A geographic data retrieval method based on adaptive indexing and related devices

Abstract

The application relates to the technical field of geographic information retrieval, and provides a geographic data retrieval method based on adaptive indexing and related equipment, which comprises the following steps: dividing a target area into multiple geographic partitions; calculating a data feature portrait of partition data for each partition data respectively, and generating a region-heterogeneous index set of the partition data based on the data feature portrait; analyzing a user retrieval request to obtain a retrieval condition set, and matching multiple candidate partition data from all partition data according to the retrieval condition set and all region-heterogeneous index sets; calculating a comprehensive correlation degree score between each candidate partition data and the user retrieval request according to the data feature portrait of each candidate partition data; and performing merging and deduplication processing on all candidate partition data according to all comprehensive correlation degree scores to obtain final retrieval data corresponding to the user retrieval request. The method can improve the accuracy of geographic data retrieval.

Description

Technical Field

[0001] This application relates to the field of geographic information retrieval technology, and in particular to a geographic data retrieval method and related equipment based on adaptive indexing. Background Technology

[0002] With the development of technologies such as satellite remote sensing, aerial photography, mobile sensing, the Internet of Things, and urban computing, geographic information systems and data warehouses have accumulated a massive amount of multimodal spatiotemporal data, typically including vector feature data, remote sensing image metadata, motion trajectory data, and sample annotation data. These data differ significantly in their organization, spatial scale, temporal characteristics, and semantic annotation systems: vector data covers geographic features with clearly defined geometric extents, such as road networks, building outlines, water system lines and surfaces, and administrative divisions; image metadata records information such as the spatial coverage of remote sensing frames, acquisition time, and sensor type; trajectory data records the continuous spatial and temporal position sequences of moving objects; and sample data is used to support model training and accuracy evaluation.

[0003] In practical applications, user search requests often include multiple constraints, such as spatial scope, temporal scope, semantic concepts, and data types. For example, retrieving vector and trajectory data belonging to the category of transportation facilities in a specific district of a city within a specified year requires joint matching of these multiple dimensions.

[0004] Existing spatiotemporal retrieval methods mainly suffer from the following shortcomings: First, single or low-dimensional indexes struggle to cover all query dimensions. Spatial indexes such as R-trees and their variants, grid indexes, quadtrees, and kd-trees primarily serve the spatial dimension; inverted indexes mainly serve the keyword dimension; and spatial-text composite indexes can support both spatial and keyword dimensions simultaneously, but they typically still struggle to support time-range queries, semantic concept expansion, and cross-modal type joint retrieval. Existing joint retrieval solutions often employ a method of multiple evaluators operating in parallel, progressively reducing the candidate set during the query phase. This mechanism is similar to the conventional approach of multi-level retrieval reduction, and the synonym resources and ontology structures relied upon by each evaluator are common techniques in this field.

[0005] Secondly, there is a discrepancy between lexical annotations and semantic intent. Semantic annotations in spatiotemporal data warehouses are mostly concrete words (such as "road," "railway," and "bridge"), while user queries often use more abstract higher-level semantic concepts (such as "transportation facilities"). This inconsistency between the lexical and conceptual representations limits the effectiveness of semantic retrieval based on inverted indexes.

[0006] Third, a globally unified index is difficult to adapt to spatially heterogeneous distributions, and existing schemes that select index types based on a single scale indicator cannot cover all dimensions of multimodal spatiotemporal retrieval. Multimodal spatiotemporal data exhibits significant differences in different spatial regions: urban core areas have dense road networks, rich semantics, and diverse modalities; suburban areas may only have a few vector elements or sporadic trajectories. Existing methods use the same index structure and fixed partitioning granularity for all regions, resulting in the construction and maintenance of complete multidimensional indexes in regions with sparse data and single modalities, leading to excessive storage and maintenance overhead for semantic inverted indexes, ontology hierarchical structures, etc.; in regions with dense data and highly heterogeneous modalities and semantics, a uniform and fixed partitioning granularity results in too many records within a single grid and reduced distinguishability. Existing approaches determine index types based on the number of records within a partition or a single density indicator, mainly targeting homogeneous or low-dimensional data, and do not address the issue of multimodal data such as vectors, images, trajectories, and samples being indexed on the same 2D basis. n GeoSOT (Geographical Coordinate Global Subdivision Grid with One-Dimensional Integer Coding) n Under the Tree framework, multi-dimensional profiles such as semantic entropy, modality mixing, and temporal density are used to determine whether to build inverted indexes, semantic hierarchical indexes, and type masks. No regional heterogeneous index set with different physical structures is formed during the construction phase.

[0007] Fourth, existing global partitioning and coding schemes such as GeoSOT mainly address the problem of unified spatial addressing, but have not systematically addressed the quantitative description of the differences in data distribution within each partition unit, nor have they provided technical solutions for selecting different index construction strategies and query execution paths based on the above differences under the same addressing framework.

[0008] As can be seen from the above, current geographic data retrieval methods suffer from low accuracy in geographic data retrieval. Summary of the Invention

[0009] This application provides a geographic data retrieval method based on adaptive indexing, which can solve the problem of low accuracy in geographic data retrieval.

[0010] In a first aspect, embodiments of this application provide a geographic data retrieval method based on an adaptive index, the geographic data retrieval method comprising: Divide the target area into multiple geographic partitions and obtain the partition data for each geographic partition; For each partition of data, a data feature profile is calculated, and a set of regional heterogeneous indexes for the partition data is generated based on the data feature profile; the data feature profile is used to describe the data characteristics of the partition data. The user's search request is parsed to obtain a set of search conditions. Based on the set of search conditions and the set of heterogeneous indexes for all regions, multiple candidate partition data are matched from all partition data. Based on the data feature profile of each candidate partition, calculate the comprehensive relevance score between each candidate partition and the user's search request; Based on all comprehensive relevance scores, all candidate partition data are merged and deduplicated to obtain the final retrieval data corresponding to the user's retrieval request.

[0011] Optionally, calculate the data feature profile of the partitioned data, including: Calculate the data density and spatial distribution dispersion of the geographic partitions corresponding to the partitioned data; Multimodal analysis was performed on the partitioned data to obtain the modal hybridity. Keyword distribution analysis was performed on the partitioned data to obtain semantic entropy; By combining data density, spatial distribution dispersion, modal hybridity, and semantic entropy as a set, a data feature profile of the partitioned data can be obtained.

[0012] Optionally, a set of regional heterogeneous indexes for partitioned data is generated based on data feature profiles, including: Calculate the depth of the partitioning hierarchy in the partitioned data; Based on data feature profiling, multiple index types for partitioned data are determined, and data indexes corresponding to each index type are generated based on the depth of the hierarchical division. By treating all data indexes as a set, we obtain a set of regional heterogeneous indexes.

[0013] Optionally, calculate the partitioning level depth of the partitioned data, including: Through the formula: ; Calculate the partitioning level depth of partitioned data ; in, For the amplitude limiting function, Indicates the baseline level. Indicates the number of records in the partition data. The target number of records per leaf grid is the desired value. Indicates the lower limit value of the hierarchy. This indicates the upper limit of the hierarchy.

[0014] Optionally, multiple index types for partitioned data can be determined based on data feature profiles, including: When the number of records in a partition exceeds the preset spatial index threshold, the spatial tree index will be used as an index type. When the number of records in the partition data is less than or equal to the preset spatial index threshold, the flat minimum bounding rectangle array index will be used as an index type. When the semantic entropy of the partitioned data is greater than the preset semantic threshold, the synonym extended inverted index will be used as an index type. When the number of leaf layer concepts in the ontology of the partitioned data exceeds the preset concept threshold, the semantic level index will be used as an index type. When the time density of partitioned data is greater than the preset time threshold, the time interval index will be used as an index type. When the modal heterogeneity of the partitioned data exceeds the preset modal threshold, the data type mask index is used as an index type.

[0015] Optionally, based on the data feature profile of each candidate partition, a comprehensive relevance score between each candidate partition and the user's search request is calculated, including: Through the formula: ; Calculate the overall relevance score ; in, Indicates the total number of index dimensions. Indicates the first Dimensional indexes for candidate partition data Output matching score, Indicates based on candidate partition data The candidate partition data was used to calculate the first The fusion weight of dimensional indexes.

[0016] Secondly, embodiments of this application provide a geographic data retrieval device based on an adaptive index, comprising: The partitioning module is used to divide the target area into multiple geographic partitions and obtain the partition data for each geographic partition. The generation module is used to calculate the data feature profile of each partition data, and generate a set of regional heterogeneous indexes for the partition data based on the data feature profile; the data feature profile is used to describe the data characteristics of the partition data. The parsing module is used to parse user search requests, obtain a set of search conditions, and match multiple candidate partition data from all partition data based on the set of search conditions and the set of heterogeneous indexes of all regions. The calculation module is used to calculate the comprehensive relevance score between each candidate partition data and the user's search request based on the data feature profile of each candidate partition data. The determination module is used to merge and deduplicate all candidate partition data based on all comprehensive relevance scores to obtain the final retrieval data corresponding to the user's retrieval request.

[0017] Thirdly, embodiments of this application provide a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the aforementioned geographic data retrieval method based on adaptive index.

[0018] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned adaptive index-based geographic data retrieval method.

[0019] The above-mentioned solution in this application has the following beneficial effects: In the embodiments of this application, the target area is divided into multiple geographical partitions, and partition data for each geographical partition is obtained. Then, for each partition data, a data feature profile is calculated, and a set of regional heterogeneous indexes for the partition data is generated based on the data feature profile. The user's search request is then parsed to obtain a set of search conditions. Based on the set of search conditions and all regional heterogeneous indexes, multiple candidate partition data are matched from all partition data. Then, based on the data feature profile of each candidate partition data, a comprehensive relevance score between each candidate partition data and the user's search request is calculated. Finally, based on all comprehensive relevance scores, all candidate partition data are merged and deduplicated to obtain the final search data corresponding to the user's search request. Obtaining the partition data of geographical partitions effectively distinguishes geographically related data at the spatial level. Adaptively generating corresponding indexes based on the data features of the partitions avoids low search accuracy caused by a single fixed indexing method. Matching multiple candidate partition data based on the user's search request and determining the final search data by calculating the comprehensive relevance score fully considers the relevance between the user's search request and the partition data, further improving the accuracy of geographical data retrieval.

[0020] Other beneficial effects of this application will be described in detail in the following detailed description section. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 A flowchart illustrating a geographic data retrieval method based on an adaptive index provided in an embodiment of this application; Figure 2 A schematic diagram of the experimental results for selecting the image conditional index structure according to an embodiment of this application; Figure 3 This is a schematic diagram illustrating the experimental results of the total number of inverted index items provided in an embodiment of this application; Figure 4 This is a schematic diagram of the experimental results of index memory usage provided in an embodiment of this application; Figure 5 This is a schematic diagram illustrating the experimental results of index construction time provided in an embodiment of this application; Figure 6 A schematic diagram illustrating experimental results of typical window query response time provided in an embodiment of this application; Figure 7 A schematic diagram of the structure of a geographic data retrieval device based on an adaptive index provided in an embodiment of this application; Figure 8 This is a schematic diagram of the structure of a terminal device provided in an embodiment of this application. Detailed Implementation

[0023] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0024] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0025] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0026] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0027] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0028] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0029] To address the issue of low accuracy in existing geographic data retrieval methods, this application provides a geographic data retrieval method based on adaptive indexing. This method divides the target area into multiple geographic partitions and obtains the partition data for each partition. Then, for each partition, a data feature profile is calculated, and a set of regional heterogeneous indexes for the partition data is generated based on the data feature profiles. Next, the user's search request is parsed to obtain a set of search conditions. Based on the search condition set and all regional heterogeneous index sets, multiple candidate partition data are matched from all partition data. Then, based on the data feature profile of each candidate partition data, a comprehensive relevance score between each candidate partition data and the user's search request is calculated. Finally, based on all comprehensive relevance scores, all candidate partition data are merged and deduplicated to obtain the final retrieved data corresponding to the user's search request. The process involves obtaining geographic partition data, which effectively distinguishes geographic data at the spatial level. Then, it adaptively generates corresponding indexes based on the data characteristics of each partition, avoiding low retrieval accuracy caused by a single fixed index method. Multiple candidate partition data are matched according to the user's retrieval request, and the final retrieval data is determined by calculating the comprehensive relevance score. This fully considers the relevance between the user's retrieval request and the partition data, further improving the accuracy of geographic data retrieval.

[0030] The following is an exemplary description of the geographic data retrieval method based on adaptive index provided in this application.

[0031] like Figure 1 As shown, the geographic data retrieval method based on adaptive indexing provided in this application includes the following steps: Step 11: Divide the target area into multiple geographic partitions and obtain the partition data for each geographic partition.

[0032] The target area can be a city, county, or other similar area. Geographical partitions are sub-regions within the target area. Partition data includes multimodal data of the geographic partition, such as vector data, image metadata, trajectory data, and sample data.

[0033] If a geographic region is a city, then vector data can be the administrative division area, the area of ​​a residential plot, the area of ​​a lake, the area of ​​a forest and cultivated land plot, etc., of that city; image metadata can be satellite remote sensing images, etc.; trajectory data can be vehicle trajectories, pedestrian trajectories, etc.; and sample data can be labels of land sample points, etc.

[0034] In some embodiments of this application, the target area can be divided into multiple geographic partitions using the GeoSOT grid, and partition data can be obtained by accessing a geographic information system or other means.

[0035] For example, using GeoSOT global latitude and longitude grid coding as the unified spatial association key, the target area is divided into several GeoSOT partitions (i.e., geographical partitions) according to a preset baseline level, and a three-stage integer quadtree recursive partitioning scheme is adopted: The first stage is degree-level subdivision (levels 0 to 9), which involves expanding the 180°×360° geospatial projection of the target area to 512°×512° and then recursively quadrilateralizing it into 1°×1° grid cells. class( ) grid side length for: ; The second stage is hierarchical meshing (levels 10 to 15), expanding the 1° cell to 64′×64′ and then recursively meshing it to 1′×1′ grid cells. class( The side length of the grid relative to a 1° cell is: ; The third stage is second-level subdivision (levels 16 to 21), which expands the 1′ cell to 64″×64″ and then recursively subdivides it into 1″×1″ grid cells. This three-stage subdivision scheme ensures that each spatial location corresponds to a unique integer code, and the code has the characteristic of preserving spatial proximity (codes with the same prefix correspond to spatially adjacent grid regions); the number of grid cells at each level of depth is strictly increased to 4 times the original, and each level corresponds to clear degree, minute, and second geographical semantics, which facilitates adaptive selection of subdivision depth based on data density profiles.

[0036] For example, multimodal spatiotemporal data includes four types: vectors, image metadata, trajectories, and samples. Each data record is uniformly abstracted into a seven-tuple. ,in It is a unique integer identifier; This is an enumeration of data types, taking values ​​from {vector, image, trajectory, sample}. The smallest bounding rectangle is represented using the WGS84 coordinate system and 32-bit floating-point precision. and These are the start and end times, encoded as integers in YYYYMMDD format. To record relevant geographic semantic annotation vocabularies; This is a set of schema attributes related to the data type. Each record is also associated with a 64-bit GeoSOT encoded value generated by the center point of the smallest bounding rectangle according to the GeoSOT encoding scheme, serving as a unified association key.

[0037] Step 12: For each partition of data, calculate the data feature profile of the partition data, and generate a set of regional heterogeneous indexes for the partition data based on the data feature profile.

[0038] The data feature profile described above is used to describe the data characteristics of partitioned data. The aforementioned set of heterogeneous regional indexes includes various indexes for partitioned data, such as semantic-level indexes and time-level indexes.

[0039] In some embodiments of this application, the steps of calculating the data feature profile of the partitioned data and generating a set of regional heterogeneous indexes for the partitioned data based on the data feature profile include: The first step is to calculate the data density and spatial distribution dispersion of the geographic partitions corresponding to the partitioned data.

[0040] The data density mentioned above is the ratio between the number of data records within a geographic partition and the area of ​​that geographic partition. The spatial distribution dispersion mentioned above is used to describe the degree of dispersion of data distribution within a geographic partition.

[0041] For example, by calculating geographic partitions Internal record number Area of ​​this geographical region The ratio between them gives the data density: ; Spatial distribution dispersion is determined by the nearest neighbor index of the record center point within the geographic partition. Representation. Specifically, establishing geographical zones. g shared within The first data record, i The center point of the smallest bounding rectangle of the record is For each center point Calculate the distance between the record and the center point of the remaining records within the partition, and take the minimum value as the nearest neighbor distance of the center point: ; The average nearest neighbor distance is obtained by averaging the nearest neighbor distances of all record center points within the geographic partition. ; Further calculations are performed on the expected average nearest neighbor distance when the record center points are randomly spatially distributed, under the conditions of the same number of records and the same partition space range. This is denoted as: ; This leads to geographical divisions. Nearest neighbor index: ; when When the actual nearest neighbor distance is less than the expected distance under random distribution, it indicates that the distance between the record center points is relatively short, and the data exhibits a spatially clustered distribution; when When, it indicates that the spatial distribution of the record center point is approximately random; when When the actual nearest neighbor distance is greater than the expected distance under random distribution, it indicates that the record center points are relatively dispersed and the data tends to be uniformly distributed. Therefore, the nearest neighbor index, by comparing the difference between the actual spatial proximity and the random spatial distribution benchmark, can characterize the degree of spatial clustering or dispersion of partitioned data and serve as the spatial distribution dispersion.

[0042] The second step is to perform multimodal analysis on the partitioned data to obtain the modal mixing degree.

[0043] The modal hybridity described above is used to describe the degree of hybridity of multimodal data in partitioned data.

[0044] For example, modal mixing can be calculated using the Shannon entropy formula. For instance, the proportion of all modal data in the partitioned data can be substituted into the Shannon entropy formula to calculate the modal mixing.

[0045] The third step is to perform keyword distribution analysis on the partitioned data to obtain semantic entropy.

[0046] The semantic entropy mentioned above is used to describe the distribution of keywords in partitioned data.

[0047] For example, geographic partitioning can be performed using Shannon entropy. The distribution of keywords within the region is analyzed. Geographical zones are defined. The different keywords appearing within constitute a keyword set. For any keyword Count the number of times the keyword appears in the geographical region. And calculate the keyword frequency based on the occurrence frequency of each keyword. Frequency distribution within this geographic region: ; After obtaining the frequency distribution of each keyword, geographical partitions are calculated based on Shannon entropy. Semantic entropy: ; in, Keywords In geographical partitioning The frequency of occurrence within a keyword. When keywords within a section are mainly concentrated in a small number of keywords, the keyword frequency distribution is relatively concentrated, and the semantic entropy is relatively high. A smaller semantic entropy indicates that the data semantics of that partition are relatively simple; when a partition contains many different keywords and the frequency of each keyword is relatively balanced, the semantic entropy is higher. A larger semantic entropy indicates that the data in that partition is more semantically rich and complex. Therefore, semantic entropy can be used to describe the complexity of keyword distribution in the partition data and can serve as one of the criteria for determining whether to build a synonym-extended inverted index.

[0048] The fourth step is to combine data density, spatial distribution dispersion, modal heterogeneity, and semantic entropy into a set to obtain a data feature profile of the partitioned data.

[0049] The fifth step is to calculate the depth of the partitioning hierarchy of the partitioned data.

[0050] Specifically, through the formula: ; Calculate the partitioning level depth of partitioned data .

[0051] in, For the amplitude limiting function, Indicates the baseline level. Indicates the number of records in the partition data. The target number of records per leaf grid is the desired value. Indicates the lower limit value of the hierarchy. This indicates the upper limit of the hierarchy.

[0052] For example, according to a preset baseline level (This embodiment takes) (Corresponding to a spatial resolution of approximately 30km × 30 km (mid-latitude region)) The GeoSOT encoded values ​​of all records are taken from the first few values. Bit prefix, to obtain GeoSOT partition set For each partition Computational data feature profiling : Data density ,in This represents the number of records within the partition. The area of ​​the partition space (measured by the number of leaf grids at the baseline level); Spatial distribution dispersion The nearest neighbor index of the record center point within a partition is defined as the ratio of the average nearest neighbor distance to the expected distance of the random distribution. Indicates clustered distribution, Indicates a random distribution. Indicates uniform dispersion; Semantic entropy ,in Keywords within the partition Frequency distribution; Modal hybridity ,in {Vector, Image, Trajectory, Sample} The percentages of the four data categories.

[0053] Time density ,in This represents the span of timestamps recorded within the partition.

[0054] The above five indicators are used together to characterize regional data, avoiding reliance on a single indicator for classification. The calculation can be completed in a single full scan, with a time complexity of O(n log n). , To record the total number.

[0055] The sixth step is to determine multiple index types for the partitioned data based on the data feature profile, and generate the data index corresponding to each index type based on the depth of the hierarchical division.

[0056] When the number of records in a partition exceeds the preset spatial index threshold, the spatial tree index will be used as an index type.

[0057] When the number of records in the partition data is less than or equal to the preset spatial index threshold, the flat minimum bounding rectangle array index is used as an index type.

[0058] When the semantic entropy of the partitioned data is greater than the preset semantic threshold, the synonym extended inverted index will be used as an index type.

[0059] When the number of leaf layer concepts in the partitioned data exceeds the preset concept threshold, the semantic level index will be used as an index type.

[0060] When the time density of partitioned data is greater than a preset time threshold, the time interval index is used as an index type.

[0061] When the modal heterogeneity of the partitioned data exceeds the preset modal threshold, the data type mask index is used as an index type.

[0062] For example, the spatial tree index is a hierarchical spatial index constructed according to the spatial range of data records. Its index items include at least the unique identifier of the data record and the corresponding minimum bounding rectangle, which is used to quickly filter data records that intersect with the spatial query range according to the spatial query range.

[0063] The flat minimum bounding rectangle array index is a spatial index formed by storing the unique identifiers of each data record within a partition and their corresponding minimum bounding rectangles in a continuous array. For geographical partitions with a small number of records, the minimum bounding rectangles in the array can be directly traversed and it can be determined whether they intersect with the query spatial range, thereby avoiding the additional overhead of building and maintaining a hierarchical spatial tree structure.

[0064] The number of ontology leaf layer concepts is the number of concepts obtained after deduplication of the domain ontology leaf layer concepts associated with each data record within the geographic partition, and is used to reflect the richness of specific semantic concepts within the geographic partition.

[0065] The synonym-expanded inverted index is a mapping relationship between keywords and unique identifiers of data records, established by expanding keywords with synonyms based on ordinary keyword inverted indexes. For keywords contained in a data record, synonyms with similar semantics are obtained from preset synonym resources, and the original keyword and its corresponding synonyms are used as index terms to establish an association with the data record, so that user search requests using different words but with similar semantics can match the corresponding data records.

[0066] The semantic hierarchy index is an index built based on the hierarchical relationships between concepts in the domain ontology. For the ontology concept associated with a data record, the index stores the relationship between the concept and its corresponding superior or subordinate concepts and the data record, so that when a user queries using a higher-level semantic concept, they can retrieve the data record belonging to its subordinate concept.

[0067] The time density refers to the number of data records contained within a unit time span within a geographic partition. For geographic partitions... The time density is expressed as: ; in, Geographical partitioning The number of records in the record, This represents the timestamp span of data records within this geographic partition.

[0068] The time interval index is an index established for the start and end time values ​​of data records, used to filter data records that overlap with the time range in the user's search request.

[0069] The data type mask index is a data index that uses different flag bits to represent different data types. In this embodiment, the data types include vector data, image metadata, trajectory data, and sample data. Corresponding flag bits can be assigned to different data types, and the data type mask records the data types contained in partitions or data records. When a user's search request includes data type conditions, the search is performed based on the corresponding data type mask to filter data records that meet the specified data type conditions.

[0070] The process of generating a data index based on the hierarchical depth of the above-mentioned layering is as follows.

[0071] First, regarding geographical partitioning g Calculate the depth of the subdivision level corresponding to the geographic partition based on the aforementioned formula. : ; Subsequently, based on the calculated This serves as the target hierarchical level for spatial indexing organization within the geographic partition. For each data record within the geographic partition, the minimum bounding rectangle of that record is obtained, and grid cells that can completely cover that minimum bounding rectangle are determined in the GeoSOT partition tree. When the minimum bounding rectangle can be determined by the first... When a single GeoSOT grid cell of a layer is fully covered, the data record is attached to the index cell corresponding to that grid cell; when the minimum bounding rectangle crosses the first... When dealing with multiple GeoSOT grids in a layer, the least common ancestor grid that can completely cover the minimum bounding rectangle is searched upwards along the GeoSOT partition tree, and the data record is attached to the index unit corresponding to the least common ancestor grid.

[0072] Each data record is attached only once in the aforementioned grid index unit, without being repeatedly written to multiple adjacent grids, thereby reducing index redundancy caused by repeated attachment of data across grids.

[0073] After spatially attaching data records, a corresponding regional heterogeneous index is constructed within the geographic partition based on the data feature profile and the aforementioned index type conditions. For partitions with more records than a preset spatial index threshold, a spatial tree index is constructed based on the minimum bounding rectangle of the attached records; for partitions with fewer than or equal to the preset spatial index threshold, a flat minimum bounding rectangle array index is constructed; when semantic entropy meets a preset condition, a synonym-extended inverted index is built based on the record's keywords and their synonyms; when the number of ontology leaf layer concepts meets a preset condition, a semantic hierarchy index is built based on the domain ontology concepts associated with the record; when temporal density meets a preset condition, a time interval index is built based on the record's start and end times; when modal heterogeneity meets a preset condition, a data type mask index is built based on the record's data type. Thus, each geographic partition only constructs a subset of index structures corresponding to its data features.

[0074] For example, a geographical region The depth of the segmentation hierarchy is calculated based on the number of records. For records in this partition If its minimum bounding rectangle lies entirely within a 14th-level GeoSOT grid Inside, it will record Connect to grid For records If its minimum bounding rectangle spans multiple GeoSOT grids at level 14, then search upwards along the GeoSOT partition tree for the minimum common ancestor grid that completely covers the minimum bounding rectangle. and only record Connect to If the partition simultaneously meets the conditions for constructing a spatial tree index, a synonym-extended inverted index, and a time interval index, then based on the record set after completing the GeoSOT spatial attachment, corresponding spatial indexes, keyword indexes, and time indexes will be established respectively, thereby forming a regional heterogeneous index set for the geographic partition.

[0075] During the query phase, the user's query space window is expanded into the corresponding GeoSOT grid code set, and the relevant geographic partitions are determined by matching the baseline GeoSOT prefix and judging the intersection between the partition space range and the query space window. Then, only the indexes that have actually been built in the relevant geographic partitions are called for retrieval, and the cross-partition candidate results are deduplicated according to the unique identifier of the record.

[0076] The seventh step is to treat all data indexes as a set, thus obtaining a set of regional heterogeneous indexes.

[0077] For example, when a partition's data feature profile deviates from the profile when the index was created due to continuous data writing, exceeding a preset threshold, only the partition's subdivision level depth and index structure subset are redefined according to the updated profile, and the partition index is partially rebuilt. The indexes of other partitions are unaffected.

[0078] GeoSOT partitioning and profiling are based on a multidimensional spatiotemporal data cube organized by a fact constellation model: GeoSOT encoding is used as the spatial dimension key, year and month encoding is used as the time dimension key, the domain ontology is used to express the semantic dimension, and four types of modalities are used as the data type dimension. Sub-fact tables are set up for the four types of data and are linked by a shared dimension key.

[0079] It should be noted that construction conditions are set for the portrait components such as semantic entropy, number of keyword types, number of ontology concepts, modality mixing, and time density: inverted indexes and semantic hierarchical indexes are not built for semantically sparse partitions, and type masks are not built for modal-single partitions. This achieves the principle of selection based on partition in physical structure, rather than filtering the uniformly constructed index only during the query stage.

[0080] Step 13: Parse the user's search request to obtain a set of search conditions, and match multiple candidate partition data from all partition data based on the set of search conditions and the set of heterogeneous indexes of all regions.

[0081] The aforementioned user retrieval request is a request for geographic data to be retrieved input by the user device, which includes information such as the spatial range, time range, keywords, semantic concepts, and data types requested by the user.

[0082] For example, algorithms such as the chi-square test can be used to process user search requests by word recognition, identifying search conditions such as spatial range, time range, keywords, semantic concepts, and data types. All identified search conditions are collected as a set to obtain a search condition set. The search condition set is then matched with the regional heterogeneous index set of each partition data. If there is identical information or the similarity is greater than a preset threshold (similarity can be calculated by cosine similarity, etc.), the match is considered successful, and the successfully matched partition data is used as candidate partition data.

[0083] In some embodiments of this application, the query space window can be expanded into a GeoSOT grid encoding set and candidate related partitions can be obtained by comparison with the baseline hierarchical prefix. At the same time, it is determined whether the spatial range of all partitions intersects with the query space window and the intersecting partitions are included in the related set. The union of the two constitutes the final related partition set, and then duplicates are removed according to the unique identifier of the record.

[0084] Receive user search requests and represent them as These correspond to five categories of conditions: spatial range, temporal range, data type, keywords, and semantic concepts, where each condition can be empty. The relevant GeoSOT partition set consists of the union of the following two parts: (a) Expanding the query spatial window into a GeoSOT grid code set and extracting each code at the baseline level. (a) The prefix at the location is compared with the partition identifier; (b) For all partitions, it is determined whether their spatial range intersects with the query spatial window, and the intersecting partitions are included in the relevant set. For conditions not specified in the query, the index of the corresponding dimension is not invoked. After the partition orientation is completed, a search is initiated only on the regional index of the relevant partition.

[0085] Step 14: Based on the data feature profile of each candidate partition data, calculate the comprehensive relevance score between each candidate partition data and the user's search request.

[0086] Specifically, through the formula: ; Calculate the overall relevance score ; in, Indicates the total number of index dimensions. Indicates the first Dimensional indexes for candidate partition data Output matching score, Indicates based on candidate partition data The candidate partition data was used to calculate the first Dimension index fusion weights: ; in, Take the spatial clustering degree (e.g., 1 minus the normalized nearest neighbor index). Take semantic entropy, Take modal hybridity, Take the normalized time density, This is a non-negative adjustable parameter. Therefore, the higher the semantic entropy of the partition, the greater the weight of the semantic dimension; the higher the degree of spatial clustering of the partition, the greater the weight of the spatial dimension. The fusion weight changes with the regional data characteristics and does not degenerate into a global constant or a binary threshold. The weighting of each dimension in the retrieval results is also adaptively adjusted according to the partition characteristics.

[0087] Step 15: Based on all comprehensive relevance scores, merge and deduplicate all candidate partition data to obtain the final retrieval data corresponding to the user's retrieval request.

[0088] For example, all candidate partition data are merged. If there are two or more identical data in the merged result, only one of them is retained to achieve deduplication. All the retained data after deduplication are sorted according to the comprehensive relevance score (e.g., sorted in descending order of comprehensive relevance score), and the sorted result is used as the final retrieval data and returned to the user device.

[0089] For each relevant partition It calls its built region index and outputs the matching scores on each non-empty dimension. Then, the candidate records are weighted and synthesized in this partition according to the formula above to obtain a comprehensive relevance score.

[0090] When partition When multiple dimensional indexes exist simultaneously, The formula above is obtained by weighting with normalized exponents. Take the normalized value of the corresponding portrait indicator; When partition If a certain dimension does not have a corresponding index, but the query includes conditions for that dimension, the weight of that dimension can be reset to zero, and other related partitions' indexes of the same dimension can be used to supplement the candidates. After deduplicating cross-partition candidate records by unique identifier, they are sorted according to their comprehensive relevance score (and distance decay factor if necessary), and returned to the top. Results.

[0091] It is worth mentioning that obtaining the geographic partition data can effectively distinguish geographic-related data at the spatial level. Then, based on the data characteristics of the partition, corresponding indexes are adaptively generated, avoiding the low retrieval accuracy caused by a single fixed index method. Multiple candidate partition data are matched according to the user's search request, and the final search data is determined by calculating the comprehensive relevance score. This fully considers the relevance between the user's search request and the partition data, further improving the accuracy of geographic data retrieval.

[0092] In addition, this application has the following beneficial effects: (1) Simplified index structure. During the construction phase, based on the data feature profile of each GeoSOT partition, the subset of index structures to be constructed for each partition is conditionally determined; for partitions with sparse semantics or single modality, semantic indexes such as inverted indexes and semantic hierarchical indexes can be omitted, thereby reducing the index storage and maintenance overhead from the perspective of physical structure.

[0093] (2) Faster response for typical queries. By using GeoSOT partitioning, queries are distributed to partitioned indexes related to spatial windows, shortening response time in typical city-level small window query scenarios; when the query window coverage is large, it can maintain comparable efficiency to the global unified indexing scheme.

[0094] (3) Coordinated attachment and retrieval. The system adopts a combination of single-point attachment using the least common ancestor grid and partition orientation, supporting cross-partition retrieval and result merging within the same GeoSOT addressing framework. The number of attached entries is consistent with the total number of records, and there are no duplicate attachments.

[0095] (4) Easy to expand and maintain. GeoSOT encoding is used as the unified association key and partition unit, supporting incremental updates; when the feature profile of partition data changes significantly, the index can be rebuilt locally only for that partition without affecting the other partitions.

[0096] The method of this application will be illustrated below with a specific example.

[0097] A synthetic dataset constructed based on the multimodal spatiotemporal data distribution characteristics of a certain city ( 33 GeoSOT partitions, automatically calculate the mean of the stratified levels. =Level 14.52, range 10-15), all relevant parameters are given accurately. The data distribution is set as follows: 6 core urban areas (semantically dense, modally diverse), 6 road network corridors (mainly trajectory-based, semantically relatively simple), and 9 suburban areas (mainly single vector-based, semantically relatively simple). In the example, 150 diverse queries and full table scan results were compared row by row for correctness verification. The globally unified index is consistent with the retrieval results of the proposed solution. The following experiments focus on comparing the differences between the two solutions in terms of index structure and query efficiency.

[0098] Experiment A: Structural savings in image-based conditional indexing structure selection. The control group used a globally unified strategy of building complete inverted indexes and ontology-level indexes in all 33 partitions, while the experimental group used the image-based conditional rules described above. Experimental results show that: the total number of inverted index entries in the full partition complete indexing scheme is 2,445,540, while the region-adaptive scheme of this invention is 1,658,938, a reduction of 32.2%; the control group built inverted indexes in all 33 partitions, while the scheme of this application only built inverted indexes in 5 partitions; the index memory usage decreased from 245.78 MB to 237.79 MB; the number of single-point attached entries is equal to the total number of records. No redundant connections.

[0099] The experimental results of selecting the conditional index structure for image representation are as follows: Figure 2 As shown in the figure, the vertical axis represents the number of partitions used to build the inverted index. The two types of bars correspond to the number of partitions used to build the inverted index under the full partition construction method and the region adaptive (i.e., the present invention) method, which are 33 partitions and 5 partitions, respectively.

[0100] Experimental results on the total number of inverted index items are as follows: Figure 3As shown in the figure, the vertical axis represents the total number of inverted index entries, in millions. The two bar charts correspond to the total number of inverted index entries under the full partition construction of a complete index and the region adaptive (i.e., the present invention) method, respectively, which are 2.45 million and 1.66 million.

[0101] The results of the index memory usage experiment are as follows Figure 4 As shown, the vertical axis represents the index memory usage in MB. The two bars correspond to the index memory usage under the full partition construction of a complete index and the region adaptive (i.e., the present invention) method, which are 245.79 and 237.79 respectively. It can be seen that the memory usage of the method in this application is reduced by 3.3%.

[0102] The experimental results of index building time are as follows: Figure 5 As shown, the vertical axis represents the index construction time in seconds (s). The two types of bars correspond to the index construction time under the full partition complete index construction and the regional adaptive (i.e., the present invention) method, which are 2.43 and 2.31, respectively.

[0103] Experiment B: Typical city window query efficiency. The control group used a globally unified index, while the experimental group used the regional heterogeneous index set of this invention. In typical city-level queries where the window is limited to a single or a few partitions, the response time of the globally unified index scheme was 14.1ms, while that of the regional adaptive scheme of this invention was 7.9ms, a speedup of approximately 1.8 times.

[0104] Typical window query response time experimental results are as follows Figure 6 As shown in the figure, the vertical axis represents the median response time in milliseconds (ms). The two types of bars represent the median response time under the global unified index and the regional adaptive (i.e., the present invention) methods, which are 14.1 and 7.9, respectively.

[0105] Experiment C: Regionalized Fusion Weighting. The Normalized Discounted Cumulative Gain (NDCG) was used to evaluate the ranking quality of search results: For each query, candidate results were ranked from highest to lowest relevance score. The degree of agreement between the results at each position in the top 20 results (denoted as NDCG@20) and the relevance labels was examined. The higher the ranking and the higher the relevance, the higher the score. The ideal ranking score for the query was then normalized to obtain an evaluation value in the [0,1] interval; a higher value indicates better ranking quality. In this experiment, the same scoring function was used to assign relevance labels to each result across 24 full-dimensional queries. The mean NDCG@20 values ​​of the two schemes—global equal weighting and regionalized fusion weighting—were compared: global equal weighting was 0.9648, and regionalized fusion weighting was 1.0000, representing a relative improvement of approximately 3.65%.

[0106] The experimental data above are all measured results on synthetic datasets, and the conclusions are valid only within the parameter range. Large-scale experiments ( The following are embodiments of the present invention: Experiment A shows that the conditional index selection of the profile reduces the number of inverted index entries by 32.2% and builds inverted indexes for only 5 / 33 partitions; Experiment B shows that the median response time for a typical window query is reduced from about 14.1ms to about 7.9ms.

[0107] Regional heterogeneous index sets support incremental updates. When writing a single record, updates are performed according to the GeoSOT partition and tier hierarchy to which it belongs. The system locates and appends data to the corresponding partition index. During batch writes, it first groups data by partition and then triggers a local update based on the range of changes in the data profile. When a partition's data feature profile deviates from the indexing point by more than a preset threshold due to continuous writes, only that partition's subdivision depth and index structure subset are redefined based on the updated profile, and the indexes of other partitions remain unchanged. This mechanism is suitable for real-time streaming data scenarios in cities and avoids the high cost of frequent full-scale reconstructions.

[0108] This method can be extended to the following query types: pure spatiotemporal queries, spatiotemporal and keyword joint queries, spatiotemporal and semantic concept joint queries, cross-modal queries, and multi-dimensional composite queries with all five conditions present simultaneously. All types of queries are organized using a unified regional profile-driven weighted scoring method, implemented by calling different index combinations, demonstrating the applicability of this method in multimodal spatiotemporal data retrieval.

[0109] Therefore, this application proposes a technical approach under the GeoSOT unified addressing framework that drives the selection of index structure subsets during the construction phase using multi-dimensional regional data feature profiling: for semantically sparse and modally singular partitions, the inverted index, semantic hierarchical index, and type mask are omitted in the physical structure; for semantically dense and modally diverse partitions, the complete multi-dimensional index is retained, and cross-partition retrieval is achieved by combining single-point attachment of the least common ancestor grid with GeoSOT partition orientation. Compared with schemes that only reduce candidates or select index types based on a single scale indicator during the query phase, the key point of this invention is that the physical index configuration is determined based on the region during the construction phase. In the main embodiment of synthetic data, the number of inverted index items is reduced by 32.2% (inverted index is built for 5 / 33 partitions), the median response time for typical window queries is reduced from approximately 14.1 ms to approximately 7.9 ms, and the regional fusion weight is optimized by optional sorting.

[0110] The following is an exemplary description of the geographic data retrieval device based on adaptive index provided in this application.

[0111] like Figure 7 As shown, this application provides a geographic data retrieval device based on an adaptive index. The geographic data retrieval device 700 based on an adaptive index includes: The partitioning module 701 is used to divide the target area into multiple geographic partitions and obtain the partition data of each geographic partition. The generation module 702 is used to calculate the data feature profile of each partition data, and generate a set of regional heterogeneous indexes of the partition data based on the data feature profile; the data feature profile is used to describe the data features of the partition data. The parsing module 703 is used to parse the user's search request, obtain the search condition set, and match multiple candidate partition data from all partition data based on the search condition set and the heterogeneous index set of all regions. The calculation module 704 is used to calculate the comprehensive relevance score between each candidate partition data and the user's search request based on the data feature profile of each candidate partition data. The determination module 705 is used to merge and deduplicate all candidate partition data based on all comprehensive relevance scores to obtain the final retrieval data corresponding to the user's retrieval request.

[0112] It should be noted that the information interaction and execution process between the above-mentioned devices / units are based on the same concept as the method embodiments of this application. For details on their specific functions and technical effects, please refer to the method embodiments section, and they will not be repeated here.

[0113] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0114] like Figure 8 As shown, an embodiment of this application provides a terminal device, wherein the terminal device D10 of this embodiment includes: at least one processor D100 ( Figure 8The diagram shows only one processor, a memory D101, and a computer program D102 stored in the memory D101 and executable on the at least one processor D100, wherein the processor D100 executes the computer program D102 to implement the steps in any of the above method embodiments.

[0115] Specifically, when the processor D100 executes the computer program D102, it divides the target area into multiple geographical partitions and obtains the partition data for each geographical partition. Then, for each partition data, it calculates a data feature profile and generates a set of regional heterogeneous indexes for the partition data based on the data feature profile. Next, it parses the user's search request to obtain a set of search conditions. Based on the search condition set and all regional heterogeneous index sets, it matches multiple candidate partition data from all partition data. Then, based on the data feature profile of each candidate partition data, it calculates a comprehensive relevance score between each candidate partition data and the user's search request. Finally, based on all comprehensive relevance scores, it merges and deduplicates all candidate partition data to obtain the final search data corresponding to the user's search request. Obtaining the partition data of geographical partitions effectively distinguishes geographically related data at the spatial level. Adaptively generating corresponding indexes based on the data characteristics of the partitions avoids low search accuracy caused by a single, fixed indexing method. Matching multiple candidate partition data based on the user's search request and determining the final search data by calculating the comprehensive relevance score fully considers the relevance between the user's search request and the partition data, further improving the accuracy of geographical data retrieval.

[0116] The processor D100 can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0117] In some embodiments, the memory D101 may be an internal storage unit of the terminal device D10, such as a hard disk or memory of the terminal device D10. In other embodiments, the memory D101 may be an external storage device of the terminal device D10, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the terminal device D10. Furthermore, the memory D101 may include both internal and external storage units of the terminal device D10. The memory D101 is used to store the operating system, applications, bootloader, data, and other programs, such as the program code of the computer program. The memory D101 can also be used to temporarily store data that has been output or will be output.

[0118] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps described in the various method embodiments above.

[0119] This application provides a computer program product that, when run on a terminal device, enables the terminal device to implement the steps described in the various method embodiments above.

[0120] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying computer program code to an adaptive index-based geographic data retrieval method device / terminal device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electrical carrier signals or telecommunication signals.

[0121] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0122] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0123] The above description is the preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this invention, and these improvements and modifications should also be considered within the scope of protection of this invention.

Claims

1. A geographic data retrieval method based on adaptive indexing, characterized in that, include: Divide the target area into multiple geographic partitions and obtain the partition data for each geographic partition; For each of the partition data, a data feature profile is calculated, and a regional heterogeneous index set of the partition data is generated based on the data feature profile; the data feature profile is used to describe the data characteristics of the partition data. The user's search request is parsed to obtain a set of search conditions. Based on the set of search conditions and the set of heterogeneous indexes for all regions, multiple candidate partition data are matched from all partition data. Based on the data feature profile of each candidate partition data, calculate the comprehensive relevance score between each candidate partition data and the user's search request; Based on all comprehensive relevance scores, all candidate partition data are merged and deduplicated to obtain the final retrieval data corresponding to the user's retrieval request.

2. The geographic data retrieval method according to claim 1, characterized in that, The calculation of the data feature profile of the partitioned data includes: Calculate the data density and spatial distribution dispersion of the geographic partitions corresponding to the partitioned data; Multimodal analysis was performed on the partitioned data to obtain the modal hybridity. Keyword distribution analysis was performed on the partitioned data to obtain semantic entropy; The data density, spatial distribution dispersion, modal heterogeneity, and semantic entropy are combined into a set to obtain the data feature profile of the partitioned data.

3. The geographic data retrieval method according to claim 2, characterized in that, The process of generating a set of regional heterogeneous indexes for the partitioned data based on the data feature profile includes: Calculate the partitioning level depth of the partitioned data; Based on the data feature profile, multiple index types of the partitioned data are determined, and a data index corresponding to each index type is generated based on the segmentation hierarchy depth. By treating all data indexes as a set, we obtain a set of regional heterogeneous indexes.

4. The geographic data retrieval method according to claim 3, characterized in that, The calculation of the partitioning level depth of the partitioned data includes: Through the formula: Calculate the partitioning depth of the partitioned data ; in, For the amplitude limiting function, Indicates the baseline level. This indicates the number of records in the partition data. The target number of records per leaf grid is the desired value. Indicates the lower limit value of the hierarchy. This indicates the upper limit of the hierarchy.

5. The geographic data retrieval method according to claim 3, characterized in that, The determination of multiple index types for the partitioned data based on the data feature profile includes: When the number of records in the partition data exceeds the preset spatial index threshold, the spatial tree index will be used as an index type. When the number of records in the partition data is less than or equal to the preset space index threshold, the flat minimum bounding rectangle array index is used as an index type. When the semantic entropy of the partitioned data is greater than a preset semantic threshold, the synonym extended inverted index will be used as an index type. When the number of ontology leaf layer concepts in the partitioned data is greater than a preset concept threshold, the semantic level index will be used as an index type. When the time density of the partitioned data is greater than a preset time threshold, the time interval index is used as an index type. When the modal heterogeneity of the partitioned data is greater than a preset modal threshold, the data type mask index is used as an index type.

6. The geographic data retrieval method according to claim 1, characterized in that, The step of calculating a comprehensive relevance score between each candidate partition data and the user's search request based on the data feature profile of each candidate partition data includes: Through the formula: Calculate the overall relevance score ; in, Indicates the total number of index dimensions. Indicates the first Dimensional indexes for candidate partition data Output matching score, Indicates based on candidate partition data The candidate partition data was used to calculate the first The fusion weight of dimensional indexes.

7. A geographic data retrieval device based on adaptive indexing, characterized in that, include: The partitioning module is used to divide the target area into multiple geographic partitions and obtain the partition data for each geographic partition. The generation module is used to calculate a data feature profile for each partition of data, and generate a set of regional heterogeneous indexes for the partition of data based on the data feature profile; the data feature profile is used to describe the data features of the partition data. The parsing module is used to parse the user's search request, obtain a set of search conditions, and match multiple candidate partition data from all partition data based on the set of search conditions and the set of heterogeneous indexes of all regions. The calculation module is used to calculate the comprehensive relevance score between each candidate partition data and the user's search request based on the data feature profile of each candidate partition data. The determination module is used to determine the final retrieval data corresponding to the user's retrieval request from all candidate partition data based on all comprehensive relevance scores.

8. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the geographic data retrieval method based on adaptive index as described in any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the geographic data retrieval method based on adaptive index as described in any one of claims 1 to 6.

Images