A method for establishing label relationships in a three-dimensional network catalog system of global observation results

Abstract

The present invention discloses a method for establishing a label relationship in a three-dimensional mesh directory system of global observation results, and belongs to the field of big data. In view of the problem that the existing hierarchical tree-like directory system is difficult to adjust and mutually compatible, resulting in the difficulty in unified management and sharing of multi-source global observation results of various centers, the present invention integrates the three dimensions of time, space and theme to establish a label association relationship, and adopts a "time-space-theme" hybrid association model to establish a deeper association relationship between global observation results. The label relationship between global observation results established by the present invention can provide a basis for establishing a three-dimensional mesh directory system for global observation results, and is compatible with the data organization structure of each center through a multi-dimensional mapping mechanism, and ultimately realizes data sharing among various centers under a unified resource directory system.

Description

Technical Field

[0001] The present invention relates to the field of big data, and in particular to a method for establishing label relationships in a three-dimensional network catalog system of global observation results. Background Art

[0002] The comprehensive observation data from major satellite resource centers and the Internet environment has shown explosive growth with the development of technology, but there are sharing barriers and inconsistent catalog systems between observation results. It is difficult to effectively obtain global strategic information by relying on the traditional centralized catalog system. The global observation results three-dimensional network catalog system uses a label-associated three-dimensional network to reconstruct the original catalog hierarchy, and is compatible with the data organization structure of each center through a multi-dimensional mapping mechanism, ultimately realizing the sharing of global observation results of each center under a unified resource catalog system.

[0003] Tags are the basic elements of a three-dimensional network directory system. The premise of building a three-dimensional network directory system is to obtain the association relationship between the tags of each global comprehensive observation result, so as to cluster and associate the global comprehensive observation results. However, the global comprehensive observation results are heterogeneous data with different sources and forms. How to construct the association relationship between the tags of these global comprehensive observation results is a technical problem that needs to be solved urgently. Summary of the invention

[0004] The purpose of the present invention is to solve the problems existing in the prior art and to provide a method for establishing label relationships in a three-dimensional network directory system of global observation results.

[0005] In order to achieve the above-mentioned invention object, the technical solution specifically adopted by the present invention is as follows:

[0006] A method for establishing label relationships of a three-dimensional network catalog system of global observation results, the steps of which are as follows:

[0007] S1. Extract the metadata of all global observation results participating in the catalog construction to obtain the time label, space label and subject label of each global observation result; the time label is the data time information corresponding to the global observation result, the space label is the place name information corresponding to the global observation result, and the subject label is the subject keyword set corresponding to the global observation result;

[0008] S2. Establish an association model from the time dimension, calculate the time topology index and time distance index of any two global observation results based on the time tag, and calculate the time similarity between any two global observation results by weighting the time topology index and the time distance index;

[0009] S3. Establish an association model from the spatial dimension, map the spatial label of each global observation result into a spatial geometric area under a unified spatial coordinate system through place name address mapping, and calculate the spatial topological index and spatial metric index of any two global observation results, and then calculate the spatial similarity between any two global observation results by weighting the spatial topological index and spatial metric index;

[0010] S4. Establish an association model from the topic dimension, map the topic label of each global observation result into a word vector through a word vector model, and calculate the topic similarity between any two global observation results based on the word vector;

[0011] S5. Traverse all global observation results participating in the construction of the directory. For any two global observation results, perform weighted summation of the corresponding time similarity, spatial similarity, and topic similarity and use them as the mixed correlation similarity between the two, thereby obtaining the label relationship between all global observation results.

[0012] Preferably, the specific implementation steps of S2 are as follows:

[0013] S21. For each global observation result, obtain its corresponding time tag, and extract the time type, time length and starting time point from the time tag through the pre-defined time class; the time type shown is divided into two categories: time point and time period. The time type with a time length less than or equal to the time length threshold is defined as a time point, and the time type with a time length greater than the time length threshold is defined as a time period;

[0014] S22. Traverse all global observation results participating in the construction of the directory, and calculate the time topology index T1 between the time classes of any two global observation results according to the time topology rule; in the time topology rule: if the time points are equal, the corresponding time topology index is 1; if the time points are not equal, the corresponding time topology index is 0; if the time points are included in the relationship between time points and time periods, the corresponding time topology index is 1; if the time points are not included in the relationship between time points and time periods, the corresponding time topology index is 0; if the time periods are equal or included in the relationship between time periods, the corresponding time topology index is 1; if the time periods are separated, the corresponding time topology index is 0; if the time periods are intersected, the corresponding time topology index calculation formula is:

[0015] T1=2*O t / (T a +T b )

[0016] Where: O t is the length of the overlap between time periods, T a, T b are two time period lengths respectively;

[0017] S23, traverse all global observation results participating in the directory construction, and calculate the time distance index T2 between the time classes of any two global observation results. The calculation formula is:

[0018] T2=(T a +T b ) / (D t +T a +T b )

[0019] Among them, D t is the center distance between the time classes corresponding to the two global observation results;

[0020] S24. For any two global observation results, calculate the temporal similarity T between them. sim :

[0021] T sim =W t1 *T1+W t2 *T2

[0022] Where: W t1 and W t2 All are weight values.

[0023] Preferably, the duration threshold is 1 year.

[0024] Preferably, the specific implementation steps of S3 are as follows:

[0025] S31. For each global observation result, obtain its corresponding spatial label, and map the place name information in the spatial label to a unified spatial coordinate system according to the place name-spatial geometry mapping relationship to obtain the spatial geometry area corresponding to each global observation result;

[0026] S32, traverse all global observation results participating in the construction of the directory, and calculate the spatial topological index S1 between the spatial geometric regions of any two global observation results according to the spatial topological rules; in the spatial topological rules: if the topological relationship between the two spatial geometric regions is an equal relationship, a containment relationship, or a contained relationship, the corresponding spatial topological index is 1; if the topological relationship between the two spatial geometric regions is an overlapping relationship, the corresponding spatial topological index is 0.8; if the topological relationship between the two spatial geometric regions is a connected relationship, the corresponding spatial topological index is 0.5; if the topological relationship between the two spatial geometric regions is a separated relationship, the corresponding spatial topological index is 0;

[0027] S33, traverse all global observation results participating in the construction of the catalog, and calculate the spatial metric index S2 between the spatial geometric areas of any two global observation results. The calculation formula is:

[0028] S2=W s21 *S 21 +W s22 *S 22

[0029] S 21 =2*S s / (S a +S b )

[0030] S 22 =(R a +R b ) / (D s +R a +R b )

[0031] Where: W s21 and W s22 are weight values, S 21 and S 22 are the area overlap index and the distance metric index respectively; S s is the overlapping area between two spatial geometric regions, D s is the minimum distance between the centers of the outer circles of two spatial geometric regions, S a and S b are the areas of two spatial geometric regions, R a and R b are the minimum enclosing circle radii of two spatial geometric regions respectively;

[0032] S34. For any two global observation results, calculate the spatial similarity S between them. sim :

[0033] S sim =W s1 *S1+W s2 *S2

[0034] Where: W s1 and W s2 All are weight values.

[0035] Preferably, before mapping the spatial labels of global observation results into spatial geometric areas, the place name information in the spatial labels is converted between common place names and standard place names in advance, and the common place names are converted into standard administrative district names or sea area names through similarity matching or a pre-constructed mapping relationship table.

[0036] Preferably, the specific implementation steps of S4 are as follows:

[0037] S41. For each global observation result, obtain its corresponding topic label, and use the pre-trained word vector model to encode each keyword in each topic label to form a word vector;

[0038] S42, traverse all global observation results participating in the directory construction, for any two global observation results' keyword sets, first pair the keywords in the two sets according to the principle of highest similarity based on word vectors, then calculate the similarity of each pair of keyword word vectors, and take the average similarity of all keyword pairs as the topic similarity F between the two global observation results sim .

[0039] Preferably, the pre-trained word vector model is a Skip-gram-based Word2Vec model.

[0040] Preferably, when encoding keywords using a pre-trained word vector model, FastText is used to handle the out-of-vocabulary (OOV) problem.

[0041] Preferably, in S5, the formula for calculating the mixed correlation similarity between any two global observation results is as follows:

[0042] Geo sim =W s *S sim +W t *T sim +W f *F sim

[0043] Where: W s is the spatial similarity S sim The weight, W t is the temporal similarity T sim The weight, W f is the topic similarity F sim The weight of .

[0044] Preferably, all global observation results participating in the construction of the directory are used as nodes of the graph network, and the edge length between the nodes in the graph network is set to be inversely proportional to the mixed correlation similarity between the corresponding two global observation results, so as to obtain a relationship network of all global observation results for similarity retrieval or visualization.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] The present invention provides a method for establishing a label relationship in a three-dimensional network directory system of global observation results, which integrates the three dimensions of time, space and theme to establish a label association relationship, and adopts a "time-space-theme" hybrid association model to establish a deeper association relationship between global observation results. The label relationship between global observation results established by the present invention can provide a basis for establishing a three-dimensional network directory system of global observation results, and through a multi-dimensional mapping mechanism, it is compatible with the data organization structure of each center, and finally realizes data sharing of each center under a unified resource directory system. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 A flowchart of the steps for establishing label relationships in a three-dimensional network catalog system for global observation results.

[0048] Figure 2 This is the principle diagram of the hybrid correlation technology for global observation results.

[0049] Figure 3 This is a table of metadata relevance results for some global observation results in the embodiments.

[0050] Figure 4 It is a metadata association network diagram of some representative global observation results in the embodiment. DETAILED DESCRIPTION

[0051] The present invention is further described and illustrated below in conjunction with the accompanying drawings and specific embodiments. The technical features of each embodiment of the present invention can be combined accordingly without conflicting with each other.

[0052] The present invention provides a method for establishing a label relationship of a three-dimensional network directory system of global observation results. The method can extract corresponding labels from all global observation results participating in the directory construction, and calculate their comprehensive association relationships from three dimensions of time, space and subject, thereby establishing a data basis for similarity analysis for constructing a three-dimensional network directory system. In a preferred embodiment of the present invention, the specific implementation steps of the method for establishing a label relationship of a three-dimensional network directory system of global observation results are as follows:

[0053] S1. Extract the metadata (including description information) of all global observation results participating in the directory construction to obtain the time label, space label and subject label of each global observation result; the time label is the data time information corresponding to the global observation result, the space label is the place name information corresponding to the global observation result, and the subject label is the subject keyword set corresponding to the global observation result.

[0054] In this invention, various observation results around the world are collectively referred to as global observation results, which are typically represented by remote sensing image data and monitoring data of various natural elements. The data time information corresponding to the global observation results is mainly the time span information of the data set, the place name information is mainly the geographical space name corresponding to the data set, and the subject keywords can be obtained from the keyword list of the data set.

[0055] For example, the East China 30m Wetland Distribution Dataset (2008) is a global observation result related to wetland monitoring data. The data time information it covers is 2008, the place name information is East China (specific provinces and cities can be further obtained from the data description), and the theme keywords are wetlands, water bodies, etc.

[0056] S2. Establish an association model from the time dimension, calculate the time topology index and time distance index of any two global observation results based on the time tag, and calculate the time similarity between any two global observation results by weighting the time topology index and the time distance index.

[0057] In this embodiment, the specific implementation steps of the above step S2 are as follows:

[0058] S21. For each global observation result, obtain its corresponding time tag, and extract the time type, time length and starting time point from the time tag through the pre-defined time class; the time type shown is divided into two categories: time point and time period. The time type with a time length less than or equal to the time length threshold is defined as a time point, and the time type with a time length greater than the time length threshold is defined as a time period.

[0059] Temporal similarity is used to measure the temporal correlation between global observation results, including temporal topological relationships and temporal distance indicators. In this step, the duration threshold can be adjusted according to actual conditions, and the duration threshold can generally be set to 1 year. Therefore, the time type of global observation results includes time points and time periods. Taking the time scale of 1 year as the standard, time lengths less than or equal to 1 year are defined as time points, and time periods greater than 1 year. The above time class is defined as Time(), which contains variables such as time type, time length, starting time point, etc., which are used to extract time indicators.

[0060] S22. Traverse all global observation results participating in the construction of the directory, and calculate the time topology index T1 between the time classes of any two global observation results according to the time topology rule.

[0061] The time topology index T1 is used to measure the relationship between metadata time. By measuring the topological relationship of the extracted time class, the time correlation of metadata can be obtained. In this embodiment, the topological rules defining two time types, time point and time period, are shown in Table 1.

[0062] Table 1 Time class topology rule table

[0063]

[0064] In the above time topology rules: if the time points are equal, the corresponding time topology index is 1; if the time points are not equal, the corresponding time topology index is 0; if the time points are included, the corresponding time topology index is 1; if the time points are not included, the corresponding time topology index is 0; if the time periods are equal or included, the corresponding time topology index is 1; if the time periods are separated, the corresponding time topology index is 0; if the time periods are intersected, the corresponding time topology index is calculated as follows:

[0065] T1=2*O t / (T a +T b )

[0066] Where: O t is the length of the overlap between time periods, T a , T b The lengths of the two time periods are respectively.

[0067] S23. Traverse all global observation results participating in the construction of the catalog, and calculate the time distance index T2 between the time classes of any two global observation results.

[0068] The temporal distance index is used to numerically reflect the time distance relationship of metadata, and is used to assist the temporal topology index to further measure the temporal similarity of metadata. The calculation formula for defining the temporal distance index T2 in this embodiment is as follows:

[0069] T2=(T a +T b ) / (D t +T a +T b )

[0070] Among them, D t is the center distance between the time classes corresponding to the two global observations, that is, the interval between the middle time points of the time span of the two global observations;

[0071] S24. For any two global observation results, calculate the temporal similarity T between them. sim (TimeSimilarity) is used to reflect the similarity of metadata time tags. In this embodiment, T is calculated sim The time dimension association model is as follows:

[0072] T sim =W t1 *T1+W t2 *T2

[0073] Where: W t1 and W t2 All are weight values.

[0074] S3. Establish an association model from the spatial dimension, map the spatial label of each global observation result into a spatial geometric area under a unified spatial coordinate system through place name and address mapping, and calculate the spatial topological index and spatial measurement index of any two global observation results, and then calculate the spatial similarity between any two global observation results by weighting the spatial topological index and spatial measurement index.

[0075] In this embodiment, the specific implementation steps of the above step S3 are as follows:

[0076] S31. For each global observation result, obtain its corresponding spatial label, and map the place name information in the spatial label to a unified spatial coordinate system according to the place name-spatial geometry mapping relationship to obtain the spatial geometry area corresponding to each global observation result.

[0077] The purpose of place name address mapping is to map the place name information contained in the spatial label of global observation results to the spatial coordinate system to obtain spatial location information. However, it should be noted that due to the complexity of the sources and forms of global observation results, there may be a large number of place name common names in some global observation results. Such place name common names have the same meaning as their corresponding standard place names, but different forms. The place name-spatial geometry mapping relationship is based on the standard place name, that is, the place name-spatial geometry mapping relationship is the mapping relationship between the standard place name and the administrative region or sea area represented by the standard place name. Therefore, before mapping the spatial label of global observation results to the spatial geometry area, the place name information in the spatial label is converted between the place name common name and the standard place name in advance. For the natural common name of the place name, there are two solutions. One is to perform similarity matching through natural language processing methods, that is, based on the comparison of a large number of subject corpora, the administrative region name or sea area name with high similarity to the natural common name of the place name is matched through the similarity comparison between words. The second method is to establish a table of common place names, extract the common place names that cannot be found in the names of administrative regions or sea areas into the "common place name mapping relationship table", and manually add the mapping relationship between the common place names and the geometric area names to the "common place name mapping relationship table". Therefore, in this embodiment, the common place names can be converted into standard administrative region names or sea area names through similarity matching or a pre-built mapping relationship table according to actual needs.

[0078] S32. Traverse all global observation results participating in the construction of the directory, and calculate the spatial topological index S1 between the spatial geometric areas of any two global observation results according to the spatial topological rules.

[0079] It should be noted that the above-mentioned spatial topological index is used to measure the relationship between metadata spaces. By measuring the topological relationship of the mapped spatial geometry, the spatial correlation of metadata can be obtained. According to the nine-intersection model, the spatial topological relationship of polygons includes equality (Equals), touches (Touches), overlaps (Overlaps), contains or is contained (Contains / Within), and disjoints (Disjoints). The spatial geometric region is essentially a geometric surface figure, so the topological relationship between two spatial geometric regions can be judged based on the topological relationship between the two geometric surface figures. In the spatial topological rules set in this embodiment: if the topological relationship between two spatial geometric regions is an equality relationship, a containment relationship, or a contained relationship, the corresponding spatial topological index is 1, if the topological relationship between the two spatial geometric regions is an overlapping relationship, the corresponding spatial topological index is 0.8, if the topological relationship between the two spatial geometric regions is a connection relationship, the corresponding spatial topological index is 0.5, if the topological relationship between the two spatial geometric regions is a disjoint relationship, the corresponding spatial topological index is 0.

[0080] S33. Traverse all global observation results participating in the construction of the catalog, and calculate the spatial measurement index S2 between the spatial geometric areas of any two global observation results.

[0081] Spatial measurement indicators can numerically measure the mutual relationship of metadata spatial geometry, including area overlap, spatial distance ratio, etc., to provide more accurate numerical indicators for spatial topological relationships. Define spatial distance as the distance between the geometric centers of spatial geometric regions. The center of the minimum enclosing circle of the spatial geometric region is actually used as its geometric center, and the geometric radius is the minimum enclosing circle radius of the spatial geometric region. Define spatial measurement indicator S2, use area overlap indicator and distance measurement indicator to calculate the indicator, use the area overlap indicator calculation formula, and use the inverse of the center distance as the distance measurement indicator to constrain its range to (0,1), which can be in the same measurement standard as other indicators. In this embodiment, the calculation formula of the above-mentioned spatial measurement indicator S2 is:

[0082] S2=W s21 *S 21 +W s22 *S 22

[0083] S 21 =2*S s / (S a +S b )

[0084] S 22 =(R a +R b ) / (D s +R a +R b )

[0085] Where: s21 and W s22 are weight values, S 21 and S 22 are the area overlap index and the distance metric, both of which range from [0,1]; S s is the overlapping area between two spatial geometric regions, D s is the minimum distance between the centers of the outer circles of two spatial geometric regions, S a and S b are the areas of two spatial geometric regions, R a and R b are the minimum enclosing circle radii of two spatial geometric regions respectively.

[0086] S34. For any two global observation results, calculate the spatial similarity S between them. sim (SpatialSimilarity), which is used to reflect the similarity of metadata spatial tags.

[0087] In this embodiment, the spatial similarity S is calculated sim The spatial dimension association model is as follows:

[0088] S sim =W s1 *S1+W s2 *S2

[0089] Where: W s1 and W s2 All are weight values.

[0090] S4. Establish an association model from the topic dimension, map the topic label of each global observation result into a word vector through a word vector model, and calculate the topic similarity between any two global observation results based on the word vector.

[0091] In this embodiment, the specific implementation steps of the above step S4 are as follows:

[0092] S41. For each global observation result, obtain its corresponding topic label, and use the pre-trained word vector model to encode each keyword in each topic label to form a word vector.

[0093] In the step, a neural network can be used to train a corpus based on the Word2Vec method to obtain a word vector, which is applied to the similarity calculation of keywords and keyword sets. The neural network structure of Word2Vec is three-layer, including an input layer, a hidden layer, and an output layer. Its training data is a corpus based on natural language used by humans. The corpus can be any meaningful human language topic. For the verification of the input layer and the output layer, Word2Vec has two models, namely the CBOW model and the Skip-gram model. The CBOW model predicts the target word based on the context, while the Skip-gram predicts the context characters based on the target word. The input layer data is the one-hot word vector obtained by preprocessing the corpus. For different languages ​​and corpus forms, the data preprocessing method is different. Generally, starting from the natural topic, a word list that retains the original topic order is obtained, and then the dictionary size containing all words is counted as the dimension of the one-hot encoding. Initially, each word is randomly assigned a vector value to obtain a one-hot word vector as the input layer data. The input layer of the CBOW model is a word vector of C dimensions V in the context, that is, a matrix of size C*V. The weight matrix sizes from the input layer to the shadow hidden layer and from the hidden layer to the output layer are V*N and N*V respectively, N is the dimension of the artificially defined dense vector, and the output layer is a V-dimensional one-hot word vector; the input layer of Skip-gram is a V-dimensional one-hot word vector, the intermediate weight matrix sizes are V*N and N*V respectively, and the output layer is the predicted C context one-hot word vector. The loss function is generally a cross entropy function, and the weight matrix is ​​updated using the gradient descent method. After the training is completed, the weight matrix W is obtained. The one-hot of each word can be multiplied by the matrix to obtain its own dense word vector, which can be used to record and represent the position of the word in the vector space. In this embodiment, the pre-trained word vector model preferably adopts the Word2Vec model based on Skip-gram.

[0094] In addition, since the corpus cannot contain all subject dictionaries, or when the word to be queried is a compound word, the problem of words not being in the vocabulary (out-of-vocabulary, OOV) will be encountered during natural language processing. The core of the OOV problem is that the word vector corresponding to the word cannot be found in the dictionary. FastText uses the N-grams entry to obtain new words, that is, to segment the new words, extract meaningful fields, and obtain the average word vector for the parts that can be found in the dictionary, which is the word vector of the new word. The FastText method uses the average word vector of the known words associated with the unknown words to represent the word vector of the unknown word, which solves the OOV problem to a certain extent. Therefore, in this embodiment, FastText can also be used to process the out-of-vocabulary (OOV) problem when the keywords are encoded with word vectors by the pre-trained word vector model.

[0095] S42, traverse all global observation results participating in the directory construction, for any two global observation results' keyword sets, first pair the keywords in the two sets according to the principle of highest similarity based on word vectors, then calculate the similarity of each pair of keyword word vectors, and use the average similarity of all keyword pairs as the topic similarity F between the two global observation results in the topic dimension association model sim .

[0096] In practical applications, the subject label of each global observation result is a set of keywords, which is extracted from the global observation result metadata containing data keyword information. The extracted keyword information can reflect the subject content represented by the data and is used to match the content of the data set.

[0097] Some subject information in metadata is expressed in the keyword field, and some contains words related to the subject content in the data title. The keyword set is extracted and stored. According to the principle of similarity between two keyword sets: the total subject relevance between two keyword sets of global observation metadata can be determined by pairing the keywords with the highest mutual relevance and finding their average relevance. Use the extracted keyword set to measure and calculate the subject similarity between metadata, and use the Gensim library to load the trained word vector model. First, screen each pair of keywords with the highest matching degree between the two keyword sets, and calculate the subject similarity of each pair of keywords after pairing, and find the average similarity. For example, the first keyword set A contains M keywords, and the second keyword set B contains N keywords (M and N can be equal or unequal). First, for each keyword in A, find the keyword with the highest word vector matching degree in B, and pair the two to form a group of keyword pairs; for a total of M groups of keyword pairs, calculate the average similarity of their word vectors, and the average similarity is the subject similarity F between the two global observation results.sim .

[0098] S5. Traverse all global observation results participating in the construction of the directory. For any two global observation results, perform weighted summation of the corresponding time similarity, spatial similarity, and topic similarity and use them as the mixed correlation similarity between the two, thereby obtaining the label relationship between all global observation results.

[0099] In this embodiment, in the above step S5, the hybrid correlation similarity between any two global observation results is calculated by adopting the "time-space-topic" hybrid correlation model, and the calculation formula is as follows:

[0100] Geo sim =W s *S sim +W t *T sim +W f *F sim

[0101] Where: W s is the spatial similarity S sim The weight, W t is the temporal similarity T sim The weight, W f is the topic similarity F sim The above three weights can be adjusted according to actual conditions. In this embodiment, the weight of the given topic similarity is preferably 0.41, the weight of the spatial similarity is 0.35, and the weight of the temporal similarity is 0.24.

[0102] Therefore, through the above steps S1 to S5, the label relationship between all global observation results can be obtained, and the data foundation for similarity analysis can be established for building a three-dimensional network directory system. Figure 2 As shown, it is a schematic diagram of the principle of the above steps S1 to S5. However, it should be noted that the technical solution in the present invention is mainly to calculate the correlation between all global observation results involved in the construction of the three-dimensional network directory system, and further constructing a three-dimensional network directory system of global observation results based on the correlation is a specific application in the next step, which is not the focus of the present invention. Figure 2 As shown, the label relationship between all global observation results obtained in the above steps S1 to S5 can be used for semantic retrieval, and can also be used for association network analysis and visualization.

[0103] Furthermore, the label relationships between all the above global observation results can be stored through a directed graph for easy retrieval in subsequent applications. Specifically, all global observation results participating in the construction of the directory can be used as nodes of the graph network, and the edge lengths between the nodes in the graph network are set to be inversely proportional to the mixed association similarity between the corresponding two global observation results, thereby obtaining a relationship network of all global observation results for similarity retrieval or visualization. Preferably, the edge lengths between the nodes in the graph network can be set to the inverse of the mixed association similarity between the two global observation results.

[0104] The visualization form can intuitively understand the distribution and aggregation of global observation metadata. Of course, when performing visualization, it is not necessary to select all global observations to participate in the visualization. Representative clusters and scattered data can be selected based on the same selection rules to form metadata samples, perform global correlation calculations, and establish a representative metadata association network.

[0105] In retrieval applications, the global observation results metadata relationship network established above can be used as the basis for retrieval, which is essentially equivalent to a recommendation task based on a graph network.

[0106] In addition, based on the label relationship between all global observation results obtained above, it can be used to dynamically expand and update the three-dimensional network directory system. For example, when updating and expanding the directory system, the correlation degree corresponding to the correlation relationship between results or labels can be calculated by the above method. If the correlation degree of a certain correlation relationship exceeds the threshold during several consecutive updates of the directory system, it is necessary to adjust its correlation relationship in the directory system.

[0107] In one embodiment of the present invention, the global observation metadata obtained by the National Earth System Science Data Center is used as the source data, and the similarities between some global observation results calculated according to the above steps S1 to S5 are as follows: Figure 3 As shown in . Taking the metadata of each global observation result as a node, the edge length between nodes in the graph network is set to be inversely proportional to the mixed association similarity between the corresponding two global observation results. The same selection rule is used to select representative clustering and scattered data to form 150 metadata samples, and the global mixed association similarity calculation is performed to establish a metadata association network, as shown in Figure 4 Through this network, global observation results can be visualized and their clustering and distribution can be analyzed, so as to facilitate the subsequent construction of a three-dimensional network directory system.

[0108] The above-described embodiment is only a preferred solution of the present invention, but it is not intended to limit the present invention. A person skilled in the relevant technical field may make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, any technical solution obtained by equivalent replacement or equivalent transformation falls within the protection scope of the present invention.

Claims

1. A method for establishing label relationships of a three-dimensional network catalog system of global observation results, characterized in that: Its steps are as follows: S1. Extract the metadata of all global observation results participating in the catalog construction to obtain the time label, space label and subject label of each global observation result; the time label is the data time information corresponding to the global observation result, the space label is the place name information corresponding to the global observation result, and the subject label is the subject keyword set corresponding to the global observation result; S2. Establish an association model from the time dimension, calculate the time topology index and time distance index of any two global observation results based on the time tag, and calculate the time similarity between any two global observation results by weighting the time topology index and the time distance index; S3. Establish an association model from the spatial dimension, map the spatial label of each global observation result into a spatial geometric area under a unified spatial coordinate system through place name address mapping, and calculate the spatial topological index and spatial metric index of any two global observation results, and then calculate the spatial similarity between any two global observation results by weighting the spatial topological index and spatial metric index; S4. Establish an association model from the topic dimension, map the topic label of each global observation result into a word vector through a word vector model, and calculate the topic similarity between any two global observation results based on the word vector; S5. Traverse all global observation results participating in the construction of the directory. For any two global observation results, perform weighted summation of the corresponding time similarity, spatial similarity, and topic similarity and use them as the mixed correlation similarity between the two, thereby obtaining the label relationship between all global observation results.

2. The method for establishing label relationships of a three-dimensional network directory system of global observation results according to claim 1, characterized in that: The specific implementation steps of S2 are as follows: S21. For each global observation result, obtain its corresponding time tag, and extract the time type, time length and starting time point from the time tag through the pre-defined time class; The time types shown are divided into two categories: time points and time periods. The time type with a time length less than or equal to the time length threshold is defined as a time point, and the time type with a time length greater than the time length threshold is defined as a time period. S22. Traverse all global observation results participating in the construction of the directory, and calculate the time topology index T1 between the time classes of any two global observation results according to the time topology rule; in the time topology rule: if the time points are equal, the corresponding time topology index is 1; if the time points are not equal, the corresponding time topology index is 0; if the time points are included in the relationship between time points and time periods, the corresponding time topology index is 1; if the time points are not included in the relationship between time points and time periods, the corresponding time topology index is 0; if the time periods are equal or included in the relationship between time periods, the corresponding time topology index is 1; if the time periods are separated, the corresponding time topology index is 0; if the time periods are intersecting, the corresponding time topology index calculation formula is: T1=2*O t / (T a +T b ) Where: O t is the length of the overlap between time periods, T a , T b are two time period lengths respectively; S23, traverse all global observation results participating in the directory construction, and calculate the time distance index T2 between the time classes of any two global observation results. The calculation formula is: T2=(T a +T b ) / (D t +T a +T b ) Among them, D t is the center distance between the time classes corresponding to the two global observation results; S24. For any two global observation results, calculate the temporal similarity T between them. sim : T sim =W t1 *T1+W t2 *T2 Where: W t1 and W t2 All are weight values.

3. The method for establishing label relationships of a three-dimensional network directory system of global observation results according to claim 2, characterized in that: The duration threshold is 1 year.

4. The method for establishing label relationships of a three-dimensional network directory system of global observation results according to claim 1, characterized in that: The specific implementation steps of S3 are as follows: S31. For each global observation result, obtain its corresponding spatial label, and map the place name information in the spatial label to a unified spatial coordinate system according to the place name-spatial geometry mapping relationship to obtain the spatial geometry area corresponding to each global observation result; S32, traverse all global observation results participating in the construction of the directory, and calculate the spatial topological index S1 between the spatial geometric regions of any two global observation results according to the spatial topological rules; in the spatial topological rules: if the topological relationship between the two spatial geometric regions is an equal relationship, a containment relationship, or a contained relationship, the corresponding spatial topological index is 1; if the topological relationship between the two spatial geometric regions is an overlapping relationship, the corresponding spatial topological index is 0.8; if the topological relationship between the two spatial geometric regions is a connected relationship, the corresponding spatial topological index is 0.5; if the topological relationship between the two spatial geometric regions is a separated relationship, the corresponding spatial topological index is 0; S33, traverse all global observation results participating in the construction of the catalog, and calculate the spatial metric index S2 between the spatial geometric areas of any two global observation results. The calculation formula is: S2=W s21 *S 21 +W s22 *S 22 S 21 =2*S s / (S a +S b ) S 22 =(R a +R b ) / (D s +R a +R b ) Where: W s21 and W s22 are weight values, S 21 and S 22 are the area overlap index and the distance metric index respectively; S s is the overlapping area between two spatial geometric regions, D s is the minimum distance between the centers of the outer circles of two spatial geometric regions, S a and S b are the areas of two spatial geometric regions, R a and R b are the minimum enclosing circle radii of two spatial geometric regions respectively; S34. For any two global observation results, calculate the spatial similarity S between them. sim : S sim =W s1 *S1+W s2 *S2 Where: W s1 and W s2 All are weight values.

5. The method for establishing label relationships of a three-dimensional network directory system of global observation results according to claim 4, characterized in that: Before mapping the spatial labels of global observation results into spatial geometric areas, the place name information in the spatial labels is converted between common place names and standard place names in advance, and the common place names are converted into standard administrative district names or sea area names through similarity matching or a pre-built mapping relationship table.

6. The method for establishing label relationships of a three-dimensional network directory system of global observation results according to claim 1, characterized in that: The specific implementation steps of S4 are as follows: S41. For each global observation result, obtain its corresponding topic label, and use the pre-trained word vector model to encode each keyword in each topic label to form a word vector; S42, traverse all global observation results participating in the directory construction, for any two global observation results' keyword sets, first pair the keywords in the two sets according to the principle of highest similarity based on word vectors, then calculate the similarity of each pair of keyword word vectors, and take the average similarity of all keyword pairs as the topic similarity F between the two global observation results sim .

7. The method for establishing label relationships of a three-dimensional network directory system of global observation results according to claim 6, characterized in that: The pre-trained word vector model is a Word2Vec model based on Skip-gram.

8. The method for establishing label relationships of a three-dimensional network directory system of global observation results according to claim 6, characterized in that: When encoding keywords using a pre-trained word embedding model, FastText is used to handle the out-of-vocabulary (OOV) problem.

9. The method for establishing label relationships of a three-dimensional network directory system of global observation results according to claim 1, characterized in that: In S5, the formula for calculating the mixed correlation similarity between any two global observation results is as follows: Geo sim =W s *S sim +W t *T sim +W f *F sim Where: W s is the spatial similarity S sim The weight, W t is the temporal similarity T sim The weight, W f is the topic similarity F sim The weight of .

10. The method for establishing label relationships of a three-dimensional network directory system of global observation results according to claim 1, characterized in that: All global observation results participating in the construction of the directory are taken as nodes of the graph network, and the edge length between the nodes in the graph network is set to be inversely proportional to the mixed correlation similarity between the corresponding two global observation results, so as to obtain a relationship network of all global observation results for similarity retrieval or visualization.

Images