Geographic entity name derivation relationship identification method, apparatus and device, medium and product

Abstract

The invention discloses a geographic entity name derivation relation identification method, device and equipment, a medium and a product, and relates to the field of geographic information science, and the method comprises the steps: obtaining a target general name, a geographic entity name and derivation element attributes; identifying whether the geographic entity name is a derived place name of the target general name or not by adopting a derivation identification model; if yes, obtaining a plurality of candidate native place names according to the derived geometric parameters; performing feature engineering processing on the geographic entity name, the derived element attribute, any candidate native place name and the native element attribute to obtain a fusion feature corresponding to the candidate native place name; according to the fusion features, adopting a native probability model to obtain the probability that the candidate native place name is the native place name of the geographic entity name; according to the method, the neural network and the machine learning model are adopted, the space-time distribution features and the semantic features are fused, and the automation degree and accuracy of identification of the derived geographical names and the derived relations of the derived geographical names are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of geographic information science, and in particular to a geographic entity name derivation relationship identification method, device, equipment, medium and product. BACKGROUND

[0002] In the fields of Geographic Information System (GIS) and bilingual map making, it is crucial to accurately understand the semantic relationships between geographic entity names (toponyms). The common name derivation relationship is an important semantic relationship, that is, the naming part of a new toponym (derived toponym, such as "Riverside Drive") is derived from the common name of another existing toponym (original toponym, such as "Sangamon River"). Identifying this relationship is of great value to ensuring the accuracy and consistency of toponym translation.

[0003] Currently, existing identification methods are mainly rule-based methods based on geographic statistics and manual summary of prior knowledge. These methods have obvious technical problems: 1. Poor generalization ability, over-reliance on manually defined fixed rules, difficulty in adapting to the diversity and complexity of toponym naming methods, and inability to effectively identify derivation patterns not covered in the rules. 2. Ignoring temporal information, existing methods usually only consider the spatial distribution and text similarity of toponyms, ignoring the important time dimension feature that toponyms have a life cycle (generation, change, abolition), resulting in insufficient basis for discrimination. 3. Low degree of automation and efficiency, rule-based methods require a lot of manual intervention and debugging when dealing with massive toponym data, have low degree of automation, and recognition efficiency needs to be improved. SUMMARY

[0004] The purpose of the present application is to provide a geographic entity name derivation relationship identification method, device, equipment, medium and product, which uses neural networks and machine learning models, and fuses spatio-temporal distribution features and semantic features, significantly improving the degree of automation and accuracy of derived toponym and its derivation relationship identification.

[0005] To achieve the above purpose, the present application provides the following solutions.

[0006] In a first aspect, the application provides a method for identifying a derivation relationship of a geographic entity name, comprising: obtaining a target common name, a geographic entity name, and a corresponding derived element attribute; the derived element attribute comprises a derived geometric parameter, a derived starting time, and a derived category attribute; using a derivation identification model to identify whether the geographic entity name is a derived place name of the target common name; the derivation identification model is a neural network model built in advance according to a first training sample set; each training sample in the first training sample set comprises a sample geographic entity name, a sample target common name, and a binary label; if the geographic entity name is not a derived place name of the target common name, the geographic entity name and the target common name have no derivation relationship; if the geographic entity name is a derived place name of the target common name, a spatial proximity analysis method is used to obtain a plurality of candidate original place names according to the derived geometric parameter, and an original element attribute corresponding to each candidate original place name is obtained; for any candidate original place name, the geographic entity name, the derived element attribute, the candidate original place name, and the original element attribute corresponding to the candidate original place name are subjected to feature engineering processing to obtain a fusion feature corresponding to the candidate original place name; the fusion feature comprises a spatiotemporal distribution feature and a semantic feature; using an original probability model to obtain a probability that the candidate original place name is an original place name of the geographic entity name according to the fusion feature; the original probability model is a machine learning model built in advance according to a second training sample set; the second training sample set comprises a sample fusion feature and a probability that a sample original place name corresponding to each sample fusion feature is an original place name of a sample geographic entity name; and determining an original place name corresponding to the geographic entity name according to the probabilities corresponding to all candidate original place names.

[0007] In a second aspect, the present application provides a geographical entity name derivation relationship identification device, comprising: an original data acquisition module, configured to acquire a target generic name, a geographical entity name, and a corresponding derived feature attribute; the derived feature attribute comprises a derived geometric parameter, a derived starting time, and a derived category attribute; a derived place name identification module, configured to identify whether the geographical entity name is a derived place name of the target generic name by using a derived identification model; the derived identification model is a neural network model built in advance according to a first training sample set; each training sample in the first training sample set comprises a sample geographical entity name, a sample target generic name, and a binary label; a non-derivation relationship determination module, configured to determine that the geographical entity name and the target generic name have no derivation relationship when the geographical entity name is not a derived place name of the target generic name; a candidate original place name generation module, configured to obtain a plurality of candidate original place names by using a spatial proximity analysis method according to the derived geometric parameter when the geographical entity name is a derived place name of the target generic name, and acquire an original feature attribute corresponding to each candidate original place name; a fusion feature extraction module, configured to perform feature engineering processing on the geographical entity name, the derived feature attribute, the candidate original place name, and the original feature attribute corresponding to the candidate original place name for any candidate original place name, to obtain a fusion feature corresponding to the candidate original place name; the fusion feature comprises a spatiotemporal distribution feature and a semantic feature; an original probability calculation module, configured to obtain a probability that the candidate original place name is an original place name of the geographical entity name by using an original probability model according to the fusion feature; the original probability model is a machine learning model built in advance according to a second training sample set; the second training sample set comprises a sample fusion feature and a probability that a sample original place name corresponding to each sample fusion feature is an original place name of a sample geographical entity name; and an original place name determination module, configured to determine an original place name corresponding to the geographical entity name according to probabilities corresponding to all candidate original place names.

[0008] In a third aspect, the present application provides a computer device, comprising: a memory, a processor, a computer program stored in the memory and executable on the processor, and the processor executes the computer program to implement the geographical entity name derivation relationship identification method described above.

[0009] In a fourth aspect, the present application provides a computer readable storage medium, having a computer program stored thereon, and the computer program is executed by a processor to implement the geographical entity name derivation relationship identification method described above.

[0010] In a fifth aspect, the present application provides a computer program product, comprising a computer program, and the computer program is executed by a processor to implement the geographical entity name derivation relationship identification method described above.

[0011] According to the specific embodiments provided in the present application, the following technical effects are disclosed.

[0012] The application provides a geographical entity name derivation relationship identification method, device, equipment, medium and product. The derivation identification model is used to automatically identify whether a geographical entity name is a derived place name of a target general name, thereby replacing a traditional method relying on a dictionary or a fixed rule, and automatically and high-precisely identifying a derived place from unstructured text. A spatial proximity analysis method is used to obtain a candidate original place name according to derivation geometric parameters, instead of relying on simple string matching or fixed range buffer analysis, thereby improving the accuracy and coverage of selection of the candidate original place name. Feature engineering processing is performed by fusing time-space distribution features and semantic features, and features such as time sequence distribution, spatial aggregation degree and semantic similarity between the candidate original place name and the geographical entity name are considered, thereby improving the accuracy of original place name identification. An original probability model is used to automatically identify the probability that the candidate original place name is an original place name of the geographical entity name, thereby realizing end-to-end derivation relationship discrimination, replacing a traditional scheme based on a rule or a simple machine learning classifier, and improving the automation degree and accuracy of derivation relationship identification. BRIEF DESCRIPTION OF DRAWINGS

[0013] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor.

[0014] Figure 1 FIG. 1 is a diagram of an application environment of a geographical entity name derivation relationship identification method according to an embodiment of the present application.

[0015] Figure 2 FIG. 2 is a flowchart of a geographical entity name derivation relationship identification method according to an embodiment of the present application.

[0016] Figure 3 FIG. 3 is a function module diagram of a geographical entity name derivation relationship identification device according to an embodiment of the present application.

[0017] Figure 4 FIG. 4 is a structural diagram of a computer device according to an embodiment of the present application. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments only constitute some embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0019] The above objects, features and advantages of the present application will become more apparent from the following detailed description considered in conjunction with the accompanying drawings and specific embodiments.

[0020] The geographic entity name derivation relationship identification method provided in the embodiments of the present application can be applied in an application environment as shown in Figure 1 The terminal 101 communicates with the server 102 through a network. The data storage system can store data required to be processed by the server 102. The data storage system can be separately arranged, integrated on the server 102, or placed on a cloud or other server. The terminal 101 can send a target common name, a geographic entity name and a corresponding derived element attribute to the server 102. After receiving the target common name, the geographic entity name and the corresponding derived element attribute, the server 102 identifies whether the geographic entity name is a derived place name of the target common name by using a derivation identification model. If yes, the server 102 further determines a native place name corresponding to the geographic entity name by using a native probability model. The server 102 can feed back the native place name corresponding to the geographic entity name to the terminal 101. In addition, in some embodiments, the geographic entity name derivation relationship identification method can also be implemented by the server 102 or the terminal 101 alone. For example, the terminal 101 can directly process the target common name, the geographic entity name and the corresponding derived element attribute to identify whether the geographic entity name is a derived place name of the target common name and determine the native place name corresponding to the derived place name. The server 102 can also obtain the target common name, the geographic entity name and the corresponding derived element attribute from the data storage system and process them to identify whether the geographic entity name is a derived place name of the target common name and determine the native place name corresponding to the derived place name.

[0021] The terminal 101 can be, but is not limited to, various desktop computers, notebook computers, smart phones, tablet computers, Internet of Things devices and portable wearable devices. The Internet of Things device can be a smart speaker, a smart television, a smart air conditioner, a smart vehicle device, etc. The portable wearable device can be a smart watch, a smart bracelet, a head-mounted device, etc. The server 102 can be implemented by an independent server or a server cluster composed of multiple servers, and can also be a cloud server.

[0022] In an exemplary embodiment, as shown in Figure 2 A geographic entity name derivation relationship identification method is provided, which is executed by a computer device, specifically by a terminal or a server or the like computer device alone or by a terminal and a server together. In the embodiments of the present application, the method is applied to the server 102 in Figure 1 The method includes the following steps 201 to 207.

[0023] Step 201, obtaining a target generic name, a geographic entity name and a corresponding derived feature attribute. The derived feature attribute includes a derived geometric parameter, a derived starting time and a derived category attribute.

[0024] Step 202, using a derived recognition model to identify whether the geographic entity name is a derived place name of the target generic name. The derived recognition model is a neural network model previously built according to a first training sample set. Each training sample in the first training sample set includes a sample geographic entity name, a sample target generic name and a binary label.

[0025] Step 203, if the geographic entity name is not a derived place name of the target generic name, the geographic entity name and the target generic name have no derived relationship.

[0026] Step 204, if the geographic entity name is a derived place name of the target generic name, using a spatial proximity analysis method according to the derived geometric parameter to obtain a plurality of candidate native place names, and obtaining a native feature attribute corresponding to each candidate native place name.

[0027] Step 205, for any candidate native place name, performing feature engineering processing on the geographic entity name, the derived feature attribute, the candidate native place name and the native feature attribute corresponding to the candidate native place name to obtain a fusion feature corresponding to the candidate native place name. The fusion feature includes a spatio-temporal distribution feature and a semantic feature.

[0028] Step 206, using a native probability model according to the fusion feature to obtain a probability that the candidate native place name is a native place name of the geographic entity name. The native probability model is a machine learning model previously built according to a second training sample set. The second training sample set includes a sample fusion feature and a probability that a sample native place name corresponding to each sample fusion feature is a native place name of a sample geographic entity name.

[0029] Step 207, determining a native place name corresponding to the geographic entity name according to probabilities corresponding to all candidate native place names.

[0030] In one example, the building method of the derived recognition model includes steps 301 to 303.

[0031] Step 301, obtaining a plurality of sample geographic entity names, and pre-setting a plurality of sample target generic names for each sample geographic entity name.

[0032] Step 302, for each sample target common name, if the sample target common name corresponds to a sample geographic entity name that is a derived place name of the sample target common name, set the binary label corresponding to the sample target common name to 1, otherwise set the binary label corresponding to the sample target common name to 0.

[0033] Step 303, according to all sample target common names and the sample geographic entity name and binary label corresponding to each sample target common name, train a neural network model to obtain the derived recognition model.

[0034] In one example, the neural network model is a bidirectional pre-training language model. Specifically, the Bidirectional Encoder Representations from Transformers (BERT) model, particularly the bert-base-uncased version, is used as a bidirectional pre-training language model for performing this task. The powerful bidirectional semantic understanding capability of the BERT model is suitable for such tasks.

[0035] In this example, different prompt template construction strategies and pre-training models are used. During model training, the BERT model is fine-tuned with sample geographic entity names, sample target common names, and binary labels. When used, the geographic entity name and target common name are used as model input, and the text classification method of the BERT model is used to identify whether the geographic entity name is a derived place name of the target common name. The hidden layer vector of the [CLS] token is obtained, and a randomly initialized linear classification layer is used for probability calculation. By setting a threshold, the probability is converted into a final binary label to represent the recognition result of whether the geographic entity name is a derived place name of the target common name. In this example, the threshold is set to 0.5, and if it exceeds the set threshold, the output is 1, indicating that the geographic entity name and the target common name have a derivation relationship. If it does not exceed the set threshold, the output is 0, indicating that the geographic entity name and the target common name have no derivation relationship.

[0036] In one example, the above step 204 can be replaced by the following steps 2041 to 2044.

[0037] Step 2041, according to the derived geometric parameters, calculate the geographic coordinates corresponding to the geographic entity name.

[0038] Step 2042, according to the geographic coordinates, use geographic coding technology to divide the geographic space of the preset range corresponding to the geographic entity name into grids to obtain the grid division result and the grid index corresponding to the grid division result.

[0039] At step 2043, according to the geographic coordinates, the grid division result, the grid index, and a preset radius, a spatial index query and distance clipping method is used to obtain the original geographic name within the preset radius range, and a category attribute corresponding to the original geographic name is obtained.

[0040] At step 2044, according to the derived category attribute and the category attribute corresponding to the original geographic name, a plurality of candidate original geographic names are obtained. Specifically, the category attribute corresponding to each candidate original geographic name is different from the derived category attribute.

[0041] In this example, the recognition result of the derived recognition model is used to locate the candidate original geographic name that is spatially adjacent and possibly derived from each recognized derived geographic name.

[0042] (1) Spatial retrieval strategy: according to the geographic coordinates corresponding to the geographic entity name, based on the first law of geography (spatially close things are more closely related), an efficient geographic coding technology GeoHash is used to perform grid division and indexing on the preset range of geographic space corresponding to the geographic entity name, and the grid division result and the grid index corresponding to the grid division result are output.

[0043] (2) Efficient retrieval and candidate set construction: based on the foregoing grid index, taking the geographic entity corresponding to the geographic entity name as the center, within a preset radius (or grid level), all other category geographic entity names are quickly retrieved and recalled as candidate original geographic names. Among them, “other categories” refers to all other category geographic entity names that are different from the derived category attribute corresponding to the geographic entity name. This step greatly reduces the calculation scale of the derived relationship recognition, converts the global “needle in a haystack” problem into a local “best match” problem, and ensures the efficiency of the subsequent complex model processing.

[0044] In one example, the original element attribute includes an original geometric parameter, an original start time, and an original category attribute. The above step 205 can be replaced by the following steps 2051 to 2057.

[0045] At step 2051, for any candidate original geographic name, according to the derived geometric parameter and the original geometric parameter corresponding to the candidate original geographic name, a topological relationship calculation method and a spatial distance calculation method are used respectively to obtain a topological relationship feature and a distance feature corresponding to the candidate original geographic name.

[0046] At step 2052, according to the derived start time and the original start time corresponding to the candidate original geographic name, a qualitative temporal reasoning method is used to obtain a time sequence feature corresponding to the candidate original geographic name.

[0047] Step 2053, according to the derived category attribute and the native category attribute corresponding to the candidate native place name, a category mapping prior method is used to obtain the category association feature corresponding to the candidate native place name.

[0048] Step 2054, according to the geographic entity name and the candidate native place name, a word sequence pair matching method is used to obtain the text feature corresponding to the candidate native place name.

[0049] Step 2055, the topological relationship feature, the distance feature and the time sequence feature are spatio-temporal feature coded to obtain the spatio-temporal distribution feature.

[0050] Step 2056, the category association feature and the text feature are text semantic feature coded to obtain the semantic feature.

[0051] Step 2057, the spatio-temporal distribution feature and the semantic feature are cross-modal feature fused to obtain the fusion feature.

[0052] In this example, the multi-dimensional feature engineering constructed comprehensively covers the topological relationship feature, the distance feature, the time sequence feature, the category association feature and the text feature corresponding to each candidate native place name. The specific construction methods of the five features are as follows.

[0053] (1) Topological relationship feature: the topological relationship between the geographic entity corresponding to the geographic entity name and the geographic entity corresponding to the candidate native place name is obtained by using the Region Connection Calculus 8 (RCC8) spatial topological relationship model, and the topological relationship feature is obtained by label encoding. RCC8 defines eight basic topological relationships, namely, disconnected (DC), externally connected (EC), partially overlapped (PO), equal (EQ), tangential proper part (TPP) and its inverse relationship, non-tangential proper part (NTPP) and its inverse relationship. RCC8 provides a solid logical foundation for qualitative spatial reasoning.

[0054] (2) Distance feature: In the geographical entity set composed of all candidate native geographical name corresponding geographical entities, the spatial distance between each geographical entity of the same category and the geographical entity corresponding to the geographical entity name is calculated. Then the geographical entities in each category are binary coded according to the spatial distance calculation result. If a geographical entity has the smallest spatial distance with the geographical entity corresponding to the geographical entity name in its category, its corresponding binary code is 1, otherwise, its corresponding binary code is 0. The distance feature reflects the spatial significance of the candidate native geographical name in the local same category entity.

[0055] (3) Time sequence feature: Introducing time dimension, according to the derived start time and the native start time corresponding to the candidate native geographical name, the sign of the start time difference (dif_start) and the end time difference (dif_end) between the geographical entity name and each candidate native geographical name in the life cycle is calculated. Here, "sign" refers to the sign (+ / -) of the value of start time difference (dif_start) and end time difference (dif_end), which is used to indicate the time sequence. This helps to determine who exists first from the diachronic perspective, providing time evidence for the derived relationship.

[0056] (4) Category association feature: the derived category attribute corresponding to the geographical entity name is combined with the corresponding native category attribute of any candidate native geographical name (such as "Road→Lake"), as the category association feature corresponding to the candidate native geographical name. The category association feature is an important category association priori knowledge.

[0057] (5) Text feature: directly using the geographical entity name and the candidate native geographical name to form the original vocabulary sequence, obtaining the text feature corresponding to the candidate native geographical name (such as "Lake" and "Lakeside"), to capture the direct or indirect derivation rules between names in word root and affix.

[0058] In this example, the category association feature and the text feature are spliced to input the BERT model and extract a two-dimensional semantic feature vector. Through one-dimensional convolution and global average pooling, it is converted into the semantic feature S'. The topological relationship feature, distance feature and time sequence feature after label encoding are spliced to obtain the spatio-temporal distribution feature T'.

[0059] In one example, the above step 2057 can be replaced by the following steps S1 to S3.

[0060] Step S1, the spatio-temporal distribution feature and the semantic feature are spliced to obtain a spliced feature.

[0061] Step S2, a channel attention mechanism is used to calculate feature weights of the spliced features, to obtain spatial-temporal distribution feature weights and semantic feature weights.

[0062] Step S3, the spatial-temporal distribution feature weights and the semantic feature weights are used to perform weighted fusion on the spatial-temporal distribution features and the semantic features, to obtain the fused features.

[0063] In this example, the generation method of the fused features includes the following steps a to c.

[0064] Step a, the semantic features S' and the spatial-temporal distribution features T' are directly spliced to perform simple fusion (SF), to obtain a mixed feature vector.

[0065] Step b, the mixed feature vector is input into a compression-excitation network (SENet). The SENet generates weights for each feature channel through global average pooling (compression) and a fully connected layer (excitation), thereby explicitly modeling the interdependence between feature channels.

[0066] Step c, the channel weights calculated by the SENet are applied to the original S' and T' features, and an attentional feature fusion (AFF) mechanism is used to perform weighted fusion, to obtain the fused features. That is, AFF(S', T') = λ·S' + (1-λ)·T'. Wherein, λ is an adaptive weight output by the SENet, used to balance the contribution degrees of the semantic and spatial-temporal distribution features.

[0067] In one example, the native probability model is a support vector machine (SVM) classifier. The SVM classifier, especially the SVM using a radial basis function (RBF) kernel function, is very powerful in handling high-dimensional and nonlinear classification problems.

[0068] The SVM classifier needs to be trained in advance according to the second training sample set. After training is completed, the fused features corresponding to any candidate native place name are input into the SVM. In this example, the output of the SVM is converted into a probability between 0 and 1 through methods such as Platt scaling. This probability represents the possibility that the current candidate native place name is the native place name corresponding to the geographic entity name.

[0069] The application adopts "prompt learning + BERT" for derived place name recognition, introduces a derived mode as a context, effectively utilizes the prior knowledge of the pre-trained language model, and significantly improves the accuracy (F1 value reaches 95.80%) of derived place name recognition. The entire process does not need to rely on a large number of artificially defined complex rules, and the derived relationship recognition is automatically completed through a data-driven neural network model and a machine learning model. The SENet-AFF method can effectively learn and fuse heterogeneous features (discrete numerical features and continuous text vector features), and enhance the adaptability of the model to complex and variable place name naming scenarios.

[0070] The application also provides an application scenario of the geographical entity name derived relationship recognition method. Specifically, the geographical entity name derived relationship recognition method provided by the embodiment can be applied to a bilingual map making scenario. The bilingual map making scenario includes three links of "geographical corpus collection - place name processing link - map publishing and updating". The derived place name - original place name recognition method provided by the embodiment belongs to the place name processing link link, is used for derived-original matching of original corpus, generates bilingual labels, improves the accuracy of place name translation, and is used for subsequent mapping, navigation, retrieval and other downstream links, realizes the consistency of homologous place name translation and bilingual Figure 1

[0071] Based on the same inventive concept, the embodiment of the application also provides a geographical entity name derived relationship recognition device for implementing the geographical entity name derived relationship recognition method. The implementation scheme of the device for solving the problem is similar to the implementation scheme described in the above method, so the specific limitations in one or more geographical entity name derived relationship recognition device embodiments provided below can be referred to the limitations of the geographical entity name derived relationship recognition method in the above, which will not be repeated here.

[0072] In one exemplary embodiment, as shown in Figure 3 a geographical entity name derived relationship recognition device is provided, including: an original data acquisition module 001, a derived place name recognition module 002, a non-derived relationship determination module 003, a candidate original place name generation module 004, a fusion feature extraction module 005, an original probability calculation module 006, and an original place name determination module 007.

[0073] ​The original data acquisition module 001 is configured to acquire a target generic name, a geographic entity name, and a corresponding derived feature attribute. The derived feature attribute includes a derived geometric parameter, a derived start time, and a derived category attribute. The derived place name identification module 002 is configured to identify whether the geographic entity name is a derived place name of the target generic name by using a derived identification model. The derived identification model is a neural network model built in advance according to a first training sample set. Each training sample in the first training sample set includes a sample geographic entity name, a sample target generic name, and a binary label. The no-derived relationship determination module 003 is configured to determine that the geographic entity name and the target generic name have no derived relationship when the geographic entity name is not a derived place name of the target generic name. The candidate native place name generation module 004 is configured to obtain a plurality of candidate native place names by using a spatial proximity analysis method according to the derived geometric parameter when the geographic entity name is a derived place name of the target generic name, and acquire a native feature attribute corresponding to each candidate native place name. The fusion feature extraction module 005 is configured to perform feature engineering processing on the geographic entity name, the derived feature attribute, the candidate native place name, and the native feature attribute corresponding to the candidate native place name for any candidate native place name, to obtain a fusion feature corresponding to the candidate native place name. The fusion feature includes a spatiotemporal distribution feature and a semantic feature. The native probability calculation module 006 is configured to obtain a probability that the candidate native place name is a native place name of the geographic entity name by using a native probability model according to the fusion feature. The native probability model is a machine learning model built in advance according to a second training sample set. The second training sample set includes a sample fusion feature and a probability that a sample native place name corresponding to each sample fusion feature is a native place name of a sample geographic entity name. The native place name determination module 007 is configured to determine a native place name corresponding to the geographic entity name according to probabilities corresponding to all candidate native place names.

[0074] In an exemplary embodiment, a computer device, which can be a server or a terminal, is provided. An internal structure diagram of the computer device can be as follows: Figure 4As shown in the figure. The computer device includes a processor, a memory, an input / output interface (I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through the system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capability. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store target common names, geographic entity names and their corresponding derived element attributes. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with the terminal outside through network connection. The computer program is executed by the processor to realize a geographic entity name derivation relationship identification method.

[0075] Those skilled in the art can understand that, Figure 4 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different component arrangement.

[0076] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to realize the steps in each of the above method embodiments.

[0077] In one exemplary embodiment, a computer readable storage medium is provided, storing a computer program, which is executed by a processor to realize the steps in each of the above method embodiments.

[0078] In one exemplary embodiment, a computer program product is provided, including a computer program, which is executed by a processor to realize the steps in each of the above method embodiments.

[0079] The beneficial effects of the present application are as follows.

[0080] 1. Automatic recognition of derived geographic names based on neural network model.

[0081] The present application is based on a neural network model, especially the combination of prompt learning and bidirectional pre-training language model (such as BERT, etc.), which is specially used for recognizing "derived geographic names" (such as "XX new village" "XX new district" etc.), and realizes the automatic and high-precision extraction of derived geographic name entities from unstructured text, replacing the traditional method relying on dictionary or fixed rules.

[0082] 2. A neighborhood candidate native place name retrieval method based on geographic statistics.

[0083] The present application proposes to automatically retrieve candidate native place names from geographic databases through spatial neighborhood analysis methods and geographic statistical characteristics (such as distance decay model, place name density distribution, etc.), rather than relying on simple string matching or fixed range buffer analysis, thereby improving the accuracy and coverage of the candidate set.

[0084] 3. Compound feature engineering fusing semantic features and spatiotemporal distribution features.

[0085] The present application constructs a fusion feature containing both of the following two types of features, which is used for the discrimination of derived relationship.

[0086] ① Semantic features: including context embedding, place name semantic similarity, etc.

[0087] ② Spatiotemporal distribution features: including time series distribution of place name appearance, spatial aggregation degree, migration trajectory, etc. dynamic geographic statistical indicators.

[0088] 4. Machine learning discrimination model based on semantic spatiotemporal distribution-attentional feature fusion (SSD-AFF).

[0089] The present application proposes the SSD-AFF method, which adaptively weights and fuses semantic features and spatiotemporal distribution features through an attention mechanism, thereby enhancing the modeling ability of the model for the interaction between the two types of features.

[0090] The fusion feature is embedded into a machine learning classifier, which realizes end-to-end derived relationship discrimination, replacing the traditional rule-based or simple machine learning classifier scheme.

[0091] 5. "Recognition-retrieval-discrimination" whole-process automation technology chain.

[0092] The present application first serializes the prompt learning guided derived place name recognition, geographic statistics driven candidate retrieval, and SSD-AFF machine learning network discrimination into a complete technical framework, realizes the whole-process automation from raw data to derived relationship discrimination, and significantly improves the efficiency and consistency in place name translation tasks.

[0093] It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant regulations.

[0094] It can be understood by those skilled in the art that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing related hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments of each method. Any reference to memory, database or other medium used in the embodiments provided by the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0095] The database involved in the embodiments provided by the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments provided by the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, etc., without being limited thereto.

[0096] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above embodiments are not described, but as long as the combination of the technical features does not exist contradictory, it should be considered as the scope of the present application.

[0097] The principles and implementation manners of the present application are described herein by using specific examples, and the above examples are only used to help understand the method of the present application and its core idea; meanwhile, for those skilled in the art, according to the idea of the present application, the specific implementation manners and application ranges will have changes. In conclusion, the content of the specification should not be understood as a limitation of the present application.

Claims

1. A method for identifying a geographical entity name derivation relationship, characterized in that, The geographic entity name derivation relationship identification method comprises: obtaining a target common name, a geographic entity name and a corresponding derived element attribute; the derived element attribute comprises a derived geometric parameter, a derived start time and a derived category attribute; using a derivation identification model to identify whether the geographic entity name is a derived place name of the target common name; the derivation identification model is a neural network model previously built according to a first training sample set; each training sample in the first training sample set comprises a sample geographic entity name, a sample target common name and a binary label; if the geographic entity name is not a derived place name of the target common name, the geographic entity name and the target common name have no derivation relationship; if the geographic entity name is a derived place name of the target common name, a spatial proximity analysis method is used to obtain a plurality of candidate original place names according to the derived geometric parameter, and an original element attribute corresponding to each candidate original place name is obtained; for any candidate original place name, feature engineering processing is performed on the geographic entity name, the derived element attribute, the candidate original place name and the original element attribute corresponding to the candidate original place name to obtain a fusion feature corresponding to the candidate original place name; the fusion feature comprises a spatiotemporal distribution feature and a semantic feature; a probability that the candidate original place name is an original place name of the geographic entity name is obtained using an original probability model according to the fusion feature; the original probability model is a machine learning model previously built according to a second training sample set; the second training sample set comprises a sample fusion feature and a probability that a sample original place name corresponding to each sample fusion feature is an original place name of a sample geographic entity name; a corresponding original place name of the geographic entity name is determined according to probabilities corresponding to all candidate original place names.

2. The method of claim 1, wherein the geographical entity name derivation relationship identification method is characterized by, The method for building the derivation identification model comprises: obtaining a plurality of sample geographic entity names and pre-setting a plurality of sample target common names for each sample geographic entity name; for any sample target common name, if a sample geographic entity name corresponding to the sample target common name is a derived place name of the sample target common name, a binary label corresponding to the sample target common name is set to 1, otherwise the binary label corresponding to the sample target common name is set to 0; a neural network model is trained according to all sample target common names and sample geographic entity names and binary labels corresponding to each sample target common name to obtain the derivation identification model.

3. The method of claim 1, wherein the geographical entity name derivation relationship identification method is characterized by, According to the derived geometric parameter, a spatial proximity analysis method is used to obtain a plurality of candidate original place names, specifically comprising: calculating geographic coordinates corresponding to the geographic entity name according to the derived geometric parameter; using a geographic coding technology, a preset range geographic space corresponding to the geographic entity name is grid divided according to the geographic coordinates to obtain a grid division result and a grid index corresponding to the grid division result; using a spatial index query and distance clipping method, an original place name within a preset radius range is obtained according to the geographic coordinates, the grid division result, the grid index and the preset radius, and a category attribute corresponding to the original place name is obtained; According to the derived category attribute and the category attribute corresponding to the original place name, a plurality of candidate original place names are obtained.

4. The method of claim 3, wherein the geographical entity name derivation relationship identification method is characterized by, The category attribute corresponding to each candidate original place name is different from the derived category attribute.

5. The geographic entity name derivation relationship identification method according to claim 1, characterized by, The original element attribute includes an original geometric parameter, an original start time, and an original category attribute; For any candidate original place name, the geographic entity name, the derived element attribute, the candidate original place name, and the original element attribute corresponding to the candidate original place name are subjected to feature engineering processing to obtain a fusion feature corresponding to the candidate original place name, specifically including: For any candidate original place name, according to the derived geometric parameter and the original geometric parameter corresponding to the candidate original place name, a topological relationship calculation method and a spatial distance calculation method are respectively used to obtain a topological relationship feature and a distance feature corresponding to the candidate original place name; According to the derived start time and the original start time corresponding to the candidate original place name, a qualitative temporal reasoning method is used to obtain a time sequence feature corresponding to the candidate original place name; According to the derived category attribute and the original category attribute corresponding to the candidate original place name, a category mapping prior method is used to obtain a category association feature corresponding to the candidate original place name; According to the geographic entity name and the candidate original place name, a word sequence pair matching method is used to obtain a text feature corresponding to the candidate original place name; The topological relationship feature, the distance feature, and the time sequence feature are subjected to spatio-temporal feature coding to obtain the spatio-temporal distribution feature; The category association feature and the text feature are subjected to text semantic feature coding to obtain the semantic feature; The spatio-temporal distribution feature and the semantic feature are subjected to cross-modal feature fusion to obtain the fusion feature.

6. The method of claim 5, wherein the geographical entity name derivation relationship identification is characterized by, The spatio-temporal distribution feature and the semantic feature are subjected to cross-modal feature fusion to obtain the fusion feature, specifically including: The spatio-temporal distribution feature and the semantic feature are subjected to feature concatenation to obtain a concatenated feature; A channel attention mechanism is used to calculate feature weights of the concatenated feature to obtain spatio-temporal distribution feature weights and semantic feature weights; The spatio-temporal distribution feature and the semantic feature are subjected to weighted fusion using the spatio-temporal distribution feature weights and the semantic feature weights to obtain the fusion feature.

7. A geographic entity name derivation relationship identification apparatus characterized by comprising: The geographic entity name derived relationship identification device includes: An original data acquisition module configured to acquire a target generic name, a geographic entity name, and a corresponding derived element attribute; the derived element attribute includes a derived geometric parameter, a derived start time, and a derived category attribute; A derived place name identification module configured to identify whether the geographic entity name is a derived place name of the target generic name using a derived identification model; the derived identification model is a neural network model previously built according to a first training sample set; each training sample in the first training sample set includes a sample geographic entity name, a sample target generic name, and a binary label; A no derived relationship determination module configured to determine that the geographic entity name and the target generic name have no derived relationship when the geographic entity name is not a derived place name of the target generic name; The candidate native place name generation module is configured to, when the geographic entity name is a derived place name of the target generic name, obtain a plurality of candidate native place names according to the derived geometric parameters by using a spatial proximity analysis method, and obtain native element attributes corresponding to each candidate native place name; The fusion feature extraction module is configured to, for any candidate native place name, perform feature engineering processing on the geographic entity name, the derived element attributes, the candidate native place name, and the native element attributes corresponding to the candidate native place name, to obtain fusion features corresponding to the candidate native place name; the fusion features include spatiotemporal distribution features and semantic features; The native probability calculation module is configured to obtain a probability that the candidate native place name is a native place name of the geographic entity name according to the fusion features by using a native probability model; the native probability model is a machine learning model that is built in advance according to a second training sample set; the second training sample set includes sample fusion features and a probability that a sample native place name corresponding to each sample fusion feature is a native place name of a sample geographic entity name; The native place name determination module is configured to determine a native place name corresponding to the geographic entity name according to probabilities corresponding to all candidate native place names.

8. A computer device comprising: A memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that the processor executes the computer program to implement the geographic entity name derived relationship identification method of any one of claims 1-6.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the geographic entity name derived relationship identification method of any one of claims 1-6.

10. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the geographic entity name derived relationship identification method of any one of claims 1-6.

Images