A database transmission optimization method and system based on a multi-dimensional label system

Abstract

The application discloses a database transmission optimization method and system based on a multi-dimensional label system, relates to the technical field of database optimization, and comprises the following steps: acquiring and constructing an initial multi-dimensional label set based on multi-source data of scientific research talents; mapping the initial multi-dimensional label set into a talent-label binary matrix, performing sparse modeling, and generating a high-dimensional sparse label matrix; acquiring demand characteristics of a to-be-matched scientific research task, and establishing a task-driven weighted label matrix based on the high-dimensional sparse label matrix; performing dimension reduction and aggregation on the weighted label matrix, generating a low-dimensional dense representation, and constructing a label association network; performing multi-dimensional adaptive matching optimization on the demand of the scientific research task based on the label association network, and generating a task-talent matching result. The application realizes accurate matching of scientific research talents and tasks through task-driven multi-dimensional label modeling and label association network construction, so as to solve the problems of high-dimensional sparsity and unexplainable matching in the prior art.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of database optimization, and more particularly, to a database transmission optimization method and system based on a multi-dimensional label system. BACKGROUND

[0002] With the cross-disciplinary development and complex cooperation mode of scientific research activities, a multi-dimensional label system is gradually introduced into the construction process of a scientific research database to depict the characteristics of scientific research personnel, such as academic achievements, cooperation relationships, research directions, and dynamic development. Through the multi-dimensional label system, fine modeling and task matching of scientific research personnel can be achieved to some extent.

[0003] However, the prior art generally faces the problem of "contradiction between label high-dimensional sparsity and retrieval efficiency" when applying the multi-dimensional label system. On the one hand, with the increase of label dimensions, the effective label distribution of different scientific research personnel under each dimension is often extremely uneven, resulting in high-dimensional sparsity of the database as a whole. On the other hand, during the execution of retrieval and matching, the high-dimensional sparse matrix will cause typical "dimension disaster", which not only increases the computational complexity and storage overhead, but also makes the matching result vulnerable to noise and missing data, thereby leading to the decline of stability and accuracy of talent recommendation.

[0004] In the above disclosed technical solution, there are at least the following technical problems: when the label dimension increases, the effective label distribution of most scientific research personnel under each dimension is uneven, resulting in high-dimensional sparsity of the database, and the retrieval and matching are prone to "dimension disaster", which is low in efficiency and unstable in result.

[0005] In view of the above problems, the present application provides a solution. SUMMARY

[0006] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present application provide a database transmission optimization method and system based on a multi-dimensional label system, which realizes accurate matching of scientific research personnel and tasks through task-driven multi-dimensional label modeling and label association network construction, to solve the problems of high-dimensional sparsity and unexplainable matching in the prior art.

[0007] To achieve the above-mentioned purposes, the present application provides the following technical solutions:

[0008] In one aspect, a scientific research database optimization method based on a multi-dimensional label system includes the following steps: obtaining and constructing an initial multi-dimensional label set based on multi-source data of scientific research talents; mapping the initial multi-dimensional label set into a talent-label binary matrix, and modeling the sparsity of missing values under different label dimensions to generate a high-dimensional sparse label matrix; obtaining the demand characteristics of a to-be-matched scientific research task, and establishing a task-driven weighted label matrix based on the high-dimensional sparse label matrix; performing dimension reduction aggregation on the weighted label matrix to generate a low-dimensional dense representation, and extracting the potential correlation between labels to construct a label correlation network; performing multi-dimensional adaptive matching optimization on the scientific research task demand based on the label correlation network to generate a task-talent matching result.

[0009] In one preferred embodiment, the initial multi-dimensional label set is mapped into a talent-label binary matrix, and the missing values under different label dimensions are modeled for sparsity to generate a high-dimensional sparse label matrix. Specifically, a mapping relationship between scientific research talents and corresponding multi-dimensional labels is established to obtain a talent-label binary matrix. Based on the row and column distribution of the talent-label binary matrix, the distribution pattern of missing labels is extracted, the missing rate under each dimension is calculated, and the missing label set with potential correlation is identified in combination with a semantic similarity matrix. Based on the missing label set, a sparse constraint graph is generated, the existence probability of the missing label is defined, the prior distribution of the missing value is generated by using Bayesian inference or graph regularization method, and a talent-label sparse matrix with probability weight is obtained. The probability sparse matrix is output as a high-dimensional sparse label matrix.

[0010] In one preferred embodiment, the missing label set with potential correlation is identified in combination with a semantic similarity matrix. Specifically, the missing rate of each label dimension in the talent-label binary matrix is calculated to form a missing distribution vector. In the semantic similarity matrix, a label set with a similarity greater than a preset similarity threshold to the high-label with high missing rate is extracted.

[0011] In one preferred embodiment, the demand characteristics of the to-be-matched scientific research task are obtained, and a task-driven weighted label matrix is established based on the high-dimensional sparse label matrix. Specifically, to-be-matched scientific research task data is obtained, the feature extraction of the text description and structured conditions of the scientific research task is performed to obtain a task demand vector. The task-label correlation is calculated based on the task demand vector and the label vector set in the high-dimensional sparse label matrix. The confidence correction coefficient of the label dimension is defined based on the high-dimensional sparse label matrix and the label missing rate. The label weighting coefficient is obtained by multiplying the task-label correlation and the confidence correction coefficient. The diagonal weight matrix is constructed based on the label weighting coefficient, and the high-dimensional sparse label matrix is right-multiplied and weighted to generate the task-driven weighted label matrix.

[0012] In a preferred embodiment, the dimension-reduced aggregation of the weighted label matrix generates a low-dimensional dense representation, specifically: based on task requirements, adaptively aggregate labels with similar semantics or task relevance in the matrix to form a low-dimensional aggregated label set; in the aggregation process, combine the label sparsity feature to suppress high missing rate labels; normalize the aggregated and sparsely constrained matrix to generate a dense vector representation of each talent in the low-dimensional space.

[0013] In a preferred embodiment, the adaptively aggregated low-dimensional label set is formed by calculating the semantic similarity and task relevance between labels based on each column of the task-driven weighted label matrix to generate a label relevance matrix; selecting a label set highly relevant to the task requirements as a candidate aggregation group according to the label relevance matrix, each aggregation group containing labels with similar semantics or task relevance; dynamically adjusting the contribution proportion of each label in aggregation according to the task weight, label missing rate and sparsity constraint; merging the labels in the candidate aggregation group according to the adaptive weight to generate a low-dimensional aggregated label vector, and repeating the process until a complete low-dimensional label set is formed.

[0014] In a preferred embodiment, the potential association between labels is extracted to construct a label association network, specifically: based on the low-dimensional dense representation, calculate the mutual information value of any two labels in the low-dimensional space to obtain an initial label association strength matrix; set a mutual information threshold based on historical data, filter the initial label association strength matrix, and retain the label pairs corresponding to elements greater than the mutual information threshold to form a preliminary label association edge set; combine the original data distribution characteristics of each label dimension in the high-dimensional sparse label matrix to correct the association strength in the preliminary label association edge set using the maximum information coefficient; construct an initial label association network with labels as nodes and the corrected mutual information value as edge weight.

[0015] In a preferred embodiment, the multi-dimensional adaptive matching optimization of the scientific research task requirements based on the label association network generates a task-talent matching result, specifically: perform feature analysis on the scientific research task requirements to extract the core labels and secondary labels required by the task to form a task label set; based on the "overall network-subnetwork module" hierarchical structure of the label association network, match the task label set with the subnetwork modules in the network to identify the target subnetwork module highly associated with the task label set; calculate the association strength matching degree of each label in the task label set and the corresponding label in the target subnetwork module; according to the association strength matching degree, filter out the candidate talents closely associated with the target subnetwork module from the scientific research database; comprehensively score the candidate talents and generate a task-talent matching result according to the comprehensive score ranking.

[0016] In another aspect, a scientific research database optimization system based on a multi-dimensional label system includes the following modules: a multi-source label construction module for obtaining and constructing an initial multi-dimensional label set based on multi-source data of scientific research talents; a sparse matrix generation module for mapping the initial multi-dimensional label set to a talent-label binary matrix and modeling the sparsity of missing values under different label dimensions to generate a high-dimensional sparse label matrix; a task-driven weighting module for obtaining the demand characteristics of a to-be-matched scientific research task and establishing a task-driven weighted label matrix based on the high-dimensional sparse label matrix; a dimension reduction aggregation and correlation extraction module for dimension reduction aggregation of the weighted label matrix to generate a low-dimensional dense representation and extract potential correlation relationships between labels to construct a label correlation network; and a multi-dimensional matching optimization module for performing multi-dimensional adaptive matching optimization of the scientific research task demand based on the label correlation network to generate a task-talent matching result.

[0017] The technical effects and advantages of the database transmission optimization method and system based on the multi-dimensional label system are as follows:

[0018] 1. The present application constructs an initial multi-dimensional label set of scientific research talents, and on this basis, performs semantic unification, missing value sparsity modeling and task-driven weighting correction, effectively solving the dimension disaster and information loss problem caused by high-dimensional sparse matrix in the prior art, making the scientific research talent feature representation more complete and robust, and providing a high-quality input basis for subsequent matching optimization.

[0019] 2. The present application introduces a task demand driven adaptive aggregation mechanism in the dimension reduction aggregation process of the weighted label matrix, and further combines the hierarchical structure of the label correlation network to perform multi-dimensional adaptive matching optimization of the task, not only improving the accuracy and robustness of the task-talent matching, but also revealing potential comprehensive relationships such as interdisciplinary ability, cooperation stability and scientific research influence, achieving the explainability and transparency of the matching result, and overcoming the deficiency of "unexplainable results" in the prior art. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 The figure is a flowchart of the scientific research database optimization method based on the multi-dimensional label system of the present application.

[0021] Figure 2 The figure is a structural diagram of the scientific research database optimization system based on the multi-dimensional label system of the present application. DETAILED DESCRIPTION

[0022] With reference to the accompanying drawings, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all the other embodiments obtained by a person of ordinary skill in the art without creative effort are within the protection scope of the present application.

[0023] Embodiment 1, Figure 1 The present application provides a scientific research database optimization method based on a multi-dimensional label system, comprising the following steps:

[0024] S1, obtaining and constructing an initial multi-dimensional label set based on multi-source data of scientific research talents;

[0025] In this embodiment, the multi-source data comprises academic achievement data, cooperation network data, scientific research direction data and dynamic development data;

[0026] The obtaining and constructing an initial multi-dimensional label set based on multi-source data of scientific research talents specifically comprises:

[0027] Obtaining multi-source data of scientific research talents, the multi-source data comprising academic achievement data, cooperation network data, scientific research direction data and dynamic development data;

[0028] Structurally analyzing the academic achievement data to extract the number of published papers, the frequency of cited papers, the type of representative achievements and the influence index of scientific research output;

[0029] Modeling the relationship of the cooperation network data to extract the cooperation frequency, the cooperation stability, the breadth of cross-institutional cooperation and the cooperation centrality index;

[0030] Performing theme clustering on the scientific research direction data to extract the discipline attribution, the research hotspot and the cross-disciplinary correlation degree;

[0031] Performing time series analysis on the dynamic development data to extract the scientific research direction migration trend, the participation degree in emerging fields and the change of scientific research activity;

[0032] Based on the above-mentioned extracted multi-source features, the label semantics is unified and an initial multi-dimensional label set is constructed.

[0033] The steps of unifying the label semantics are as follows:

[0034] Based on the semantic similarity calculation of the same or similar labels in the multi-source data, a label similarity matrix is generated;

[0035] Performing aggregation operation on the label similarity matrix, and merging the labels with a semantic similarity greater than a preset similarity threshold;

[0036] For the semantic fuzzy or ambiguous labels, the disambiguation processing is performed based on the context relationship and the demand of the scientific research task.

[0037] On the basis of the merging and disambiguation, a unified label semantic framework is generated and serves as a benchmark for subsequent label expansion and dynamic optimization.

[0038] The semantic similarity calculation is specifically:

[0039] Supposing that the label set is , the vector representation is generated by a semantic embedding model ;

[0040]

[0041] wherein, is the semantic similarity, is the transpose of the vector representation of the label , is the vector representation of the label , and is the modulus of the vector.

[0042] S2, the initial multi-dimensional label set is mapped into a talent-label binary matrix, and the missing values under different label dimensions are modeled for sparsity, to generate a high-dimensional sparse label matrix, specifically:

[0043] A mapping relationship between the scientific research talents and the corresponding multi-dimensional labels is established, to obtain a talent-label binary matrix M, wherein represents that the talent i has the label j, represents that the label is missing;

[0044] Based on the row and column distribution of the talent-label binary matrix M, the distribution mode of the missing labels is extracted, the missing rate under each dimension is calculated, and the missing label set with potential correlation is identified in combination with the semantic similarity matrix S;

[0045] Based on the missing label set, a sparse constraint graph is generated, wherein the node set V represents the talents and the labels, and the edge set E represents the potential completion paths identified through the semantic similarity; the sparse constraint graph simultaneously retains the missing label positions and their adjacency relationships;

[0046] The existence probability of the missing labels is defined on the sparse constraint graph, the prior distribution of the missing values is generated by using the Bayesian inference or the graph regularization method, to obtain a talent-label sparse matrix with a probability weight; wherein the missing label is represented by a probability value in the interval [0, 1];

[0047] The probability sparse matrix is output as the high-dimensional sparse label matrix. ​

[0048] The combined semantic similarity matrix S identifies a set of missing labels with potential associations, specifically:

[0049] Calculate the missing rate of each label dimension in the talent-label binary matrix to form a missing distribution vector , wherein represents the missing rate of label j;

[0050] In the semantic similarity matrix , extract a set of labels with high missing rate with a similarity greater than a preset similarity threshold;

[0051] Based on the missing distribution vector and the set of labels with a similarity threshold, generate a pair of potentially associated labels , and calculate the association weight of the pair of potentially associated labels;

[0052] The pair of potentially associated labels and their weights are used as input for subsequent sparse constraint graph construction.

[0053] The missing rate is specifically:

[0054]

[0055] wherein is the missing rate of label j, is the value of the i-th scientific researcher in the talent-label binary matrix in the j-th label dimension, is the total number of scientific researchers, is an indicator function, taking the value 1 when , otherwise taking the value 0.

[0056] The association weight is specifically:

[0057]

[0058] wherein is the association weight, is the missing constraint, is the label semantic similarity.

[0059] S3, obtain the demand characteristics of the scientific task to be matched, and based on the high-dimensional sparse label matrix, establish a task-driven weighted label matrix;

[0060] In this embodiment, the demand characteristics of the scientific task to be matched are obtained, and based on the high-dimensional sparse label matrix, a task-driven weighted label matrix is established, specifically:

[0061] Obtain scientific task data to be matched, extract features from the text description and structured conditions of the scientific task, and obtain a task demand vector;

[0062] calculating task-label relevance based on the task demand vector and the set of label vectors in the high-dimensional sparse label matrix;

[0063] defining a confidence correction coefficient of label dimension based on the high-dimensional sparse label matrix and the label missing rate;

[0064] multiplying the task-label relevance and the confidence correction coefficient to obtain a label weighting coefficient;

[0065] constructing a diagonal weight matrix based on the label weighting coefficient, and right multiplying the high-dimensional sparse label matrix to generate a task-driven weighted label matrix;

[0066] The confidence correction coefficient is specifically:

[0067]

[0068] wherein, the confidence correction coefficient is, the missing rate of label j, a preset hyperparameter, the time interval during which label j is last supported by evidence, a front label priority set based on historical experience.

[0069] The task-driven weighted label matrix is specifically:

[0070]

[0071] wherein, the task-driven weighted label matrix is, the high-dimensional sparse label matrix, a diagonal matrix composed of label weights, the task-label relevance.

[0072] S4, dimension reduction aggregation is performed on the weighted label matrix to generate a low-dimensional dense representation, and potential correlation between labels is extracted to construct a label correlation network;

[0073] In the embodiment, the dimension reduction aggregation is performed on the weighted label matrix to generate a low-dimensional dense representation, and the dimension reduction aggregation is specifically:

[0074] The task-driven weighted label matrix is taken as an input matrix, wherein each row of the matrix represents a scientific research talent, each column represents a label weight, and the matrix element contains task-driven label weight information;

[0075] Based on the task demand, labels with similar semantics or task relevance in the matrix are adaptively aggregated to form an aggregated low-dimensional label set;

[0076] Incorporate the label sparsity feature during the aggregation process, and weight the high missing rate labels to prevent low-dimensional representation bias caused by the loss of sparse information;

[0077] The matrix adjusted by aggregation and sparsity constraint is normalized to generate a dense vector representation of each talent in the low-dimensional space.

[0078] The adaptive aggregation of the matrix based on the task demand is performed on the labels with similar semantics or task correlation to form an aggregated low-dimensional label set, specifically:

[0079] Based on each column (label vector) of the task-driven weighted label matrix, the semantic similarity and task correlation between labels are calculated to generate a label correlation matrix;

[0080] According to the label correlation matrix, a label set highly related to the task demand is selected as the candidate aggregation group, and each aggregation group contains labels with similar semantics or task correlation;

[0081] For the labels in the candidate aggregation group, the contribution proportion of each label in the aggregation is dynamically adjusted according to the task weight, label missing rate and sparsity constraint, to ensure that the key task labels have the maximum influence in the low-dimensional representation;

[0082] The labels in the candidate aggregation group are merged according to the adaptive weight to generate an aggregated low-dimensional label vector, and the process is repeated until a complete low-dimensional label set is formed.

[0083] In this embodiment, the contribution proportion of each label in the aggregation is dynamically adjusted according to the task weight, label missing rate and sparsity constraint, specifically:

[0084] The task correlation vector of each label in the candidate aggregation group is taken as the initial contribution proportion, reflecting the preliminary importance of each label to the task matching;

[0085] According to the missing rate or sparsity feature of each label in the high-dimensional weighted label matrix, the initial contribution proportion is corrected to reduce the influence of labels with high missing rate or high sparsity in the aggregation, and the key label weight is highlighted;

[0086] The priority or importance of the current scientific research task is taken as the task weight factor, and the contribution proportion of each label is dynamically adjusted according to the task demand, so that the labels highly related to the current task increase in proportion in the aggregation, and the proportion of low correlation labels decreases;

[0087] The dynamically adjusted label contribution proportion vector is normalized to keep the aggregation total weight uniform, ensuring the numerical stability and comparability of the aggregated low-dimensional label vector.

[0088] Potential correlation between the extracted labels, constructing a label correlation network, specifically:

[0089] Based on the low-dimensional dense representation, the mutual information value of any two labels in the low-dimensional space is calculated, and an initial label correlation strength matrix is obtained, and the matrix element value reflects the dependence relationship of two labels at the probability distribution level;

[0090] Based on historical data, set the mutual information threshold, filter the initial label correlation strength matrix, and keep the label pairs corresponding to the elements greater than the mutual information threshold to form a preliminary label correlation edge set, which is determined based on the probability correlation characteristics contained in the low-dimensional dense representation;

[0091] Combined with the original data distribution characteristics of each label dimension in the high-dimensional sparse label matrix, the maximum information coefficient is used to correct the correlation strength in the preliminary label correlation edge set, and the correction process takes into account linear and nonlinear correlation characteristics;

[0092] Taking the label as the node and the corrected mutual information value as the edge weight, an initial label correlation network is constructed, and the network structure reflects the probability correlation of the label extracted by the low-dimensional dense representation and the corrected relationship of the original data nonlinear correlation;

[0093] Based on the community discovery algorithm, the initial label correlation network is clustered, and the label communities with strong probability correlation are identified, and these communities are used as subnetwork modules of the label correlation network, so that the label correlation network forms a hierarchical structure of "overall network-subnetwork module". The labels in each community have high nonlinear synergy in scientific research attributes, and provide probability correlation basis for subsequent multi-dimensional adaptive matching optimization.

[0094] The mutual information value, specifically:

[0095]

[0096] The specific calculation formula of the maximum information coefficient is:

[0097]

[0098] Among them, is the mutual information value, , are the marginal probability distributions of X and Y, is the joint probability distribution of X and Y, is the maximum information coefficient, is the discretized mutual information, , are the discretization interval numbers of X and Y, is the total number of samples, is a constant (usually 0.6).

[0099] S5, performing multi-dimensional adaptive matching optimization on the scientific research task demand based on the label association network to generate a task-talent matching result.

[0100] In this embodiment, the label association network is used to perform multi-dimensional adaptive matching optimization on the scientific research task demand to generate a task-talent matching result, specifically as follows:

[0101] The features of the scientific research task demand are analyzed, and the core labels and secondary labels required by the task are extracted to form a task label set. The core labels correspond to the key scientific research attributes of the task, and the secondary labels are auxiliary scientific research attributes.

[0102] Based on the "overall network-subnetwork module" hierarchical structure of the label association network, the task label set is matched with the subnetwork modules in the network to identify target subnetwork modules that are highly associated with the task label set. The label synergy in the target subnetwork module provides the basis for matching.

[0103] The association strength matching degree of each label in the task label set and the corresponding label in the target subnetwork module is calculated. The association strength matching degree is determined based on the edge weight in the network and the probability association characteristics between labels, and the core labels are given higher weights in combination with the task demand.

[0104] According to the association strength matching degree, candidate talents closely associated with the target subnetwork module are selected from the scientific research database. The multi-dimensional labels of the candidate talents need to be adaptively matched with the task label set in terms of probability distribution and nonlinear association.

[0105] The candidate talents are given a comprehensive score, and the scoring indicators include core label matching degree, secondary label matching degree, and label community synergy adaptation degree. The label community synergy adaptation degree refers to the label synergy characteristics in the target subnetwork module.

[0106] According to the comprehensive score, a task-talent matching result is generated. The result preferentially outputs the best adaptive talent with the highest score, along with the matching details in each dimension.

[0107] The association strength matching degree is specifically as follows:

[0108]

[0109] The comprehensive score is specifically as follows:

[0110]

[0111] wherein, is the association strength matching degree of the ith label, is a preset weight coefficient (the weight of the core label is greater than that of the secondary label, and the weight of the core label is ​ the sum of the secondary labels the ratio of the sum of the secondary labels is set according to the importance of the task requirements, is the edge weight of the label in the target sub-network module and the corresponding task label, is the probability correlation value between labels (calculated by mutual information), is the comprehensive score, , , are respectively the preset weights of the core label matching degree, the secondary label matching degree, and the label community synergy adaptation degree, is the core label matching degree, is the secondary label matching degree, is the label community synergy adaptation degree.

[0112] Embodiment 2, Figure 2 The application provides a scientific research database optimization system based on a multi-dimensional label system, which comprises the following modules:

[0113] A multi-source label construction module is configured to obtain and construct an initial multi-dimensional label set based on multi-source data of scientific research talents;

[0114] A sparse matrix generation module is configured to map the initial multi-dimensional label set into a talent-label binary matrix, and model the missing values in different label dimensions for sparse modeling, thereby generating a high-dimensional sparse label matrix;

[0115] A task-driven weighting module is configured to obtain the demand characteristics of a scientific research task to be matched, and establish a task-driven weighted label matrix based on the high-dimensional sparse label matrix;

[0116] A dimension reduction aggregation and correlation extraction module is configured to perform dimension reduction aggregation on the weighted label matrix to generate a low-dimensional dense representation, and extract the potential correlation between labels to construct a label correlation network;

[0117] A multi-dimensional matching optimization module is configured to perform multi-dimensional adaptive matching optimization on the scientific research task demand based on the label correlation network, thereby generating a task-talent matching result.

[0118] The above formulas are all dimensionless values, and the formulas are obtained by collecting a large amount of data to simulate a formula closest to the real situation, and the preset parameters in the formula are set by a person skilled in the art according to the actual situation.

[0119] The above embodiments can be realized by software, hardware, firmware or any combination thereof, in whole or in part. When realized by software, the above embodiments can be realized in the form of a computer program product in whole or in part.

[0120] Those skilled in the art can understand that the modules and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized in electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0121] In addition, each functional module in each embodiment of the present application can be integrated in one processing module, or each module can exist physically alone, or two or more modules can be integrated in one module.

[0122] The above is merely specific embodiments of the present application, but the protection scope of the present application is not limited thereto, any skilled in the art within the technical scope disclosed in the present application can easily think of changes or replacements, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0123] Finally: the above is only the preferred embodiment of the present application, and is not used to limit the present application, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for database transfer optimization based on a multi-dimensional tag system, characterized in that, Includes the following steps: Acquire and construct an initial multidimensional label set based on multi-source data of scientific research talents; The initial multidimensional label set is mapped to a talent-label binary matrix, and the sparsity of missing values ​​under different label dimensions is modeled to generate a high-dimensional sparse label matrix. To obtain the requirements features of the research tasks to be matched, and to establish a task-driven weighted label matrix based on a high-dimensional sparse label matrix, the following steps are taken: obtain the research task data to be matched, extract features from the text description and structured conditions of the research tasks to obtain the task requirement vector; calculate the task-label correlation based on the task requirement vector and the set of label vectors in the high-dimensional sparse label matrix. The confidence correction coefficient for the label dimension is defined based on the high-dimensional sparse label matrix and the label missing rate; The label weighting coefficient is obtained by multiplying the task-label correlation with the confidence correction coefficient; a diagonal weight matrix is ​​constructed based on the label weighting coefficient, and the high-dimensional sparse label matrix is ​​right-multiplied and weighted to generate a task-driven weighted label matrix. The weighted label matrix is ​​reduced in dimension and aggregated to generate a low-dimensional dense representation. The potential relationships between labels are extracted to construct a label association network. Based on the label association network, multi-dimensional adaptation and matching optimization is performed on the research task requirements to generate task-talent matching results.

2. The database transmission optimization method based on multi-dimensional label system according to claim 1, characterized in that, The process involves mapping the initial multidimensional label set to a talent-label binary matrix and performing sparsity modeling on missing values ​​under different label dimensions to generate a high-dimensional sparse label matrix. Specifically: By establishing a mapping relationship between scientific research talents and their corresponding multidimensional labels, a talent-label binary matrix is ​​obtained. Based on the row and column distribution of the talent-tag binary matrix, the distribution pattern of missing tags is extracted, the missing rate under each dimension is calculated, and the set of missing tags with potential associations is identified by combining the semantic similarity matrix. Based on the set of missing labels, a sparse constraint graph is generated, the existence probability of missing labels is defined, and a prior distribution of missing values ​​is generated using Bayesian inference or graph regularization methods to obtain a talent-label sparse matrix with probability weights. The probability sparse matrix is ​​output as a high-dimensional sparse label matrix.

3. The method of claim 2, wherein, The method of identifying a set of missing labels with potential associations by combining semantic similarity matrices is as follows: Calculate the missing rate for each label dimension in the talent-label binary matrix to form a missing distribution vector; In the semantic similarity matrix, extract the set of tags whose similarity to tags with high missing rates is greater than a preset similarity threshold.

4. The database transmission optimization method based on multi-dimensional label system according to claim 3, characterized in that, The step of dimensionality reduction and aggregation of the weighted label matrix to generate a low-dimensional dense representation specifically involves: Based on task requirements, the semantically similar or task-related tags in the matrix are adaptively aggregated to form a low-dimensional tag set after aggregation. During the aggregation process, the sparsity feature of the labels is combined to perform weighted suppression on labels with high missing rates; The aggregated and sparsely constrained matrices are normalized to generate a dense vector representation of each talent in a low-dimensional space.

5. The database transmission optimization method based on a multi-dimensional tag system according to claim 4, characterized in that, The adaptive aggregation of semantically similar or task-related tags in the matrix based on task requirements to form a low-dimensional tag set after aggregation is specifically as follows: Based on each column of the task-driven weighted label matrix, the semantic similarity and task relevance between labels are calculated to generate a label relevance matrix. Based on the tag relevance matrix, a set of tags that are highly relevant to the task requirements is selected as a candidate aggregation group. Each aggregation group contains tags that are semantically similar or task-related. For labels within candidate aggregation groups, the contribution ratio of each label in the aggregation is dynamically adjusted based on task weight, label missing rate, and sparsity constraints. The labels within the candidate aggregation group are merged according to adaptive weights to generate an aggregated low-dimensional label vector, and this process is repeated until a complete low-dimensional label set is formed.

6. The database transmission optimization method based on a multi-dimensional tag system according to claim 5, characterized in that, The extraction of potential relationships between tags and the construction of a tag association network specifically involves: Based on the low-dimensional dense representation, the mutual information value of any two labels in the low-dimensional space is calculated to obtain the initial label association strength matrix; Based on historical data, a mutual information threshold is set, and the initial label association strength matrix is ​​filtered to retain the label pairs corresponding to elements with mutual information greater than the mutual information threshold, thus forming a preliminary label association edge set. Combining the original data distribution characteristics of each label dimension in the high-dimensional sparse label matrix, the maximum information coefficient is used to correct the association strength of the initial label association edge set; An initial label association network is constructed using labels as nodes and modified mutual information values ​​as edge weights.

7. The database transmission optimization method based on a multi-dimensional tag system according to claim 6, characterized in that, The label association network-based multidimensional adaptation and matching optimization of scientific research task requirements generates task-talent matching results, specifically as follows: The characteristics of scientific research task requirements are analyzed to extract the core and secondary tags required for the task, forming a task tag set; Based on the "overall network-sub-network module" hierarchical structure of the label association network, the task label set is matched with the sub-network modules in the network to identify the target sub-network modules that are highly associated with the task label set. Calculate the association strength matching degree between each label in the task label set and the corresponding label in the target sub-network module; Based on the correlation strength matching degree, candidates closely related to the target sub-network module are screened from the scientific research database; Candidates are comprehensively evaluated and ranked according to their comprehensive scores to generate task-talent matching results.

8. A system using a scientific research database optimization method based on a multi-dimensional tagging system as described in any one of claims 1-7, characterized in that, Includes the following modules: Multi-source label building module: used to acquire and build an initial multi-dimensional label set based on multi-source data of scientific research talents; Sparse matrix generation module: used to map the initial multidimensional label set into a talent-label binary matrix, and to model the sparsity of missing values ​​under different label dimensions to generate a high-dimensional sparse label matrix; Task-driven weighted module: used to obtain the requirements features of the scientific research tasks to be matched, and to establish a task-driven weighted label matrix based on a high-dimensional sparse label matrix; Dimensionality Reduction Aggregation and Association Extraction Module: This module is used to perform dimensionality reduction aggregation on the weighted label matrix, generate a low-dimensional dense representation, extract potential associations between labels, and construct a label association network. Multidimensional Matching Optimization Module: Used to perform multidimensional adaptation matching optimization based on label association network for scientific research task requirements, and generate task-talent matching results.

Images