Address data matching method and related equipment

Abstract

The embodiment of the invention provides an address data matching method and related equipment, and belongs to the technical field of geographic information services. The method comprises the following steps: constructing an address annotation corpus according to input address information data and a preset address database; the method comprises the following steps: generating a geographic information embedding vector according to a preset geographic information knowledge graph, performing address element analysis in combination with an address annotation corpus to obtain an address element sequence so as to construct a dictionary tree, and performing similarity screening through a spatial hierarchical matching algorithm to obtain a similar address set; generating an address embedding vector matrix through a preset word embedding vector model, and performing feature extraction through a preset semantic feature extraction model to obtain semantic-level similar features; according to input address information data, multi-dimensional character similarity matching is carried out to obtain character-level similar features, then weighted fusion is carried out in combination with semantic-level similar features, and target matching address data is determined according to a weighted fusion result. According to the embodiment of the invention, the address data matching accuracy and efficiency can be improved.

Description

Technical Field

[0001] This application relates to the field of geographic information service technology, and in particular to an address data matching method and related equipment. Background Technology

[0002] Input address information is related to production and daily life information. However, due to the lack of a unified standard for the storage and representation of address information, there are significant differences in description methods and address structures between data from different sources, which limits the use and sharing of data and prevents the formation of effective communication relationships between data. Related technologies mainly rely on manual comparison or comparison based on overlaying vector layers for address matching, which has low matching efficiency and accuracy.

[0003] In summary, the technical problems existing in the relevant technologies need to be improved. Summary of the Invention

[0004] The main objective of this application is to propose an address data matching method and related equipment, which can effectively improve the accuracy and efficiency of address data matching.

[0005] To achieve the above objectives, one aspect of this application proposes an address data matching method, the method comprising: Construct an address annotation corpus based on input address information data and a preset address database; Based on a pre-defined geographic information knowledge graph, graph embedding processing is performed to generate geographic information embedding vectors. Based on the address annotation corpus and the geographic information embedding vector, address elements are parsed using a preset deep learning model to obtain an address element sequence. A trie is constructed based on the address element sequence, and then a similarity filter is performed using a spatial hierarchical matching algorithm based on the trie to obtain a set of similar addresses. Based on the set of similar addresses, a corresponding address embedding vector matrix is ​​generated using a preset word embedding vector model. Then, a preset semantic feature extraction model is used to extract features from the address embedding vector matrix to obtain semantic-level similarity features. Multidimensional character similarity matching is performed based on the input address information data to obtain character-level similarity features; The semantic-level similarity features and the character-level similarity features are weighted and fused together, and the target matching address data is determined based on the weighted fusion result.

[0006] In some embodiments, the step of parsing address elements using a preset deep learning model based on the address annotation corpus and the geographic information embedding vector to obtain an address element sequence includes: Input address information data is fed into a pre-trained language model for vector space mapping to obtain address vector representation data; A multimodal vector is generated based on the address vector representation data and the geographic information embedding vector; The multimodal vector is input into the first bidirectional long short-term memory network model for semantic information extraction to obtain the associated sequence vector; The address element sequence is obtained by identifying address elements using a conditional random field model based on the associated sequence vector and the address annotation corpus.

[0007] In some embodiments, the step of constructing a trie based on the address element sequence, and then performing similarity filtering based on the trie using a spatial hierarchical matching algorithm to obtain a set of similar addresses includes: The trie is constructed according to the address element sequence and a preset address hierarchy. Based on the trie, candidate addresses are retrieved according to the hierarchical structure to obtain preset candidate addresses; The preset candidate addresses are matched and filtered using a preset matching strategy to obtain the set of similar addresses; wherein the preset matching strategy includes an exact matching strategy and a fuzzy matching strategy.

[0008] In some embodiments, the step of generating a corresponding address embedding vector matrix based on the similar address set using a preset word embedding vector model, and then extracting features from the address embedding vector matrix using a preset semantic feature extraction model to obtain semantic-level similarity features includes: The address embedding vector matrix is ​​generated based on the set of similar addresses using a fast text model; The address embedding vector matrix is ​​input into the second bidirectional long short-term memory network model for feature extraction to obtain global semantic features; The global semantic features are processed by a convolutional neural network model to obtain local semantic features; The global semantic features and the local semantic features are summed, and the summation result is then input into the attention module to generate a semantic feature vector. The semantic distance is calculated based on the semantic feature vector, and then the semantic similarity features are determined based on the calculated semantic distance data.

[0009] In some embodiments, the step of performing multidimensional character similarity matching based on the input address information data to obtain character-level similarity features includes: Based on preset Chinese address data, preset text feature statistics are performed to construct a character fuzzy replacement dictionary. Then, based on the input address information data, the edit distance similarity data is calculated through the character fuzzy replacement dictionary. The preset text features include misspellings, similar-looking characters, and homophones. Based on the input address information data, characters are split according to a preset text splitting method, and then character structure similarity is calculated to obtain character structure similarity data; wherein, the preset text splitting method includes character radical and stroke splitting; The similarity of the characters is calculated based on the pinyin information corresponding to the input address information data to obtain pinyin similarity data; Input operation similarity is calculated based on the input address information data and the preset input method keyboard layout information to obtain keyboard input similarity data; The preset weight parameters are dynamically determined based on the address level, and then the comprehensive character similarity data is calculated based on the preset weight parameters, the edit distance similarity data, the character structure similarity data, the pinyin similarity data, and the keyboard input similarity data. When the comprehensive character similarity data is determined to be greater than or equal to a preset similarity threshold, the character-level similarity features are obtained based on the comprehensive character similarity data.

[0010] In some embodiments, the semantic-level similarity feature includes candidate address semantic vectors, and the character-level similarity feature includes candidate address character similarity data; The step of performing weighted fusion based on the semantic-level similarity features and the character-level similarity features, and then determining the target matching address data based on the weighted fusion result, includes: Semantic similarity data is calculated using a cosine similarity algorithm based on the candidate address semantic vector and the input address semantic vector corresponding to the input address information data; The address matching score data is obtained by weighted fusion of the semantic similarity data and the candidate address character similarity data. The target matching address data is determined based on the address matching score data.

[0011] To achieve the above objectives, another aspect of this application provides an address data matching apparatus, the apparatus comprising: The first module is used to construct an address annotation corpus based on the input address information data and a preset address database. The second module is used to perform graph embedding processing based on a preset geographic information knowledge graph to generate geographic information embedding vectors. The third module is used to parse address elements based on the address labeling corpus and the geographic information embedding vector using a preset deep learning model to obtain an address element sequence. The fourth module is used to construct a trie based on the address element sequence, and then perform similarity filtering based on the trie using a spatial hierarchical matching algorithm to obtain a set of similar addresses; The fifth module is used to generate a corresponding address embedding vector matrix based on the set of similar addresses using a preset word embedding vector model, and then extract features from the address embedding vector matrix using a preset semantic feature extraction model to obtain semantic-level similarity features. The sixth module is used to perform multi-dimensional character similarity matching based on the input address information data to obtain character-level similarity features; The seventh module is used to perform weighted fusion based on the semantic-level similarity features and the character-level similarity features, and then determine the target matching address data based on the weighted fusion result.

[0012] To achieve the above objectives, another aspect of this application provides an electronic device, the electronic device comprising: At least one processor; At least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor performs the method described above.

[0013] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0014] To achieve the above objectives, another aspect of this application provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method. The embodiments of this application include at least the following beneficial effects: This application provides an address data matching method, apparatus, electronic device, storage medium, and program product. This solution constructs an address labeling corpus by inputting address information and a preset address database, and performs graph modeling based on a preset address information knowledge graph to generate geographic information embedding vectors. Then, based on the address labeling corpus and geographic information embedding vectors, address elements are parsed using a preset learning model to obtain an address element sequence. Next, the embodiments of this invention construct a trie based on the address element sequence, and then perform similarity filtering using a spatial hierarchical matching algorithm based on the trie to obtain a set of similar addresses. Based on the set of similar addresses, a corresponding address embedding vector matrix is ​​generated using a preset word embedding vector model. Then, a preset semantic feature extraction model is used to extract features from the address embedding vector matrix to obtain semantic-level similarity features. Simultaneously, the embodiments of this invention perform multi-dimensional character similarity matching based on input address information data to obtain character-level similarity features. Weighted fusion is then performed based on the semantic-level similarity features and the character-level similarity features, and the target matching address data is determined based on the weighted fusion result, thus achieving address data matching. It is readily understood that this embodiment of the invention introduces a pre-defined geographic information knowledge graph to facilitate the understanding of geographic information within addresses by a pre-defined deep learning model, thereby improving the accuracy of address data matching. Simultaneously, by constructing a trie and employing spatial hierarchical matching, it effectively enhances the retrieval efficiency of address data, thus significantly improving the efficiency of address data matching. Furthermore, this embodiment of the invention utilizes multi-dimensional character similarity matching, combining character-level and semantic-level similarity features to determine target address data, effectively improving the fault tolerance and recall rate of address data matching, thereby enhancing the accuracy of address data matching. Attached Figure Description

[0015] Figure 1 This is a flowchart of the address data matching method provided in an embodiment of the present invention; Figure 2 This is a flowchart provided by an embodiment of the present invention, which describes how to parse address elements using a preset deep learning model based on an address annotation corpus and geographic information embedding vectors to obtain a sequence of address elements. Figure 3 This is a model framework diagram of ALBERT-BiLSTM-CRF provided in the embodiments of the present invention; Figure 4 This is a flowchart provided by an embodiment of the present invention, which describes the construction of a trie based on an address element sequence, and then the similarity filtering based on the trie using a spatial hierarchical matching algorithm to obtain a set of similar addresses. Figure 5 This is a schematic diagram of constructing a Trie tree based on address elements provided in an embodiment of the present invention; Figure 6This is a flowchart provided by an embodiment of the present invention, which generates a corresponding address embedding vector matrix based on a set of similar addresses using a preset word embedding vector model, and then extracts features from the address embedding vector matrix using a preset semantic feature extraction model to obtain semantic-level similarity features. Figure 7 This is a flowchart provided by an embodiment of the present invention for performing multi-dimensional character similarity matching based on input address information data to obtain character-level similarity features; Figure 8 This is a flowchart provided by an embodiment of the present invention, which performs weighted fusion based on semantic-level similarity features and character-level similarity features, and then determines the target matching address data based on the weighted fusion result; Figure 9 This is a schematic diagram of the overall steps of address data matching provided in an embodiment of the present invention; Figure 10 This is a schematic diagram of the overall architecture for address data matching provided in an embodiment of the present invention; Figure 11 This is a schematic diagram of the address data matching device provided in an embodiment of the present invention; Figure 12 This is a schematic diagram of the hardware structure of the electronic device provided in an embodiment of the present invention. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit it. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this application; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this application as detailed in the appended claims.

[0017] It is understood that the terms “first,” “second,” etc., used in this application may be used herein to describe various concepts, but unless otherwise stated, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another. For example, without departing from the scope of the embodiments of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the words “if,” “when,” or “in response to a determination” as used herein may be interpreted as “when…” or “when…” or “in response to a determination.”

[0018] As used in this application, the terms "at least one", "multiple", "each", "any", etc., "at least one" includes one, two or more, "multiple" includes two or more, "each" refers to each of the corresponding multiples, and "any" refers to any one of the multiples.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0020] Before providing a detailed description of the embodiments of this application, some of the nouns and terms involved in the embodiments of this application will be explained first. The nouns and terms involved in the embodiments of this application are subject to the following interpretations.

[0021] Natural Language Processing (NLP): NLP is an important research direction in the field of artificial intelligence. It integrates knowledge from multiple disciplines such as linguistics, computer science, machine learning, and mathematics. It encompasses two main aspects: natural language understanding and natural language generation. Its research content includes multiple levels such as characters, words, phrases, sentences, paragraphs, and texts, serving as a bridge between machine language and human language. Its aim is to enable machines to understand, interpret, and generate human language, achieving effective communication between humans and machines, and enabling computers to perform tasks such as language translation, sentiment analysis, and text summarization. NLP allows computers to transform input language into meaningful symbols and relationships, and then process them according to the intended purpose.

[0022] Input address information is related to production and daily life information. However, due to the lack of a unified standard for the storage and representation of address information, there are significant differences in description methods and address structures between data from different sources, which limits the use and sharing of data and prevents the formation of effective communication relationships between data. Related technologies mainly rely on manual comparison or comparison based on overlaying vector layers for address matching, which has low matching efficiency and accuracy.

[0023] In view of this, this application provides an address data matching method, apparatus, electronic device, storage medium, and program product. This solution constructs an address labeling corpus by inputting address information and a preset address database, and performs graph modeling based on a preset address information knowledge graph to generate geographic information embedding vectors. Then, based on the address labeling corpus and geographic information embedding vectors, address elements are parsed using a preset learning model to obtain an address element sequence. Next, this embodiment constructs a trie based on the address element sequence, and then performs similarity filtering using a spatial hierarchical matching algorithm based on the trie to obtain a set of similar addresses. Based on the similar address set, a corresponding address embedding vector matrix is ​​generated using a preset word embedding vector model. Then, a preset semantic feature extraction model is used to extract features from the address embedding vector matrix to obtain semantic-level similarity features. Simultaneously, this embodiment performs multi-dimensional character similarity matching based on the input address information data to obtain character-level similarity features. Weighted fusion is performed based on the semantic-level similarity features and the character-level similarity features, and the target matching address data is determined based on the weighted fusion result, thus achieving address data matching and effectively improving the accuracy and efficiency of address data matching.

[0024] The address data matching method provided in this application relates to the field of geographic information service technology. The address data matching method provided in this application can be applied to a terminal, a server, or software running on a terminal or server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, or vehicle terminal, but is not limited to these. The server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The server can also be a node server in a blockchain network. The software can be an application implementing the address data matching method, but is not limited to the above forms.

[0025] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics devices, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0026] Figure 1 This is an optional flowchart of the address data matching method provided in the embodiments of this application. Figure 1 The method may include, but is not limited to, steps S110 to S170.

[0027] Step S110: Construct an address annotation corpus based on the input address information data and the preset address database.

[0028] Step S120: Perform graph embedding processing based on the preset geographic information knowledge graph to generate geographic information embedding vectors.

[0029] Step S130: Based on the address labeling corpus and geographic information embedding vectors, address elements are parsed using a preset deep learning model to obtain an address element sequence.

[0030] Step S140: Construct a trie based on the address element sequence, and then use a spatial hierarchical matching algorithm to filter similar addresses based on the trie to obtain a set of similar addresses.

[0031] Step S150: Generate the corresponding address embedding vector matrix based on the set of similar addresses using a preset word embedding vector model, and then extract features from the address embedding vector matrix using a preset semantic feature extraction model to obtain semantic-level similarity features.

[0032] Step S160: Perform multi-dimensional character similarity matching based on the input address information data to obtain character-level similarity features.

[0033] Step S170: Perform weighted fusion based on semantic-level similarity features and character-level similarity features, and then determine the target matching address data based on the weighted fusion result.

[0034] In the operation of this specific embodiment, the present invention first constructs an address annotation corpus based on input address information data and a preset address database. Specifically, the input address information data in this embodiment refers to the addresses that need to be matched and analyzed. Additionally, the preset address database stores several address information data sets for matching with the input address information data. Correspondingly, since Chinese addresses lack natural separators (such as null characters) between different address element levels and lack a unified organizational standard, data matching becomes more difficult. Therefore, the present invention first parses the address data. This process involves disassembling the non-standard structured data and organizing it according to a unified standard. For example, the Chinese address organization rules constructed in this embodiment are shown in Table 1 below: Table 1

[0035] Next, this embodiment of the invention annotates the corpus using the BIOES method, where B (begin) represents the start of an entity, I (intermediate) represents the middle part of an entity, O (other) represents irrelevant characters, E (end) represents the end of an entity, and S (single) represents an entity with only one character. Examples of corpus annotation are shown in Table 2 below: Table 2

[0036] Next, this embodiment of the invention performs graph embedding processing based on a preset geographic information knowledge graph to generate geographic information embedding vectors. Specifically, the preset geographic information knowledge graph in this embodiment refers to an external geographic information knowledge graph, which captures the spatial hierarchy, adjacency, and subordinate relationships of structured geographic entities such as administrative divisions, place names, streets, and landmarks through graph modeling. Specifically, this embodiment uses a graph embedding algorithm to encode nodes and their structural information into low-dimensional vector representations to generate knowledge embeddings. This knowledge embedding is cross-fused with the address text vector in the feature fusion layer, achieving joint modeling of geographic hierarchical information and linguistic semantics, effectively improving the model's ability to resolve complex situations such as ambiguity, ambiguity, and homonyms, and enhancing its generalization robustness.

[0037] For example, in this embodiment of the invention, a preset geographic information knowledge graph is first constructed based on the hierarchy, adjacency, or subordinate relationships between geographic entities and nodes at different levels. In this embodiment, the node set V represents geographic entities at different levels, such as provinces, cities, districts, streets, communities, roads, buildings, etc., and the edge set E represents the hierarchy, adjacency, or subordinate relationships between nodes, for example, "xx province → xx city → xx district". Correspondingly, the knowledge graph is formalized as a directed graph, as shown in equation (1) below: (1) The rules governing the relationships between nodes and edges are shown in Table 3 below: Table 3

[0038] Next, this embodiment of the invention performs knowledge embedding generation, for each node. Generate initial vector And structured geographic information is mapped through a mapping function. This is transformed into low-dimensional knowledge embedding, as shown in equation (2): (2) In feature fusion, the text vector is used. With knowledge embedding Combined, a weighted fusion is adopted, as shown in equation (3) below: (3) Where, in the formula To integrate the weights, the optimal value can be selected through cross-validation.

[0039] For example, in the feature fusion generation example, when the input address information in this embodiment of the invention is: Room h, Unit f, Floor g, Building e, Dd Building, No. Cc, Bb Road, Aa Community, Zz District, Yy City, Xx Province, the presence or absence of a knowledge graph is compared in Table 4 below: Table 4

[0040] In this embodiment of the invention, through the example of fusion vectors, the address resolution model can understand not only the meaning of the text, but also the geographical hierarchy and spatial relationships.

[0041] Next, in this embodiment of the invention, address elements are parsed using a preset deep learning model based on the address labeling corpus and geographic information embedding vectors to obtain an address element sequence. Specifically, in this embodiment of the invention, the geographic information embedding vectors and corpus data from the address labeling corpus are input into the preset deep learning model for semantic information extraction and address element recognition, thereby obtaining the address element sequence. The preset deep learning model is constructed using a BiLSTM model and a CRF model. Further, in this embodiment of the invention, a trie is constructed based on the address element sequence, and then a similarity filter is performed using a spatial hierarchical matching algorithm based on the trie to obtain a set of similar addresses. Specifically, to improve the storage and retrieval efficiency of massive amounts of data, in this embodiment of the invention, a trie is built for the parsed standard addresses (address element sequence) at levels such as province, city, district, and street. Then, in this embodiment of the invention, spatial hierarchical matching is used to filter similar addresses, and candidate addresses are sequentially retrieved according to the hierarchical structure of the parsed standard addresses to obtain a set of similar addresses.

[0042] Furthermore, in this embodiment of the invention, a corresponding address embedding vector matrix is ​​generated based on a set of similar addresses using a preset word vector model. Then, a preset semantic feature extraction model is used to extract features from the address embedding vector matrix to obtain semantically similar features. Specifically, this embodiment generates an embedding vector matrix corresponding to each address element in the set of similar addresses using a preset word vector model. Then, based on the obtained address embedding vector matrix, a preset semantic feature extraction model constructed from a bidirectional long short-term memory network (BiLSTM) and a neural convolutional network (CNN) is used for feature extraction and filtering to determine candidate addresses with similar feature distances and their corresponding semantic feature vectors, thus obtaining semantically similar features.

[0043] Meanwhile, this embodiment of the invention performs multi-dimensional character similarity matching based on the input address information data to obtain character-level similarity features. Specifically, to address the problem of decreased matching accuracy in Chinese address matching due to subtle differences in characters (misspellings, similar-looking characters, homophones, input errors, etc.), this embodiment of the invention constructs a multi-dimensional fusion matching strategy based on character similarity. By integrating features from multiple levels, a dynamic weighted fusion model is constructed, and combined with a Trie tree index, to determine the similarity of characters between candidate addresses and input addresses, achieving efficient matching and obtaining character-level similarity features. Finally, this embodiment of the invention performs weighted fusion based on semantic-level similarity features and character-level similarity features, and then determines the target matching address data based on the weighted fusion result. In particular, this embodiment of the invention effectively improves the accuracy of address matching by combining character similarity and semantic features for weighted fusion, and then using the weighted fusion result for address matching analysis.

[0044] Reference Figure 2 To improve the accuracy and robustness of address resolution, in some embodiments of the present invention, address element resolution is performed using a preset deep learning model based on an address annotation corpus and geographic information embedding vectors to obtain an address element sequence, including but not limited to the following steps: Step S210: Input the input address information data into the pre-trained language model for vector space mapping to obtain address vector representation data.

[0045] Step S220: Generate a multimodal vector based on the address vector representation data and the geographic information embedding vector.

[0046] Step S230: Input the multimodal vector into the first bidirectional long short-term memory network model to extract semantic information and obtain the associated sequence vector.

[0047] Step S240: Based on the associated sequence vector and the address labeling corpus, address elements are identified using a conditional random field model to obtain an address element sequence.

[0048] In this specific embodiment, the present invention first inputs the address information data into a pre-trained language model for vector space mapping to obtain address vector representation data. Specifically, the present invention inputs the address into the ALBERT model, mapping the Chinese place name address to the vector space to obtain address vector representation data. The present invention uses the ALBERT model to segment the input raw characters, then generates three embedding vectors: Token Embedding, Segment Embedding, and Position Embedding. These three vectors are added element-wise as input for the next step. Next, the present invention generates a multimodal vector based on the address vector representation data and the geographic information embedding vector, and then inputs the multimodal vector into a first bidirectional long short-term memory network model for semantic information extraction to obtain an associated sequence vector. Specifically, the present invention combines the embedded vector of structured geographic information with the address vector representation, and inputs the fused multimodal vector into a BiLSTM model to extract contextual features using a bidirectional structure, improving the ability to identify semantic boundaries between address elements, and outputting a globally context-aware high-dimensional vector sequence, i.e., the associated sequence vector. Furthermore, in this embodiment of the invention, address element identification is performed using a Conditional Random Field (CRF) model based on the associated sequence vectors and the address-labeled corpus to obtain an address element sequence. Specifically, this embodiment of the invention constructs a CRF model, using the sequence vectors output by BiLSTM as input, and trains the label sequence of address elements using the BIOES-labeled corpus. Simultaneously, to further integrate knowledge graph information, geographic embedding features are introduced into the feature function. This allows structured geographic information to be explicitly incorporated into sequence labeling decisions. The joint probability distribution of the final sequence is represented by the following equation (4): (4) Where, in the formula The label transfer feature function is represented by the fusion of geographic embedding vectors, reflecting the spatial / hierarchical contextual constraints between labels; It represents the state feature function of the current label and the input, which strengthens the modeling of the spatial relationship between address elements and their location in the map; Indicates position Embedding of geographical knowledge graphs associated with corresponding characters; A normalization factor is used to ensure that the sum of probabilities is 1. Accordingly, in this embodiment of the invention, the preset deep learning model includes an ALBERT-BiLSTM-CRF model, which is constructed from an ALBERT model, a BiLSTM model, and a CRF model, and its architecture is as follows: Figure 3 As shown.

[0049] Reference Figure 4 To improve matching efficiency and accuracy, in some embodiments of the present invention, a trie is constructed based on the address element sequence, and then a similarity filter is performed using a spatial hierarchical matching algorithm based on the trie to obtain a set of similar addresses, including but not limited to the following steps: Step S310: Construct a trie according to the address element sequence and the preset address hierarchy.

[0050] Step S320: Search for candidate addresses according to the hierarchical structure of the trie to obtain preset candidate addresses.

[0051] Step S330: Match and filter preset candidate addresses using a preset matching strategy to obtain a set of similar addresses. The preset matching strategy includes an exact matching strategy and a fuzzy matching strategy.

[0052] In this specific implementation, the present invention first constructs a trie according to a preset address hierarchy based on the address element sequence, then retrieves candidate addresses according to the hierarchical structure of the trie to obtain preset candidate addresses, and then performs matching and filtering of the preset candidate addresses through a preset matching strategy to obtain a set of similar addresses. Specifically, as shown... Figure 5 As shown, this embodiment of the invention constructs a trie based on the parsed address element sequence according to the corresponding hierarchical division. Nodes with the same prefix in the trie share a common prefix, saving storage space and enabling rapid location of similar address sets based on hierarchical prefixes. Next, this embodiment of the invention sequentially searches for candidate addresses according to the hierarchical structure. During the matching process, the administrative division level strictly employs precise character matching to ensure the accuracy of basic location. Accordingly, this embodiment of the invention includes a preset matching strategy and a fuzzy matching strategy. Due to the complexity of Chinese addresses and the existence of multiple names, for address elements other than administrative divisions, this embodiment of the invention combines precise matching and fuzzy matching strategies. When precise matching fails, fuzzy matching is used to find addresses with higher similarity. Simultaneously, this embodiment of the invention also includes a hierarchical backtracking strategy. When no similar address is found at the current level, backtracking matching is performed, i.e., returning to the previous level of address elements for rematching, thereby further narrowing the subsequent calculation range and improving matching efficiency.

[0053] It is easy to understand that, compared to the inverted index mechanism, which relies solely on keywords for fast retrieval and easily ignores hierarchical and contextual relationships, the spatial hierarchy matching and hierarchy backtracking in this embodiment of the invention can preserve address hierarchy structure information, more accurately capture multi-level address matching relationships, and reduce the risk of incorrect matching.

[0054] Reference Figure 6To further improve the accuracy of address data matching, in some embodiments of the present invention, a corresponding address embedding vector matrix is ​​generated based on a set of similar addresses using a preset word embedding vector model. Then, a preset semantic feature extraction model is used to extract features from the address embedding vector matrix to obtain semantic-level similarity features, including but not limited to the following steps: Step S410: Generate an address embedding vector matrix based on the set of similar addresses using a fast text model.

[0055] Step S420: Input the address embedding vector matrix into the second bidirectional long short-term memory network model for feature extraction to obtain global semantic features.

[0056] Step S430: Process the global semantic features using a convolutional neural network model to obtain local semantic features.

[0057] Step S440: Summate the global semantic features and local semantic features, and then input the summation result into the attention module to generate a semantic feature vector.

[0058] Step S450: Calculate the semantic distance based on the semantic feature vector, and then determine the semantic similarity features based on the calculated semantic distance data.

[0059] In this specific embodiment, the present invention first generates an address embedding vector matrix based on a set of similar addresses using a FastText model. Specifically, the present invention generates an embedding vector matrix corresponding to address elements using a FastText model. That is, the address embedding vector matrix. (wherein) | indicates the number of address elements in the training corpus. (This represents the dimension of the word embedding vector). Unlike the Word2Vec model, the FastText model represents each address element based on sub-word representations, that is, by decomposing the address element into several character n-grams (for example, the address "Xx" can be decomposed into...).<X,Xx,x> The weighted sum of these n-gram embedding vectors yields the overall vector representation of the address elements. Finally, for any given Chinese address parsed from its address elements, the word embedding vector matrix can be accessed via an index lookup table. The corresponding word embedding vector submatrix was retrieved from the data. ,(in This represents the number of address elements in the Chinese address. For example, given a Chinese address "xx city xx district xx road xx number", the parsed address elements are "xx city", "xx district", "xx road", and "xx number". Using the FastText model, the corresponding word embedding vectors can be obtained as follows: , , , Combined to form an address embedding vector matrix ,(in ).

[0060] Next, in this embodiment of the invention, the address embedding vector matrix is ​​input into a second bidirectional long short-term memory network model for feature extraction to obtain global semantic features. These global semantic features are then processed by a convolutional neural network model to obtain local semantic features. Finally, the global and local semantic features are summed, and the summation result is input into an attention module to generate a semantic feature vector. Specifically, this embodiment of the invention extracts latent semantic features contained in Chinese addresses using an RCnn-t model based on the obtained word embedding matrix (address embedding vector matrix). Specifically, this embodiment of the invention first inputs the word embedding matrix into a bidirectional long short-term memory network (BiLSTM) to obtain global contextual features, i.e., global semantic features. Correspondingly, the bidirectional long short-term memory network is an improved model of the RNN model, and compared with the unidirectional LSTM model, it has a stronger ability to capture contextual information. Specifically, this embodiment of the invention provides input address elements... Training is performed in two sequences, forward and backward, to obtain the hidden states for the forward pass. and the hidden state of the back Then, the spliced ​​bidirectional hidden state Stacking at different times yields the final global semantic feature vector. .

[0061] Next, this embodiment of the invention processes the obtained global context features using a convolutional neural network (CNN) model. The CNN model localizes the overall data and extracts features from the local data using convolutional kernel functions, thus better capturing the local features of address elements. For example, this embodiment uses a convolutional kernel with height w and width d (i.e., a two-dimensional convolutional kernel, adapting to the dimension of the input features) to process the global semantic feature vector. Convolutional operations are performed, followed by max pooling to reduce the dimensionality and sample the features. Finally, a fully connected layer maps the local feature vectors to a high-dimensional semantic space, yielding the final local semantic feature vectors. Next, in this embodiment of the invention, the global features obtained by BiLSTM and the local features obtained by CNN are summed, and the calculation formula is shown in the following formula (5): (5) Where, in the formula The global semantic feature vector representing the current address. V represents the local semantic feature vector of an address, which is the semantic feature of the address.

[0062] Then, the semantic features obtained in this embodiment of the invention are input into the attention module. The attention module can assign relatively higher weights to more important features, highlighting key features, reducing interference from redundant information, and improving the expressive power of the features. For example, the relevant calculation formula is shown in equation (6) below: (6) Where, in the formula , It is the transpose of the weight matrix obtained during training. It is a semantic feature vector obtained through the Attention module.

[0063] Finally, this embodiment of the invention calculates semantic distance based on semantic feature vectors, and then determines semantic-level similarity features based on the calculated semantic distance data. Specifically, this embodiment of the invention obtains the similarity between two addresses by calculating Euclidean distance; the higher the similarity, the higher the degree of address matching. The Euclidean distance calculation formula in this embodiment of the invention is shown in equation (7) below: (7) Where, in the formula b and 'a' are semantic feature vectors from different addresses.

[0064] Accordingly, in this embodiment of the invention, the corresponding candidate addresses and corresponding semantic feature vectors are determined by filtering from the set of familiar addresses using the calculated semantic distance data, thereby obtaining semantic-level similarity features.

[0065] Reference Figure 7 To alleviate the problem of decreased matching accuracy due to character differences, in some embodiments of the present invention, multi-dimensional character similarity matching is performed based on input address information data to obtain character-level similarity features, including but not limited to the following steps: Step S510: Perform preset text feature statistics based on preset Chinese address data, construct a character fuzzy replacement dictionary, and then calculate the edit distance similarity data based on the input address information data using the character fuzzy replacement dictionary. The preset text features include misspellings, similar-looking characters, and homophones.

[0066] Step S520: Based on the input address information data, perform character segmentation according to a preset character segmentation method, and then calculate the character structure similarity to obtain character structure similarity data. The preset character segmentation method includes character radical and stroke segmentation.

[0067] Step S530: Calculate the pinyin similarity based on the pinyin information corresponding to the input address information data to obtain pinyin similarity data.

[0068] Step S540: Calculate the similarity of input operations based on the input address information data and the preset input method keyboard layout information to obtain keyboard input similarity data.

[0069] Step S550: Dynamically determine the preset weight parameters based on the address level, and then calculate the comprehensive character similarity data based on the preset weight parameters, edit distance similarity data, character structure similarity data, pinyin similarity data, and keyboard input similarity data.

[0070] Step S560: When the comprehensive character similarity data is determined to be greater than or equal to the preset similarity threshold, character-level similarity features are obtained based on the comprehensive character similarity data.

[0071] In this specific embodiment, the present invention first performs preset text feature statistics on preset Chinese address data to construct a character fuzzy replacement dictionary, and then calculates edit distance similarity data based on the input address information data using the character fuzzy replacement dictionary. Specifically, the preset text features in this embodiment include misspellings, similar-looking characters, and homophones. Accordingly, based on a large amount of Chinese address data, the present invention statistically analyzes common misspellings, similar-looking characters, and homophones to construct a character fuzzy replacement dictionary, as shown in Table 5 below: Table 5

[0072] Accordingly, in the matching process, this embodiment of the invention first performs precise matching on non-administrative division address elements. If no result is found, the dictionary is used for fuzzy replacement expansion to generate candidate variants, thereby increasing the tolerance for character errors.

[0073] Next, in this embodiment of the invention, characters are split according to a preset character splitting method based on the input address information data, and then character structure similarity is calculated to obtain character structure similarity data between the input address information and the candidate address. Specifically, the preset character splitting method in this embodiment includes character radical and stroke splitting. In this embodiment, the Chinese character radical and stroke splitting method is used to calculate character structure similarity. For example, suppose the stroke sets of character A and character B are respectively... and The radicals are respectively and Structural similarity The calculation formula is shown in equation (8) below: (8) Where, in the formula If the radicals are the same, the value is 1; otherwise, the value is 0. This is the weight parameter, with a value range of (0,1).

[0074] Further, in this embodiment of the invention, the pinyin similarity is calculated based on the pinyin information corresponding to the input address information data to obtain pinyin similarity data between the input address and the candidate address. Specifically, the pinyin of a character consists of three parts: initial consonant, final vowel, and tone. In this embodiment of the invention, when calculating the pinyin similarity S_pinyin, the matching scores of the initial consonant, final vowel, and tone are calculated separately and weighted to obtain the pinyin similarity data. Next, in this embodiment of the invention, the input operation similarity is calculated based on the input information data and the preset input method keyboard layout information to obtain keyboard input similarity data. Specifically, in this embodiment of the invention, the character keyboard distance D_keyboard is calculated based on the input method keyboard layout, and the keyboard input similarity data is defined as shown in the following formula (9): (9) Where, in the formula To adjust the parameters, the weight of the keyboard distance is controlled.

[0075] Furthermore, in this embodiment of the invention, a preset weight parameter is dynamically determined based on the address level. This preset weight parameter, combined with edit distance similarity data, character structure similarity data, pinyin similarity data, and keyboard input similarity data, is used to calculate a comprehensive character similarity data. Specifically, this embodiment of the invention dynamically adjusts the character similarity weights for different address levels, i.e., the preset weight parameter. In this embodiment, the edit distance similarity is... Character structure similarity is The similarity of the pinyin is Keyboard input similarity is Comprehensive character similarity The definition is shown in equation (10): (10) Wherein, the weights Dynamically adjusted according to address level, satisfying the following formula (11): (11) For example, in the administrative division level The weight is relatively large, and the weights of other elements in the secondary address level should be appropriately increased.

[0076] Finally, this embodiment of the invention determines whether the comprehensive character similarity data meets a preset similarity threshold. If the comprehensive character similarity data is greater than or equal to the preset similarity threshold, then character-level similarity features are obtained based on the comprehensive character similarity data. Specifically, this embodiment of the invention combines Trie tree indexing for hierarchical matching and calculates the comprehensive matching score of candidate addresses. If there is no matching result in the current layer, the previous layer is backtracked, and the threshold is relaxed for fuzzy matching. For example, when the input address "xx district yi x road x number" is matched with "xx district yi x road x number" in the database, because "yi" and "yi" are similar in shape, similar in pinyin, and close in keyboard distance, the character similarity S_char is calculated to reach the threshold, and it is determined to be a match.

[0077] Reference Figure 8 To further improve the accuracy of address data matching, in some embodiments of the present invention, semantic-level similarity features include candidate address semantic vectors, and character-level similarity features include candidate address character similarity data. Accordingly, a weighted fusion is performed based on the semantic-level similarity features and the character-level similarity features, and then the target matching address data is determined based on the weighted fusion result, including but not limited to the following steps: Step S610: Based on the candidate address semantic vector and the input address semantic vector corresponding to the input address information data, calculate the semantic similarity data using the cosine similarity algorithm.

[0078] Step S620: Perform weighted fusion of semantic similarity data and candidate address character similarity data to obtain address matching score data.

[0079] Step S630: Determine the target matching address data based on the address matching score data.

[0080] In this specific embodiment, the present invention first calculates semantic similarity data based on the semantic vectors of the candidate addresses and the semantic vectors of the input addresses corresponding to the input address information data using a cosine similarity algorithm. Then, it performs weighted fusion of the semantic similarity data and the character similarity data of the candidate addresses to obtain address matching score data, thereby determining the target matching address data based on the address matching score data. Specifically, after completing the character similarity calculation (S_char) and semantic feature extraction (V_a, V_b), the present invention weighted fusion of the two types of similarity to obtain the final matching score S_total, thereby improving the matching accuracy. In this embodiment, the present invention first calculates semantic embedding similarity. For the semantic vectors V_a and V_b of the input address a and the candidate address b, the semantic similarity S_semantic is calculated using cosine similarity, as shown in the following formula (12): (12) Where, · represents the vector dot product, Represents the Euclidean norm of a vector.

[0081] Then, the embodiments of the present invention calculate the weighted fusion of character and semantic similarity, and perform a weighted fusion of the character similarity S_char and the semantic similarity S_semantic to obtain the final matching score S_total, as shown in the following formula (13): (13) where, in the formula is the fusion weight, which can be adaptively adjusted according to the address level and the historical matching effect. For example, at the administrative region level, the weight of the character similarity can be increased, and at the street and house number level, the semantic weight can be appropriately increased.

[0082] Finally, the embodiments of the present invention perform candidate address sorting and matching determination, calculate S_total for each candidate in the candidate address set, sort them from high to low according to S_total, and select the address with the highest score as the final matching result. Alternatively, the embodiments of the present invention set a threshold T. If S_total≥T, it is determined that the matching is successful; otherwise, it returns unmatched. For example, an example of the matching result in the embodiments of the present invention is shown in Table 6 below: Table 6

[0083] As can be seen from the above table, after adding the semantic embedding similarity, S_total distinguishes "Yi x Road" and "Yi x Road" more clearly, and the final matching result is more accurate. Among them, the character similarity S_char mainly captures errors such as typos, similar-shaped characters, and homophonic characters, while the semantic similarity S_semantic supplements the context and address-level semantic information.

[0084] Next, in combination with a specific address data matching scenario, the solution of the embodiments of the present invention will be introduced and described in detail: Exemplarily, referring to Figure 9 and Figure 10In navigation address search applications, it is often necessary to match and search the address information input by the user. Specifically, this embodiment of the invention first constructs a corpus based on multi-source Chinese addresses and trains a word segmentation model to realize address element parsing. Simultaneously, this embodiment of the invention introduces a geographic information knowledge graph to generate knowledge embeddings by integrating structured data such as administrative divisions, assisting the model in understanding geographic hierarchy and spatial semantics. Next, this embodiment of the invention constructs a Trie tree and uses a spatial hierarchy matching algorithm to filter similar address sets based on the character similarity of the parsed standard addresses. Further, this embodiment of the invention obtains address element embedding vectors through the FastTest model, transforming each address into a vector matrix representation. Then, this embodiment of the invention extracts global and local semantic features of the addresses based on BiLSTM and convolutional neural network models, superimposes them, and inputs them into the Attention module. Finally, it calculates feature similarity using the Euclidean distance algorithm to determine candidate addresses with similarity. Furthermore, in this embodiment of the invention, based on the semantic feature vector of the candidate address, multi-dimensional similarity information at the character level is fused to construct a character fuzzy replacement dictionary, character structure similarity, pinyin distance and input method key position fault tolerance mechanism to obtain the final address semantic features. Then, the address similarity between each candidate address and the input address is calculated to determine the matching score and matching result, and obtain the target address that best matches the address information input by the user.

[0085] For example, in address information management systems, it is often necessary to update and maintain address information, which requires matching and updating newly input address information with historical address information in the database. Specifically, this embodiment of the invention first constructs an address annotation corpus based on new Chinese address information and address information in historical data, and then uses an external geographic information knowledge graph to parse Chinese address elements, obtaining an address element sequence. Next, this embodiment of the invention constructs a Trie tree based on the address element sequence, and then uses a spatial hierarchical matching algorithm to filter and obtain a set of similar addresses. Then, this embodiment of the invention obtains the corresponding address element embedding vectors through the FastTest model, and then uses a BiLSTM model, a CNN model, and an Attention module to extract address semantic features. The similarity between the address semantic features of each address in the similar address set is calculated using the Euclidean distance algorithm, thereby obtaining several candidate addresses and their corresponding address semantic feature vectors. Simultaneously, this embodiment of the invention performs character-level fuzzy feature extraction on each candidate address and the input new address information to perform multi-dimensional fusion of character similarity, obtaining the character similarity between the new address information and each candidate address. Finally, in this embodiment of the invention, the character similarity data and the semantic similarity data determined by the semantic feature vector are fused together to determine the final matching score and the address information that needs to be updated.

[0086] It is readily understood that this embodiment of the invention combines a multi-layered natural language learning model with an ALBERT-BiLSTM-CRF model for address parsing. This not only parses multi-source data into a unified organizational structure but also significantly enhances the ability to express and recognize address semantics. Simultaneously, this embodiment introduces an external geographic knowledge graph, transforming structured information such as administrative divisions and geographic entities into knowledge embeddings, which are then fused with the model's semantic features. This helps the model understand the spatial hierarchy and ambiguous expressions of elements within an address, further improving the accuracy and robustness of the parsing. Furthermore, this embodiment establishes a Trie tree to store massive amounts of hierarchical address data. Based on this, it fully considers address character similarity features and uses spatial hierarchical matching and hierarchical backtracking to narrow the computational scope of address matching, improving the storage efficiency and organizational retrieval efficiency of address data, thus contributing to the overall performance improvement of the algorithm. Accordingly, in the address matching stage, this embodiment starts from the unique hierarchical structure of the address data, first filtering addresses to be matched through character matching to narrow the subsequent computational scope. Then, by combining a bidirectional long short-term memory network and a convolutional neural network to obtain global and local semantic features of the address, and introducing an Attention module to highlight key information, the expressive power of the features is improved, thereby enhancing the performance and accuracy of the algorithm. Simultaneously, this embodiment of the invention fully considers common ambiguous situations in Chinese addresses, such as character misspellings, similar-looking characters, and homophones, and constructs a similarity calculation framework combining character similarity and semantic embedding. Specifically, this embodiment constructs a common character replacement rule table, introduces multi-dimensional features such as character pinyin distance and character structure similarity, and jointly models these features with FastText word vectors and Attention semantic vectors during the matching stage, thereby obtaining a more accurate matching score through weighted fusion. This mechanism effectively alleviates the problem of matching failures due to subtle character differences, and is particularly suitable for scenarios such as user input errors and spelling differences in heterogeneous source addresses, greatly improving the algorithm's fault tolerance and matching recall rate.

[0087] Please see Figure 11 This application also provides an address data matching device that can implement the above method. The device includes: The first module 710 is used to construct an address annotation corpus based on the input address information data and the preset address database.

[0088] The second module 720 is used to perform graph embedding processing based on a preset geographic information knowledge graph to generate geographic information embedding vectors.

[0089] The third module 730 is used to parse address elements based on the address labeling corpus and geographic information embedding vectors using a preset deep learning model to obtain an address element sequence.

[0090] The fourth module 740 is used to construct a trie based on the address element sequence, and then use a spatial hierarchical matching algorithm to filter similar addresses based on the trie to obtain a set of similar addresses.

[0091] The fifth module 750 is used to generate a corresponding address embedding vector matrix based on a set of similar addresses using a preset word embedding vector model, and then extract features from the address embedding vector matrix using a preset semantic feature extraction model to obtain semantic-level similarity features.

[0092] The sixth module 760 is used to perform multi-dimensional character similarity matching based on the input address information data to obtain character-level similarity features.

[0093] Module 770 is used to perform weighted fusion based on semantic-level similarity features and character-level similarity features, and then determine the target matching address data based on the weighted fusion result.

[0094] It is understood that the content of the above method embodiments is applicable to the present device embodiments. The specific functions implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0095] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0096] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0097] Please see Figure 12 , Figure 12 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes: The processor 810 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory 820 can be implemented as a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 820 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 820 and is called and executed by the processor 810 using the methods described in the embodiments of this application. The input / output interface 830 is used to implement information input and output; The communication interface 840 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 850 transmits information between various components of the device (e.g., processor 810, memory 820, input / output interface 830, and communication interface 840); The processor 810, memory 820, input / output interface 830 and communication interface 840 are connected to each other within the device via bus 850.

[0098] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0099] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0100] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0101] It is understood that the content of the above method embodiments is applicable to the embodiments of this program product. The specific functions implemented by the embodiments of this program product are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0102] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0103] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0104] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0105] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0106] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0107] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0108] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0109] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0110] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0111] Furthermore, the functional units in the various embodiments of this application 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.

[0112] 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, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0113] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. An address data matching method, characterized in that, The method includes the following steps: Construct an address annotation corpus based on input address information data and a preset address database; Based on a pre-defined geographic information knowledge graph, graph embedding processing is performed to generate geographic information embedding vectors. Based on the address annotation corpus and the geographic information embedding vector, address elements are parsed using a preset deep learning model to obtain an address element sequence. A trie is constructed based on the address element sequence, and then a similarity filter is performed using a spatial hierarchical matching algorithm based on the trie to obtain a set of similar addresses. Based on the set of similar addresses, a corresponding address embedding vector matrix is ​​generated using a preset word embedding vector model. Then, a preset semantic feature extraction model is used to extract features from the address embedding vector matrix to obtain semantic-level similarity features. Multidimensional character similarity matching is performed based on the input address information data to obtain character-level similarity features; The semantic-level similarity features and the character-level similarity features are weighted and fused together, and the target matching address data is determined based on the weighted fusion result.

2. The method according to claim 1, characterized in that, The step of parsing address elements using a preset deep learning model based on the address annotation corpus and the geographic information embedding vector to obtain an address element sequence includes: Input address information data is fed into a pre-trained language model for vector space mapping to obtain address vector representation data; A multimodal vector is generated based on the address vector representation data and the geographic information embedding vector; The multimodal vector is input into the first bidirectional long short-term memory network model for semantic information extraction to obtain the associated sequence vector; The address element sequence is obtained by identifying address elements using a conditional random field model based on the associated sequence vector and the address annotation corpus.

3. The method according to claim 1, characterized in that, The step of constructing a trie based on the address element sequence, and then performing similarity filtering using a spatial hierarchical matching algorithm based on the trie to obtain a set of similar addresses, includes: The trie is constructed according to the address element sequence and a preset address hierarchy. Based on the trie, candidate addresses are retrieved according to the hierarchical structure to obtain preset candidate addresses; The preset candidate addresses are matched and filtered using a preset matching strategy to obtain the set of similar addresses; wherein the preset matching strategy includes an exact matching strategy and a fuzzy matching strategy.

4. The method according to claim 1, characterized in that, The step of generating a corresponding address embedding vector matrix based on the similar address set using a preset word embedding vector model, and then extracting features from the address embedding vector matrix using a preset semantic feature extraction model to obtain semantic-level similarity features includes: The address embedding vector matrix is ​​generated based on the set of similar addresses using a fast text model; The address embedding vector matrix is ​​input into the second bidirectional long short-term memory network model for feature extraction to obtain global semantic features; The global semantic features are processed by a convolutional neural network model to obtain local semantic features; The global semantic features and the local semantic features are summed, and the summation result is then input into the attention module to generate a semantic feature vector. The semantic distance is calculated based on the semantic feature vector, and then the semantic similarity features are determined based on the calculated semantic distance data.

5. The method according to claim 1, characterized in that, The step of performing multi-dimensional character similarity matching based on the input address information data to obtain character-level similarity features includes: Based on preset Chinese address data, preset text feature statistics are performed to construct a character fuzzy replacement dictionary. Then, based on the input address information data, the edit distance similarity data is calculated through the character fuzzy replacement dictionary. The preset text features include misspellings, similar-looking characters, and homophones. Based on the input address information data, characters are split according to a preset text splitting method, and then character structure similarity is calculated to obtain character structure similarity data; wherein, the preset text splitting method includes character radical and stroke splitting; The similarity of the characters is calculated based on the pinyin information corresponding to the input address information data to obtain pinyin similarity data; Input operation similarity is calculated based on the input address information data and the preset input method keyboard layout information to obtain keyboard input similarity data; The preset weight parameters are dynamically determined based on the address level, and then the comprehensive character similarity data is calculated based on the preset weight parameters, the edit distance similarity data, the character structure similarity data, the pinyin similarity data, and the keyboard input similarity data. When the comprehensive character similarity data is determined to be greater than or equal to a preset similarity threshold, the character-level similarity features are obtained based on the comprehensive character similarity data.

6. The method according to claim 1, characterized in that, The semantic-level similarity features include candidate address semantic vectors, and the character-level similarity features include candidate address character similarity data; The step of performing weighted fusion based on the semantic-level similarity features and the character-level similarity features, and then determining the target matching address data based on the weighted fusion result, includes: Semantic similarity data is calculated using a cosine similarity algorithm based on the candidate address semantic vector and the input address semantic vector corresponding to the input address information data; The address matching score data is obtained by weighted fusion of the semantic similarity data and the candidate address character similarity data. The target matching address data is determined based on the address matching score data.

7. An address data matching device, characterized in that, The device includes: The first module is used to construct an address annotation corpus based on the input address information data and a preset address database. The second module is used to perform graph embedding processing based on a preset geographic information knowledge graph to generate geographic information embedding vectors. The third module is used to parse address elements based on the address labeling corpus and the geographic information embedding vector using a preset deep learning model to obtain an address element sequence. The fourth module is used to construct a trie based on the address element sequence, and then perform similarity filtering based on the trie using a spatial hierarchical matching algorithm to obtain a set of similar addresses; The fifth module is used to generate a corresponding address embedding vector matrix based on the set of similar addresses using a preset word embedding vector model, and then extract features from the address embedding vector matrix using a preset semantic feature extraction model to obtain semantic-level similarity features. The sixth module is used to perform multi-dimensional character similarity matching based on the input address information data to obtain character-level similarity features; The seventh module is used to perform weighted fusion based on the semantic-level similarity features and the character-level similarity features, and then determine the target matching address data based on the weighted fusion result.

8. An electronic device, characterized in that, include: At least one processor; At least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the method 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 method of any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 6.

Images