A government affair data address alignment method and system based on multi-level address mapping

Abstract

The application relates to the technical field of government affair data processing, in particular to a government affair data address alignment method and system based on multi-level address mapping, wherein the method comprises the following steps: collecting two kinds of government affair system addresses respectively and preprocessing; establishing a standard address coding hierarchical system by fusing according to address levels; performing address semantic alignment based on address text data through hierarchical and semantic feature fusion, ambiguous address identification and attention weight calibration mechanism; performing address space alignment by calculating the spatial overlap degree of the semantic matching address pair through the multi-dimensional space feature of the overlapping area, the center point distance and the area difference; and respectively implementing standard address coding according to the standard address coding hierarchical system. Through accurate semantic alignment and space alignment, the application realizes the standardized output of different government affair system addresses.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of government data processing, and in particular to a government data address alignment method and system based on multi-level address mapping. BACKGROUND

[0002] In the current field of government data management, various government systems usually use independent address division and coding systems for data storage and management. For example, some government systems divide addresses according to administrative jurisdiction logic and assign corresponding codes according to the administrative level from large to small, in order to realize the classification and management of relevant data such as civil affairs and social security; some government systems define and code addresses according to the jurisdiction division standard of public security work, for the data organization of police work, public security management, etc.

[0003] For the integration of various government data address coding systems, there is no standardized and systematic general method, and in most cases, the integration mode relies on manual guidance: the staff needs to first sort out the rules and coverage of general government address coding (such as street, community, grid level) and public security management coding (such as public security area, responsibility area level), and then manually establish the corresponding relationship between the two types of coding by comparing address description information one by one. The whole integration process lacks automated technical support, neither a unified mapping model for associating the two types of coding is constructed, nor technical means capable of automatically recognizing and resolving address semantic differences are introduced, and the integration effect completely depends on the meticulousness of manual operation and the familiarity with the two types of coding system. The existing integration method has the following defects: first, the integration efficiency is extremely low, the process of manually comparing and establishing the corresponding relationship between the codes is time-consuming and laborious, and it is difficult to adapt to the actual demand of large data volume and wide coverage. When the address information is updated (such as the addition of a community or the adjustment of the range of a grid or responsibility area), a large-scale manual matching needs to be carried out again, and the updating efficiency lags seriously; second, the integration accuracy is difficult to guarantee, and manual operation is easily affected by subjective factors, such as misunderstanding of the two types of coding rules and address description information recognition errors, which may lead to errors in the corresponding relationship between the codes, and further affect the reliability of subsequent data fusion applications. SUMMARY

[0004] To solve one of the above-mentioned problems in the prior art, the present application provides a government data address alignment method and system based on multi-level address mapping.

[0005] To achieve the above object, the application provides a government data address alignment method based on multi-level address mapping, comprising the following steps: S1, collecting and preprocessing a first government system address and a second government system address respectively to obtain address text data; S2, obtaining each level contained in the first government system address and the second government system address, and establishing a standard address coding level system according to the levels; S3, performing address semantic alignment: based on the address text data, calculating the semantic similarity between the first government system address and the second government system address through level and semantic feature fusion, ambiguous address identification and attention weight calibration mechanism, if the semantic similarity is greater than a first threshold, outputting a semantic matching address pair; S4, performing address space alignment: based on the semantic matching address pair, fusing multi-dimensional space features of overlapping area, center point distance and area difference, and dynamically calculating the spatial overlap degree of the semantic matching address pair by using an adaptive coefficient, if the spatial overlap degree is greater than a second threshold, outputting a spatial matching address pair; S5, according to the standard address coding level system, implementing standard address coding on the first government system address and the second government system address in the spatial matching address pair respectively.

[0006] The application also provides a government data address alignment system based on multi-level address mapping, comprising: a collection and preprocessing module for collecting and preprocessing a first government system address and a second government system address respectively to obtain address text data; a system construction module for obtaining each level contained in the first government system address and the second government system address, and establishing a standard address coding level system according to the levels; an address semantic alignment module for performing address semantic alignment: based on the address text data, calculating the semantic similarity between the first government system address and the second government system address through level and semantic feature fusion, ambiguous address identification and attention weight calibration mechanism, if the semantic similarity is greater than a first threshold, outputting a semantic matching address pair; an address space alignment module for performing address space alignment: based on the semantic matching address pair, fusing multi-dimensional space features of overlapping area, center point distance and area difference, and dynamically calculating the spatial overlap degree of the semantic matching address pair by using an adaptive coefficient, if the spatial overlap degree is greater than a second threshold, outputting a spatial matching address pair; an address coding module for implementing standard address coding on the first government system address and the second government system address in the spatial matching address pair respectively according to the standard address coding level system.

[0007] The beneficial effects of the present application are embodied in that the multi-level address mapping based government data address alignment method and system provided by the present application realizes standardized alignment output of different government system addresses through precise semantic alignment and spatial alignment, realizes automation and high precision of alignment, greatly improves processing efficiency, reduces dependence on manual intervention, and effectively makes up for the defects of low efficiency and poor accuracy of the prior art. In addition, the present application provides reliable input for hierarchical splitting through the mutual coordination of various functional modules and the fusion of gating mechanism high-quality address features, the results of hierarchical splitting allow ambiguity recognition and similarity calculation to focus on the same level association, the interference items removed by ambiguity recognition reduce the invalid load of similarity calculation, and the attention mechanism optimizes the core level weight based on the results of the foregoing modules, solving the problem that a single module or a simple combination of modules cannot cope with complex address scenarios, and being able to adaptively process semantic alignment of address texts with different description specifications, different hierarchical integrity and different ambiguity. Furthermore, the present application realizes spatial information complementation through a multi-source map system, eliminates system deviation through coordinate calibration, dynamically adjusts weights through adaptive coefficients, adapts different scale address pairs through weight adaptation, and realizes comprehensive judgment through multi-physical quantity integration, greatly improving the accuracy and scene adaptability of spatial alignment, and effectively solving the defects of data deviation, fixed weight and single dimension in the prior art. BRIEF DESCRIPTION OF DRAWINGS

[0008] Figure 1 The multi-level address mapping based government data address alignment method flowchart provided by the present application is provided in Embodiment 1 of the present application.

[0009] Figure 2 The method flowchart for performing address semantic alignment provided by the present application is provided in Embodiment 1 of the present application.

[0010] Figure 3 The method flowchart for performing address spatial alignment provided by the present application is provided in Embodiment 1 of the present application.

[0011] Figure 4 The multi-level address mapping based government data address alignment system structure schematic diagram provided by the present application is provided in Embodiment 1 of the present application. DETAILED DESCRIPTION

[0012] The technical solutions in the embodiments of the present application will be described in detail below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor are within the scope of protection of the present application.

[0013] Embodiment 1

[0014] This embodiment provides a method for aligning government data addresses based on multi-level address mapping, such as... Figure 1 As shown, it includes the following steps:

[0015] Step S1: Collect the addresses of the first and second government service systems respectively and preprocess them to obtain address text data. Specifically, this involves collecting address data from the first government service system (such as the code, name, and jurisdiction description of streets / communities / grids) and address data from the second government service system (the code, name, and jurisdiction boundary information of security districts and responsibility areas), cleaning up invalid data, and standardizing the text description format. For example, the first government service system can be the general government service system of various levels of departments, and the second government service system can be the government service system that leads public security management.

[0016] Step S2: Obtain the various levels contained in the first and second government system addresses, and merge them to establish a standard address coding hierarchy system. Specifically, combining the levels of the two government systems, and based on the hierarchical logic of "province-city-district-street-community-grid-security area-responsibility area-building-unit-household," formulate generation rules and assignment standards for each segment of the code to form a unique standard address code, which serves as the unified output format for address alignment results. For example, a 12-segment standard address code can be set.

[0017] Step S3: Perform address semantic alignment: Based on address text data, through hierarchical and semantic feature fusion, ambiguous address identification, and attention weight calibration mechanisms, calculate the semantic similarity between the first and second government system addresses. If the semantic similarity is greater than a first threshold, the output is a semantically matched address pair. Specifically, different addresses follow different hierarchical logics. For example, general government addresses follow the hierarchical logic of province, city, district, street, community, and grid, while public security management system addresses follow the hierarchical logic of province, city, district, public security district, and responsibility district. There is no unified description standard for the two systems. In addition, address texts generally have problems such as incomplete descriptions (e.g., missing hierarchical information) and non-standard expressions (e.g., using abbreviations or aliases). There are also ambiguities such as "same name, different areas," for example, the same community name may belong to different streets. These characteristics make it easy for direct matching of the two types of address texts to fail due to hierarchical mismatch, description differences, or ambiguity interference, which is the core difficulty in achieving semantic alignment in this scenario. Therefore, this embodiment can build a semantic understanding network, input two types of address text data, and let the model learn the name association rules of the two types of addresses (such as the semantic correspondence between "XX Community" and "XX Responsibility Area"), automatically identify ambiguous information in the address description (addresses with the same name but different areas), and filter out semantically matching address pairs.

[0018] In one alternative implementation, such asFigure 2 As shown, step S3 is implemented by the following steps:

[0019] Step S31: For the first and second government system addresses in the address text data, a hierarchical feature vector is extracted according to the integrity of the address text data, a semantic feature vector is extracted based on a pre-trained language model, and a comprehensive address feature vector is generated by fusing the hierarchical feature vector and the semantic feature vector through a gating fusion mechanism. Specifically, step S31 can extract hierarchical features and semantic features and dynamically fuse them through a gating mechanism, and adaptively adjust the weights of the two types of features according to the actual situation of the address text. For example, for an address with complete hierarchy, the gating mechanism can increase the weight of the hierarchical features to ensure the accuracy of the hierarchy; for an address with missing hierarchy but rich semantics, the gating mechanism can increase the weight of the semantic features, and capture the core semantic association through the BERT semantic model, thereby generating address features that take into account both hierarchy and semantics, avoiding the limitations of single feature comparison.

[0020] In an optional implementation, step S31 can be implemented by the following steps:

[0021] Step S311: Based on the integrity of each data hierarchical structure in the address text data, a hierarchical index vector is generated, wherein the elements of the hierarchical index vector are used to indicate whether each hierarchy exists; specifically, the dimension of the hierarchical index vector can be determined according to the number of address levels, if a certain hierarchy exists in the address, then the corresponding element of the hierarchical index vector is 1, otherwise it is 0. For example, if a complete description of a government address is "XX province-XX city-XX district-XX street-XX community-XX grid", then the hierarchical index vector is a 6-dimensional vector (corresponding to 6 levels), and the existing hierarchy is filled with 1 and the missing hierarchy is filled with 0, i.e. the hierarchical index vector is [1, 1, 1, 1, 1, 1], which corresponds to the address of "province, city, district, street, community, grid" respectively.

[0022] Step S312: According to the hierarchical index vector and the corresponding hierarchical embedding matrix, a hierarchical feature vector is generated; specifically, the hierarchical features and semantic features can be extracted according to the first and second government system address texts, and then the two types of features can be dynamically fused through a gating mechanism to obtain two types of government address features, which can be calculated in the following way:

[0023] ;

[0024] ;

[0025] wherein, is the first government system address hierarchical feature vector, is the first government system hierarchical embedding matrix, a first government system level index vector, a second government system address level feature vector, a second government system address level embedding matrix, a second government system address level index vector.

[0026] In step S313, the address text data is input into the pre-trained language model to generate a semantic feature vector. The semantic feature vector can be generated by the following calculation method:

[0027] ;

[0028] ;

[0029] wherein, a first government system address semantic feature vector, a pre-trained language model, a current first government system address text, a second government system address semantic feature vector, a current second government system address text.

[0030] In step S314, the level feature vector and the semantic feature vector are input into a gating unit, and the level feature vector and the semantic feature vector are weighted and fused using a pre-calculated gating coefficient, and a comprehensive address feature vector is output. The high-quality address feature fused by the gating mechanism of the embodiment can provide reliable input for the next step of level splitting.

[0031] In an optional embodiment, the level feature vector and the semantic feature vector are weighted and fused using a pre-calculated gating coefficient in step S314, which is realized by the following method: the gating unit calculates the gating coefficient using an activation function; the gating coefficient is used as the weight of the level feature vector, and the complementary value of the gating coefficient is used as the weight of the semantic feature vector, wherein the complementary value of the gating coefficient and the gating coefficient add up to 1; the level feature vector and the semantic feature vector are weighted and summed using the weight of the level feature vector and the weight of the semantic feature vector to obtain a comprehensive address feature vector.

[0032] In an optional embodiment, step S314 can be realized by the following calculation method:

[0033] ;

[0034] ;

[0035] ;

[0036] ;

[0037] wherein, is a first government system gating coefficient, is an activation function, is a first government system first weight matrix, is a first government system second weight matrix, is a first government system bias vector, is a second government system gating coefficient, is a second government system first weight matrix, is a second government system second weight matrix, is a second government system bias vector, is a first government system address feature, is a second government system address feature.

[0038] Step S32: According to the preset address level index, the integrated address feature vector is split into sub-feature vectors corresponding to different levels; Specifically, the preset address level index can be the aforementioned level index vector, which can split the address features according to the level to focus on the semantic association comparison between addresses of the same level, reduce the error caused by cross-level comparison, and make the subsequent calculation more targeted. According to the first government system address level index vector and the second government system address level index vector, the first government system address features and the second government system address features can be split according to the level dimension to obtain government address features of each level, which can be calculated in the following way:

[0039] ;

[0040] ;

[0041] wherein, is a level slicing function; is the i-th element in the first government system level index vector, i is the i-th element in the first government system level index vector, is the first government system address feature of the i-th level, i is the first government system level index, i (e.g. corresponding to XX province-XX city-XX district-XX street-XX community-XX grid); is the i-th element in the second government system level index vector, is the i-th element in the second government system level index vector, j is the second government system address feature of the i-th level, is the second government system level index, j j ​​(for example, corresponding to -XX police district - XX responsibility area - XX building - XX unit - XX household). The result of the hierarchical splitting of the present embodiment can focus the subsequent ambiguity recognition and similarity calculation on the same level association.

[0042] Step S33: Based on the sub-feature vectors, the semantic consistency between adjacent levels is calculated, and the ambiguity score of the address pair is calculated combined with the character edit distance, to identify and filter the ambiguous address pairs; specifically, step S33 calculates the semantic consistency between adjacent levels to judge the semantic association of the internal levels of the address (low association may exist contradictory description), combined with the character edit distance to capture the character difference of the same name address, both of which together build the ambiguity score, accurately mark the ambiguous address pairs, and exclude the combinations that are easy to cause matching errors in advance, reducing the invalid load of subsequent calculation.

[0043] In an optional implementation, step S33 is specifically implemented by the following steps:

[0044] Step S331, respectively calculate the hierarchical semantic consistency of the first government system address and the hierarchical semantic consistency of the second government system address, the hierarchical semantic consistency of the first government system address is the average value of the cosine similarity between the feature vectors of the adjacent levels in the first government system address, and the hierarchical semantic consistency of the second government system address is the average value of the cosine similarity between the feature vectors of the adjacent levels in the second government system address; Specifically, the following methods can be used to calculate:

[0045] ;

[0046] ;

[0047] Wherein, represents the hierarchical semantic consistency of the first government system address, represents the average value calculation, represents the calculation of cosine similarity, represents the hierarchical semantic consistency of the second government system address; , , and respectively represent take 1, 2, 5 and 6, , , and respectively represent take 1, 2, 5 and 6.

[0048] ​​Step S332, calculate the character residual term, which is based on the character edit distance between the address texts with the same name; specifically, the character residual term can be calculated according to the following manner:

[0049]

[0050] wherein, is the character residual term, is the edit distance function, is the other government address text with the same name as the current first government system address text .

[0051] Step S333, calculate the ambiguity score based on the hierarchical semantic consistency of the first government system address, the hierarchical semantic consistency of the second government system address, and the character residual term, wherein the hierarchical semantic consistency contributes to the calculation of the ambiguity score in a negative correlation manner, and the character residual term contributes to the calculation of the ambiguity score in a positive correlation manner; specifically, the ambiguity score can be calculated according to the following manner:

[0052]

[0053] wherein, is the ambiguity score, and are the first and second ambiguity score coefficients, respectively.

[0054] Step S334, compare the calculated ambiguity score with a preset ambiguity threshold value; if the ambiguity score is higher than the ambiguity threshold value, mark it as an ambiguous address pair. Specifically, if wherein θ represents the ambiguity threshold value, mark it as an “ambiguous address pair”, otherwise mark it as a “non-ambiguous address pair”.

[0055] Step S34: for the filtered non-ambiguous address pairs, calculate the semantic similarity using the attention mechanism; if the semantic similarity is greater than a first threshold value, output the semantic matching address pair. Specifically, in step S34, the hierarchical weight can be calibrated by combining the attention mechanism, the weight of the core level (such as community, responsibility area, etc.) can be automatically increased, and the interference of the secondary level (such as province, city, etc.) can be weakened, and then the total similarity can be calculated by combining the weight and the hierarchical semantic similarity, so as to ensure that the matching result can truly reflect the actual correlation degree of the two types of addresses.

[0056] In one specific embodiment, according to the labeled “non-ambiguous address pair”, the semantic similarity of the two government systems is calculated and the matching address pair is screened by combining the attention mechanism; the specific calculation manner can be as follows:

[0057]

[0058] ​​​ ;

[0059]

[0060] wherein, is the calibrated first government system level weight, is the attention mechanism, is the calibrated second government system level weight, is the first government system-second government system semantic similarity, is the weight of the i th first government system level, is the weight of the j th second government system level. If , wherein is the first threshold, the semantic matching address pair is included, otherwise it is excluded.

[0061] Step S4: Perform address space alignment: based on the semantic matching address pair, the multi-dimensional spatial features of the overlapping area, the center point distance and the area difference are fused, and the spatial overlap degree of the semantic matching address pair is dynamically calculated by using an adaptive coefficient, and if the spatial overlap degree is greater than a second threshold, the output is a spatial matching address pair; Specifically, the core feature of the precise alignment scene of the address space of different government systems is the multi-source nature of the spatial data source and the difference of the address jurisdiction range. At the same time, the area size of the two types of addresses is significantly different, and part of the address jurisdiction boundary is irregular in shape, and it is difficult to accurately judge the matching degree by only using a single overlapping area or center point distance. Therefore, the spatial overlap degree of the address jurisdiction range of the two government addresses is verified by using multi-dimensional spatial features in this embodiment, the spatial matching of the address pair is completed, and the alignment result is generated.

[0062] In an optional implementation, as shown in Figure 3 , step S4 can be implemented by the following specific steps:

[0063] Step S41: Based on the semantic matching address pair, the spatial data corresponding to the first government system address and the second government system address is obtained and calibrated by using a multi-source map system; Specifically, the multi-source map system can include map systems such as a geographic information system (GIS), Gaode map, Baidu map, etc., and the coordinate systems used by the two types of systems are different, which can easily lead to non-uniformity of the spatial data reference, so calibration fusion is also needed. In an optional implementation, step S41 can include:

[0064] Step S411, calling the first map system to obtain the jurisdictional area boundary range corresponding to the first government system address and the second government system address in the semantically matched address; specifically, the first map system is used to determine the jurisdictional boundary, which can adopt GIS, and through GIS, the latitude and longitude range of the jurisdictional area of each of the two government addresses (i.e., the minimum latitude, maximum latitude, minimum longitude, and maximum longitude corresponding to each address) is extracted according to the address text to form the boundary range.

[0065] Step S412, calling the second map system to obtain the center point coordinates corresponding to the first government system address and the second government system address in the semantically matched address; the second map system is used to determine the address center point coordinates, which can adopt map systems such as Gaode Map and Baidu Map, and through calling the Gaode Map interface, the high-precision center point spatial coordinates (i.e., the specific latitude and longitude values) of the two government addresses are obtained according to the same address text to obtain the spatial basic data of the two types of addresses.

[0066] Step S413, converting the jurisdictional area boundary range and the center point coordinates to the same coordinate system for calibration processing through a preset coordinate conversion rule. Specifically, the latitude and longitude range extracted by GIS and the center point coordinates obtained by Gaode Map can be converted to the same coordinate system through a preset coordinate conversion rule (based on the industry common coordinate system conversion standard) to eliminate the coordinate deviation between different systems, and the calibrated latitude and longitude range and center point coordinates are obtained.

[0067] S42: calculating the spatial physical quantity data based on the calibrated data in step S41, and obtaining a multi-dimensional address pair feature vector through operation and coding according to the spatial physical quantity data, wherein the spatial physical quantity data at least includes: the total area of the jurisdictional area corresponding to the first government system address, the total area of the jurisdictional area corresponding to the second government system address, the overlapping area of the first government system address and the second government system address, and the distance between the two center points of the first government system address and the second government system address; specifically, in an optional implementation, step S42 can be implemented through the following steps:

[0068] Step S421, calculating the total area of the jurisdictional area corresponding to the first government system address, the total area of the jurisdictional area corresponding to the second government system address, and the overlapping area of the first government system address and the second government system address according to the calibrated jurisdictional area boundary range; specifically, based on the calibrated latitude and longitude range and the calibrated center point coordinates, the overlapping part of the jurisdictional areas of the two government addresses is first determined through the spatial intersection calculation function of GIS, and then the area of the overlapping part is calculated, and the total areas of the jurisdictional areas of the two government addresses are calculated respectively.

[0069] Step S422, calculate the distance between the two center points of the first government system address and the second government system address according to the calibrated center point coordinates; through the above two steps, the dual-dimensional requirements of "overlapping area reflecting jurisdiction range correlation degree and center point distance reflecting position core degree" can be considered to realize spatial precise alignment.

[0070] Step S423, calculate the area difference value and the area sum value using the total area of the jurisdiction area corresponding to the first government system address and the total area of the jurisdiction area corresponding to the second government system address; specifically, the calculation method of the area difference value is , and the calculation method of the area sum value is , wherein is the total area of the jurisdiction area of the first government system address, is the total area of the jurisdiction area of the second government system address.

[0071] Step S424, calculate the normalized distance using the distance between the two center points of the first government system address and the second government system address and the preset maximum distance threshold; specifically, the calculation method of the normalized distance is , wherein is the distance between the two center point coordinates of the calibrated government addresses, is the preset maximum distance threshold, which can be set to 5000 meters by default.

[0072] Step S425, encode the address pair feature vector based on the area difference value, the area sum value and the normalized distance. Specifically, the address pair feature vector can be encoded as: , wherein is the overlapping area. In this way, the area difference of the two government addresses, the normalized center point distance and the overlapping area ratio are integrated and encoded as the address pair feature vector, converting the scattered physical quantities into unified calculation inputs to realize the organic integration of multi-dimensional spatial information.

[0073] S43: input the address pair feature vector into the adaptive coefficient calculation neural network to output the first adaptive coefficient and the second adaptive coefficient; specifically, the first and second adaptive coefficients can be calculated through the first and second adaptive coefficient weight matrices, the first and second adaptive coefficient bias vectors, combined with the ReLU activation function and the Sigmoid activation function, so that the system can dynamically allocate weights according to the area difference degree and the center point distance of different address pairs, for example, increase the area related weight for the address pair with small area difference, and reduce the distance attenuation influence for the address pair with short center point distance, to realize data-driven dynamic adaptation. The calculation method of the first and second adaptive coefficients according to the address pair feature vector can be as follows:

[0074] ;

[0075] wherein, is an activation function, and are first and second adaptive coefficients, respectively, and are first and second adaptive coefficient weight matrices, respectively, and are first and second adaptive coefficient bias vectors, respectively.

[0076] S44: Calculate the spatial overlap degree according to the first adaptive coefficient, the second adaptive coefficient and the spatial physical quantity data; in an optional embodiment, step S44 can be implemented by the following method: taking the ratio of the overlap area to the smaller total area in the first administrative system address and the second administrative system address as the basic overlap degree indicator, adjusting the weight of the basic overlap degree indicator by the first adaptive coefficient, and combining the second adaptive coefficient and the normalized distance as the factor of the exponential function, multiplying the weighted and adjusted basic overlap degree indicator by the exponential function to obtain the spatial overlap degree. The specific calculation formula can be as follows:

[0077] ;

[0078] wherein, is an adaptive spatial overlap degree, is a minimum value function, k is a basic balance coefficient, is an exponential function. In this step, the adaptive spatial overlap degree is calculated by combining the adaptive coefficients, which not only retains the basic correlation indicator of the "overlap area to the smaller total area ratio", but also adjusts the weight of the area correlation by the first adaptive coefficient and controls the distance decay intensity by the second adaptive coefficient, solving the problem of weight distortion when matching different scale address pairs (such as small grid and large responsibility area).

[0079] S45: Compare the spatial overlap degree with the preset second threshold value, if the spatial overlap degree is greater than the second threshold value, it is determined as a spatial matching address pair. Specifically, if the adaptive spatial overlap degree is greater than the preset second threshold value (for example, the default setting is 0.7), it is determined that the address pair is a "spatial matching address pair"; otherwise, it is determined as "spatial mismatch".

[0080] Step S5: According to the standard address coding hierarchy system, the first government system address and the second government system address in the spatial matching address pair are respectively implemented standard address coding. Specifically, the standard address coding (for example, 12 segment standard address coding) set in step S2 can be allocated for the spatially matched address pair, and the final alignment result containing the first government system address text, the second government system address text, the 12 segment coding, the spatial overlap degree and the center point distance is generated.

[0081] The government data address alignment method based on multi-level address mapping provided by the embodiment realizes the standardized alignment output of different government system addresses through accurate semantic alignment and spatial alignment, realizes the automation and high precision of alignment, greatly improves the processing efficiency, reduces the dependence on manual intervention, and effectively makes up for the defects of low efficiency and poor accuracy of the prior art. In addition, the high-quality address features of the gate control mechanism fusion provided by the mutual coordination of each functional module provide reliable input for hierarchical splitting, the results of hierarchical splitting allow ambiguity recognition and similarity calculation to focus on the same level association, the interference items removed by ambiguity recognition reduce the invalid load of similarity calculation, and the attention mechanism optimizes the core level weight based on the results of the foregoing modules. The problem that a single module or a simple combination of modules cannot cope with complex address scenarios is solved, and the semantic alignment of address texts with different description specifications, different hierarchical integrity and different ambiguity situations can be adaptively processed. Furthermore, the embodiment realizes spatial information complementation through a multi-source map system, eliminates system deviation through coordinate calibration, adjusts the weight of different scale address pairs through adaptive coefficients, and realizes comprehensive judgment through the integration of multiple physical quantities, greatly improving the accuracy and scene adaptability of spatial alignment, and effectively solving the defects of data deviation, fixed weight and single dimension in the prior art.

[0082] The embodiment also provides a government data address alignment system based on multi-level address mapping, which is used to implement the aforementioned government data address alignment method based on multi-level address mapping. The content has been described in the foregoing method, and will not be described here. Only the system structure is simply described. As shown in Figure 4 The government data address alignment system based on multi-level address mapping of the embodiment specifically includes:

[0083] The acquisition and preprocessing module 401 is used to acquire and preprocess the first government system address and the second government system address respectively, and obtain address text data.

[0084] The system construction module 402 is used to obtain each level contained in the first government system address and the second government system address, and fuse and establish a standard address coding hierarchy system according to the level.

[0085] The address semantic alignment module 403 is configured to perform address semantic alignment. Based on the address text data, the semantic similarity between the first government system address and the second government system address is calculated through hierarchical and semantic feature fusion, ambiguous address identification, and attention weight calibration mechanism. If the semantic similarity is greater than a first threshold value, the output is a semantic matching address pair.

[0086] The address space alignment module 404 is configured to perform address space alignment. Based on the semantic matching address pair, the spatial overlap degree of the semantic matching address pair is dynamically calculated by fusing multi-dimensional space features such as overlapping area, center point distance, and area difference, and using an adaptive coefficient. If the spatial overlap degree is greater than a second threshold value, the output is a spatial matching address pair.

[0087] The address coding module 405 is configured to implement standard address coding on the first government system address and the second government system address in the spatial matching address pair according to a standard address coding hierarchical system.

[0088] The government data address alignment system based on multi-level address mapping provided in the embodiment realizes standardized alignment output of addresses of different government systems through precise semantic alignment and spatial alignment, realizes automation and high precision of alignment, greatly improves processing efficiency, reduces dependence on manual intervention, and effectively makes up for the defects of low efficiency and poor accuracy of the prior art. In addition, the high-quality address features provided by the mutual coordination of various functional modules and the gating mechanism fusion provide reliable input for hierarchical splitting. The results of hierarchical splitting allow ambiguous identification and similarity calculation to focus on the same level association. The interference items removed by ambiguous identification reduce the invalid load of similarity calculation. The attention mechanism optimizes the core level weight based on the results of the foregoing modules, solves the problem that a single module or a simple combination of modules cannot cope with complex address scenarios, and can adaptively process semantic alignment of address texts with different description specifications, different hierarchical integrity, and different ambiguity. Furthermore, the embodiment realizes spatial information complementation through a multi-source map system, eliminates system deviation through coordinate calibration, dynamically adjusts the weight of different scale address pairs through an adaptive coefficient, and realizes comprehensive judgment through multi-physical quantity integration, which greatly improves the accuracy and scene adaptability of spatial alignment and effectively solves the defects of data deviation, fixed weight, and single dimension in the prior art.

[0089] The embodiment also provides a computer readable storage medium including computer instructions, which, when executed on an electronic device, cause the electronic device to perform the government data address alignment method based on multi-level address mapping as described above.

[0090] In the description of the embodiments of the present application, the terms "first", "second", etc. are used only to describe purposes and can not be understood as indicating or implying relative importance or implicitly indicating the number of the technical features indicated. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features.

[0091] In the description of the embodiments of the present application, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0092] Although the embodiments of the present application have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, alternatives, and variations can be made thereto without departing from the principles and spirit of the application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for aligning government data addresses based on multi-level address mapping, characterized in that, include: S1: Collect the addresses of the first and second government systems respectively, and preprocess them to obtain address text data; S2: Obtain the first government system address and the various levels contained in the second government system address, and merge them according to the levels to establish a standard address coding hierarchy system; S3: Perform address semantic alignment: Based on the address text data, calculate the semantic similarity between the first government system address and the second government system address through hierarchical and semantic feature fusion, ambiguous address identification and attention weight calibration mechanism. If the semantic similarity is greater than the first threshold, output a semantically matched address pair. S4: Perform address space alignment: Based on the semantically matched address pair, by fusing multi-dimensional spatial features such as overlapping area, center point distance and area difference, and using adaptive coefficients to dynamically calculate the spatial overlap of the semantically matched address pair, if the spatial overlap is greater than the second threshold, then the output is a spatially matched address pair; S5: According to the standard address coding hierarchy, standard address coding is applied to the first government system address and the second government system address in the spatial matching address pair, respectively; Specifically, step S3 includes: S31: For the first government system address and the second government system address in the address text data, extract hierarchical feature vectors based on the integrity of the address text data, extract semantic feature vectors based on a pre-trained language model, and fuse the hierarchical feature vectors and the semantic feature vectors using a gating fusion mechanism to generate a comprehensive address feature vector; S32: Based on the preset address level index, the comprehensive address feature vector is split into sub-feature vectors corresponding to different levels; S33: Based on the sub-feature vector, calculate the semantic consistency between adjacent levels, and combine it with the character edit distance to calculate the ambiguity score of the address pair in order to identify and filter ambiguous address pairs; S34: For the filtered unambiguous address pairs, use an attention mechanism to calculate the semantic similarity. If the semantic similarity is greater than the first threshold, output a semantically matched address pair. Step S4 specifically includes: S41: Based on the semantic matching address pair, use a multi-source map system to obtain the spatial data corresponding to the first government system address and the second government system address respectively, and perform calibration; S42: Calculate the space physical quantity data based on the data calibrated in step S41, and perform operations and encoding on the space physical quantity data to obtain a multi-dimensional address-pair feature vector; S43: Input the address pair feature vector into the adaptive coefficient calculation neural network, and output the first adaptive coefficient and the second adaptive coefficient; S44: Calculate the spatial overlap based on the first adaptive coefficient, the second adaptive coefficient, and the spatial physical quantity data; S45: Compare the spatial overlap with a preset second threshold. If the spatial overlap is greater than the second threshold, then determine that the spatial address pair is a matching address pair.

2. The method for aligning government data addresses based on multi-level address mapping according to claim 1, characterized in that, Step S31 specifically includes: Based on the integrity of each data hierarchy in the address text data, a hierarchy index vector is generated, wherein the elements of the hierarchy index vector are used to indicate whether each hierarchy exists; The hierarchical feature vector is generated based on the hierarchical index vector and the corresponding hierarchical embedding matrix; The address text data is input into the pre-trained language model to generate the semantic feature vector; The hierarchical feature vector and the semantic feature vector are input into the gating unit, and the hierarchical feature vector and the semantic feature vector are weighted and fused using pre-calculated gating coefficients to output the comprehensive address feature vector.

3. The method for aligning government data addresses based on multi-level address mapping according to claim 2, characterized in that, The step of using pre-calculated gating coefficients to weight and fuse the hierarchical feature vector and the semantic feature vector specifically includes: The gating unit calculates the gating coefficient using an activation function; The gate coefficient is used as the weight of the hierarchical feature vector, and the complementary value of the gate coefficient is used as the weight of the semantic feature vector, wherein the complementary value of the gate coefficient is added to the gate coefficient to get 1; The comprehensive address feature vector is obtained by weighting and summing the hierarchical feature vector and the semantic feature vector using the weights of the hierarchical feature vector and the semantic feature vector.

4. The method for aligning government data addresses based on multi-level address mapping according to claim 1, characterized in that, Step S33 specifically includes: The hierarchical semantic consistency of the first government system address and the hierarchical semantic consistency of the second government system address are calculated respectively. The hierarchical semantic consistency of the first government system address is the average cosine similarity between the feature vectors of adjacent levels in the first government system address, and the hierarchical semantic consistency of the second government system address is the average cosine similarity between the feature vectors of adjacent levels in the second government system address. Calculate the character residual term, which is derived based on the character edit distance between address texts with the same name; The ambiguity score is calculated based on the hierarchical semantic consistency of the first government system address, the hierarchical semantic consistency of the second government system address, and the character residual term, wherein the hierarchical semantic consistency contributes to the calculation of the ambiguity score in a negatively correlated manner, and the character residual term contributes to the calculation of the ambiguity score in a positively correlated manner. The calculated ambiguity score is compared with a preset ambiguity threshold. If the ambiguity score is higher than the ambiguity threshold, it is marked as the ambiguous address pair.

5. The method for aligning government data addresses based on multi-level address mapping according to claim 1, characterized in that, The spatial physical quantity data includes at least: the total area of ​​the jurisdiction corresponding to the first government system address, the total area of ​​the jurisdiction corresponding to the second government system address, the overlapping area of ​​the first government system address and the second government system address, and the distance between the two center points of the first government system address and the second government system address.

6. The method for aligning government data addresses based on multi-level address mapping according to claim 5, characterized in that, Step S41 specifically includes: Call the first map system to obtain the jurisdictional boundary ranges corresponding to the first government system address and the second government system address in the semantic matching address; Call the second map system to obtain the center point coordinates corresponding to the first government system address and the second government system address in the semantic matching address; By using preset coordinate transformation rules, the boundary range of the managed area and the coordinates of the center point are uniformly transformed to the same coordinate system for calibration.

7. The method for aligning government data addresses based on multi-level address mapping according to claim 6, characterized in that, Step S42 specifically includes: Calculate the total area of ​​the jurisdiction corresponding to the first government system address, the total area of ​​the jurisdiction corresponding to the second government system address, and the overlapping area of ​​the first government system address and the second government system address based on the calibrated jurisdiction boundary range. Calculate the distance between the two center points of the first government system address and the second government system address based on the calibrated center point coordinates; The area difference and the total area are calculated using the total area of ​​the jurisdiction corresponding to the first government system address and the total area of ​​the jurisdiction corresponding to the second government system address; The normalized distance is calculated using the distance between the two central points of the first government system address and the second government system address, as well as a preset maximum distance threshold. The address pair feature vector is obtained by encoding based on the area difference value, the total area value, and the normalized distance.

8. The method for aligning government data addresses based on multi-level address mapping according to claim 7, characterized in that, Step S44 specifically includes: The basic overlap degree index is obtained by using the ratio of the overlapping area to the smaller total area of ​​the first government system address and the second government system address as the basic overlap degree index, adjusting the weight of the basic overlap degree index by a power of the first adaptive coefficient, and combining the second adaptive coefficient and the normalized distance as factors of the exponential function. The weighted basic overlap degree index is then multiplied by the exponential function to obtain the spatial overlap degree.

9. A government data address alignment system based on multi-level address mapping, characterized in that, include: The data acquisition and preprocessing module is used to collect the addresses of the first government system and the second government system respectively, and preprocess them to obtain address text data; The system construction module is used to obtain the various levels contained in the first government system address and the second government system address, and to merge them according to the levels to establish a standard address coding hierarchy system; The address semantic alignment module is used to perform address semantic alignment: based on the address text data, through hierarchical and semantic feature fusion, ambiguous address identification and attention weight calibration mechanism, the semantic similarity between the first government system address and the second government system address is calculated; if the semantic similarity is greater than a first threshold, the output is a semantically matched address pair. The address space alignment module is used to perform address space alignment: based on the semantically matched address pair, it integrates multi-dimensional spatial features such as overlapping area, center point distance and area difference, and dynamically calculates the spatial overlap of the semantically matched address pair using adaptive coefficients. If the spatial overlap is greater than a second threshold, the output is a spatially matched address pair. The address encoding module is used to implement standard address encoding for the first government system address and the second government system address in the spatial matching address pair according to the standard address encoding hierarchy system. Specifically, the address semantic alignment module is used for: extracting hierarchical feature vectors based on the integrity of the address text data for the first government system address and the second government system address in the address text data; extracting semantic feature vectors based on a pre-trained language model; fusing the hierarchical feature vectors and the semantic feature vectors using a gating fusion mechanism to generate a comprehensive address feature vector; splitting the comprehensive address feature vector into sub-feature vectors corresponding to different levels according to a preset address level index; calculating the semantic consistency between adjacent levels based on the sub-feature vectors; and calculating the ambiguity score of address pairs by combining the character edit distance, so as to identify and filter ambiguous address pairs; and calculating the semantic similarity of the filtered unambiguous address pairs using an attention mechanism. If the semantic similarity is greater than a first threshold, the output is a semantically matched address pair. The address space alignment module is specifically used for: based on the semantically matched address pair, using a multi-source map system to obtain and calibrate the spatial data corresponding to the first government system address and the second government system address respectively; calculating spatial physical quantity data based on the calibrated data, and performing calculations and encoding based on the spatial physical quantity data to obtain a multi-dimensional address pair feature vector; inputting the address pair feature vector into an adaptive coefficient calculation neural network, and outputting a first adaptive coefficient and a second adaptive coefficient; calculating the spatial overlap based on the first adaptive coefficient, the second adaptive coefficient, and the spatial physical quantity data; comparing the spatial overlap with a preset second threshold, and if the spatial overlap is greater than the second threshold, then determining it as the spatially matched address pair.

Images