bees, responsible for formulating optimization strategies; set the nodes with quality scores equal to the preset threshold as follower bees, responsible for transmitting optimization instructions; set the nodes with quality scores lower than the preset threshold as execution bees, responsible for implementing optimization operations; construct a node optimization association network using the similarity feature matrix and the dependence feature matrix, and design a collaborative optimization rule set based on the node optimization association network; Start the optimization process based on the collaborative optimization rule set. The command bee analyzes the feature pattern of the first-quality label to generate an optimization instruction set. The follower bee plans the optimization path through the node optimization association network to generate a task assignment plan. The execution bee implements feature optimization on the second-quality label according to the task assignment plan; After the optimization is completed, input the optimization result back into the feature extraction module for quality evaluation and update the label quality grading sequence; when the improvement value of the quality score after continuous multi-round optimization is less than the preset improvement threshold, end the optimization process and output the final label data set; Construct a deep feature enhancement network, which includes a feature mapping layer, a feature enhancement layer, and a feature reconstruction layer. The feature mapping layer uses a non-linear activation function to map the label comprehensive feature matrix to a high-dimensional latent space. The feature enhancement layer calculates the attention weights between feature vectors and performs weighted fusion based on the attention weights to achieve feature enhancement. The feature reconstruction layer uses a deconvolution operation to remap the enhanced features back to the original feature space; Use the cosine similarity method to calculate the semantic similarity between labels to obtain a similarity feature matrix. At the same time, introduce a time series association analysis method to construct a directed weighted graph through the time series pattern and co-occurrence frequency of label appearances to obtain a dependence feature matrix; Fuse the similarity feature matrix and the dependence feature matrix with the original label comprehensive feature matrix to construct a label quality evaluation model. The label quality evaluation model calculates the feature quality of the label itself, the similarity relationship between labels, and the dependence relationship between labels respectively, and obtains the initial quality score through weighted summation; Perform normalization processing on the initial quality score, and use the density clustering method to automatically determine the quality grade division threshold, and generate a label quality grading sequence containing quality grade information.

2. The method according to claim 1, wherein The semantic similarity between tags is calculated using the cosine similarity method to obtain a similarity feature matrix. At the same time, a temporal correlation analysis method is introduced to construct a directed weighted graph through the temporal pattern and co-occurrence frequency of tag appearances to obtain a dependence feature matrix, including: Obtain a set of tag texts, input the set of tag texts into a pre-trained word vector model to obtain an initial word vector corresponding to each tag text; for tags containing multiple words in the set of tag texts, calculate the term frequency-inverse document frequency value of each word in the tag as a weight coefficient, and perform weighted averaging on the initial word vectors of each word according to the weight coefficient to obtain an overall semantic feature vector of the tag. Based on the overall semantic feature vector, calculate the cosine similarity between any two tags in the set of tag texts, and fill the cosine similarity into the corresponding row and column positions of the similarity feature matrix to obtain the semantic similarity feature matrix of the set of tag texts. The dimension of the semantic similarity feature matrix is the product of the total number of tags and the total number of tags. Obtain the temporal data of the set of tag texts, set a time window, analyze the co-occurrence pattern of the set of tag texts within the time window, count the co-occurrence frequency of tag pairs, record the relative appearance order of tags, and calculate the time interval distribution of tag pairs; based on the co-occurrence frequency, the relative appearance order, and the time interval distribution, perform weighted combination using a preset weight coefficient to obtain the temporal dependence strength of tag pairs. Construct a directed weighted graph according to the temporal dependence strength, map the tags in the set of tag texts to the nodes of the directed weighted graph, determine the connection direction between nodes based on the relative appearance order of tag pairs, and use the temporal dependence strength as the weight value of the corresponding connection to obtain a directed temporal weighted graph describing the temporal association of tags; construct a dependence feature matrix based on the connection relationship and weight value of the directed temporal weighted graph.

3. The method according to claim 1, wherein Fuse the similarity feature matrix and the dependence feature matrix with the original tag comprehensive feature matrix to construct a tag quality evaluation model. The tag quality evaluation model calculates the quality of the features of the tag itself, the similarity relationship between tags, and the dependence relationship between tags respectively, and obtains an initial quality score through weighted summation, including: Perform weighted averaging on the semantic similarity feature and the structural similarity feature in the similarity feature matrix to obtain an overall similarity feature between tags, and perform weighted averaging on the hierarchical dependence feature and the association dependence feature in the dependence feature matrix to obtain an overall dependence feature between tags. Perform normalization processing on the language normality feature, the professional term normality feature, and the semantic integrity feature in the original tag comprehensive feature matrix to obtain a normalized feature, and perform logarithmic transformation processing on the word frequency distribution feature and the query coverage feature to obtain a statistical feature. Concatenate and combine the overall similarity feature between tags, the overall dependence feature between tags, the normalized feature, and the statistical feature to construct a feature fusion matrix, and construct a tag quality evaluation model based on the feature fusion matrix. Calculate the quality score of the label's own features based on the label quality assessment model, where the quality score of the label's own features is obtained by weighted summation of the normalized features and the statistical features; Calculate the similarity relationship score between labels based on the label quality assessment model, where the similarity relationship score between labels is obtained by threshold screening of the comprehensive similarity features between labels; Calculate the label dependency relationship score based on the label quality assessment model, where the label dependency relationship score is obtained by weighted calculation of the comprehensive dependency features between labels; Perform weighted summation on the quality score of the label's own features, the similarity relationship score between labels, and the label dependency relationship score to obtain the initial label quality score.

4. The method according to claim 1, characterized in that, Based on the collaborative optimization rule set, start the optimization process. The instruction bee analyzes the feature pattern of the first quality label to generate an optimization instruction set. The follower bee generates a task assignment plan through the node optimization association network for optimization path planning. The execution bee performs feature optimization on the second quality label according to the task assignment plan, including: Construct a node optimization association network and establish a collaborative optimization rule set. The collaborative optimization rule set includes a feature extraction rule, an optimization instruction generation rule, and a quality optimization rule. Based on the collaborative optimization rule set, start the label optimization process; The instruction bee analyzes the first quality label according to the feature extraction rule, obtains semantic features through word vector representation, obtains structural features through hierarchical relationships, obtains usage features through application frequencies, and performs weighted combination on the semantic features, the structural features, and the usage features to generate label feature data; The instruction bee processes the label feature data based on the optimization instruction generation rule, inputs the label feature data into the feature vector space for clustering analysis, extracts feature patterns to establish a feature pattern library, and generates an optimization instruction set according to the feature pattern library; The follower bee receives the optimization instruction set, calculates the optimization distance and optimization cost between nodes through the node optimization association network, selects the optimal optimization path using a heuristic algorithm, and generates a task assignment plan based on the optimal optimization path and node optimization capabilities; Based on the quality optimization rule and the task assignment plan, the execution bee analyzes the semantic content and word vector features of the second quality label, selects words with similar semantic content from a preset synonym library for supplementation, and simultaneously expands the dimensional information of the word vector features to generate an optimized label.

5. The method according to claim 4, characterized in that The follower bee receives the optimization instruction set, calculates the optimization distance and optimization cost between nodes through the node optimization association network, selects the optimal optimization path using a heuristic algorithm, and generates a task assignment plan based on the optimal optimization path and node optimization capabilities, including: [[ID= Calculate the cosine similarity of the node label word vectors to obtain the semantic distance, calculate the hierarchical span between nodes and the depth of the common ancestor node in the label hierarchical tree to obtain the structural distance, and calculate the co-occurrence frequency and complementarity degree between nodes to obtain the functional distance; Multiply the semantic distance, the structural distance, and the functional distance by their corresponding weights respectively and sum them up to generate a comprehensive optimization distance matrix for the node optimization association network; Obtain the task queue length and resource occupancy of the node to get the node load value, analyze the time resources of the node to get the time constraint value, and calculate the optimization cost based on the comprehensive optimization distance matrix, the node load value, and the time constraint value; Set the initial pheromone distribution according to the priority information of the optimization instruction set, generate the state transition probability by combining the optimization cost and the node optimization ability, and perform path search based on the state transition probability; Update the local pheromone in real time during the path search process, and record the historical optimal path information. When the search iteration times reach the preset threshold, select the optimal optimization path; Decompose the optimization task based on the optimal optimization path and the node processing ability, generate a sub-task set and determine the dependency relationship between sub-tasks, and calculate the resource requirements for each sub-task; Formulate a task scheduling strategy according to the sub-task set, the dependency relationship, and the resource requirements, and generate a task allocation plan including the execution order and parallelism.

6. The method according to claim 5, characterized in that, Calculating the cosine similarity of the node label word vectors to obtain the semantic distance, calculating the hierarchical span between nodes and the depth of the common ancestor node in the label hierarchical tree to obtain the structural distance, and calculating the co-occurrence frequency and complementarity degree between nodes to obtain the functional distance includes: Obtain the node label text, input the node label text into the pre-trained word vector model to generate the word vector corresponding to the node label text; calculate the cosine similarity between the word vectors of node labels to obtain the semantic distance between the node labels; The formula for calculating the semantic distance is as follows: Among them, sim semantic is the semantic similarity, t i , t j are two label nodes, is the corresponding word vector; Obtain the pre-constructed label hierarchical tree, locate the node positions in the label hierarchical tree based on the semantic distance; calculate the hierarchical span between nodes, and the hierarchical span is determined by the path length between nodes; identify the common ancestor nodes between nodes and calculate the depth of the common ancestor nodes; perform weighted combination of the hierarchical span and the depth to obtain the structural distance between node labels; The formula for calculating the structural distance is as follows: where α is the weight coefficient of the structural distance, path(t i , t j ) is the hierarchical span between nodes, maxPath is the maximum path length in the tree, depth(LCA) is the depth of the nearest common ancestor node, and maxDepth is the maximum depth of the tree; Determine the associated node pairs based on the semantic distance and the structural distance, and count the co-occurrence frequency of the associated node pairs; extract the usage scenario information of the associated node pairs and calculate the complementarity degree of node labels in the usage scenario; perform weighted combination of the co-occurrence frequency and the complementarity degree to obtain the functional distance between node labels; The formula for calculating the functional distance is as follows: where t i is the i-th label node to be calculated, t j is the j-th label node to be calculated, β is the weight coefficient of the functional distance, coot(t i , t j ) is the co-occurrence times of the label pair, ∑ k cooc(t k ) is the total co-occurrence times of all label pairs, t k is the k-th pair of label combinations in the label set, is the distribution vector of the label t i in all scenarios, is the distribution vector of the label t j in all scenarios, is the scenario vector of the label, is the Jensen-Shannon divergence of the scenario vector.

7. An intelligent swarm-based quality optimization management system for multi-source tag data, which is used to implement the method described in any one of the foregoing claims 1-6, characterized in that Includes: The first unit is used to obtain multi-source tag data input, and respectively extract the comprehensive feature vectors of each tag through a feature extraction module. The comprehensive feature vectors include semantic feature vectors, association feature vectors, and timeliness feature vectors. The comprehensive feature vectors are input into a feature fusion layer to generate a tag comprehensive feature matrix; the tag comprehensive feature matrix is processed by a deep feature enhancement network, and a similarity feature matrix and a dependency feature matrix between tags are calculated; Based on each feature matrix, a tag quality evaluation model is constructed, the initial quality score of each tag is calculated, and a tag quality grading sequence is generated; The second unit is used to map the tag data into bee colony network nodes according to the tag quality grading sequence, and perform three-level partitioning on the nodes based on the initial quality score: the nodes with quality scores higher than the preset threshold are set as instruction bees, responsible for formulating optimization strategies; the nodes with quality scores equal to the preset threshold are set as follower bees, responsible for transmitting optimization instructions; the nodes with quality scores lower than the preset threshold are set as execution bees, responsible for implementing optimization operations; a node optimization association network is constructed by using the similarity feature matrix and the dependency feature matrix, and a collaborative optimization rule set is designed based on the node optimization association network; The third unit is used to start the optimization process based on the collaborative optimization rule set. The instruction bee analyzes the feature pattern of the first quality tag to generate an optimization instruction set. The follower bee generates a task assignment plan through the node optimization association network for optimization path planning. The execution bee performs feature optimization on the second quality tag according to the task assignment plan; After the optimization is completed, the optimization result is re-input into the feature extraction module for quality evaluation and the tag quality grading sequence is updated; when the improvement value of the quality score after continuous multi-round optimization is less than the preset improvement threshold, the optimization process ends and the final tag data set is output.

8. An electronic device, characterized in that, It includes: A processor; A memory for storing instructions executable by the processor; Wherein, the processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, the method according to any one of claims 1 to 6 is implemented.

Images