Place name address translation system and matching method based on artificial intelligence and special name library

Abstract

The invention relates to the technical field of geographical name address translation, and particularly discloses a geographical name address translation system and matching method based on artificial intelligence and a special name library. Based on a to-be-translated geographical name address text input by a user, multiple candidate geographical entity information descriptions are extracted from a geographical entity information library; and performing deep context semantic interaction matching analysis on the to-be-translated geographical name address text and the information description of each candidate geographical entity by utilizing a natural language processing technology, and performing translation on the to-be-translated geographical name address text by taking the matched standard geographical name address description of the candidate geographical entity as a basis and combining translation knowledge in a special name library. And performing target language translation matching and standardization processing on the standard place name address description to generate a final target language translation place name address. According to the application, the geographic entity information base is utilized to assist in identifying the fuzzy or non-standard place name address input by the user, so that reliable geographic entity anchoring can be provided for the translation process, and the accuracy and consistency of translation are improved.

Description

Technical Field

[0001] The present application relates to the technical field of geographical name and address translation, and more specifically, to a geographical name and address translation system and matching method based on artificial intelligence and a special and common name database. Background Art

[0002] With the acceleration of the globalization process and the increasing frequency of international exchanges, the demand for accurate and efficient cross-language information communication has increased significantly. Especially in the field of geographical information, geographical names and addresses, as the key carriers for describing specific geographical locations, play a crucial role in cross-border trade, international logistics, tourism navigation, multilingual map services, and geographical information system (GIS) applications, etc.

[0003] Traditional translation methods often rely on general dictionaries or simple transliteration rules, and it is difficult to deal with the unique composite structure and cultural differences of geographical names and addresses, which easily lead to semantic deviation or information loss. For example, the geographical names and addresses to be translated input by users often have various forms, which may include abbreviations, aliases, common names, historical names, and even spelling mistakes or non-standard expressions. When directly performing literal translation based on a general translation model or dictionary matching, it is often difficult to capture the uniqueness of geographical names and addresses as specific geographical entities and their standard expressions in the target language, which easily leads to translation errors, ambiguities, or even loss of geographical reference meaning, seriously affecting the accurate transmission and application efficiency of cross-language geographical information.

[0004] Therefore, it is necessary to provide an optimized geographical name and address translation system and matching method based on artificial intelligence and a special and common name database to solve the above technical problems. Summary of the Invention

[0005] In order to solve the above technical problems, the present application is proposed.

[0006] According to one aspect of the present application, there is provided a geographical name and address translation matching method based on artificial intelligence and a special and common name database, which includes:[[]] Obtaining the geographical name and address text to be translated input by the user and the specified target translation language; Constructing a geographical entity information database, and extracting multiple candidate geographical entity information descriptions from the geographical entity information database based on the geographical name and address text to be translated; Performing matching analysis on the multiple candidate geographical entity information descriptions and the geographical name and address text to be translated respectively to obtain multiple candidate geographical entity confidence scores; Extracting the standard geographical name and address description in the candidate geographical entity information description corresponding to the maximum value from the multiple candidate geographical entity confidence scores as the optimal matching standard geographical name and address description; Segment and perform named entity recognition and annotation on the optimal matching standard toponym address description to obtain a sequence of toponym address word segments with component annotations for translation. Based on the target translation language, use the proper name library and common name library to perform translation matching on the sequence of toponym address word segments with component annotations to obtain the target language translated toponym address.

[0007] According to another aspect of the present application, there is provided a toponym address translation system based on artificial intelligence and proper / common name libraries, which includes: A user input module for obtaining the toponym address text to be translated and the specified target translation language input by the user. A geographical entity information extraction module for constructing a geographical entity information library and extracting multiple candidate geographical entity information descriptions from the geographical entity information library based on the toponym address text to be translated. A matching analysis module for respectively performing matching analysis on the multiple candidate geographical entity information descriptions and the toponym address text to be translated to obtain multiple candidate geographical entity confidence scores. An optimal matching selection module for extracting the standard toponym address description in the candidate geographical entity information description corresponding to the maximum value from the multiple candidate geographical entity confidence scores as the optimal matching standard toponym address description. A toponym address annotation module for segmenting and performing named entity recognition and annotation on the optimal matching standard toponym address description to obtain a sequence of toponym address word segments with component annotations for translation. A toponym address translation matching module for performing translation matching on the sequence of toponym address word segments with component annotations based on the target translation language, using the proper name library and common name library to obtain the target language translated toponym address.

[0008] Advantageous effects: Compared with the prior art, the toponym address translation system and matching method based on artificial intelligence and proper / common name libraries provided by the present application extract multiple candidate geographical entity information descriptions from the geographical entity information library based on the toponym address text to be translated by the user, and use natural language processing technology to perform in-depth context semantic interaction matching analysis between the toponym address text to be translated and each candidate geographical entity information description to determine the candidate geographical entity that best meets the user's input intention. Furthermore, based on the standard toponym address description of the matched candidate geographical entity, combined with the translation knowledge in the proper / common name library, perform target language translation matching and normalization processing on the standard toponym address description to generate the final target language translated toponym address. This method helps to provide a reliable geographical entity anchor for the translation process and improve the accuracy and consistency of translation by using the geographical entity information library to assist in identifying fuzzy or non-standard toponym addresses input by the user. Description of the Drawings

[0009] The above and other objects, features, and advantages of the present application will become more apparent by describing the embodiments of the present application in more detail with reference to the accompanying drawings. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation to the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0010] Figure 1 It is a flowchart of a method for translating and matching geographical names and addresses based on artificial intelligence and a specialized common name library according to an embodiment of the present application.

[0011] Figure 2 It is a data flow diagram of a method for translating and matching geographical names and addresses based on artificial intelligence and a specialized common name library according to an embodiment of the present application.

[0012] Figure 3 It is a flowchart of sub-step S3 of a method for translating and matching geographical names and addresses based on artificial intelligence and a specialized common name library according to an embodiment of the present application.

[0013] Figure 4 It is a flowchart of sub-step S32 of a method for translating and matching geographical names and addresses based on artificial intelligence and a specialized common name library according to an embodiment of the present application.

[0014] Figure 5 It is a block diagram of a system for translating and matching geographical names and addresses based on artificial intelligence and a specialized common name library according to an embodiment of the present application. Detailed implementation manners

[0015] As shown in the present application and the claims, unless the context clearly indicates an exception, words such as "a", "an", "one", and / or "the" are not specifically singular and may also include the plural. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of the clearly identified steps and elements, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0016] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, however, any number of different modules can be used and run on the user terminal and / or the server. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0017] Flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the operations before or below do not necessarily need to be executed precisely in order. On the contrary, according to needs, various steps can be executed in reverse order or simultaneously. At the same time, other operations can also be added to these processes, or one or several steps can be removed from these processes.

[0018] Next, exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. It should be understood that the present application is not limited by the exemplary embodiments described herein.

[0019] It should be noted in advance that all relevant processing of the data in the present application is carried out on the premise of complying with the corresponding data protection regulations and policies of the location and obtaining authorization from the corresponding authority managers.

[0020] Figure 1 FIG. is a flowchart of a method for translating and matching geographical names and addresses based on artificial intelligence and a special and common name database according to an embodiment of the present application. Figure 2 FIG. is a data flow diagram of a method for translating and matching geographical names and addresses based on artificial intelligence and a special and common name database according to an embodiment of the present application. As Figure 1 and Figure 2 shown, the method for translating and matching geographical names and addresses based on artificial intelligence and a special and common name database includes the steps of: S1, obtaining a geographical name and address text to be translated input by a user and a specified target translation language; S2, constructing a geographical entity information database, and based on the geographical name and address text to be translated, extracting a plurality of candidate geographical entity information descriptions from the geographical entity information database; S3, respectively matching and analyzing the plurality of candidate geographical entity information descriptions with the geographical name and address text to be translated to obtain a plurality of candidate geographical entity confidence scores; S4, extracting the standard geographical name and address description in the candidate geographical entity information description corresponding to the maximum value from the plurality of candidate geographical entity confidence scores as the optimal matching standard geographical name and address description; S5, performing word segmentation and named entity recognition and annotation on the optimal matching standard geographical name and address description to obtain a sequence of component-annotated geographical name and address word segments to be translated; S6, based on the target translation language, using a special name database and a common name database to perform translation matching on the sequence of component-annotated geographical name and address word segments to be translated to obtain a target language translated geographical name and address.

[0021] In the above method for translating and matching geographical names and addresses based on artificial intelligence and a special and common name database, in step S1, a geographical name and address text to be translated input by a user and a specified target translation language are obtained. It should be understood that the core goal of geographical name and address translation is to meet the cross-language expression needs of users for specific geographical entities, and there are significant differences in the naming rules of geographical entities in different language systems. Therefore, in order to ensure the directivity and controllability of the translation process, the present application is based on the design of a user interaction interface, and receives the text to be translated and the target language parameter submitted by the user through a front-end input module or an API interface, and uses them as the initial input for subsequent processing. Specifically, the user can provide the original geographical name and address through form input or file upload, and select the target language (such as English), so as to provide clear boundary conditions for subsequent semantic parsing and translation processing.

[0022] In the process of specific implementation, users usually submit the text of place names and addresses to be translated to the system through a graphical interface or a program interface. For example, in application scenarios such as map services, logistics scheduling, or multilingual GIS systems, users may fill in the name of a specific location, such as "No. 1, Zhongguancun Avenue, Haidian District, Beijing", in a manual input manner, or upload a file containing multiple address entries in batches for centralized cross-language conversion. At the same time, users also need to clearly specify the target translation language, such as English, French, Spanish, etc., so that the system can perform differential processing according to the selected language rules. To meet the requirements in different usage scenarios, the system needs to support multiple input forms, including but not limited to single-line text box input, multi-line text editor, file upload control, etc., and at the same time be compatible with the import methods of multiple data formats such as JSON, CSV, XML, etc. This flexibility not only improves the applicability of the system but also provides a good foundation for subsequent automated processing.

[0023] At the interface level, the system usually exposes the place name and address translation service externally in the form of an API (Application Programming Interface), allowing third-party platforms or internal modules to call. For example, an international logistics company may integrate this translation module into its order processing system. When a user fills in the shipping or receiving address, the system automatically converts the Chinese address into the target language version for generating international shipping labels or customer notification emails. In this case, the user input does not directly come from the keyboard operation of the end user but is passed by other system modules through a standard protocol. Therefore, the system must have reliable parameter parsing capabilities to be able to identify and extract the address fields and language identifiers therein. Specifically, the API interface should support GET or POST request methods, accept a request body containing the original address text and the target language code, and return a structured response result to reserve a unified data interface for subsequent processing.

[0024] In addition, in addition to traditional text input methods, modern place name and address translation systems are gradually introducing emerging technologies such as speech recognition and image OCR (Optical Character Recognition) as supplementary input means. For example, users can take pictures of envelopes, signs, or map screenshots containing foreign addresses with a mobile device. The system uses OCR technology to extract the text content in the image and processes it as the text to be translated. In addition, for some special groups or specific usage environments, the system also supports voice input functions. Users can quickly enter address information through natural language instructions, and the system converts the audio stream into text form through a speech recognition engine. Although these diverse input channels increase the complexity of system implementation, they also significantly improve the user experience and usage efficiency, enabling the place name and address translation service to cover more diverse usage scenarios.

[0025] In terms of specifying language parameters, the system's support for users to select the target translation language also has diversity. A common way is to let users select from a preset list of languages through a drop-down menu or tab, with each language corresponding to a standardized language code (e.g., en for English, es for Spanish, ja for Japanese, etc.). Another way is through an intelligent recommendation mechanism that automatically infers the target language based on the user's geographical location, browser settings, or historical preferences, thereby reducing the user's operation burden. For example, if the user is currently located in Germany and the browser language setting is German, the system can default the translation target to German unless the user actively changes this setting. This intelligent language selection strategy helps to enhance the system's friendliness and adaptability, especially for map services or travel navigation applications open to the public.

[0026] It is worth noting that since place names often carry specific cultural backgrounds and naming habits, there are significant differences in the ways of expressing place names among different language systems. For example, Chinese place names are usually arranged in a hierarchical structure of "Province - City - District - Street", while English place names tend to follow the order of "Street, City, State, Country", and commas are often used to separate the components. In addition, some regions incorporate directional words (such as North, South), administrative division types (such as County, Province), or official naming marks (such as Saint, Mount) into place names. All these details need to be preserved or appropriately adjusted during the translation process. Therefore, when the system receives user input, it not only needs to identify the semantic content of the address itself but also record its original language attribute for reference to corresponding grammar rules and naming norms in the subsequent translation process.

[0027] To ensure the integrity and accuracy of the input information, the system also needs to have a certain fault tolerance mechanism and data verification ability. For example, when the content entered by the user contains spelling mistakes, abbreviations, or non-standard expressions, the system should be able to identify potential problems through a built-in error correction module or fuzzy matching algorithm and prompt the user for confirmation or correction. In addition, for situations where the address text is too long, the format is not standardized, or it contains illegal characters, the system should also provide a corresponding feedback mechanism to prevent subsequent processing failures due to abnormal input. Although this data preprocessing ability does not belong to the core scope of this step, it is an important support for ensuring the overall system stability.

[0028] In the above method for translating and matching place names and addresses based on artificial intelligence and a specialized common name database, in step S2, a geographic entity information library is constructed, and based on the text of the place name and address to be translated, multiple candidate geographic entity information descriptions are extracted from the geographic entity information library. It should be understood that in this application, it is considered that the text of the place name and address to be translated input by the user may adopt a non-standard format, contain abbreviations, aliases, common names, or even have errors, and there are differences from the standard geographic entity descriptions, resulting in easy translation errors, ambiguities, or even loss of geographical reference meaning when directly performing literal translation. Therefore, in order to accurately identify the potential standard geographic entities that the user input text may point to, this application constructs a geographic entity information library containing a large amount of standardized data of geographic entities, and based on the search and matching technology of the text of the place name and address to be translated, retrieves and extracts multiple candidate geographic entity information descriptions related to the input text from this geographic entity information library, providing an initial candidate set for subsequent geographic entity anchoring. Specifically, the geographic entity information library pre-stores detailed data of geographic entities in a specific area, including their standard place name and address descriptions, entity types (such as provinces, cities, districts, streets, buildings, scenic spots, etc.), administrative division information (such as subordination relationships), geographic coordinates, various aliases, common names, former names, and associated entity descriptions (information of surrounding or superior associated entities); after receiving the text of the place name and address to be translated input by the user, using a retrieval algorithm based on word vector similarity, performs retrieval and matching based on partial keywords on the text of the place name and address to be translated and each candidate geographic entity information description in the geographic entity information library, and retrieves multiple candidate geographic entity information descriptions potentially related to the text of the place name and address to be translated as a candidate set. Through this method, it is possible to effectively handle the ambiguity and non-standard expressions of the user input. Even if the user inputs a non-standard or fuzzy place name and address, the closest standard place name and address description can be found through matching with the geographic entity information library, providing an accurate geographic entity basis for subsequent translation.

[0029] In the above method for matching and translating place names and addresses based on artificial intelligence and a specialized common name database, in step S3, the multiple candidate geographical entity information descriptions are respectively matched and analyzed with the place name and address text to be translated to obtain multiple candidate geographical entity confidence scores. Specifically, since the traditional literal matching method cannot distinguish the semantic differences between "Chaoyangmen" as a place name and "Chaoyang District" as an administrative region in "South Street of Chaoyangmen", and it is difficult to handle spelling mistakes (such as miswriting "Hu" as "Hu"). Therefore, in order to more accurately determine the specific geographical entity pointed to by the place name and address to be translated entered by the user, the present application further introduces the ability of deep semantic understanding. By respectively performing in-depth semantic understanding and interaction analysis on each candidate geographical entity information description and the place name and address text to be translated, the semantic association strength between each candidate entity and the place name and address text to be translated is quantified, thereby breaking through the limitation of surface character similarity and achieving accurate matching based on deep semantics. Among them, Figure 3 is a flowchart of sub-step S3 of the method for matching and translating place names and addresses based on artificial intelligence and a specialized common name database according to an embodiment of the present application. As Figure 3 shown, step S3 includes the steps of: S31, respectively extracting the context semantic features of the candidate geographical entity information description and the place name and address text to be translated to obtain a candidate geographical entity information description context semantic encoding vector and a place name and address context semantic encoding vector to be translated; S32, performing semantic transfer inference interaction encoding on the place name and address context semantic encoding vector to be translated and the candidate geographical entity information description context semantic encoding vector to obtain a place name and address - candidate geographical entity semantic interaction response inference encoding vector; S33, performing feature decoding on the place name and address - candidate geographical entity semantic interaction response inference encoding vector to obtain the candidate geographical entity confidence score.

[0030] Specifically, in a specific example of the present application, the step S31 includes: performing context semantic encoding based on the BERT model on the candidate geographical entity information description and the to-be-translated place name and address text respectively to obtain the candidate geographical entity information description context semantic encoding vector and the to-be-translated place name and address context semantic encoding vector. Specifically, in order to effectively identify the semantic equivalence between non-standard expressions input by the user (such as the semantic equivalence between "Jingshi Nine Gates" and the standard name "Nine Gates of the Inner City of Beijing") and the implicit context associations (such as "bridge" may refer to a bridge entity or a regional name in different contexts), based on the context awareness ability of the pre-trained language model, through the bidirectional Transformer architecture of the BERT model, context semantic encoding is performed on the candidate geographical entity information description and the to-be-translated place name and address text respectively. Utilizing the deep semantic understanding ability obtained by the BERT model through pre-training on a large-scale corpus, after tokenizing and word embedding processing the candidate geographical entity information description and the to-be-translated place name and address text respectively, context semantic features at the word level and sentence level in the text are extracted through a multi-layer self-attention mechanism, and the deep semantic information and context relevance of the text are mined, thereby generating a high-dimensional context semantic vector representation, and obtaining the candidate geographical entity information description context semantic encoding vector and the to-be-translated place name and address context semantic encoding vector. In this way, the candidate geographical entity information description and the to-be-translated place name and address text can be mapped to a unified semantic space, laying a foundation for subsequent fine-grained semantic interaction matching.

[0031] Specifically, in step S32, semantic transfer inference interaction encoding is performed on the context semantic encoding vector of the to-be-translated place name and address text and the context semantic encoding vector of the candidate geographical entity information description to obtain a to-be-translated place name and address - candidate geographical entity semantic interaction response inference encoding vector. It should be understood that in this application, it is considered that simple vector cosine similarity calculation can only capture the similarity between the candidate geographical entity information description and the to-be-translated place name and address text at the global macro semantic level, and it is difficult to reveal the in-depth interaction relationship between the two in terms of micro semantic details and context association (for example, it is difficult to handle the high similarity between "Huangpu River" and "Huangpu District" in some feature dimensions). Therefore, in order to more accurately measure the semantic fit between the candidate geographical entity and the to-be-translated place name and address, this application introduces a neural network model to perform semantic transfer inference interaction modeling on the context semantic encoding vector of the to-be-translated place name and address text and the context semantic encoding vector of the candidate geographical entity information description, so as to deeply explore the in-depth interaction relationship between the candidate geographical entity and the to-be-translated place name and address in terms of micro semantic details and context association. By continuously aggregating and updating the feature responses of the two in terms of micro semantic details, a more accurate to-be-translated place name and address - candidate geographical entity semantic interaction response inference encoding vector is generated, providing a more accurate semantic basis for the subsequent calculation of the candidate geographical entity confidence score. Among them, Figure 4 is a flowchart of sub-step S32 of the place name and address translation matching method based on artificial intelligence and a specialized and common name database according to an embodiment of the present application. As Figure 4 shown, step S32 includes the steps: S321, performing local fine-grained semantic interaction response encoding on the context semantic encoding vector of the to-be-translated place name and address text and the context semantic encoding vector of the candidate geographical entity information description to obtain a sequence of to-be-translated place name and address - candidate geographical entity local semantic interaction response encoding matrices; S322, performing local semantic interaction feature inference transfer encoding on the sequence of to-be-translated place name and address - candidate geographical entity local semantic interaction response encoding matrices to obtain the to-be-translated place name and address - candidate geographical entity semantic interaction response inference encoding vector.

[0032] More specifically, step S321 includes: First, performing an ordered arrangement based on the eigenvalue size on the context semantic encoding vector of the to-be-translated place name and address text and the context semantic encoding vector of the candidate geographical entity information description to obtain an ordered encoding vector of the to-be-translated place name semantic features and an ordered encoding vector of the candidate geographical entity semantic features, which is expressed by the formula: ; ; Among them, represents the context semantic encoding vector of the to-be-translated place name and address text, Represents the context semantic encoding vector for describing candidate geographical entity information, Represents the operation of sorting the vector elements, Represents the ordered arrangement encoding vector of the semantic features of the place name and address to be translated, Represents the ordered arrangement encoding vector of the semantic features of candidate geographical entities.

[0033] Through ordered arrangement, the attention is focused on the distribution pattern and relative intensity of the feature values, rather than the initial position of the features, so that the model obtains invariance or equivariance to the input arrangement, making the context semantic encoding vectors of the place name and address text to be translated from different sources and the context semantic encoding vectors for describing candidate geographical entity information have a more consistent representation form in the feature space, generating the ordered arrangement encoding vector of the semantic features of the place name and address to be translated and the ordered arrangement encoding vector of the semantic features of candidate geographical entities, thereby significantly improving the stability of semantic matching, highlighting the importance of key semantic features, suppressing the interference of noise features, and further improving the calculation accuracy of the confidence score of candidate geographical entities.

[0034] Then, perform equal-granularity feature segmentation on the ordered arrangement encoding vector of the semantic features of the place name and address to be translated and the ordered arrangement encoding vector of the semantic features of candidate geographical entities to obtain a sequence of ordered encoding vectors of local semantic features of the place name and address to be translated and a sequence of ordered encoding vectors of local semantic features of candidate geographical entities, which is expressed by the formula: ; ; Wherein, , , and respectively represent the 1st, 2nd, the th and the th ordered encoding vectors of local semantic features of the place name and address to be translated in the sequence of ordered encoding vectors of local semantic features of the place name and address to be translated, is the number of ordered encoding vectors of local semantic features of the place name and address to be translated, , , and respectively represent the 1st, 2nd, the th and the th ordered encoding vectors of local semantic features of candidate geographical entities in the sequence of ordered encoding vectors of local semantic features of candidate geographical entities, represents the feature segmentation function.

[0035] It should be understood that there may be differences in the ordered arrangement encoding vectors of the semantic features of the to-be-translated place names and addresses from different sources and the ordered arrangement encoding vectors of the semantic features of the candidate geographical entities. Directly performing an overall match is likely to overlook local semantic details, while equal-granularity segmentation can decompose complex semantics into comparable basic units to facilitate capturing more detailed semantic correspondence relationships. Therefore, in this application, the ordered arrangement encoding vector of the semantic features of the to-be-translated place names and addresses and the ordered arrangement encoding vector of the semantic features of the candidate geographical entities are decomposed into sequences of ordered encoding vectors of local semantic features of the to-be-translated place names and addresses with equal granularity and sequences of ordered encoding vectors of local semantic features of the candidate geographical entities, so as to more accurately locate the semantic association points between the to-be-translated text and the candidate geographical entities. Even in the case of abbreviations, aliases, or spelling mistakes, potential correct correspondence relationships can be identified through local matching, thereby significantly improving the accuracy of semantic matching and optimizing the usage efficiency of computing resources while ensuring the accuracy of semantic analysis.

[0036] Finally, each group of corresponding ordered encoding vectors of local semantic features of the to-be-translated place names and addresses and ordered encoding vectors of local semantic features of the candidate geographical entities in the sequences of ordered encoding vectors of local semantic features of the to-be-translated place names and addresses and the sequences of ordered encoding vectors of local semantic features of the candidate geographical entities are input into the semantic interaction response inference unit to obtain a sequence of to-be-translated place name - candidate geographical entity local semantic interaction response encoding matrices, which is expressed by the formula: ; Wherein, represents a linear transformation matrix, represents the ReLU activation function, represents matrix multiplication, represents and the to-be-translated place name - candidate geographical entity local semantic interaction response encoding matrix between.

[0037] Since the semantic matching between the to-be-translated place names and addresses and the candidate geographical entities involves various complex semantic relationships, such as similarity, difference, alignment relationship, etc., therefore, in this application, the ordered encoding vectors of local semantic features of the to-be-translated place names and addresses and the ordered encoding vectors of local semantic features of the candidate geographical entities are input into the semantic interaction response inference unit to quantify and encode the semantic interaction patterns between the ordered encoding vectors of local semantic features of the to-be-translated place names and addresses and the ordered encoding vectors of local semantic features of the candidate geographical entities within a specific eigenvalue range, generating a sequence of to-be-translated place name - candidate geographical entity local semantic interaction response encoding matrices, thereby more accurately describing the semantic association therebetween and providing a more reliable semantic basis for the subsequent calculation of the matching confidence score.

[0038] More specifically, in a specific example of the present application, the step S322 includes: inputting the sequence of the to-be-translated place name address-candidate geographical entity local semantic interaction response encoding matrix into an LSTM transfer encoding module with an attention mechanism to obtain the to-be-translated place name address-candidate geographical entity semantic interaction response inference encoding vector, which is represented by the formula: ; ; ; Wherein, represents the matrix flattening operation, represents the to-be-translated place name address-candidate geographical entity local semantic interaction response encoding vector obtained by unfolding, represents the exponential function operation with base e, is the feature modulation function based on the attention mechanism, represents and the to-be-translated place name address-candidate geographical entity local semantic interaction response encoding matrix between, represents and the to-be-translated place name address-candidate geographical entity local semantic interaction response encoding matrix between, LSTM model, represents the to-be-translated place name address-candidate geographical entity semantic interaction response inference encoding vector.

[0039] That is, through the LSTM transfer encoding module with an attention mechanism, the cross-interval dependencies and global context between the to-be-translated place name address-candidate geographical entity local semantic interaction response encoding matrices are deeply mined, so as to more accurately capture the overall semantic interaction relationship between them, enabling the model to better handle fuzzy or non-standard expressions when facing complex place name address translation tasks, significantly improving the accuracy and consistency of place name address translation. The to-be-translated place name address-candidate geographical entity semantic interaction response inference encoding vector generated in this way has stronger expressive ability and robustness, and can provide a more reliable semantic basis for subsequent translation matching and normalization processing, thereby improving the performance of the entire translation system.

[0040] Specifically, considering that the segmentation granularity of equal-granularity feature segmentation affects the regional granularity size of the to-be-translated place name address-candidate geographical entity local semantic interaction response encoding matrix, it will also directly affect the interaction structure representation between the sequence of the to-be-translated place name address local semantic feature ordered encoding vectors and the candidate geographical entity local semantic feature ordered encoding vectors. Based on this, in a preferred example of the present application, the step S322 includes: First, based on the semantic interaction response connection density between each group of corresponding to-be-translated place name address local semantic feature ordered encoding vectors and candidate geographical entity local semantic feature ordered encoding vectors, perform feature sparsity constraint on each to-be-translated place name address-candidate geographical entity local semantic interaction response encoding matrix in the sequence of the to-be-translated place name address-candidate geographical entity local semantic interaction response encoding matrix to obtain an optimized sequence of the to-be-translated place name address-candidate geographical entity local semantic interaction response encoding matrix; Then, input the optimized sequence of the to-be-translated place name address-candidate geographical entity local semantic interaction response encoding matrix into the LSTM transfer encoding module with an attention mechanism to obtain the to-be-translated place name address-candidate geographical entity semantic interaction response inference encoding vector.

[0041] Specifically, since the to-be-translated place name address-candidate geographical entity local semantic interaction response encoding matrix is a cross-regional interaction response between two local areas, it essentially expresses the interaction space metric based on its row vectors, and the length of the row vector is also an indication of the regional constraint intensity. If the regional constraint intensity, that is, the length of the row vector is introduced as the regional constraint intensity, then the low-dimensional space metric representation of the to-be-translated place name address-candidate geographical entity local semantic interaction response encoding matrix, that is, the Frobenius norm should follow the Poisson distribution relationship: ; wherein, represents the factorial, represents the Frobenius norm, represents the length of the row vector of represents the exponential function with the natural constant as the base, represents the Poisson distribution parameter, which characterizes the expected frequency of the to-be-translated place name address-candidate geographical entity local semantic interaction. Thus, the Poisson distribution parameter can be solved.

[0042] In this way, in the case of generating a Poisson distribution process with the local interaction constraint intensity of and the mean expectation of , with and Describe the edge association representation of the interactive space metric, and further determine that the semantic interaction response connection density between two local to-be-translated place names and candidate geographical entities is: ; Wherein, represents and the semantic interaction response connection density between, represents taking the absolute value, represents calculating the two-norm.

[0043] Then, use the semantic interaction response connection density to perform parameter iteration optimization on the local interaction constraint strength : ; Wherein, represents the optimized .

[0044] Finally, use the optimized to re-constrain the sparsity of the to-be-translated place name address - candidate geographical entity local semantic interaction response coding matrix : ; Wherein, represents the corresponding optimized to-be-translated place name address - candidate geographical entity local semantic interaction response coding matrix.

[0045] That is to say, under the condition of strictly ensuring that the expected degree is , determine the regularization of the overall interactive space metric through the regularization constraint of the mean expected connection probability fluctuation of each row, so that the semantic interaction information within the regional interaction structure can avoid the risk of regional overfitting, improve the overall expression efficiency of the sequence of the to-be-translated place name address - candidate geographical entity local semantic interaction response coding matrix, and further improve the accuracy and robustness of the subsequent global semantic interaction transfer aggregation based on the LSTN model.

[0046] Specifically, in step S33, the inference coding vector of the to-be-translated place name address-candidate geographical entity semantic interaction response is feature-decoded to obtain the confidence score of the candidate geographical entity. In a specific example of the present application, step S33 includes: inputting the inference coding vector of the to-be-translated place name address-candidate geographical entity semantic interaction response into a feature decoder based on a multi-layer perceptron to obtain the confidence score of the candidate geographical entity. That is, in order to convert the inference coding vector of the to-be-translated place name address-candidate geographical entity semantic interaction response into a comparable scalar score, the present application constructs a feature decoder by stacking fully-connected layers and activation functions (such as ReLU, Sigmoid) based on the non-linear fitting ability of the multi-layer perceptron (MLP). Specifically, by inputting the inference coding vector of the to-be-translated place name address-candidate geographical entity semantic interaction response into an MLP model with two hidden layers, the non-linear transformation ability of the MLP is used to perform non-linear transformation on it, so as to extract implicit information such as the strong correlation between the candidate geographical entity and the to-be-translated place name address, and the consistency of administrative division context. After passing through Dropout regularization, a normalized confidence score s∈[0,1] is output as a quantitative indicator of the matching degree between the candidate geographical entity and the to-be-translated place name address. In this way, the semantic matching degree between the candidate entity and the user input address can be quantified, providing a differentiable decision basis for the selection of the optimal entity, and solving the problem of mis-matching caused by traditional similarity calculation methods ignoring the semantic level.

[0047] In the above method for matching place name addresses in translation based on artificial intelligence and a specialized and common name database, in step S4, the standard place name address description in the candidate geographical entity information description corresponding to the maximum value is extracted from the confidence scores of the multiple candidate geographical entities as the optimal matching standard place name address description. That is, in order to select the only standard geographical entity that best represents the user's intention from potential candidate entities, the present application is based on the principle of maximum confidence selection. By traversing all candidate geographical entities and their associated confidence score lists, the maximum candidate geographical entity confidence score is extracted, and the candidate geographical entity information description corresponding to the highest candidate geographical entity confidence score is located. The content of the "standard place name address description" field is extracted from it as the benchmark for this translation task. In this way, from the possibly ambiguous or non-standard user input, the geographical entity with a canonical representation behind it can be determined, and its standard address description is used as the only input for subsequent translation processing, avoiding ambiguity and uncertainty.

[0048] In the above method for translating and matching geographical names and addresses based on artificial intelligence and a specialized and common name library, in step S5, word segmentation and named entity recognition and annotation are performed on the optimal matching standard geographical name and address description to obtain a sequence of to-be-translated geographical name and address word segments with component annotations. Specifically, since geographical addresses have a certain internal structure, including different types of geographical components such as administrative divisions, street names, house numbers, building names, etc., and the translation methods and rules for different components may be different (for example, administrative divisions may have established translations, street names may require transliteration plus common names, and building names may require querying a specific word library). Therefore, in order to structurally analyze the optimal matching standard geographical name and address description and generate a sequence of translatable word segments, this application is based on the component analysis method in linguistics, and hierarchically annotates the optimal matching standard geographical name and address description through a pre-trained named entity recognition (NER) model to identify each component in the geographical name and address. In the embodiments of this application, a BiLSTM-CRF model is used to identify components such as administrative divisions (such as "Beijing City"), common names (such as "Avenue"), and proper names (such as "Zhongguancun") in the optimal matching standard geographical name and address description, and output an annotation sequence (such as "[Administrative Division] Beijing City / [Administrative Division] Haidian District / [Proper Name] Zhongguancun / [Common Name] Avenue"), so as to provide a structured input with clear semantic roles for subsequent address translation and avoid component misalignment caused by overall translation.

[0049] In the above method for translating and matching place names and addresses based on artificial intelligence and a proper and common name library, in step S6, based on the target translation language, the proper name library and the common name library are used to perform translation matching on the sequence of the to-be-translated place name and address word segments with component annotations to obtain the target language translated place name and address. It should be understood that since proper names (such as "Yangtze River") need to follow transliteration or historical conventional translation methods ("Yangtze River"), and common names (such as "River") need to be freely translated according to the norms of the target language ("River"), traditional unified translation strategies will lead to the loss of cultural attributes or grammar errors. Therefore, in order to balance translation norms and cultural preservation, based on the principle of "separating proper names from common names" in translation studies, by constructing a bilingual-aligned proper name library and common name library, each word segment in the sequence of the to-be-translated place name and address word segments with component annotations is respectively matched with the target language expression. Specifically, first, the sequence of the to-be-translated place name and address word segments with component annotations is traversed. According to the semantic roles (such as administrative divisions, proper names, common names, etc.) annotated for each to-be-translated place name and address word segment, the proper name library is called to match proper nouns (such as "Zhongguancun → Zhongguancun"), or the common name library is called to match general terms (such as "Avenue → Avenue"). Finally, the target language translation results of each to-be-translated place name and address word segment found are combined and formatted according to the writing habits and order of the place name and address in the target language to form the final target language translated place name and address string. In this way, the professional translation knowledge in the field is fully utilized, ensuring the accuracy, standardization, and practicality of the translation results, and effectively solving the deficiencies of the general translation method in place name and address translation.

[0050] In summary, the method for translating and matching place names and addresses based on artificial intelligence and a proper and common name library according to the embodiments of the present application is elucidated. It extracts multiple candidate geographical entity information descriptions from the geographical entity information library based on the to-be-translated place name and address text input by the user, and uses natural language processing technology to perform in-depth context semantic interaction matching analysis between the to-be-translated place name and address text and each candidate geographical entity information description to determine the candidate geographical entity that most conforms to the user's input intention. Furthermore, based on the standard place name and address description of the matched candidate geographical entity, combined with the translation knowledge in the proper and common name library, target language translation matching and normalization processing are performed on the standard place name and address description to generate the final target language translated place name and address. This method helps to provide a reliable geographical entity anchor for the translation process by using the geographical entity information library to assist in identifying fuzzy or non-standard place names and addresses input by the user, improving the accuracy and consistency of the translation.

[0051] Furthermore, a place name and address translation system based on artificial intelligence and a proper and common name library is also provided.

[0052] Figure 5The block diagram of a place name and address translation system based on artificial intelligence and a specialized and common name database according to an embodiment of the present application. As Figure 5 shown, the place name and address translation system 100 based on artificial intelligence and a specialized and common name database according to an embodiment of the present application includes: a user input module 110, configured to obtain a place name and address text to be translated and a specified target translation language input by a user; a geographical entity information extraction module 120, configured to construct a geographical entity information database, and based on the place name and address text to be translated, extract multiple candidate geographical entity information descriptions from the geographical entity information database; a matching analysis module 130, configured to perform matching analysis on the multiple candidate geographical entity information descriptions and the place name and address text to be translated respectively to obtain multiple candidate geographical entity confidence scores; an optimal matching selection module 140, configured to extract the standard place name and address description in the candidate geographical entity information description corresponding to the maximum value from the multiple candidate geographical entity confidence scores as the optimal matching standard place name and address description; a place name and address annotation module 150, configured to perform word segmentation and named entity recognition annotation on the optimal matching standard place name and address description to obtain a sequence of component-annotated place name and address word segments to be translated; and a place name and address translation matching module 160, configured to perform translation matching on the sequence of component-annotated place name and address word segments to be translated by using a specialized name database and a common name database based on the target translation language to obtain a target language translated place name and address.

[0053] The basic principles of the present invention have been described above in conjunction with specific embodiments. However, it should be noted that the advantages, benefits, effects, etc. mentioned in the present invention are only examples and not limitations, and it cannot be considered that these advantages, benefits, effects, etc. are essential for each embodiment of the present invention. In addition, the specific details of the above embodiments are only for the purpose of illustration and easy understanding, rather than limitations, and the above details do not limit the present invention to necessarily adopt the above specific details to implement.

[0054] In the above embodiments, the descriptions of each embodiment have their own emphases. For parts not detailed or recorded in a certain embodiment, reference can be made to the relevant descriptions of other embodiments. In the several embodiments provided by the present invention, it should be understood that the disclosed system and method can be implemented in other ways. For example, the system embodiments described above are only illustrative. For example, the unit division is only a logical function division, and there may be other division methods in actual implementation. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0055] It is obvious to those skilled in the art that the present invention is not limited to the details of the above-described exemplary embodiments, and the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention. Therefore, in any aspect, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Accordingly, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be embraced by the present invention. Any reference signs in the claims should not be construed as limiting the claim concerned.

[0056] In addition, it is obvious that the word "comprising" does not exclude other elements or steps, and the singular does not exclude the plural. The multiple elements stated in the system claims can also be implemented by one element through software or hardware.

[0057] Finally, it should be noted that the above description has been given for purposes of illustration and description. In addition, the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the technical solutions are modified or equivalently replaced with reference to the preferred embodiments, they do not depart from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for translating and matching place names and addresses based on artificial intelligence and a specialized and general name database, characterized in that Including: Obtain the text of the place name address to be translated and the specified target translation language input by the user; Construct a geographic entity information library, and based on the text of the place name address to be translated, extract multiple candidate geographic entity information descriptions from the geographic entity information library; Perform matching analysis on the multiple candidate geographic entity information descriptions and the text of the place name address to be translated respectively to obtain multiple candidate geographic entity confidence scores; Extract the standard place name address description in the candidate geographic entity information description corresponding to the maximum value from the multiple candidate geographic entity confidence scores as the optimal matching standard place name address description; Perform word segmentation and named entity recognition annotation on the optimal matching standard place name address description to obtain a sequence of word segments of the place name address to be translated with component annotations; Based on the target translation language, use a proper name library and a common name library to perform translation matching on the sequence of word segments of the place name address to be translated with component annotations to obtain the target language translated place name address.

2. The method for translating and matching geographical names and addresses based on artificial intelligence and a specialized common name database according to claim 1, wherein The geographic entity information description includes a standard place name address description, an entity type, administrative division information, geographic coordinates, an alias, a common name, a former name, and an associated entity description.

3. The method for translating and matching geographical names and addresses based on artificial intelligence and a specialized and general name library according to claim 2, wherein Performing matching analysis on the multiple candidate geographic entity information descriptions and the text of the place name address to be translated respectively to obtain multiple candidate geographic entity confidence scores, including: Respectively extract the context semantic features of the candidate geographic entity information description and the text of the place name address to be translated to obtain a candidate geographic entity information description context semantic coding vector and a place name address to be translated context semantic coding vector; Perform semantic transfer inference interactive coding on the place name address to be translated context semantic coding vector and the candidate geographic entity information description context semantic coding vector to obtain a place name address - candidate geographic entity semantic interaction response inference coding vector; Perform feature decoding on the place name address - candidate geographic entity semantic interaction response inference coding vector to obtain the candidate geographic entity confidence score.

4. The method for translating and matching geographical names and addresses based on artificial intelligence and a specialized and general name database according to claim 3, wherein Respectively extract the context semantic features of the candidate geographic entity information description and the text of the place name address to be translated to obtain a candidate geographic entity information description context semantic coding vector and a place name address to be translated context semantic coding vector, including: Perform context semantic coding based on the BERT model on the candidate geographic entity information description and the text of the place name address to be translated respectively to obtain the candidate geographic entity information description context semantic coding vector and the place name address to be translated context semantic coding vector.

5. The method for translating and matching geographical names and addresses based on artificial intelligence and a specialized and general name database according to claim 4, characterized in that Perform semantic transfer inference interactive coding on the place name address to be translated context semantic coding vector and the candidate geographic entity information description context semantic coding vector to obtain a place name address - candidate geographic entity semantic interaction response inference coding vector, including: Perform local fine-grained semantic interaction response coding on the place name address to be translated context semantic coding vector and the candidate geographic entity information description context semantic coding vector to obtain a sequence of place name address - candidate geographic entity local semantic interaction response coding matrices; Perform local semantic interaction feature inference and transfer encoding on the sequence of the to-be-translated place name address - candidate geographical entity local semantic interaction response encoding matrix to obtain the to-be-translated place name address - candidate geographical entity semantic interaction response inference encoding vector.

6. The method for translating and matching geographical names and addresses based on artificial intelligence and a specialized and general name database according to claim 5, wherein Perform local fine-grained semantic interaction response encoding on the to-be-translated place name address text context semantic encoding vector and the candidate geographical entity information description context semantic encoding vector to obtain a sequence of the to-be-translated place name address - candidate geographical entity local semantic interaction response encoding matrix, including: Perform an ordered arrangement based on the eigenvalue magnitudes on the to-be-translated place name address text context semantic encoding vector and the candidate geographical entity information description context semantic encoding vector to obtain an ordered encoding vector of the to-be-translated place name address semantic features and an ordered encoding vector of the candidate geographical entity semantic features; Perform equal-grained feature segmentation on the ordered encoding vector of the to-be-translated place name address semantic features and the ordered encoding vector of the candidate geographical entity semantic features to obtain a sequence of the to-be-translated place name address local semantic feature ordered encoding vectors and a sequence of the candidate geographical entity local semantic feature ordered encoding vectors; Input each group of corresponding to-be-translated place name address local semantic feature ordered encoding vectors and candidate geographical entity local semantic feature ordered encoding vectors in the sequences of the to-be-translated place name address local semantic feature ordered encoding vectors and the candidate geographical entity local semantic feature ordered encoding vectors into a semantic interaction response inference unit to obtain a sequence of the to-be-translated place name address - candidate geographical entity local semantic interaction response encoding matrix.

7. The method for translating and matching geographical names and addresses based on artificial intelligence and a specialized common name database according to claim 6, characterized in that Perform local semantic interaction feature inference and transfer encoding on the sequence of the to-be-translated place name address - candidate geographical entity local semantic interaction response encoding matrix to obtain the to-be-translated place name address - candidate geographical entity semantic interaction response inference encoding vector, including: Input the sequence of the to-be-translated place name address - candidate geographical entity local semantic interaction response encoding matrix into an LSTM transfer encoding module with an attention mechanism to obtain the to-be-translated place name address - candidate geographical entity semantic interaction response inference encoding vector.

8. The method for translating and matching geographical names and addresses based on artificial intelligence and a specialized and general name database according to claim 6, wherein Perform local semantic interaction feature inference and transfer encoding on the sequence of the to-be-translated place name address - candidate geographical entity local semantic interaction response encoding matrix to obtain the to-be-translated place name address - candidate geographical entity semantic interaction response inference encoding vector, including: Based on the semantic interaction response connection density between each group of corresponding to-be-translated place name address local semantic feature ordered encoding vectors and candidate geographical entity local semantic feature ordered encoding vectors, perform feature sparsity constraint on each to-be-translated place name address - candidate geographical entity local semantic interaction response encoding matrix in the sequence of the to-be-translated place name address - candidate geographical entity local semantic interaction response encoding matrix to obtain an optimized sequence of the to-be-translated place name address - candidate geographical entity local semantic interaction response encoding matrix; Input the sequence of the optimized to-be-translated place name and address - candidate geographical entity local semantic interaction response encoding matrix into the LSTM transfer encoding module with an attention mechanism to obtain the to-be-translated place name and address - candidate geographical entity semantic interaction response inference encoding vector.

9. The method for translating and matching geographical names and addresses based on artificial intelligence and a specialized and general name database according to claim 3, wherein Perform feature decoding on the to-be-translated place name and address - candidate geographical entity semantic interaction response inference encoding vector to obtain the candidate geographical entity confidence score, including: Input the to-be-translated place name and address - candidate geographical entity semantic interaction response inference encoding vector into a feature decoder based on a multi-layer perceptron to obtain the candidate geographical entity confidence score.

10. A geographical name and address translation system based on artificial intelligence and a specialized and general name database, characterized in that, Including: A user input module for obtaining the to-be-translated place name and address text input by the user and the specified target translation language; A geographical entity information extraction module for constructing a geographical entity information database and extracting multiple candidate geographical entity information descriptions from the geographical entity information database based on the to-be-translated place name and address text; A matching analysis module for respectively performing matching analysis on the multiple candidate geographical entity information descriptions and the to-be-translated place name and address text to obtain multiple candidate geographical entity confidence scores; An optimal matching selection module for extracting the standard place name and address description in the candidate geographical entity information description corresponding to the maximum value from the multiple candidate geographical entity confidence scores as the optimal matching standard place name and address description; A place name and address annotation module for performing word segmentation and named entity recognition annotation on the optimal matching standard place name and address description to obtain a sequence of component-annotated to-be-translated place name and address word segments; A place name and address translation matching module for performing translation matching on the sequence of component-annotated to-be-translated place name and address word segments based on the target translation language using a proper name library and a common name library to obtain the target language translated place name and address.

Images