A data product query aggregation method based on a data grid

Abstract

The application discloses a data product query aggregation method based on a data grid, and is characterized in that the method comprises the following steps: step 1, data product tree structure creation, including metadata design and tree structure management; step 2, decomposition query, based on the tree structure and the metadata, the data product query is decomposed into a structured query and an unstructured query, and parallel execution of the query is performed; step 3, query aggregation, multi-party query and the results of the corresponding sub-queries are converged, and are arranged and returned from high to low according to matching similarity. The application carries out fine design on metadata of a data product, proposes structured attributes and unstructured attributes of the metadata, and constructs a blood relationship of the data product into a tree structure for management. The application proposes a novel and innovative scheme for metadata design of the data grid and association management of the data product, especially greatly reduces a query range for the data product query, improves query efficiency, and is intuitive and expandable.

Description

Technical Field

[0001] This invention relates to the field of big data technology, and in particular to a data product query and aggregation method based on data grids. Background Technology

[0002] A data grid is a distributed computing and data management architecture. Based on domain ownership of data and the principle of data as a product, the data grid has a dynamically expandable topology. In a data grid, each domain manages its own business data and builds data products with domain insights. The data product is the smallest architectural quantum in the data grid, encapsulating the corresponding analytical data and the infrastructure required to run the data product.

[0003] As the scale of data grids continues to expand, queries of data products distributed across different fields will face performance challenges. The following problems exist with data product queries within data grids: (1) Data product queries are based on data product management, requiring an intuitive data product association management method; (2) Queries of products within the data grid lack an efficient implementation based on specific topological relationships.

[0004] Therefore, it is necessary to provide a data product query aggregation method based on data grids to solve the above-mentioned technical problems. Summary of the Invention

[0005] The purpose of this section is to outline some aspects of the embodiments of the present invention and to briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section, as well as in the abstract and title of the present application, to avoid obscuring the purpose of this section, the abstract and title of the invention. Such simplifications or omissions shall not be used to limit the scope of the present invention.

[0006] Therefore, the technical problem to be solved by this invention is how to construct a hierarchical management structure based on the distributed data architecture characteristics of data grids and the data flow dependencies between data products, so as to achieve efficient querying of data products.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a data product query and aggregation method based on data grids, characterized by comprising the following steps:

[0008] Step 1: Create the data product tree structure, including metadata design and tree structure management;

[0009] Step 2: Decompose the query. Based on the tree structure and metadata, the data product query is decomposed into structured query and unstructured query, and the queries are executed in parallel.

[0010] Step 3: Query aggregation. The results of multiple queries and corresponding subqueries are aggregated and returned in descending order of matching similarity.

[0011] As a preferred embodiment of the data product query and aggregation method based on data grid of the present invention, step 1, data product tree structure creation, proposes a new optimization scheme for metadata design and constructs the tree structure of data products as an attribute of metadata. Specifically, it includes step 1-1 metadata design and step 1-2 tree structure management.

[0012] As a preferred embodiment of the data product query and aggregation method based on data grid of the present invention, the metadata design in step 1-1 divides the data product metadata into two levels: a structured level and an unstructured level.

[0013] (1) Structured level

[0014] The metadata at the structural level consists of quantifiable and directly queried information. This level includes the following information:

[0015] {

[0016] Data types

[0017] Numerical range

[0018] Data volume (dimensions)

[0019] Time range (timestamp)

[0020] Geographic location information

[0021] (Field) Industry Classification

[0022] Downstream data product tree

[0023] Data permissions

[0024] Other necessary quantifiable attributes

[0025] }

[0026] (2) Unstructured level

[0027] The non-structural level, also known as the semantic description and understanding level, contains the following information:

[0028] {

[0029] Keywords

[0030] Data source description

[0031] Data usage description

[0032] Domain Description

[0033] }

[0034] As a preferred embodiment of the data product query and aggregation method based on data grid of the present invention, step 1-2, tree structure management, constructs the data dependencies between data products into a decision tree structure data model. Specifically, the method for constructing the data dependency tree structure for each data product in the data grid includes the following:

[0035] (1) Initialize the empty subtree;

[0036] (2) Use the current data product node as the root node of the current subtree;

[0037] (3) Backtrack to find the predecessor node of the current node, which is the subordinate child node of the root node;

[0038] (4) Backtrack step by step, adding subsequent child nodes for each child node;

[0039] (5) The current subtree is completed when all leaf nodes are backtracked;

[0040] (6) Fill the “Downstream Data Product Tree” description in the metadata of the data product with the complete tree structure constructed.

[0041] As a preferred embodiment of the data product query and aggregation method based on data grid of the present invention, step 2, decomposing the query, is implemented in parallel by decomposing the query of data products through the already constructed data product tree structure and metadata. The method for querying data products is as follows:

[0042] (1) Constructing queries: including structured query data and unstructured query data;

[0043] (2) Query decomposition: Decompose the query into multiple queries and process them in parallel;

[0044] (3) Execute subqueries: perform subtree queries based on the complete tree hierarchy.

[0045] As a preferred embodiment of the data product query aggregation method based on data grid of the present invention, step 3, query aggregation, gathers the results of multiple queries and corresponding subqueries, and returns them in descending order of matching similarity. Specifically, in the query of each data product, the matching degree of structured queries is s1, and the matching degree of unstructured queries is s2. Then, the matching score S of the data product is:

[0046] S = α*s1 + β*s2;

[0047] The ranges of s1, s2, α and β are all [0,1], and α+β=1 is used as a proportional parameter, with the default α=β=0.5;

[0048] Optionally, α and β can be adaptively adjusted as weighting parameters based on the actual mesh effect:

[0049] (1) During grid operation, the original ratio parameters α and β are set to 0.5 by default, which means that the filtering ratio of structured and unstructured queries is the same;

[0050] (2) Randomly select a query, and uniformly sample α and β, with a value range of [a, b] and a value interval of m. The default value of a is 0.1, the default value of b is 0.9, and the default value of m is 0.1.

[0051] (3) Grid search sorting based on uniform sampling;

[0052] (4) After sorting each weight ratio, obtain the top N candidate results (N defaults to 1);

[0053] (5) The results obtained from multiple weight ratios are summarized and deduplicated. For the same candidate value, only the candidate value whose corresponding α and β are closest to 0.5 is retained.

[0054] (6) Return the deduplicated results to the user;

[0055] (7) Record the user's click confirmation event and record the corresponding weight ratio;

[0056] (8) After multiple simulations, the weight parameter selected most often or the average value of the corresponding parameter clicked multiple times will be used as the optimal scaling parameter to update the grid system;

[0057] (9) As the grid continues to expand and the metadata continues to be enriched, dynamic updates will be carried out periodically.

[0058] The beneficial effects of this invention are as follows: The data product query aggregation method based on data grid proposed in this invention constructs a tree-structured management model between data products based on the data dependencies between them. It designs the metadata of data products into structured and unstructured attributes for subsequent data queries. For specific query requirements, it decomposes and executes parallel subqueries, and finally returns the results of multiple subqueries arranged according to their matching degree. The advantages are as follows:

[0059] (1) Tree-structured data dependency representation has a clear and intuitive hierarchical structure and is suitable for dynamic data grid topology relationships;

[0060] (2) The structured attributes of metadata are used for quick filtering, and the unstructured attributes are used for similarity matching. The structured query results and the unstructured matching results are considered at the same time to improve the overall matching performance.

[0061] (3) Layer-by-layer data query and narrowing of the query subtree range greatly improve query efficiency. Attached Figure Description

[0062] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0063] in:

[0064] Figure 1 A flowchart illustrating a data product query and aggregation method based on a data grid, as provided in an embodiment of the present invention;

[0065] Figure 2 This invention provides a topology diagram of the data grid components in an embodiment of the data grid-based data product query and aggregation method.

[0066] Figure 3 This invention provides a schematic diagram illustrating the upstream and downstream data dependencies of a data product based on a data grid-based data product query and aggregation method, as illustrated in one embodiment.

[0067] Figures 4a-4d A tree structure management example diagram of the data product query aggregation method based on data grid provided in one embodiment of the present invention;

[0068] Figures 5a-5c This is a query decomposition diagram of a data product query aggregation method based on a data grid, provided by an embodiment of the present invention;

[0069] Figure 6 This is a schematic diagram of a data product node in a data grid-based data product query and aggregation method provided by an embodiment of the present invention. Detailed Implementation

[0070] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0071] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0072] Example 1

[0073] Reference Figure 1-6 The first embodiment of the present invention provides a data product query and aggregation method based on a data grid, characterized by comprising the following steps:

[0074] Step 1: Create the data product tree structure, including metadata design and tree structure management;

[0075] Step 2: Decompose the query. Based on the tree structure and metadata, the data product query is decomposed into structured query and unstructured query, and the queries are executed in parallel.

[0076] Step 3: Query aggregation. The results of multiple queries and corresponding subqueries are aggregated and returned in descending order of matching similarity.

[0077] The specific implementation method includes the following: Step 1: Data product tree structure creation, including metadata design and tree structure management;

[0078] like Figure 2 As shown, the data grid, as a distributed data management architecture, most intuitively displays its distributed dynamic topology as domain-based data ownership and data as a product. The data grid incorporates four fundamental principles: federated computing governance and self-service data service platforms embody global rules and operational infrastructure; domain-based ownership divides the organization into multiple autonomous data domains; and data domains flow between each other through the global minimum architectural quantum—the data product. Figure 3 As shown, data products themselves rely on domain data processing to obtain analytical data that can directly provide services. Multiple data products can also be cascaded through subscription to build upper-level data products. For example, if a business line needs to build a machine learning prediction model with several related data, it needs to use data variables from multiple underlying data products as input.

[0079] Step 1-1: Metadata Design

[0080] When data products need to interact, a data contract is required to confirm the information about the data products. A data contract includes metadata information related to the data itself, such as associated data tables, as well as service information as the product itself, such as service level indicators. The data contract is essentially a user manual for the data product, helping downstream data products decide whether to subscribe. Metadata is crucial for establishing relationships between data products; therefore, the necessary attributes for the metadata structure must be designed first.

[0081] The metadata of the data product designed in this invention is divided into two levels: a structured level and an unstructured level.

[0082] (3) Structured level

[0083] The metadata at the structural level consists of quantifiable and directly queried information. This level includes the following information:

[0084] {

[0085] Data types

[0086] Numerical range

[0087] Data volume (dimensions)

[0088] Time range (timestamp)

[0089] Geographic location information

[0090] (Field) Industry Classification

[0091] Downstream data product tree

[0092] Data permissions

[0093] Other necessary quantifiable attributes

[0094] }

[0095] (4) Unstructured level

[0096] The non-structural level, also known as the semantic description and understanding level, contains the following information:

[0097] {

[0098] Keywords

[0099] Data source description

[0100] Data usage description

[0101] Domain Description

[0102] }

[0103] The above list of some necessary attributes is intended to illustrate the differences between structured and unstructured metadata and to provide suggested values. These can be reduced or removed based on actual needs during the actual development of the grid.

[0104] Steps 1-2: Tree Structure Management

[0105] One native way to represent the relationships between data products is based on the data lineage description contained in the metadata of the data contract. The current data product, as a descendant node of the upstream data product, contains cascading analytical data. This is a representation and description of the data flow from the data source to the data product, such as... Figure 3 From left to right. This method models the structural relationship between data products using reverse data flow, that is, from... Figure 3 Data dependency backtracking is performed from right to left. The data flow between data products or the association between data products in the data grid is an intuitive directed graph structure, and the data products are basically directed acyclic graph structures. Therefore, the data dependencies between data products can be constructed as a decision tree structure data model.

[0106] For each data product in the data grid, the specific methods for constructing the data dependency tree structure include the following:

[0107] (1) Initialize the empty subtree;

[0108] (2) Use the current data product node as the root node of the current subtree;

[0109] (3) Backtrack to find the predecessor node of the current node, which is the subordinate child node of the root node;

[0110] (4) Backtrack step by step, adding subsequent child nodes for each child node;

[0111] (5) The current subtree is completed when all leaf nodes are backtracked;

[0112] (6) Fill the “Downstream Data Product Tree” description in the metadata of the data product with the complete tree structure constructed.

[0113] like Figures 4a-4d The image shows an example of a data product dependency tree representation formed by the tree management method of this invention. The grid currently contains data products {A, B, C, D, E, F, G, H, I}. The relationships between the data products are represented by right arrows '>', 'X>Y' represents data product X depending on data product Y, and square brackets '[]' represent parallel branching nodes. Several possible relationships are listed below:

[0114] (1) Figure 4a Independent data product nodes

[0115] (2) Figure 4b H>A>B>C

[0116] (3) Figure 4c :F>[D][E]

[0117] (4) Figure 4d : G>[B>C][F>[D][E]][I]

[0118] As described above, only a few possible tree structure descriptions are listed. In actual grid implementation, the tree structure is constructed based on the full data products. This is a superior description method for dynamic topological relationships in data grids, which are based on data products as the architectural quantum, and has significant advantages.

[0119] Clear hierarchical relationships: The tree structure clearly illustrates the hierarchical relationships between data products. Each data product can be associated with multiple sub-data products, forming a hierarchical structure that helps in understanding the organizational relationships between data.

[0120] Easy to manage and maintain: The tree structure makes managing and maintaining data products easier. The tree structure allows for tracking and locating specific data products, facilitating the organization and management of large amounts of data.

[0121] Step 2: Decompose the query

[0122] Managing data dependencies between data products using a tree structure and writing them into the structured attributes of metadata serves two purposes: firstly, to provide a clear hierarchical representation of the dynamic topology of distributed data products and facilitate maintenance and management; and more importantly, to significantly promote the querying of data products.

[0123] Queries targeting data products can be divided into two levels: the application level and the construction level. The application level primarily targets the direct users of the data product, such as users who only need the direct analytical results of a specific data product. The construction level, on the other hand, is geared towards the builders of the data product, such as domain engineers. This invention focuses on the construction level.

[0124] When building a new data product, it is often necessary to rely on data from other domains. Data grids are divided by domains and each domain manages a large amount of related data within the current business scope. Therefore, the ability to quickly locate a domain and further locate a related data product within that domain is crucial for the development of data products in a data grid architecture.

[0125] Based on the data product tree structure management of this invention, the following method for data product query is proposed:

[0126] (1) Constructing a query

[0127] The query construction step involves clarifying the required data source information. This also includes two levels of query information construction, consistent with the metadata design: structured query data and unstructured query data.

[0128] Structured query data limits the quantifiable and filterable data information, such as data types being integers and time ranges from 2022 to 2023.

[0129] Unstructured query data consists of descriptive information, such as "data on patient information from a certain hospital".

[0130] (2) Query decomposition

[0131] Modeling data products typically involves multiple data sources. For example, if a current data product needs to predict health indicators using height and weight data, its machine learning model can be simply described as Z = aX + bY + c, where X represents height data, Y represents weight data, a, b, and c are empirical parameters, and Z is the prediction indicator. This requires output data from two upstream data products. In actual development, data requirements are more complex. To improve query efficiency and accuracy, the above query is broken down into multiple queries processed in parallel.

[0132] Specifically, X and Y can be decomposed into two first-level queries, that is, queries can be constructed separately for the data requirements of X and Y.

[0133] Optionally, the scope of the first-level query can be restricted based on the current domain's permission scope, i.e., the query scope can be narrowed and a new subquery can be constructed.

[0134] (3) Execute subquery

[0135] The multi-faceted subqueries have been constructed, and the tree structure can be used to further optimize query space and efficiency. For example... Figure 5a This is a topology diagram for querying data products without a tree structure. In the native data grid architecture, although there are relationships between data products, for query operations, all data products belong to the same dimension. Data search requires filtering from the metadata of all N data products. As the scale of the data grid continues to increase, the distributed topology puts pressure on the network and efficiency of queries.

[0136] Tree structure management based on the present invention, such as Figure 5b As shown, the data product subtree structure in the entire grid can be summarized into a complete data grid-level tree structure and hierarchically managed. All subqueries can be performed layer by layer, and when the relevant subtree is obtained, parallel queries of the subtrees can be performed again, which greatly improves efficiency.

[0137] Specifically, structured queries with subqueries can be filtered directly, while unstructured queries with subqueries can be filtered using similarity. The specific implementation technology is not limited; for example, the Elasticsearch search engine can perform both structured and unstructured queries simultaneously.

[0138] like Figure 5c Therefore, we can combine practical considerations to optimize subtree queries in multiple ways:

[0139] (1) At the query level, reduce the scope of query subtrees.

[0140] For example, based on role permissions, the query subtree for which the user lacks permission can be deleted; based on domain classification, only the query tree for the relevant domain can be retained.

[0141] (2) Tree construction level, multi-level query information aggregation

[0142] For the complete tree structure of all data products at the data grid level, the data descriptions of each layer containing data products are summarized, and the total descriptive feature vector of that layer is extracted. When executing subqueries, the initial query level can be located based on vector matching first. For example, if the query is located at level 3, a breadth-first search can be performed from the level 3 node downwards to further reduce the query scope and improve efficiency. For actual business logic, a depth-first search can also be performed after aggregation and extraction based on the number of tree branches.

[0143] Step 3: Query Aggregation

[0144] The results of multiple queries and their corresponding subqueries are aggregated and returned in descending order of matching similarity. Specifically, for each data product, if the matching degree of structured queries is s1 and the matching degree of unstructured queries is s2, then the matching score for that data product is:

[0145] S=α*s1+β*s2

[0146] The ranges of s1, s2, α and β are all [0,1], and α+β=1 is used as a proportional parameter, with the default α=β=0.5;

[0147] In this invention, α and β, as weight adjustment parameters, can be adaptively adjusted according to the actual mesh effect, such as by performing A / B testing to select the optimal combination. Optionally, this invention provides an implementation method, such as... Figure 6 As shown, ABCDE represent data products, specifically:

[0148] (1) During grid operation, the original ratio parameters α and β are set to 0.5 by default, which means that the filtering ratio of structured and unstructured queries is the same;

[0149] (2) Randomly select a query, and uniformly sample α and β, with a value range of [a, b] and a value interval of m. The default value of a is 0.1, the default value of b is 0.9, and the default value of m is 0.1.

[0150] (3) Grid search sorting based on uniform sampling;

[0151] (4) After sorting each weight ratio, obtain the top N candidate results, where N is 1 by default;

[0152] (5) The results obtained from multiple weight ratios are summarized and deduplicated. For the same candidate value, only the candidate value whose corresponding α and β are closest to 0.5 is retained.

[0153] (6) Return the deduplicated results to the user;

[0154] (7) Record the user's click confirmation event and record the corresponding weight ratio;

[0155] (8) After multiple simulations, the weight parameter selected most often or the average value of the corresponding parameter clicked multiple times will be used as the optimal scaling parameter to update the grid system;

[0156] (9) As the grid continues to expand and the metadata continues to be enriched, dynamic updates will be carried out periodically.

[0157] Domain engineers select appropriate data for subscription development based on the query results. Once development is complete, the tree structure of the current data product is automatically built and populated back into the metadata of that data product, while simultaneously updating the complete tree structure information at the data grid level.

[0158] It is worth noting that each data product retrieved is a node in the complete data grid tree, and the upstream and downstream flow relationships are intuitive. Domain engineers can use the upstream data products of the subscribed data products for in-depth data tracing or the downstream data products for upper-layer application development as a reference.

[0159] In summary, this invention proposes a data product query and aggregation method based on a data grid. It features meticulous design of data product metadata, proposing structured and unstructured attributes for metadata and managing the lineage relationships of data products as a tree structure. For data product development, it decomposes data products into multi-party sub-queries, utilizing the complete tree structure within the constructed data grid for layer-by-layer structured query filtering and unstructured query matching. Furthermore, it proposes optimization schemes to narrow the query scope. The results of the multi-party sub-queries are returned based on similarity ranking, allowing domain engineers to subscribe and develop as needed. This invention offers novel and innovative solutions for data grid metadata design and data product association management, significantly narrowing the query scope and improving query efficiency, especially for data product queries. It is intuitive and scalable.

[0160] It should be understood that numerous specific implementation decisions can be made during the development of any practical implementation, such as in any engineering or design project. Such development efforts may be complex and time-consuming, but for those skilled in the art who benefit from this disclosure, the development effort will be a routine work of design, manufacturing, and production without requiring much experimentation.

[0161] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A data product query and aggregation method based on data grids, characterized in that, Includes the following steps: Step 1: Create the data product tree structure, including metadata design and tree structure management; Step 2: Decompose the query. Based on the tree structure and metadata, the data product query is decomposed into structured query and unstructured query, and the queries are executed in parallel. Step 3: Query aggregation, which combines the results of multiple queries and their corresponding subqueries, and returns them in descending order of matching similarity; Step 3, query aggregation, gathers the results of multiple queries and their corresponding subqueries, and returns them arranged in descending order of matching similarity. Specifically, for each data product, the matching degree of structured queries is s1, and the matching degree of unstructured queries is s2. Then, the matching score S of the data product is: S = α*s1 + β*s2; The ranges of s1, s2, α and β are all [0, 1], and α+β=1 is used as a proportional parameter, with the default α=β=0.5; α and β, as weight adjustment parameters, can be adaptively adjusted according to the actual mesh effect: (1) During grid operation, the original ratio parameters α and β are set to 0.5 by default, which means that the filtering ratio of structured and unstructured queries is the same; (2) Randomly select a query, and uniformly sample α and β, with a value range of [a, b] and a value interval of m. The default value of a is 0.1, the default value of b is 0.9, and the default value of m is 0.

1. (3) Grid search sorting based on uniform sampling; (4) After sorting each weight ratio, obtain the top N candidate results, where N is 1 by default; (5) The results obtained from multiple weight ratios are summarized and deduplicated. For the same candidate value, only the candidate value whose corresponding α and β are closest to 0.5 is retained; (6) Return the deduplicated results to the user; (7) Record the user's click confirmation event and record the corresponding weight ratio; (8) After multiple simulations, the weight parameter that is selected most often or the average value of the corresponding parameter after multiple clicks will be used as the optimal ratio parameter to update the grid system; (9) As the grid expands and the metadata becomes richer, it will be dynamically updated periodically.

2. The data product query and aggregation method based on data grid according to claim 1, characterized in that, Step 1, the creation of the data product tree structure, proposes a new optimization scheme for the design of metadata and constructs the tree structure of the data product as an attribute of metadata, including step 1-1 metadata design and step 1-2 tree structure management.

3. The data product query and aggregation method based on data grid according to claim 2, characterized in that, Step 1-1, the metadata design, divides the data product metadata into two layers: a structured layer and an unstructured layer. The structured layer includes metadata that can be quantified and directly filtered and queried. This layer contains the following information: { Data types Numerical range Data volume Time range Geographic location information Industry Classification Downstream data product tree Data permissions } The unstructured layer, also known as the semantic description and understanding layer, contains the following information: { Keywords Data source description Data usage description Domain Description }。 4. The data product query and aggregation method based on data grid according to claim 3, characterized in that, Step 1-2, tree structure management, constructs a decision tree structure data model for the data dependencies between data products. The specific methods for constructing the data dependency tree structure for each data product in the data grid include the following: Initialize an empty subtree; Use the current data product node as the root node of the current subtree; Backtrack to find the predecessor node of the current node, which is a subordinate child node of the root node; Backtrack level by level, adding subsequent child nodes to each child node; The current subtree is completed only after backtracking to all leaf nodes; Fill the "Downstream Data Product Tree" description in the metadata of this data product with the completed tree structure.

5. The data product query and aggregation method based on data grid according to claim 1, characterized in that, Step 2, decomposing the query, is implemented in parallel by decomposing the query of data products using the already constructed data product tree structure and metadata. The method for querying data products is as follows: (1) Query construction: including structured query data and unstructured query data; (2) Query decomposition: Decompose the query into multiple queries and process them in parallel; (3) Execute subqueries: query subtrees level by level based on the complete tree hierarchy.

Images