A multi-caliber data management method, device and equipment based on knowledge graph

Abstract

The present invention provides a multi-caliber data management method, device and equipment based on knowledge graph, which relates to the field of data management technology, including obtaining historical knowledge graph; obtaining multi-caliber target data; obtaining a data screening model; using the entity corresponding to the target data as the target core node; obtaining an initial knowledge graph; optimizing the initial knowledge graph to obtain a target knowledge graph of multi-caliber data. The present invention learns the relationship between the core node and the leaf node in the knowledge graph, and the relationship between the entity corresponding to the core node and the entity corresponding to the leaf node by constructing a decision tree model to predict the entity that should be used as the core node in the current multi-caliber data, thereby determining the knowledge graph model corresponding to the current multi-caliber data, thereby generating a knowledge graph. Compared with the prior art, the knowledge graph model corresponding to the multi-caliber data can be accurately and efficiently determined, and the complex data relationship can be displayed in a more intuitive graphical manner.

Description

Technical Field

[0001] The present invention relates to the field of data management technology, and in particular to a multi-caliber data management method, device and equipment based on a knowledge graph. Background Art

[0002] Multi-caliber data refers to multiple data sets from different sources and in different formats or perspectives. These data may involve the same entity or phenomenon, but due to differences in their sources, collection methods, statistical calibers, time spans or other factors, there are differences and diversity between the data.

[0003] In order to better manage and utilize multi-caliber data, knowledge graphs are usually used to effectively integrate multi-caliber data and clearly express the relationship between different data points through knowledge graphs. Considering that the node types, edge types, and the relationship between nodes and edges in knowledge graphs in different fields are different, it is necessary to accurately determine the corresponding knowledge graph model based on multi-caliber data and bring multi-caliber data into the model to generate a knowledge graph. Considering that multi-caliber data has many types and large differences in data formats, it is difficult to accurately determine the corresponding knowledge graph model for multi-caliber data. Summary of the invention

[0004] The purpose of the present invention is to provide a multi-caliber data management method, device and equipment based on knowledge graph to improve the above problems. In order to achieve the above purpose, the technical solution adopted by the present invention is as follows:

[0005] In a first aspect, the present application provides a multi-caliber data management method based on a knowledge graph, comprising:

[0006] Obtain a historical knowledge graph, wherein the historical knowledge graph includes a core node and multiple leaf nodes, and entities corresponding to the core node and the leaf nodes each carry multiple parameter information, wherein the parameter information is respectively type, source, format, and update frequency;

[0007] Acquire multi-caliber data and perform preprocessing to obtain multi-caliber target data, wherein each data in the multi-caliber target data carries parameter information;

[0008] Based on the core nodes and leaf nodes in the historical knowledge graph, a preset decision tree model is trained, and when the preset model indicators meet the set conditions, the training is stopped to obtain a data screening model;

[0009] Inputting the multi-caliber target data into the data screening model, the data screening model determines the target data from the multi-caliber target data, and uses the entity corresponding to the target data as the target core node;

[0010] Determine a target model from a plurality of preset knowledge graph models based on the target core node, and import the multi-caliber target data into the target model to obtain an initial knowledge graph;

[0011] The initial knowledge graph is optimized to obtain a target knowledge graph of the multi-caliber data.

[0012] In the second aspect, the present application also provides a multi-caliber data management device based on a knowledge graph, comprising:

[0013] A first acquisition unit is used to acquire a historical knowledge graph, wherein the historical knowledge graph includes a core node and multiple leaf nodes, and entities corresponding to the core node and the leaf nodes each carry multiple parameter information, wherein the parameter information is respectively type, source, format, and update frequency;

[0014] A second acquisition unit is used to acquire multi-caliber data and perform preprocessing to obtain multi-caliber target data, wherein each data in the multi-caliber target data carries parameter information;

[0015] A training unit, used to train a preset decision tree model based on the core nodes and leaf nodes in the historical knowledge graph, and when the preset model indicators meet the set conditions, stop the training to obtain a data screening model;

[0016] A first input unit is used to input the multi-caliber target data into the data screening model, the data screening model determines the target data from the multi-caliber target data, and uses the entity corresponding to the target data as the target core node;

[0017] A first determination unit is used to determine a target model from a plurality of preset knowledge graph models based on the target core node, and import the multi-caliber target data into the target model to obtain an initial knowledge graph;

[0018] An optimization unit is used to optimize the initial knowledge graph to obtain a target knowledge graph of the multi-caliber data.

[0019] In a third aspect, the present application also provides a multi-caliber data management device based on a knowledge graph, including:

[0020] Memory for storing computer programs;

[0021] A processor is used to implement the steps of the multi-caliber data management method based on the knowledge graph when executing the computer program.

[0022] In a fourth aspect, the present application also provides a readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above-mentioned multi-caliber data management method based on the knowledge graph are implemented.

[0023] The beneficial effects of the present invention are:

[0024] The present invention constructs a decision tree model to learn the relationship between core nodes and leaf nodes in the knowledge graph, as well as the relationship between entities corresponding to core nodes and entities corresponding to leaf nodes, to predict the entities that should be core nodes in the current multi-caliber data, thereby determining the knowledge graph model corresponding to the current multi-caliber data, thereby generating a knowledge graph. Compared with the existing technology, it can accurately and efficiently determine the knowledge graph model corresponding to multi-caliber data, and display complex data relationships in a more intuitive graphical manner.

[0025] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or be understood by implementing the embodiments of the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description, claims, and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0027] Figure 1 It is a schematic diagram of the process of the multi-caliber data management method based on the knowledge graph described in an embodiment of the present invention;

[0028] Figure 2 Schematic diagram of the target knowledge graph structure described in an embodiment of the present invention;

[0029] Figure 3 A schematic diagram of a portion of the knowledge graph corresponding to type 1 described in an embodiment of the present invention;

[0030] Figure 4 This is a schematic diagram of a portion of the knowledge graph corresponding to type 2 described in an embodiment of the present invention;

[0031] Figure 5 It is a schematic diagram of the structure of a multi-caliber data management device based on a knowledge graph according to an embodiment of the present invention;

[0032] Figure 6Schematic diagram of the structure of a multi-caliber data management device based on a knowledge graph described in an embodiment of the present invention.

[0033] Markings in the figure:

[0034] 10. First acquisition unit; 20. Second acquisition unit; 30. Training unit; 40. First input unit; 50. First determination unit; 60. Optimization unit; 800. Multi-caliber data management device based on knowledge graph; 801. Processor; 802. Memory; 803. Multimedia component; 804. I / O interface; 805. Communication component. DETAILED DESCRIPTION

[0035] In order to make the purpose, 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 in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0036] It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings. At the same time, in the description of the present invention, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0037] Embodiment 1:

[0038] This embodiment provides a multi-caliber data management method based on a knowledge graph.

[0039] See also Figure 1 , the figure shows that the method includes step S10, step S20, step S30, step S40, step S50 and step S60.

[0040] Step S10. Obtain a historical knowledge graph, the historical knowledge graph includes a core node and multiple leaf nodes, the entities corresponding to the core node and the leaf nodes carry multiple parameter information, and the parameter information is respectively type, source, format and update frequency;

[0041] Specifically, knowledge graphs can be divided into multiple types according to their application fields, structures, and data sources. For example, medical knowledge graphs, financial knowledge graphs, corporate knowledge graphs, social network knowledge graphs, event tracking knowledge graphs, and geographic knowledge graphs, etc. The relationship between the core nodes and leaf nodes in different types of knowledge graphs is not the same.

[0042] Step S20. Acquire multi-aperture data and perform preprocessing to obtain multi-aperture target data, each of which carries parameter information;

[0043] Specifically, multi-caliber data includes structured data, unstructured data and semi-structured data. Among them, structured data can be directly used as the source of nodes and relationships, while unstructured data and semi-structured data need to undergo some cleaning processing before they can be directly applied as entities to the knowledge graph.

[0044] Specifically, step S20 specifically includes step S21, step S22, step S23, step S24, step S25, step S26, step S27, step S28, step S29, step S210 and step S211:

[0045] Step S21. Perform cleaning processing on the multi-caliber data to obtain a plurality of first data, wherein the cleaning processing includes removing noise, eliminating duplication and processing missing values;

[0046] Step S22. Perform format normalization processing on the plurality of first data to obtain a plurality of second data;

[0047] Step S23. Filtering the semi-structured data and the unstructured data from the second data as a plurality of third data;

[0048] Step S24. Performing division processing based on the type of the third data to obtain a plurality of division units;

[0049] Specifically, considering that both semi-structured data and unstructured data contain a large amount of useless information, it is necessary to perform word segmentation screening to determine relevant information and delete useless information.

[0050] Step S25. Calculate the occurrence frequency of each of the multiple division units to obtain frequency information;

[0051] Specifically, the occurrence frequency of relevant information is usually high, and a higher weight can be set for the division unit with a higher occurrence frequency.

[0052] Step S26. Determine the location of each division unit in the plurality of division units and obtain location information;

[0053] Specifically, the probability of relevant information appearing in a separate paragraph or at the end of a paragraph is much higher than that in other parts, and a higher weight can be set for the division units appearing in a separate paragraph or at the end of a paragraph.

[0054] Step S27. Determine the property of each of the multiple division units to obtain property information;

[0055] Specifically, the property information is the part of speech that characterizes the division unit, which includes nouns, verbs, adjectives, and adverbs. Statistics show that entities in the knowledge graph are usually nouns, and a few are verbs, adjectives, adverbs, and other parts of speech. Therefore, when setting weights, a higher weight is set for the division unit that is a noun.

[0056] Step S28. Perform weighted calculation based on frequency information, location information and property information to obtain weight information of each division unit;

[0057] Step S29. Screening the multiple division units based on the weight information to obtain multiple screened target screening units;

[0058] Step S210. Processing data based on multiple target screening units;

[0059] Step S211. Composing multi-caliber target data based on the structured data and processed data in the second data;

[0060] Specifically, weighted calculation is performed according to the weight information corresponding to the frequency information, position information and property information of each division unit to obtain the weight information of each division unit. Multiple target division units can be screened out through the weight information, and the screened target division units are added as processing data to the multi-caliber target data for subsequent calculations.

[0061] Step S30. Train the preset decision tree model based on the core nodes and leaf nodes in the historical knowledge graph. When the preset model indicators meet the set conditions, stop the training and obtain the data screening model;

[0062] Specifically, the present application adopts a decision tree (LightGBM) model. The advantage of this model is that it uses a leaf splitting algorithm as a tree growth strategy. This strategy finds a leaf with the largest splitting gain from all current leaves each time, then splits and repeats the cycle. The leaf splitting algorithm produces smaller errors than the horizontal splitting algorithm and has higher positioning accuracy. The model also uses a histogram algorithm. When constructing the tree, the histogram information is used to calculate the splitting gain and quickly select the optimal splitting point, effectively reducing the amount of calculation. The model also uses a gradient-based single-sided sampling algorithm. The single-sided sampling algorithm excludes most of the samples with small gradients and only uses the remaining samples to calculate the information gain. It can reduce the amount of data while ensuring the model positioning accuracy, effectively reducing the workload of calculation and improving the model calculation speed.

[0063] In this application, the grid search method can be used to adjust the three important parameter values ​​in the decision tree model. By setting the value range of these three parameters, the grid search will search for the optimal parameters of the model within the selected interval. The final grid search results will select the values ​​of the three parameters corresponding to the largest evaluation index value in the selected interval, thereby obtaining a data screening model based on the values ​​of these three parameters.

[0064] Specifically, step S30 specifically includes step S31, step S32, step S33, step S34, step S35, step S36, step S37, step S38, step S39, step S310, step S311 and step S312:

[0065] Step S31. randomly divide multiple leaf nodes into three parts, respectively as training set, test set and validation set, and add core nodes to the training set, test set and validation set;

[0066] Step S32. Obtain the value range of each model parameter;

[0067] Specifically, the model parameters considered in this application are learning rate, maximum depth and number of iterations. Among them, the learning rate determines the amplitude of the model parameter update in each iteration. A larger learning rate will cause the model parameters to be greatly updated in each iteration, while a smaller learning rate will cause the parameter update amplitude to decrease; the maximum depth can be used to limit the depth of each tree to prevent the model from over-learning the details of the training data, thereby reducing the risk of overfitting. Limiting the depth of the tree can reduce the complexity of the model and improve the generalization ability of the model, which is especially important when processing high-dimensional data. Deeper trees usually pay more attention to capturing the details of the training data, which may lead to better fitting effects on the training set, but also increase the model's sensitivity to noise; an increase in the number of iterations usually improves the performance of the model, but increasing the number of iterations may also cause the model to perform well on the training set but increase the risk of overfitting on the test set. At the same time, a larger number of iterations will also make the training time of the model longer and the model training efficiency lower.

[0068] Step S33. First generation operation: randomly generating a plurality of initial model parameters based on the initial range of each model parameter;

[0069] Step S34. Selection operation: randomly selecting an initial model parameter from multiple initial model parameters of each model parameter for combination to obtain multiple parameter combinations, wherein at most two initial model parameters in any two parameter combinations have the same value;

[0070] Step S35. Adjustment operation: adjusting the decision tree model based on each parameter combination to obtain multiple initial screening models;

[0071] Step S36: Bring in the training set to the initial screening model for training, and calculate the model index to obtain multiple initial model indexes;

[0072] Specifically, in the present application, by adjusting the three model parameters of the decision tree, multiple different initial screening models will be obtained. It is necessary to calculate the model indicators corresponding to each initial screening model to determine the initial screening model with better performance as the basis for subsequent screening.

[0073] Specifically, step S36 specifically includes step S361, step S362, step S363, step S364, step S365, step S366, step S367 and step S368:

[0074] Step S361. Obtain initial weight ranges corresponding to a plurality of parameter information;

[0075] Specifically, the parameter information in this application includes type, source, format and update frequency. Each parameter information corresponds to multiple categories in different scenarios. For example, in the company business scenario, the type may include financial data, human data, sales data, and customer data, etc.; the source may include internal systems, customer feedback, and public data, etc.; the format may include structured data, semi-structured data, and streaming data, etc.; the update frequency may include real-time update, weekly update, and monthly update, etc. There are differences in the parameter information corresponding to different scenarios, and no special restrictions are made here.

[0076] Step S362. Based on the initial weight range and the initial screening model, generate the weight corresponding to each entity and each category in each parameter information to obtain multiple weight combinations;

[0077] Step S363. Calculate the comprehensive weight values ​​of the core node and each leaf node in the training set based on the weight combination to obtain multiple comprehensive weight values;

[0078] Step S364. Determine the maximum comprehensive weight value from the multiple comprehensive weight values, and use the node corresponding to the maximum comprehensive weight value as the initial node;

[0079] Step S365. Determine a secondary comprehensive weight value from the multiple comprehensive weight values, where the secondary comprehensive weight value is only less than the maximum comprehensive weight value;

[0080] Step S366. Calculate the similarity between the entity corresponding to the initial node and the entity corresponding to the core node to obtain a first value;

[0081] Step S367. Calculate the difference between the maximum comprehensive weight and the secondary comprehensive weight to obtain a second value;

[0082] Step S368. Based on the first value, the second value and the preset weight ratio, calculate and obtain the initial model index;

[0083] Specifically, according to the initial screening model, the weight corresponding to each entity and each category in each parameter information is determined from the initial weight range, and based on this, the corresponding comprehensive weight value of the core node and each leaf node is calculated. It is generally believed that the comprehensive weight value of all leaf nodes is smaller than the comprehensive weight of the core node, and the greater the difference between the comprehensive weight value of the core node and the largest comprehensive weight value among the leaf nodes, the more it can reflect that the current initial screening model has good core node screening capabilities.

[0084] In addition, it is also necessary to consider the similarity between the entity corresponding to the core node and the entity corresponding to the current maximum comprehensive weight value. The higher the similarity, the more accurate the current initial screening model has.

[0085] The comprehensive weight value and similarity are considered as model indicators to measure the screening performance of the initial screening model.

[0086] The calculation formula of the initial model index is:

[0087] E=λ1*sim(X1,X h )+λ2*(S1-S2)

[0088] Among them, E is the initial model index; λ1 is the first weight ratio; λ2 is the second weight ratio; X1 is the first node; X h is the core node; sim() is the cosine similarity calculation function; S1 is the maximum comprehensive weight value; S2 is the secondary comprehensive weight value.

[0089] Step S37. Determine operation: determine the optimal screening model from multiple initial screening models, the optimal screening model is the model corresponding to the optimal parameter combination, and the optimal parameter combination is the parameter combination corresponding to the maximum initial model index;

[0090] Step S38. Comparison operation: compare each target model parameter in the optimal parameter combination with the corresponding initial range. When there is a target model parameter that belongs to the boundary of the corresponding initial range, adjust the initial range to obtain an updated initial range.

[0091] Specifically, each model parameter has a preset initial range, and the optimal parameter is found within the initial range as the model parameter. When the optimal parameter is at the boundary of the initial range, it is considered that the current initial range may limit the parameter value, so the initial range needs to be updated.

[0092] Specifically, step S38 specifically includes step S381, step S382, step S383, step S384, step S385, step S386, step S387 and step S388:

[0093] Step S381. Take the target model parameters in the optimal parameter combination that belong to the corresponding initial range boundary as the parameters to be adjusted;

[0094] Step S382. Determine a suboptimal model index from all initial model indexes, where the suboptimal model index is only smaller than the maximum model index;

[0095] Step S383. Obtain the target model parameter corresponding to the parameter to be adjusted in the target parameter combination corresponding to the suboptimal model indicator as a reference parameter;

[0096] Step S384. Calculate the difference between the parameter to be adjusted and the reference parameter to obtain a target difference;

[0097] Step S385. Compare the target difference value with the second set threshold value to obtain a parameter comparison result;

[0098] Step S386. Determine the magnification factor based on the parameter comparison result;

[0099] Step S387. Calculate the product of the target difference and the expansion multiple to obtain the expansion range;

[0100] Step S388. Generate an updated initial range based on the initial range and the expanded range;

[0101] Specifically, a suboptimal parameter combination second only to the optimal parameter combination is determined from all parameter combinations, the parameters in the optimal parameter combination located on the boundary of the initial range are used as parameters to be adjusted, and reference parameters of the same category as the parameters to be adjusted are determined from the suboptimal parameter combinations, and the difference between the parameters to be adjusted and the reference parameters is calculated. When the difference is large, it is necessary to expand the initial range by a larger range, and when the difference is small, it is only necessary to expand the initial range by a smaller range. There is a difference in the corresponding expansion multiples in the two cases. The product of the expansion multiple and the difference is calculated as the required expanded range, and finally the required expanded range is superimposed on the initial range to obtain the adjusted initial range.

[0102] Step S39. Repeat the first generation operation, the selection operation, the adjustment operation, the introduction operation, the determination operation and the comparison operation until the model parameters in the optimal parameter combination all belong to the corresponding initial range, and use the initial screening model corresponding to the optimal parameter combination as the screening model to be tested;

[0103] Specifically, the optimal parameter is determined again within the updated initial range, and the relationship between the parameter and the updated initial range is compared, until the optimal parameter is within the initial range, the optimal parameter is used as the final parameter of the model.

[0104] Step S310. Bring the test set and the validation set into the screening model to be tested for prediction, and calculate the model index based on the prediction results to obtain the first model index and the second model index;

[0105] Step S311. Calculate the difference between the first model index and the second model index to obtain a target difference;

[0106] Step S312. When the target difference is less than the first set threshold, the screening model to be tested is used as a data screening model;

[0107] Specifically, based on the model trained with the training set, you need to use the validation set and test set to verify whether the model parameters are reasonable. The validation set and test set can test the performance of the model on unknown data. If the model index of the validation set is significantly smaller than the model index of the test set, it means that the model has the disadvantage of overfitting and the model parameters need to be adjusted.

[0108] Step S40. Input the multi-caliber target data into the data screening model, the data screening model determines the target data from the multi-caliber target data, and uses the entity corresponding to the target data as the target core node;

[0109] Specifically, by inputting multi-caliber target data into a trained data screening model for screening, the entity data that should be used as the core node can be determined more accurately.

[0110] Step S50. Determine a target model from a plurality of preset knowledge graph models based on the target core node, and import multi-caliber target data into the target model to obtain an initial knowledge graph;

[0111] Specifically, a knowledge graph model is determined based on the predicted target core nodes, and all multi-caliber target data are imported into the knowledge graph model to obtain an initial knowledge graph.

[0112] Step S60: Optimize the initial knowledge graph to obtain a target knowledge graph of multi-caliber data;

[0113] Specifically, considering that the generated initial knowledge graph is automatically generated according to the preset knowledge graph model, its architecture is relatively fixed and it is difficult to perfectly adapt to the current multi-caliber target data. Therefore, the knowledge graph can be optimized to significantly improve its quality and practicality.

[0114] Specifically, step S60 specifically includes step S61, step S62, step S63, step S64, step S65, step S66, step S67 and step S68:

[0115] Step S61. Extract entities and relationships based on the initial knowledge graph to form triples, where the triples include a first triple and a second triple. The first triple is a combination of a first entity, a mutual relationship, and a second entity, and the second triple is a combination of a first attribute, a second attribute, and a third attribute corresponding to the first entity, the mutual relationship, and the second entity.

[0116] Step S62: based on the graph embedding technology, convert the first triplet and the second triplet into a first vector and a second vector respectively;

[0117] Step S63. Process the first vector and the second vector based on the word2vec model to obtain a third vector and a fourth vector;

[0118] Step S64. Send the third vector and the fourth vector to the convolution layer to obtain a fifth vector and a sixth vector;

[0119] Specifically, multiple filters are used to perform convolution operations on the input third vector and fourth vector, so as to obtain shallow semantic features of each vector.

[0120] Step S65. Send the fifth vector and the sixth vector to the interaction layer to generate a first eigenvector and a second eigenvector;

[0121] Step S66. Obtain a seventh vector based on the first eigenvector and the second eigenvector, where the seventh vector represents the correlation between the fifth vector and the sixth vector of the input interaction layer;

[0122] Specifically, the interaction layer includes a bidirectional long short-term memory encoding layer, an attention interaction layer and a bidirectional long short-term memory synthesis layer, wherein the bidirectional long short-term memory encoding layer is used to encode the embedded representation of the two input vectors, and generate the semantic feature representation of a single word and contextual attribute information, namely, the first feature vector and the second feature vector; the attention interaction layer is used to calculate the attention weight of the bidirectional sequential encoding of the input, obtain the local correlation between the two input vectors, and then identify and combine their related semantics according to the attention weight; the bidirectional long short-term memory synthesis layer is used to encode the local semantics and context of the two vectors, so as to further determine the overall semantic relationship between the two vectors, and finally calculate the average and maximum value of the overall semantic representation of the two input vectors, respectively, and concatenate all the vectors to generate the seventh vector.

[0123] The process of obtaining the seventh vector is:

[0124] L7=[mean(L'1),max(L'1),mean(L'2),max(L'2)]

[0125] Among them, L7 is the seventh vector; mean(L'1) is the average value of the first eigenvector; max(L'1) is the maximum value of the first eigenvector; mean(L'2) is the average value of the second eigenvector; max(L'2) is the maximum value of the second eigenvector.

[0126] Step S67: Send the seventh vector to the classification layer to obtain scoring information;

[0127] Step S68. Based on the score information, different optimization methods are used to optimize the initial knowledge graph to obtain the target knowledge graph;

[0128] Specifically, the optimization method required for the current initial knowledge graph is determined according to the amount of scoring information, and corresponding optimization is performed to obtain the target knowledge graph.

[0129] Step S70: Obtain query information, where the query information carries a query target and a query relationship;

[0130] Step S80. Determine the query core node based on the query target;

[0131] Step S90. Determine query leaf nodes based on the query relationship, and the number of query leaf nodes is at least 1;

[0132] Step S100. Generate a query knowledge graph model based on the query core node and the query leaf node;

[0133] Specifically, Figure 2 As shown in the figure, it is a schematic diagram of the structure of a target knowledge graph. As can be seen from the figure, each node in the graph is connected to multiple nodes, and the types of connections between different nodes are different. When you need to query a certain category of connections, you can use the query information of the core node plus the category to get the corresponding partial knowledge graph diagram. Figure 2 For example, when the input carries the query information of "A+Type 2", the corresponding output will be Figure 3 The schematic diagram of part of the knowledge graph shown in the figure; when the input carries the query information of "A+Type 3", the corresponding output will be Figure 4 The schematic diagram of part of the knowledge graph is shown. The format and specific composition of the query information can be set according to actual needs and are not particularly limited here.

[0134] Through the above query steps, complex data relationships can be displayed in an intuitive graphical manner, and efficient query based on the knowledge graph can efficiently obtain specific information of a single or multiple sets of data, and can quickly identify and locate data and management differences under different calibers.

[0135] Embodiment 2:

[0136] like Figure 5 As shown, this embodiment provides a multi-caliber data management device based on a knowledge graph, and the device includes:

[0137] The first acquisition unit 10 is used to acquire a historical knowledge graph, which includes a core node and multiple leaf nodes. The entities corresponding to the core nodes and the leaf nodes carry multiple parameter information, which are type, source, format and update frequency.

[0138] The second acquisition unit 20 is used to acquire multi-aperture data and perform preprocessing to obtain multi-aperture target data, each of which carries parameter information;

[0139] The training unit 30 is used to train the preset decision tree model based on the core nodes and leaf nodes in the historical knowledge graph. When the preset model indicators meet the set conditions, the training is stopped to obtain the data screening model;

[0140] A first input unit 40 is used to input the multi-caliber target data into the data screening model, and the data screening model determines the target data from the multi-caliber target data and uses the entity corresponding to the target data as the target core node;

[0141] A first determination unit 50 is used to determine a target model from a plurality of preset knowledge graph models based on a target core node, and import multi-caliber target data into the target model to obtain an initial knowledge graph;

[0142] The optimization unit 60 is used to optimize the initial knowledge graph to obtain a target knowledge graph of multi-caliber data.

[0143] In a specific implementation disclosed in the present application, the training unit 30 includes:

[0144] A partitioning unit is used to randomly partition multiple leaf nodes into three parts, which are used as training set, test set and validation set respectively, and add core nodes to the training set, test set and validation set;

[0145] A third acquisition unit is used to obtain a value range of each model parameter;

[0146] A first generating unit, used for a first generating operation: randomly generating a plurality of initial model parameters based on an initial range of each model parameter;

[0147] A first selection unit is used for a selection operation: randomly selecting an initial model parameter from a plurality of initial model parameters of each model parameter for combination to obtain a plurality of parameter combinations, wherein at most two initial model parameters in any two parameter combinations have the same value;

[0148] A first adjustment unit is used for adjustment operation: adjusting the decision tree model based on each parameter combination to obtain multiple initial screening models;

[0149] The first input unit is used for input operation: the training set is input into the initial screening model for training, and the model index is calculated to obtain multiple initial model indexes;

[0150] A second determination unit is used for determining an operation: determining an optimal screening model from a plurality of initial screening models, wherein the optimal screening model is a model corresponding to an optimal parameter combination, wherein the optimal parameter combination is a parameter combination corresponding to a maximum initial model index;

[0151] A first comparison unit is used for comparison operation: comparing each target model parameter in the optimal parameter combination with the corresponding initial range, and when there is a target model parameter that belongs to the corresponding initial range boundary, adjusting the initial range to obtain an updated initial range;

[0152] A repeating unit, used to repeat the first generation operation, the selection operation, the adjustment operation, the introduction operation, the determination operation and the comparison operation until the model parameters in the optimal parameter combination all belong to the corresponding initial range, and the initial screening model corresponding to the optimal parameter combination is used as the screening model to be tested;

[0153] The second input unit is used to respectively input the test set and the validation set into the screening model to be tested for prediction, and calculate the model index based on the prediction results to obtain the first model index and the second model index;

[0154] A first calculation unit is used to calculate the difference between the first model index and the second model index to obtain a target difference;

[0155] The first is a unit, which is used to use the screening model to be detected as a data screening model when the target difference is less than a first set threshold.

[0156] In a specific embodiment disclosed in the present application, the first bringing-in unit includes:

[0157] A fourth acquisition unit, used to acquire initial weight ranges corresponding to the plurality of parameter information respectively;

[0158] A second generating unit is used to generate a weight corresponding to each entity and each category in each parameter information based on the initial weight range and the initial screening model to obtain multiple weight combinations;

[0159] A first obtaining unit is used to calculate the comprehensive weight value of the core node and each leaf node in the training set based on the weight combination to obtain multiple comprehensive weight values;

[0160] A third determining unit is used to determine a maximum comprehensive weight value from a plurality of comprehensive weight values, and use a node corresponding to the maximum comprehensive weight value as an initial node;

[0161] a fourth determining unit, configured to determine a secondary comprehensive weight value from the plurality of comprehensive weight values, the secondary comprehensive weight value being only less than the maximum comprehensive weight value;

[0162] A second calculation unit, used to calculate the similarity between the entity corresponding to the initial node and the entity corresponding to the core node, to obtain a first value;

[0163] A third calculation unit is used to calculate the difference between the maximum comprehensive weight and the secondary comprehensive weight value to obtain a second value;

[0164] The fourth calculation unit is used to calculate the initial model index based on the first value, the second value and a preset weight ratio.

[0165] In a specific implementation disclosed in the present application, the first comparison unit includes:

[0166] The second unit is used to take the target model parameters belonging to the corresponding initial range boundary in the optimal parameter combination as the parameters to be adjusted;

[0167] a fifth determining unit, used to determine a suboptimal model index from all initial model indexes, where the suboptimal model index is only less than the maximum model index;

[0168] A fifth acquisition unit is used to acquire, as a reference parameter, a target model parameter corresponding to the parameter to be adjusted in the target parameter combination corresponding to the suboptimal model indicator;

[0169] A fifth calculation unit, used for calculating the difference between the parameter to be adjusted and the reference parameter to obtain a target difference;

[0170] A second comparison unit is used to compare the target difference value with a second set threshold value to obtain a parameter comparison result;

[0171] A sixth determining unit, configured to determine an enlargement multiple based on the parameter comparison result;

[0172] The sixth calculation unit is used to calculate the product of the target difference and the expansion multiple to obtain the expansion range;

[0173] The third generating unit is configured to generate an updated initial range based on the initial range and the expanded range.

[0174] In a specific implementation manner disclosed in the present application, the second acquisition unit 20 includes:

[0175] A cleaning unit, used for performing cleaning processing on the multi-caliber data to obtain a plurality of first data, wherein the cleaning processing includes removing noise, eliminating duplication and processing missing values;

[0176] A normalization unit, used for performing format normalization processing on the plurality of first data to obtain a plurality of second data;

[0177] A first screening unit, configured to screen the semi-structured data and the unstructured data from the second data as a plurality of third data;

[0178] A first processing unit, configured to perform division processing based on the type of the third data to obtain a plurality of division units;

[0179] a seventh calculation unit, configured to calculate the occurrence frequency of each of the plurality of division units to obtain frequency information;

[0180] a seventh determining unit, configured to determine a location of each of the plurality of dividing units to obtain location information;

[0181] an eighth determining unit, configured to determine a property of each of the plurality of dividing units to obtain property information;

[0182] an eighth calculation unit, configured to perform weighted calculation based on the frequency information, the location information and the property information to obtain weight information of each division unit;

[0183] A second screening unit is used to screen the multiple division units based on the weight information to obtain a plurality of screened target screening units;

[0184] A first composition unit, for composing and processing data based on a plurality of target screening units;

[0185] The second composition unit is used to compose multi-caliber target data based on the structured data and processed data in the second data.

[0186] In a specific implementation disclosed in the present application, the optimization unit 60 includes:

[0187] An extraction unit, configured to extract entities and relations based on the initial knowledge graph to form triples, wherein the triples include a first triple and a second triple, wherein the first triple is a combination of a first entity, a mutual relationship, and a second entity, and the second triple is a combination of a first attribute, a second attribute, and a third attribute corresponding to the first entity, the mutual relationship, and the second entity;

[0188] a conversion unit, configured to convert the first triplet and the second triplet into a first vector and a second vector respectively based on a graph embedding technology;

[0189] A second processing unit is used to process the first vector and the second vector based on a word2vec model to obtain a third vector and a fourth vector;

[0190] A first sending unit, used for sending the third vector and the fourth vector to the convolution layer to obtain a fifth vector and a sixth vector;

[0191] A second sending unit, used for sending the fifth vector and the sixth vector to the interaction layer to generate a first eigenvector and a second eigenvector;

[0192] A second obtaining unit is used to obtain a seventh vector based on the first eigenvector and the second eigenvector, where the seventh vector represents the correlation between the fifth vector and the sixth vector of the input interaction layer;

[0193] A third sending unit, used for sending the seventh vector to the classification layer to obtain scoring information;

[0194] The third obtaining unit is used to optimize the initial knowledge graph using different optimization methods based on the level of scoring information to obtain the target knowledge graph.

[0195] In a specific embodiment disclosed in the present application, the device includes:

[0196] A sixth acquisition unit, used to acquire query information, where the query information carries a query target and a query relationship;

[0197] An eighth determining unit, configured to determine a query core node based on a query target;

[0198] A ninth determining unit, configured to determine a query leaf node based on the query relationship, wherein the number of the query leaf node is at least 1;

[0199] The fourth generation unit is used to generate a query knowledge graph model based on the query core node and the query leaf node.

[0200] It should be noted that, regarding the device in the above embodiment, the specific manner in which each module performs the operation has been described in detail in the embodiment of the method, and will not be elaborated here.

[0201] Embodiment 3:

[0202] Corresponding to the above method embodiment, this embodiment also provides a multi-caliber data management device based on a knowledge graph. The multi-caliber data management device based on a knowledge graph described below and the multi-caliber data management method based on a knowledge graph described above can refer to each other.

[0203] Figure 6 is a block diagram of a multi-caliber data management device 800 based on a knowledge graph according to an exemplary embodiment. Figure 6 As shown, the multi-caliber data management device 800 based on the knowledge graph may include: a processor 801, a memory 802. The multi-caliber data management device 800 based on the knowledge graph may also include one or more of a multimedia component 803, an I / O interface 804, and a communication component 805.

[0204] Among them, the processor 801 is used to control the overall operation of the multi-caliber data management device 800 based on the knowledge graph to complete all or part of the steps in the multi-caliber data management method based on the knowledge graph. The memory 802 is used to store various types of data to support the operation of the multi-caliber data management device 800 based on the knowledge graph. These data may include, for example, instructions for any application or method operating on the multi-caliber data management device 800 based on the knowledge graph, as well as application-related data, such as contact data, messages sent and received, pictures, audio, video, etc. The memory 802 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, disk or optical disk. The multimedia component 803 may include a screen and an audio component. The screen may be, for example, a touch screen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signal may be further stored in the memory 802 or sent via the communication component 805. The audio component also includes at least one speaker for outputting audio signals. The I / O interface 804 provides an interface between the processor 801 and other interface modules, and the above-mentioned other interface modules can be keyboards, mice, buttons, etc. These buttons can be virtual buttons or physical buttons. The communication component 805 is used for wired or wireless communication between the multi-caliber data management device 800 based on the knowledge graph and other devices. Wireless communication, such as Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G or 4G, or a combination of one or more of them, so the corresponding communication component 805 can include: Wi-Fi module, Bluetooth module, NFC module.

[0205] In an exemplary embodiment, the multi-caliber data management device 800 based on the knowledge graph can be implemented by one or more application-specific integrated circuits (Application Specific Integrated Circuit, referred to as ASIC), digital signal processors (Digital Signal Processing Device, referred to as DSP), digital signal processing devices (Digital Signal Processing Device, referred to as DSPD), programmable logic devices (Programmable Logic Device, referred to as PLD), field programmable gate arrays (Field Programmable Gate Array, referred to as FPGA), controllers, microcontrollers, microprocessors or other electronic components to execute the above-mentioned multi-caliber data management method based on the knowledge graph.

[0206] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, and when the program instructions are executed by a processor, the steps of the multi-caliber data management method based on the knowledge graph are implemented. For example, the computer-readable storage medium can be the memory 802 including the program instructions, and the program instructions can be executed by the processor 801 of the multi-caliber data management device 800 based on the knowledge graph to complete the multi-caliber data management method based on the knowledge graph.

[0207] Embodiment 4:

[0208] Corresponding to the above method embodiment, a readable storage medium is also provided in this embodiment. The readable storage medium described below and the multi-caliber data management method based on the knowledge graph described above can refer to each other.

[0209] A readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the multi-caliber data management method based on a knowledge graph of the above-mentioned method embodiment.

[0210] The readable storage medium may specifically be a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, or other readable storage medium that can store program codes.

[0211] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0212] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A multi-caliber data management method based on knowledge graph, characterized in that: include: Obtain a historical knowledge graph, wherein the historical knowledge graph includes a core node and multiple leaf nodes, and entities corresponding to the core node and the leaf nodes each carry multiple parameter information, wherein the parameter information is respectively type, source, format, and update frequency; Acquire multi-caliber data and perform preprocessing to obtain multi-caliber target data, wherein each data in the multi-caliber target data carries parameter information; Based on the core nodes and leaf nodes in the historical knowledge graph, a preset decision tree model is trained, and when the preset model indicators meet the set conditions, the training is stopped to obtain a data screening model; Inputting the multi-caliber target data into the data screening model, the data screening model determines the target data from the multi-caliber target data, and uses the entity corresponding to the target data as the target core node; Determine a target model from a plurality of preset knowledge graph models based on the target core node, and import the multi-caliber target data into the target model to obtain an initial knowledge graph; Optimizing the initial knowledge graph to obtain a target knowledge graph of the multi-caliber data; The preset decision tree model is trained based on the core nodes and leaf nodes in the historical knowledge graph. When the preset model indicators meet the set conditions, the training is stopped to obtain the data screening model. The decision tree model includes three model parameters, including: The plurality of leaf nodes are randomly divided into three parts, which are used as a training set, a test set and a validation set respectively, and the core node is added to the training set, the test set and the validation set; Get the value range of each model parameter; First generation operation: randomly generating a plurality of initial model parameters based on the initial range of each model parameter; Selection operation: randomly selecting an initial model parameter from multiple initial model parameters of each model parameter for combination to obtain multiple parameter combinations, wherein at most two initial model parameters in any two parameter combinations have the same value; Adjustment operation: adjusting the decision tree model based on each parameter combination to obtain multiple initial screening models; Bring-in operation: bring the training set into the initial screening model for training, and calculate the model index to obtain multiple initial model indexes; Determine operation: determine the optimal screening model from multiple initial screening models, the optimal screening model is a model corresponding to the optimal parameter combination, and the optimal parameter combination is a parameter combination corresponding to the maximum initial model index; Comparison operation: compare each target model parameter in the optimal parameter combination with the corresponding initial range. When a target model parameter belongs to the boundary of the corresponding initial range, adjust the initial range to obtain an updated initial range. Repeating the first generating operation, the selecting operation, the adjusting operation, the bringing-in operation, the determining operation and the comparing operation until all model parameters in the optimal parameter combination belong to the corresponding initial range, and taking the initial screening model corresponding to the optimal parameter combination as the screening model to be detected; The test set and the validation set are respectively brought into the screening model to be tested for prediction, and the model index is calculated based on the prediction results to obtain a first model index and a second model index; Calculating the difference between the first model indicator and the second model indicator to obtain a target difference; When the target difference is less than a first set threshold, the to-be-detected screening model is used as the data screening model; Among them, the import operation: import the training set into the initial screening model for training, and calculate the model index to obtain multiple initial model indexes, each parameter information corresponds to multiple categories, including: Obtain the initial weight ranges corresponding to multiple parameter information; Based on the initial weight range and the initial screening model, a weight corresponding to each entity and each category in each parameter information is generated to obtain multiple weight combinations; Calculate the comprehensive weight values ​​of the core node and each leaf node in the training set based on the weight combination to obtain multiple comprehensive weight values; Determine a maximum comprehensive weight value from multiple comprehensive weight values, and use a node corresponding to the maximum comprehensive weight value as an initial node; Determine a secondary comprehensive weight value from the multiple comprehensive weight values, wherein the secondary comprehensive weight value is only less than the maximum comprehensive weight value; Calculating the similarity between the entity corresponding to the initial node and the entity corresponding to the core node to obtain a first value; Calculate the difference between the maximum comprehensive weight and the secondary comprehensive weight to obtain a second value; The initial model index is calculated based on the first value, the second value and a preset weight ratio; The initial knowledge graph is optimized to obtain the target knowledge graph of the multi-caliber data, and then includes: Acquire query information, wherein the query information carries a query target and a query relationship; Determine a query core node based on the query target; Determine a query leaf node based on the query relationship, where the number of the query leaf node is at least 1; A query knowledge graph model is generated based on the query core nodes and query leaf nodes.

2. The multi-caliber data management method based on knowledge graph according to claim 1 is characterized in that , optimizing the initial knowledge graph to obtain the target knowledge graph of the multi-caliber data, including: Extracting entities and relationships based on the initial knowledge graph to form triples, the triples comprising a first triple and a second triple, the first triple being a combination of a first entity, a mutual relationship, and a second entity, and the second triple being a combination of a first attribute, a second attribute, and a third attribute corresponding to the first entity, the mutual relationship, and the second entity; Based on the graph embedding technology, the first triplet and the second triplet are respectively converted into a first vector and a second vector; Processing the first vector and the second vector based on a word2vec model to obtain a third vector and a fourth vector; Sending the third vector and the fourth vector to a convolutional layer to obtain a fifth vector and a sixth vector; Sending the fifth vector and the sixth vector to an interaction layer to generate a first eigenvector and a second eigenvector; A seventh vector is obtained by encoding the first feature vector and the second feature vector, where the seventh vector represents the correlation between the fifth vector and the sixth vector of the input interaction layer; Sending the seventh vector to the classification layer to obtain scoring information; Based on the rating information, different optimization methods are used to optimize the initial knowledge graph to obtain the target knowledge graph.

3. The multi-caliber data management method based on knowledge graph according to claim 1 is characterized in that Based on the first value, the second value and the preset weight ratio, the initial model index is calculated, and the calculation formula of the initial model index is: E=λ1*sim(X1,X h )+λ2*(S1-S2) Among them, E is the initial model index; λ1 is the first weight ratio; λ2 is the second weight ratio; X1 is the first node; X h is the core node; sim() is the cosine similarity calculation function; S1 is the maximum comprehensive weight value; S2 is the secondary comprehensive weight value.

4. The multi-caliber data management method based on knowledge graph according to claim 2 is characterized in that , based on the first feature vector and the second feature vector, a seventh vector is obtained by encoding, and the seventh vector is obtained by: L7=[mean(L'1),max(L'1),mean(L'2),max(L'2)] Among them, L7 is the seventh vector; mean(L'1) is the average value of the first eigenvector; max(L'1) is the maximum value of the first eigenvector; mean(L'2) is the average value of the second eigenvector; max(L'2) is the maximum value of the second eigenvector.

5. A multi-caliber data management device based on knowledge graph, characterized in that: include: A first acquisition unit is used to acquire a historical knowledge graph, wherein the historical knowledge graph includes a core node and multiple leaf nodes, and entities corresponding to the core node and the leaf nodes each carry multiple parameter information, wherein the parameter information is respectively type, source, format, and update frequency; A second acquisition unit is used to acquire multi-caliber data and perform preprocessing to obtain multi-caliber target data, wherein each data in the multi-caliber target data carries parameter information; A training unit, used to train a preset decision tree model based on the core nodes and leaf nodes in the historical knowledge graph, and when the preset model indicators meet the set conditions, stop the training to obtain a data screening model; A first input unit is used to input the multi-caliber target data into the data screening model, the data screening model determines the target data from the multi-caliber target data, and uses the entity corresponding to the target data as the target core node; A first determination unit is used to determine a target model from a plurality of preset knowledge graph models based on the target core node, and import the multi-caliber target data into the target model to obtain an initial knowledge graph; An optimization unit, used to optimize the initial knowledge graph to obtain a target knowledge graph of the multi-caliber data; The training units include: A partitioning unit is used to randomly partition multiple leaf nodes into three parts, which are used as training set, test set and validation set respectively, and add core nodes to the training set, test set and validation set; A third acquisition unit is used to obtain a value range of each model parameter; A first generating unit, used for a first generating operation: randomly generating a plurality of initial model parameters based on an initial range of each model parameter; A first selection unit is used for a selection operation: randomly selecting an initial model parameter from a plurality of initial model parameters of each model parameter for combination to obtain a plurality of parameter combinations, wherein at most two initial model parameters in any two parameter combinations have the same value; A first adjustment unit is used for adjustment operation: adjusting the decision tree model based on each parameter combination to obtain multiple initial screening models; The first input unit is used for input operation: the training set is input into the initial screening model for training, and the model index is calculated to obtain multiple initial model indexes; A second determination unit is used for determining an operation: determining an optimal screening model from a plurality of initial screening models, wherein the optimal screening model is a model corresponding to an optimal parameter combination, wherein the optimal parameter combination is a parameter combination corresponding to a maximum initial model index; A first comparison unit is used for comparison operation: comparing each target model parameter in the optimal parameter combination with the corresponding initial range, and when there is a target model parameter that belongs to the corresponding initial range boundary, adjusting the initial range to obtain an updated initial range; A repeating unit, used to repeat the first generation operation, the selection operation, the adjustment operation, the introduction operation, the determination operation and the comparison operation until the model parameters in the optimal parameter combination all belong to the corresponding initial range, and the initial screening model corresponding to the optimal parameter combination is used as the screening model to be tested; The second input unit is used to respectively input the test set and the validation set into the screening model to be tested for prediction, and calculate the model index based on the prediction results to obtain the first model index and the second model index; A first calculation unit is used to calculate the difference between the first model index and the second model index to obtain a target difference; The first is used as a unit, for taking the screening model to be detected as a data screening model when the target difference is less than a first set threshold value; The first bringing-in unit includes: A fourth acquisition unit, used to acquire initial weight ranges corresponding to the plurality of parameter information respectively; A second generating unit is used to generate a weight corresponding to each entity and each category in each parameter information based on the initial weight range and the initial screening model to obtain multiple weight combinations; A first obtaining unit is used to calculate the comprehensive weight value of the core node and each leaf node in the training set based on the weight combination to obtain multiple comprehensive weight values; A third determining unit is used to determine a maximum comprehensive weight value from a plurality of comprehensive weight values, and use a node corresponding to the maximum comprehensive weight value as an initial node; a fourth determining unit, configured to determine a secondary comprehensive weight value from the plurality of comprehensive weight values, the secondary comprehensive weight value being only less than the maximum comprehensive weight value; A second calculation unit, used to calculate the similarity between the entity corresponding to the initial node and the entity corresponding to the core node, to obtain a first value; A third calculation unit is used to calculate the difference between the maximum comprehensive weight and the secondary comprehensive weight value to obtain a second value; A fourth calculation unit, configured to calculate an initial model indicator based on the first value, the second value and a preset weight ratio; The device comprises: A sixth acquisition unit, used to acquire query information, where the query information carries a query target and a query relationship; An eighth determining unit, configured to determine a query core node based on a query target; A ninth determining unit, configured to determine a query leaf node based on the query relationship, wherein the number of the query leaf node is at least 1; The fourth generation unit is used to generate a query knowledge graph model based on the query core node and the query leaf node.

6. A multi-caliber data management device based on knowledge graph, characterized in that: include: Memory for storing computer programs; A processor, used to implement the steps of the multi-caliber data management method based on the knowledge graph as described in any one of claims 1 to 4 when executing the computer program.

7. A readable storage medium, characterized in that: The readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the multi-caliber data management method based on a knowledge graph as described in any one of claims 1 to 4.

Images