CN117312478B - Address positioning method, device, electronic equipment and storage medium - Google Patents

Abstract

The invention provides an address positioning method, an address positioning device, electronic equipment and a storage medium, belonging to the technical field of positioning and searching, wherein the method comprises the following steps: determining a geographic grid range and an embedded representation vector of a query address based on the query address input by a user; determining a plurality of candidate address information matched with the geographical grid range from an address coordinate library, and determining an embedded representation vector of each candidate address information; the address coordinate library is constructed based on all address text data of the space region range and corresponding geographic coordinate information; and carrying out matching analysis on the embedded representation vector of the query address and the embedded representation vector of each piece of candidate address information, and determining the geographic position information of the query address. The invention can realize effective address positioning, improve the address inquiry matching efficiency and improve the address positioning precision.

Description

Address positioning method, device, electronic equipment and storage medium

Technical field

The present invention relates to the technical field of positioning and search, and in particular to an address positioning method, device, electronic equipment and storage medium.

Background technique

With the rapid development of Internet technology and intelligent positioning technology, in more and more Internet business scenarios (such as logistics and transportation, intelligent navigation services), users are required to enter address text information to obtain the corresponding actual geographical location information. In order to solve the above problems, currently the address resolution (Geocoding) algorithm is usually used, which parses the actual geographical location corresponding to the text address from the geocoding database based on the address information input by the user.

However, due to limitations of existing Geocoding technology, the number of Points of Interest (POIs) in the geocoding database is limited and cannot cover all text addresses entered by users. Moreover, the text addresses entered by users may also be irregular. , it is common to encounter text addresses that do not appear in the geocoding database. For text addresses that do not appear in the geocoding database, it is generally necessary to find nearby related POIs one by one in the geocoding database, and then output an approximate geographical location through interpolation and other methods. If the nearby POI cannot be queried, or the query is inaccurate, then the positioning cannot be performed or the positioning is invalid, and the efficiency of address query matching is also relatively low.

Contents of the invention

The present invention provides an address positioning method, device, electronic equipment and storage medium to solve the defects in the prior art that the geographical location information provided by Geocoding technology has low accuracy and the efficiency of address query and matching is also low.

The present invention provides an address positioning method, which includes:

Based on the query address input by the user, determine the geographical grid range and embedded representation vector of the query address;

Determine multiple candidate address information matching the geographical grid range from the address coordinate library, and determine the embedded representation vector of each candidate address information; the address coordinate library is based on all address texts of the spatial area range Data and corresponding geographical coordinate information are constructed;

Matching analysis is performed on the embedded representation vector of the query address and the embedded representation vector of each candidate address information to determine the geographical location information of the query address.

According to an address positioning method provided by the present invention, determining the geographical grid range and embedded representation vector of the query address based on the query address input by the user includes:

Input the query address input by the user into the positioning prediction neural network model, and obtain the geographical grid range and embedded representation vector of the query address output by the positioning prediction neural network model;

The positioning prediction neural network model is trained based on text address samples and their corresponding sample labels.

According to an address positioning method provided by the present invention, the positioning prediction neural network model includes a basic backbone sub-network, a first prediction sub-network and a second prediction sub-network; the query address input by the user is input into the positioning prediction Neural network model to obtain the geographical grid range and embedded representation vector of the query address output by the positioning prediction neural network model, including:

Input the query address input by the user into the basic backbone sub-network to obtain the text semantic vector of the query address output by the basic backbone sub-network; the basic backbone sub-network is constructed based on the Transformer encoder;

Input the textual semantic vector of the query address into the first prediction sub-network to obtain the geographical grid range of the query address output by the first prediction sub-network; the first prediction sub-network is based on full connection Constructed by layer neural network;

Input the textual semantic vector of the query address into the second prediction sub-network to obtain the embedded representation vector of the query address output by the second prediction sub-network; the second prediction sub-network is based on a sentence vector pool It is constructed by layer neural network and fully connected layer neural network.

According to an address positioning method provided by the present invention, the step of matching and analyzing the embedded representation vector of the query address and the embedded representation vector of each candidate address information to determine the geographical location information of the query address includes:

Using a similarity matching method, calculate the distance between the embedded representation vector of the query address and the embedded representation vector of each of the candidate address information to determine the distance of the embedded representation vector of the query address relative to each of the candidate address information. The embedding represents the similarity information of the vector;

From each of the candidate address information, determine the target candidate address information corresponding to the maximum value in the similarity information;

According to the target candidate address information, the geographical location information of the query address is obtained.

According to an address positioning method provided by the present invention, before inputting the query address input by the user into the positioning prediction neural network model, the method further includes:

Use the text address sample and its corresponding sample label as a set of training samples to obtain multiple sets of training samples;

The multiple sets of training samples are used to train the positioning prediction neural network model.

According to an address positioning method provided by the present invention, using the multiple sets of training samples to train a positioning prediction neural network model includes:

For any set of training samples, input the training samples into the positioning prediction neural network model, and output the prediction probability corresponding to the training samples;

Using a preset loss function, calculate the loss value based on the predicted probability corresponding to the training sample and the sample label in the training sample;

Based on the loss value, adjust the model parameters of the positioning prediction neural network model until the number of model training times reaches a preset number;

The model parameters obtained when the number of model training times reaches the preset number of times are used as model parameters of the trained positioning prediction neural network model.

According to an address positioning method provided by the present invention, before determining the geographical grid range and embedded representation vector of the query address based on the query address input by the user, the method further includes:

Divide the spatial area into geographical grids, number each divided geographical grid, and determine the grid number of each geographical grid;

Associate all address text data and corresponding geographic coordinate information in the spatial area with the grid number of each geographic grid, and obtain the associated information after association processing;

Based on the associated information, construct geographic grid space index rules for all address text data and corresponding geographic coordinate information, and determine embedded representation vectors of all address text data;

The geographical grid spatial index rules, the embedded representation vectors of all address text data, the all address text data and the corresponding geographical coordinate information are stored in a spatial database to obtain the address coordinate library.

The invention also provides an address positioning device, including:

A processing module configured to determine the geographical grid range and embedded representation vector of the query address based on the query address input by the user;

A matching module, configured to determine multiple candidate address information that matches the geographical grid range from the address coordinate library, and determine an embedded representation vector of each candidate address information; the address coordinate library is based on a spatial area All address text data in the range and corresponding geographical coordinate information are constructed;

A positioning module, configured to perform matching analysis on the embedded representation vector of the query address and the embedded representation vector of each candidate address information, and determine the geographical location information of the query address.

The present invention also provides an electronic device, including a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements any of the above address positioning methods. .

The present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, it implements any of the above address positioning methods.

The present invention also provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the computer program implements any of the above address positioning methods.

The address positioning method, device, electronic equipment and storage medium provided by the present invention pre-construct an address coordinate library by utilizing all address text data in the spatial area and corresponding geographical coordinate information, and use the semantic feature embedding representation of the address text to improve the address. The text query matching efficiency enables that after receiving the query address input by the user, by determining the geographical grid range and embedded representation vector of the query address, multiple addresses matching the geographical grid range can be quickly matched from the address coordinate library. candidate address information and the embedded representation vector of each candidate address information, and then perform matching analysis on the embedded representation vector of the query address and the embedded representation vector of each candidate address information, and obtain the query address based on the successfully matched candidate address information. Geographical location information not only achieves effective address positioning and improves the efficiency of address query matching, but also improves the accuracy of address positioning.

Description of drawings

In order to explain the present invention or the technical solutions in the prior art more clearly, the drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are of the present invention. For some embodiments of the invention, those of ordinary skill in the art can also obtain other drawings based on these drawings without exerting creative efforts.

Figure 1 is one of the flow diagrams of the address positioning method provided by the present invention;

Figure 2 is the second schematic flow chart of the address positioning method provided by the present invention;

Figure 3 is a schematic structural diagram of the address positioning device provided by the present invention;

Figure 4 is a schematic diagram of the physical structure of the electronic device provided by the present invention.

Detailed ways

In order to make the purpose, technical solutions and advantages of the present invention more clear, the technical solutions in the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention. , not all examples. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without making creative efforts fall within the scope of protection of the present invention.

In the description of the invention, it should be noted that, unless otherwise clearly stated and limited, the terms "installation", "connection" and "connection" should be understood in a broad sense. For example, it can be a fixed connection or a detachable connection. , or integrally connected; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium; it can be an internal connection between two components. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

The address positioning method, device, electronic equipment and storage medium of the present invention will be described below with reference to Figures 1-4.

Figure 1 is one of the flow diagrams of the address positioning method provided by the present invention. As shown in Figure 1, it includes: step 110, step 120 and step 130.

Step 110: Based on the query address input by the user, determine the geographical grid range and embedded representation vector of the query address;

Step 120: Determine multiple candidate address information matching the geographical grid range from the address coordinate library, and determine the embedded representation vector of each candidate address information; the address coordinate library is based on all address text data and corresponding spatial area ranges. Constructed with geographical coordinate information;

Step 130: Perform matching analysis on the embedded representation vector of the query address and the embedded representation vector of each candidate address information to determine the geographical location information of the query address.

Specifically, the query address described in the embodiment of the present invention refers to the text information input by the user on the front-end human-computer interaction interface to describe the geographical location that the user wants to query, which may specifically include province, city, district, street /Town and other regional level place names or detailed address information.

The embedded representation vector described in the embodiment of the present invention refers to the vector representation obtained by extracting text features of the query address input by the user through a text feature extraction algorithm, which can be used to characterize the relevant semantic information of the query address text.

The geographical grid range described in the embodiment of the present invention refers to the geographical grid range to which the query address belongs obtained by using a geographical location prediction algorithm or model to predict and query the query address input by the user.

The address coordinate database described in the embodiment of the present invention is an address database constructed in advance based on all address text data in the spatial area and corresponding geographical coordinate information. This database stores all address data and the text semantic features corresponding to each address data. Embedding represents vector data.

In the embodiment of the present invention, in step 110, by receiving the query address information input by the user on the front end, the text features of the query address input by the user are extracted, and the geographical location prediction and embedding vector representation are performed based on the extracted text features. Thus, the geographical grid range and embedded representation vector of the query address are obtained.

Further, in the embodiment of the present invention, in step 120, according to the obtained geographical grid range, multiple candidates matching the geographical grid range can be found from the pre-constructed address coordinate library address information, and determine the embedding representation vector of each candidate address information.

Further, in the embodiment of the present invention, in step 130, after determining multiple candidate address information and corresponding embedded representation vectors matching the geographical grid range, the embedded representation vector of the query address is compared with each Matching analysis can be performed on the embedding representation vectors of each candidate address information. By calculating the similarity between the embedding representation vector of the query address and the embedding representation vector of each candidate address information, the embedding with the highest similarity to the embedding representation vector of the query address can be matched. Representation vector, from which the candidate address information corresponding to the matched embedded representation vector is deduced, so that the candidate address information is finally used as the geographical location information of the query address, thereby completing the rapid and accurate positioning of the query address input by the user.

The address positioning method of the embodiment of the present invention pre-constructs an address coordinate library by using all address text data in the spatial area and corresponding geographical coordinate information, and uses the semantic feature embedding representation of the address text to improve the address text query matching efficiency, so that in After receiving the query address input by the user, by determining the geographical grid range and embedded representation vector of the query address, multiple candidate address information matching the geographical grid range and each candidate address information can be quickly matched from the address coordinate library. The embedded representation vector of the candidate address information is then matched and analyzed between the embedded representation vector of the query address and the embedded representation vector of each candidate address information, and the geographical location information of the query address is obtained based on the successfully matched candidate address information, which is effective in achieving Address positioning, while improving the efficiency of address query matching, also improves the accuracy of address positioning.

Based on the contents of the above embodiments, as an optional embodiment, based on the query address input by the user, determining the geographical grid range and embedded representation vector of the query address includes:

Input the query address input by the user into the positioning prediction neural network model, and obtain the geographical grid range and embedded representation vector of the query address output by the positioning prediction neural network model;

The location prediction neural network model is trained based on text address samples and their corresponding sample labels.

Specifically, the positioning prediction neural network model described in the embodiment of the present invention is obtained by training a deep neural network model based on text address samples and their corresponding sample labels. Based on the powerful learning ability and feature extraction ability of the deep neural network, the Identify different text address information, learn the association between address text features and its real geographical grid range information, and learn the regular relationship between the address text features converted into embedded representation vectors of their text semantic features, thereby outputting text at the same time The geographical grid range corresponding to the address and the corresponding embedding representation vector.

It should be noted that, for the embedding representation vector, a fixed neural network structure can be used in the positioning prediction neural network model to perform a vector embedding representation operation on the text address, thereby extracting the embedded representation vector of the text address.

Among them, the deep neural network can be composed of a Transformer neural network model including a multi-head attention mechanism, a pooling layer neural network, a front connection layer neural network, etc. It can also use other geographical grid ranges used to accurately identify and output text addresses. and neural networks embedding representation vectors, which are not specifically limited in the present invention.

The sample labels described in the embodiments of the present invention are predetermined based on the text address samples, and correspond one-to-one with the text address samples. In other words, each text address sample in the training sample is preset to carry a corresponding sample label. Among them, the sample labels include labels required for training the model to predict the geographical grid range, and labels required for training the model to generate embedding representation vectors.

Further, in the embodiment of the present invention, by inputting the query address input by the user into the trained positioning prediction neural network model, the deep neural network is used to identify and process the query address input by the user, and at the same time, the query address is obtained. The geographic grid range and embedded representation vector are used to prepare for subsequent address query matching.

The method of the embodiment of the present invention uses a deep neural network model with a double-headed output to simultaneously calculate the geographical grid range to which the query address belongs and the query address embedding representation vector used for subsequent address matching calculations. Through the model, only the query needs to be The address is processed once, which greatly reduces the tedious and time-consuming processing steps, and can effectively improve the effect of text address query matching and positioning.

Based on the content of the above embodiment, as an optional embodiment, the positioning prediction neural network model includes a basic backbone subnetwork, a first prediction subnetwork and a second prediction subnetwork; the query address input by the user is input into the positioning prediction neural network model, and the geographic grid range and embedded representation vector of the query address output by the positioning prediction neural network model are obtained, including:

Input the query address input by the user into the basic backbone subnetwork, and obtain the text semantic vector of the query address output by the basic backbone subnetwork; the basic backbone subnetwork is built based on the Transformer encoder;

Input the textual semantic vector of the query address into the first prediction sub-network to obtain the geographical grid range of the query address output by the first prediction sub-network; the first prediction sub-network is constructed based on a fully connected layer neural network;

Input the textual semantic vector of the query address into the second prediction sub-network to obtain the embedded representation vector of the query address output by the second prediction sub-network; the second prediction sub-network is based on the sentence vector pooling layer neural network and the fully connected layer neural network Constructed.

Specifically, the basic backbone subnetwork described in the embodiment of the present invention is used to extract text features of the query address input by the user and output the text semantic vector of the query address, which can also be described as a backbone network. It is built based on the Transformer encoder, which can be built using the basic Roberta-wwm pre-trained model.

Among them, the basic version of Roberta-wwm model is stacked by 12 Transformer encoder modules. Each Transformer encoder is composed of 12 multi-head self-attention modules, 1 feedforward neural network and 1 residual connection. The multi-head self-attention mechanism divides the input sequence into multiple subspaces and performs self-attention calculations on each subspace. In this way, the model can focus on information from different subspaces simultaneously to gain a more comprehensive understanding of the input sequence. In terms of training strategy, Roberta-wwm removes the next sentence prediction task proposed in the language model BERT, and uses a dynamic whole word mask pre-training mechanism to update model parameters.

The first prediction sub-network described in the embodiment of the present invention is used to predict the geographical location information, such as the geographical grid number, to which the query address text belongs based on the text semantic vector of the query address. It can also be described as a geographical grid prediction sub-network. It is constructed based on a fully connected layer neural network. Specifically, it can be constructed using a two-layer fully connected neural network. The two fully connected layers are connected through the Tanh activation function.

The second prediction sub-network described in the embodiment of the present invention is used to further perform feature mapping on the text semantic vector according to the text semantic vector of the query address, and obtain the embedded representation vector of the query address for subsequent embedded representation vectors. Matching, which can also be described as a query matching subnetwork. It is constructed based on the sentence vector pooling layer neural network and the fully connected layer neural network. Specifically, it can be composed of 1 sentence vector pooling layer and 1 fully connected layer.

Optionally, in the embodiment of the present invention, when the user inputs a piece of query address text information, two special identifiers [CLS] and [SEP] can be added at the beginning and end respectively. [CLS] is located at the beginning of the sentence and is used to aggregate the text semantic vector representation of the entire text and serves as the input of the first prediction sub-network. The [SEP] identifier is used to identify the end of the sentence. After that, a word segmenter is used to split the input query address into several independent characters, and then input into the basic backbone subnetwork Roberta-wwm model for feature extraction to obtain the text semantic vector of the query address.

It should be noted that in this embodiment of the present invention, the spatial area to which the address database belongs is divided into geographical grids in advance, and each grid is represented by a unique number.

Further, in the embodiment of the present invention, the textual semantic vector of the [CLS] identifier of the query address can be input to the first prediction sub-network, and processed by the two-layer fully connected layer neural network in the first prediction sub-network, And after the second fully connected layer, a softmax function is used to convert the output vector into the probability that the query address belongs to each geographical grid number, and the result with the highest probability is output, thereby obtaining the geographical grid range of the query address, that is, the query address is obtained The geographical grid number to which it belongs.

Further, in the embodiment of the present invention, the text semantic vector of the query address is simultaneously input to the sentence vector pooling layer and the fully connected layer in the second prediction sub-network. Through the sentence vector pooling layer, the basic backbone sub-network is The semantic vector of each character in the output query address is pooled into a fixed-dimensional vector representation, and then a fully connected layer is used to further project the features to obtain the embedded representation vector of the query address that is ultimately used for subsequent similarity matching calculations.

The method of the embodiment of the present invention uses the basic backbone sub-network, the first prediction sub-network and the second prediction sub-network to construct a deep neural network model with dual-head output, which can simultaneously calculate the geographical grid number to which the query address belongs and the address used for it. The semantic feature embedding representation vector for similarity calculation is used to filter out the candidate address set from the global address coordinate library using the predicted geographical grid, avoiding similarity calculation with all addresses and improving query matching efficiency; and using the embedded representation of the query address Vectors are used to calculate similarity to improve the query matching effect.

Based on the contents of the above embodiments, as an optional embodiment, a matching analysis is performed on the embedded representation vector of the query address and the embedded representation vector of each candidate address information to determine the geographical location information of the query address, including:

Using a similarity matching method, calculate the distance between the embedded representation vector of the query address and the embedded representation vector of each candidate address information to determine the similarity information of the embedded representation vector of the query address relative to the embedded representation vector of each candidate address information;

From each candidate address information, determine the target candidate address information corresponding to the maximum value in the similarity information;

According to the target candidate address information, the geographical location information of the query address is obtained.

Specifically, the target candidate address information described in the embodiment of the present invention refers to the candidate address information with the greatest similarity determined through similarity matching calculation between the embedded representation vector of the query address and the embedded representation vector of each candidate address information. .

In embodiments of the present invention, by calculating the Euclidean distance between the embedded representation vector of the query address and the embedded representation vector of each candidate address information, similarity ranking can be performed according to the calculated Euclidean distance. The smaller the Euclidean distance, the greater the similarity; the greater the Euclidean distance, the smaller the similarity. In this way, the similarity information of the embedded representation vector of the query address with respect to the embedded representation vector of each candidate address information can be determined, and then the candidate address information can be sorted in order of similarity.

Further, in the embodiment of the present invention, from each candidate address information, the target candidate address information corresponding to the maximum value in the similarity information is determined, and the target candidate address information is used as the geographical location information of the query address, so as to Realize the positioning of the query address.

The method of the embodiment of the present invention filters out a set of candidate addresses from the global address coordinate library, and uses the embedded representation vector of the query address and the embedded representation vector of the candidate address to perform similarity calculations, thus avoiding the need to perform similarity calculations with all global addresses. It is cumbersome, improves the efficiency of query address matching, and improves the effect of address positioning.

Based on the contents of the above embodiments, as an optional embodiment, before inputting the query address input by the user into the positioning prediction neural network model, the method further includes:

Use text address samples and their corresponding sample labels as a set of training samples to obtain multiple sets of training samples;

Multiple sets of training samples are used to train the positioning prediction neural network model.

Specifically, in the embodiment of the present invention, before inputting the query address input by the user into the positioning prediction neural network model, the positioning prediction neural network model needs to be trained to obtain a trained positioning prediction neural network model.

In the embodiment of the present invention, the positioning prediction neural network model is trained using training set data. The specific training process is as follows:

The text address samples and their corresponding sample labels are used as a set of training samples. For different text address samples, multiple sets of training samples can be obtained.

In the embodiment of the present invention, there is a one-to-one correspondence between text address samples and the sample tags they carry.

Then, after obtaining multiple sets of training samples, the multiple sets of training samples are sequentially input into the positioning prediction neural network model, and the multiple sets of training samples are used to train the positioning prediction neural network model, that is:

The text address samples and their corresponding sample labels in each group of training samples are simultaneously input into the positioning prediction neural network model. According to the prediction results of each output in the positioning prediction neural network model, the positioning prediction is performed by calculating the loss function value. The model parameters in the neural network model are adjusted, and when the preset training termination conditions are met, the entire training process of the positioning prediction neural network model is finally completed, and the trained positioning prediction neural network model is obtained.

The method of the embodiment of the present invention uses text address samples and their corresponding sample labels as a set of training samples, and uses multiple sets of training samples to train the positioning prediction neural network model, which is beneficial to improving the performance of the trained positioning prediction neural network model. Model accuracy.

Based on the contents of the above embodiments, as an optional embodiment, multiple sets of training samples are used to train the positioning prediction neural network model, including:

For any set of training samples, input the training samples into the positioning prediction neural network model and output the prediction probability corresponding to the training samples;

Use the preset loss function to calculate the loss value based on the predicted probability corresponding to the training sample and the label in the training sample;

Based on the loss value, adjust the model parameters of the positioning prediction neural network model until the number of model training times reaches the preset number;

The model parameters obtained when the number of model training times reaches the preset number are used as the model parameters of the trained positioning prediction neural network model.

Specifically, the preset loss function described in the embodiment of the present invention refers to the loss function preset in the positioning prediction neural network model, which is used for model evaluation; the preset threshold refers to the threshold value preset by the model, which is used for model evaluation. Obtain the minimum loss value and complete the model training; the preset number of times refers to the preset maximum number of iterative training times for the model.

After obtaining multiple sets of training samples, for any set of training samples, the text address samples and their corresponding sample labels in each set of training samples are simultaneously input to the positioning prediction neural network model, and the predicted probability of the corresponding result of the training sample is output.

On this basis, the preset loss function is used to calculate the loss value based on the predicted probability corresponding to the training sample and the sample label corresponding to the training sample.

Further, after the loss value is calculated, the training process ends. Then use the BackPropagation (BP) algorithm to adjust the model parameters of the positioning prediction neural network model based on the loss value to update the weight parameters of each layer of the model in the positioning prediction neural network model, and then proceed to the next time Training, and so on iteratively perform model training.

During the training process, if the training results for a certain set of training samples meet the preset training termination conditions, such as the corresponding calculated loss value is less than the preset threshold, or the current iteration number reaches the preset number, the model's loss value If it can be controlled within the convergence range, the model training ends. At this time, the obtained model parameters can be used as model parameters of the trained positioning prediction neural network model, then the positioning prediction neural network model training is completed, thereby obtaining the trained positioning prediction neural network model.

The method of the embodiment of the present invention uses multiple sets of training samples to repeatedly and iteratively train the positioning prediction neural network model, and controls the loss value of the positioning prediction neural network model within the convergence range, thereby helping to improve the accuracy of the model output results. Improve the accuracy of address location prediction results.

Based on the contents of the above embodiments, as an optional embodiment, before determining the geographical grid range and embedded representation vector of the query address based on the query address input by the user, the method further includes:

Divide the spatial area into geographical grids, number each divided geographical grid, and determine the grid number of each geographical grid;

Associate all address text data and corresponding geographic coordinate information in the spatial area with the grid number of each geographic grid, and obtain the associated information after association processing;

Based on the associated information, construct geographic grid spatial index rules for all address text data and corresponding geographic coordinate information, and determine the embedded representation vectors of all address text data;

The geographical grid spatial index rules, the embedded representation vectors of all address text data, all address text data and the corresponding geographical coordinate information are stored in the spatial database to obtain the address coordinate library.

Specifically, in the embodiment of the present invention, before determining the geographical grid range and embedding representation vector of the query address based on the query address input by the user, the address coordinate library needs to be pre-constructed.

In the embodiment of the present invention, the space area range refers to the space area range covering the required query address, which can be a specific space area range on the earth or a global area range.

In the embodiment of the present invention, first, the spatial area range is divided into geographical grids, and each divided geographical grid is uniquely numbered, and a unique grid number is set for each geographical grid in turn.

Furthermore, all address text data and corresponding geographical coordinate information in the spatial area are associated with the grid number of each geographical grid, so that the grid number of each geographical grid corresponds to its related address data. and geographical coordinate information (such as latitude and longitude information), to construct the association information between the address text data and the corresponding geographical coordinate information and each geographical grid number.

Further, in the embodiment of the present invention, based on the above-mentioned associated information, geographical grid space index rules are constructed for all address text data and corresponding geographical coordinate information, so that under the geographical grid space index rules, the geographical grid can be The grid number can be indexed and queried to the corresponding geographical coordinate information.

At the same time, all address texts in the address coordinate library are input into the basic backbone subnetwork and query matching subnetwork of the aforementioned positioning prediction neural network model, thereby obtaining the embedded representation vectors of all address text data.

Finally, in the embodiment of the present invention, the above-mentioned geographical grid spatial index rules, the embedded representation vectors of all address text data, all address text data and the corresponding geographical coordinate information are stored in the spatial database, thereby constructing the address coordinates library.

The method of the embodiment of the present invention uses the powerful learning ability and feature extraction ability of the deep neural network to directly store address data and geographical coordinate data into the database, and uses semantic features to embed representation vectors for query matching without the need to construct complex manual matching. Rules, compared with traditional methods based on statistical word segmentation, address models, and manual matching rules, the method of constructing an address coordinate library in the present invention is simpler and more efficient.

Figure 2 is a second schematic flow chart of the address positioning method provided by the present invention. As shown in Figure 2, in a specific embodiment of the present invention, first, by receiving the query address input by the front-end user, the query address is first input to The backbone network obtains the text semantic vector of the query address. Then, the textual semantic vector of the query address is input to the geographical grid prediction subnetwork and the query matching subnetwork respectively, and the geographical grid number to which the query address belongs is predicted through the geographical grid prediction subnetwork, and the query matching subnetwork is used to output the A vector embedding representation of the query address.

Further, in this embodiment, the geographical grid coordinate range of the grid is obtained according to the geographical grid number to which the query address belongs. Then, use spatial query to select all address data within the geographical grid coordinate range from the address coordinate database as candidate address information, and a set of candidate address embedding representation vectors corresponding to each candidate address information, which contains each candidate address. Embedding representation vector of information.

Further, in this embodiment, the embedded representation vector of the query address and the set of candidate address embedded representation vectors are input to the similarity calculation network layer for similarity matching analysis, and the embedded representation vector of the query address and each candidate address are calculated. The embedding of information represents the similarity of vectors, determines the best matching address and coordinate calculation results, and completes the positioning of the query address input by the user.

Compared with existing text address positioning technology, the method of the embodiment of the present invention can simultaneously calculate the geographical grid information to which the query address belongs and the semantic feature embedding representation vector used for query matching. Compared with using two deep neural networks of the same scale respectively, Network model prediction methods for geographic grids and query matching can reduce processing steps and save computing resources.

Compared with the method of only using a query matching model based on a deep neural network for global retrieval, the present invention can effectively reduce the number of candidate addresses to be query matched through geographical grid screening, and improve the calculation efficiency. In addition, if the query address has not been stored in the address coordinate library, this method can output the geographical range to which the query address belongs based on the predicted geographical grid number.

The address positioning device provided by the present invention will be described below. The address positioning device described below and the address positioning method described above can be referenced correspondingly.

Figure 3 is a schematic structural diagram of the address positioning device provided by the present invention. As shown in Figure 3, it includes:

The processing module 310 is used to process the module and determine the geographical grid range and embedded representation vector of the query address based on the query address input by the user;

Matching module 320, used to determine multiple candidate address information that matches the geographical grid range from the address coordinate library, and determine the embedded representation vector of each candidate address information; the address coordinate library is all address text based on the spatial area range Data and corresponding geographical coordinate information are constructed;

The positioning module 330 is used to perform matching analysis on the embedded representation vector of the query address and the embedded representation vector of each candidate address information, and determine the geographical location information of the query address.

The address positioning device described in this embodiment can be used to execute the above embodiment of the address positioning method. Its principles and technical effects are similar and will not be described again here.

The address positioning device according to the embodiment of the present invention pre-constructs an address coordinate library by using all address text data in the spatial area and corresponding geographical coordinate information, and uses the semantic feature embedding representation of the address text to improve address text query matching efficiency, so that in After receiving the query address input by the user, by determining the geographical grid range and embedded representation vector of the query address, multiple candidate address information matching the geographical grid range and each candidate address information can be quickly matched from the address coordinate library. The embedded representation vector of the candidate address information is then matched and analyzed between the embedded representation vector of the query address and the embedded representation vector of each candidate address information, and the geographical location information of the query address is obtained based on the successfully matched candidate address information, which is effective in achieving Address positioning, while improving the efficiency of address query matching, also improves the accuracy of address positioning.

Figure 4 is a schematic diagram of the physical structure of an electronic device provided by the present invention. As shown in Figure 4, the electronic device may include: a processor (processor) 410, a communications interface (Communications Interface) 420, a memory (memory) 430 and a communication bus 440 , wherein the processor 410, the communication interface 420, and the memory 430 complete communication with each other through the communication bus 440. The processor 410 can call logical instructions in the memory 430 to execute the address positioning method provided by each of the above methods. The method includes: based on the query address input by the user, determining the geographical grid range and embedded representation vector of the query address; Determine multiple candidate address information matching the geographical grid range from the address coordinate library, and determine the embedded representation vector of each candidate address information; the address coordinate library is based on all address texts of the spatial area range The embedded representation vector of the query address and the embedded representation vector of each candidate address information are matched and analyzed to determine the geographical location information of the query address.

In addition, the above-mentioned logical instructions in the memory 430 can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention essentially or the part that contributes to the existing technology or the part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including Several instructions are used to cause a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disk and other media that can store program code. .

On the other hand, the present invention also provides a computer program product. The computer program product includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can Executing the address positioning method provided by each of the above methods, the method includes: based on the query address input by the user, determining the geographical grid range and embedded representation vector of the query address; determining the geographical grid range from the address coordinate library Match multiple candidate address information, and determine the embedded representation vector of each candidate address information; the address coordinate library is constructed based on all address text data in the spatial area range and the corresponding geographical coordinate information; all The embedded representation vector of the query address and the embedded representation vector of each candidate address information are matched and analyzed to determine the geographical location information of the query address.

In yet another aspect, the present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. The computer program is implemented when executed by a processor to perform the address positioning method provided by each of the above methods. The method includes: Based on the query address input by the user, determine the geographical grid range and embedded representation vector of the query address; determine multiple candidate address information matching the geographical grid range from the address coordinate library, and determine each of the The embedded representation vector of the candidate address information; the address coordinate library is constructed based on all address text data in the spatial area range and the corresponding geographical coordinate information; the embedded representation vector of the query address and each of the candidate address information Perform matching analysis on the embedded representation vector to determine the geographical location information of the query address.

The device embodiments described above are only illustrative. The units described as separate components may or may not be physically separated. The components shown as units may or may not be physical units, that is, they may be located in One location, or it can be distributed across multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Persons of ordinary skill in the art can understand and implement the method without any creative effort.

Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the part of the above technical solution that essentially contributes to the existing technology can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM/RAM, magnetic disc, optical disk, etc., including a number of instructions to cause a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments or certain parts of the embodiments.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that it can still be used Modifications are made to the technical solutions described in the foregoing embodiments, or equivalent substitutions are made to some of the technical features; however, these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims (7)

1. An address positioning method, characterized by comprising: Based on the query address input by the user, determine the geographical grid range and embedded representation vector of the query address; Determine multiple candidate address information matching the geographical grid range from the address coordinate library, and determine the embedded representation vector of each candidate address information; the address coordinate library is based on all address texts of the spatial area range Data and corresponding geographical coordinate information are constructed; Perform matching analysis on the embedded representation vector of the query address and the embedded representation vector of each candidate address information to determine the geographical location information of the query address; Wherein, based on the query address input by the user, determining the geographical grid range and embedded representation vector of the query address includes: Input the query address input by the user into the positioning prediction neural network model, and obtain the geographical grid range and embedded representation vector of the query address output by the positioning prediction neural network model; The positioning prediction neural network model is trained based on text address samples and their corresponding sample labels; Wherein, the positioning prediction neural network model includes a basic backbone sub-network, a first prediction sub-network and a second prediction sub-network; the query address input by the user is input into the positioning prediction neural network model to obtain the positioning prediction The geographical grid range and embedded representation vector of the query address output by the neural network model include: Inputting the query address input by the user into the basic backbone sub-network to obtain a text semantic vector of the query address output by the basic backbone sub-network; the basic backbone sub-network is constructed based on a Transformer encoder; Input the textual semantic vector of the query address into the first prediction sub-network to obtain the geographical grid range of the query address output by the first prediction sub-network; the first prediction sub-network is based on full connection Constructed by layer neural network; Input the textual semantic vector of the query address into the second prediction sub-network to obtain the embedded representation vector of the query address output by the second prediction sub-network; the second prediction sub-network is based on a sentence vector pool Constructed by layer neural network and fully connected layer neural network; Wherein, before determining the geographical grid range and embedded representation vector of the query address based on the query address input by the user, the method further includes: Divide the spatial area into geographical grids, number each divided geographical grid, and determine the grid number of each geographical grid; Associate all address text data and corresponding geographic coordinate information in the spatial area with the grid number of each geographic grid, and obtain the associated information after association processing; Based on the associated information, construct geographic grid space index rules for all address text data and corresponding geographic coordinate information, and determine embedded representation vectors of all address text data; The geographical grid spatial index rules, the embedded representation vectors of all address text data, the all address text data and the corresponding geographical coordinate information are stored in a spatial database to obtain the address coordinate library. 2. The address positioning method according to claim 1, characterized in that the embedded representation vector of the query address and the embedded representation vector of each candidate address information are matched and analyzed to determine the query address. Geolocation information, including: Using a similarity matching method, calculate the distance between the embedded representation vector of the query address and the embedded representation vector of each of the candidate address information to determine the distance of the embedded representation vector of the query address relative to each of the candidate address information. The embedding represents the similarity information of the vector; From each of the candidate address information, determine the target candidate address information corresponding to the maximum value in the similarity information; According to the target candidate address information, the geographical location information of the query address is obtained. 3. The address positioning method according to claim 1, characterized in that, before inputting the query address input by the user into the positioning prediction neural network model, the method further includes: Use the text address sample and its corresponding sample label as a set of training samples to obtain multiple sets of training samples; The multiple sets of training samples are used to train the positioning prediction neural network model. 4. The address positioning method according to claim 3, characterized in that said utilizing the plurality of sets of training samples to train a positioning prediction neural network model includes: For any set of training samples, input the training samples into the positioning prediction neural network model, and output the prediction probability corresponding to the training samples; Using a preset loss function, calculate the loss value based on the predicted probability corresponding to the training sample and the sample label in the training sample; Based on the loss value, adjust the model parameters of the positioning prediction neural network model until the number of model training times reaches a preset number; The model parameters obtained when the number of model training times reaches the preset number of times are used as model parameters of the trained positioning prediction neural network model. 5. An address positioning device, characterized by comprising: A processing module configured to determine the geographical grid range and embedded representation vector of the query address based on the query address input by the user; A matching module, configured to determine multiple candidate address information that matches the geographical grid range from the address coordinate library, and determine an embedded representation vector of each candidate address information; the address coordinate library is based on a spatial area All address text data in the range and corresponding geographical coordinate information are constructed; A positioning module, configured to perform matching analysis on the embedded representation vector of the query address and the embedded representation vector of each candidate address information, and determine the geographical location information of the query address; Among them, the processing module is specifically used for: Input the query address input by the user into the positioning prediction neural network model, and obtain the geographical grid range and embedded representation vector of the query address output by the positioning prediction neural network model; The positioning prediction neural network model is trained based on text address samples and their corresponding sample labels; Wherein, the positioning prediction neural network model includes a basic backbone sub-network, a first prediction sub-network and a second prediction sub-network; the processing module is also specifically used for: Input the query address input by the user into the basic backbone sub-network to obtain the text semantic vector of the query address output by the basic backbone sub-network; the basic backbone sub-network is constructed based on the Transformer encoder; Input the textual semantic vector of the query address into the first prediction sub-network to obtain the geographical grid range of the query address output by the first prediction sub-network; the first prediction sub-network is based on full connection Constructed by layer neural network; Input the textual semantic vector of the query address into the second prediction sub-network to obtain the embedded representation vector of the query address output by the second prediction sub-network; the second prediction sub-network is based on a sentence vector pool Constructed by layer neural network and fully connected layer neural network; Wherein, the device is specifically used for: Divide the spatial area into geographical grids, number each divided geographical grid, and determine the grid number of each geographical grid; Associate all address text data and corresponding geographic coordinate information in the spatial area with the grid number of each geographic grid, and obtain the associated information after association processing; Based on the associated information, construct geographic grid space index rules for all address text data and corresponding geographic coordinate information, and determine embedded representation vectors of all address text data; The geographical grid spatial index rules, the embedded representation vectors of all address text data, the all address text data and the corresponding geographical coordinate information are stored in a spatial database to obtain the address coordinate library. 6. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that when the processor executes the program, it implements claim 1 Go to the address positioning method described in any one of 4. 7. A non-transitory computer-readable storage medium on which a computer program is stored, characterized in that when the computer program is executed by a processor, the address positioning method according to any one of claims 1 to 4 is implemented.