Geographic object data processing method and geographic object value evaluation model training method

Abstract

The invention discloses a geographic object data processing method, a geographic object value evaluation model training method and device, electronic equipment and a computer storage medium. The geographic object data processing method comprises the steps that a to-be-evaluated target geographic object is determined; obtaining attribute data of the target geographic object, wherein the attribute data comprises feature data of the target geographic object, feature data of a region where the target geographic object is located, ownership transfer time sequence data of a region object where the target geographic object is located, and data of other geographic objects in a relationship between the position of the target geographic object and the region where the target geographic object is located; and inputting the attribute data into a pre-trained geographic object value evaluation model to obtain a value evaluation result for the target geographic object. According to the geographic object data processing method, the efficiency of geographic object value evaluation is improved, and meanwhile computing resources are reduced.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a geographic object data processing method, a geographic object value assessment model training method, device, electronic device and computer storage medium. Background Art

[0002] With the development of the economy, transactions of geographic objects (such as real estate, garages, etc.) are becoming more and more frequent. Geographic object transactions mainly refer to the transfer of ownership of geographic objects. Taking real estate transactions as an example, real estate valuation is particularly important in the process of real estate transactions. It is a key technology for realizing asset digitization. The accuracy of real estate valuation is of great significance to real estate transactions and is also an important prerequisite for the digitization of asset transactions. In the increasingly complex real estate evaluation, accurately evaluating the value of real estate has become particularly challenging.

[0003] When the ownership of a geographic object is transferred, such as house sales, house rentals, etc., it is usually necessary to evaluate the value of the geographic object. In the scenario of evaluating the value of a geographic object, one is to rely on manual calculation to evaluate the value of the geographic object. Manual calculation valuation is not only long and costly, but also has large evaluation differences due to the experience and preferences of appraisers; the other is to use a geographic object evaluation model to evaluate the value of the geographic object, and use digital form to estimate the value of the geographic object. Although it can improve the accuracy of online geographic object value evaluation compared to manual calculation, when using the existing geographic object evaluation model to evaluate the value of the geographic object, it is necessary to collect a large amount of disordered related data, and the collected data will vary from user to user. This results in this method not only occupying a large amount of computing resources, time-consuming and labor-intensive, but also there will be large differences between the valuation results of the geographic object and the actual transaction results, and between the results of use by different users, resulting in the situation that the geographic object transaction cycle is too long or the transaction cannot be completed.

[0004] Therefore, how to improve the efficiency of geographic object value assessment while reducing computing resources is an urgent problem that needs to be solved. Summary of the invention

[0005] The embodiment of the present application provides a geographic object data processing method to improve the efficiency of target geographic object value assessment while reducing computing resources.

[0006] The embodiment of the present application provides a method for processing geographic object data, including: determining a target geographic object to be evaluated; obtaining attribute data of the target geographic object, wherein the attribute data includes: characteristic data of the target geographic object, characteristic data of the region where the target geographic object is located, time series data of object ownership transfer in the region where the target geographic object is located, and other geographic object data in the relationship between the target geographic object position and the region where the target geographic object is located; inputting the attribute data into a pre-trained geographic object value assessment model to obtain a value assessment result for the target geographic object; wherein the geographic object value assessment model is trained using the following method: obtaining training sample data, wherein the training sample data includes attribute data of multiple geographic objects, wherein the attribute data of each geographic object includes: characteristic data of the geographic object, characteristic data of the region where the geographic object is located, time series data of object ownership transfer in the region where the geographic object is located, and other geographic object data in the relationship between the geographic object position and the region where the geographic object is located; inputting the training sample data into a pre-trained geographic object value assessment model to obtain a value assessment result for the target geographic object; wherein the geographic object value assessment model is trained using the following method: obtaining training sample data, wherein the training sample data includes attribute data of multiple geographic objects, wherein the attribute data of each geographic object includes: characteristic data of the geographic object, characteristic data of the region where the geographic object is located, time series data of object ownership transfer in the region where the geographic object is located, and other geographic object data in the relationship between the geographic object position and the region where the geographic object is located; Classify into numerical data, text data, time series data and geographic information data; input the training sample data into an initial model including an encoding module and a decoding module for iterative training, and the training process is as follows: input the attribute data of each geographic object into the corresponding encoding module according to the above classification category for encoding to generate a data vector; input the data vectors output by each encoding module into the sub-decoder corresponding to each encoding module, and at the same time, fuse the data vectors output by each encoding module and input them into the global decoder; obtain evaluation results by decoding through each sub-decoder, calculate each residual between each evaluation result and the target result, and feed back each residual to the corresponding encoder to adjust the parameters; obtain evaluation results by decoding through the global decoder, calculate the total residual between the evaluation result and the target result, and feed back the total residual to each encoder to adjust the parameters; iteratively execute the encoding and decoding process until the total residual meets the set value.

[0007] Optionally, in the training of the geographic object value assessment model, other geographic object data in the relationship between the geographic object location and the region in which it is located includes: geographic object relationship data and region relationship data; The geographic object relationship data is used to represent the relationship data between a selected geographic object and surrounding associated geographic objects, wherein the surrounding associated geographic objects are geographic objects whose distance from the selected geographic object is within a first preset distance range and have an associated influence on the user in the selected geographic object; The area relationship data is used to represent the relationship data between an area where a selected geographical object is located and a neighboring area, and the neighboring area is an area whose distance from the area where the selected geographical object is located is within a second preset distance range.

[0008] Optionally, the relationship data between the selected geographic object and surrounding associated geographic objects is obtained in the following manner: Obtaining the position data of the selected geographic object as a selected geographic object position vector; Determine the surrounding associated geographic objects and their location data as surrounding associated geographic object data vectors; According to the selected geographic object position vector and the surrounding associated geographic object data vector, relationship data between the selected geographic object and the surrounding associated geographic objects is determined.

[0009] Optionally, the relationship data between the selected geographic object and surrounding associated geographic objects is obtained in the following manner: Acquire the position data of the selected geographic object as the selected geographic object position vector; Determine surrounding associated geographical objects of the selected geographical object; For each of the surrounding associated geographic objects, determine its position data as a position vector of each surrounding associated geographic object; Calculate the difference between the position vector of each of the surrounding associated geographic objects and the position vector of the selected geographic object as a position difference vector; After linear processing, each position difference vector is added to the corresponding position vector of the surrounding associated geographic object to obtain a geocoded vector corresponding to each surrounding associated geographic object; All geocoded vectors are fused through a multi-head attention mechanism to obtain the relationship data between the selected geographic object and the surrounding associated geographic objects.

[0010] Optionally, the location data includes longitude and latitude; The position difference vector includes a longitude difference and a latitude difference.

[0011] Optionally, the location data includes longitude and latitude; The step of calculating the difference between the position vector of each of the surrounding associated geographic objects and the position vector of the selected geographic object as a position difference vector comprises: Calculating the longitude difference and latitude difference between each of the surrounding associated geographic objects and the selected geographic object; The longitude difference and the latitude difference are normalized and mapped into a code value of a set dimension to obtain the position difference vector.

[0012] Optionally, the relationship data between the area where the selected geographical object is located and the neighboring areas are obtained in the following manner: Obtaining characteristic data of other geographical objects and time series data of ownership transfer in the area where the selected geographical object is located as the selected area data vector; Determine the neighborhood area; Obtaining geographic object feature data and ownership transfer time series data of the neighboring area as a neighboring area data vector; According to the selected area data vector and the neighbor area data vector, the relationship data between the area where the selected geographic object is located and the neighbor area is obtained.

[0013] Optionally, the step of obtaining characteristic data of other geographical objects and time series data of ownership transfer in the area where the selected geographical object is located as the selected area data vector includes: Classify the characteristic data of other geographical objects in the area where the selected geographical object is located and the time series data of ownership transfer into text data and time series data, and for each geographical object in the other geographical objects, concatenate the text data and the time series data as a sub-vector, and combine the sub-vectors of all other geographical objects as the selected area data vector; The step of obtaining the geographic object feature data and ownership transfer time series data of the neighboring area as a neighboring area data vector includes: The characteristic data of the geographical objects in the neighboring area and the time series data of the ownership transfer are classified into text data and time series data. For each geographical object in the neighboring area, the text data and the time series data are spliced ​​as a sub-vector, and the sub-vectors of all geographical objects are combined as the data vector of the neighboring area.

[0014] The embodiment of the present application also provides a training method for a geographic object value assessment model, including: obtaining training sample data, the training sample data including attribute data of multiple geographic objects, the attribute data of each geographic object including: geographic object feature data, feature data of the region where the geographic object is located, time series data of object ownership transfer in the region where the geographic object is located, and other geographic object data in the relationship between the geographic object position and the region where it is located; classifying the training sample data into numerical data, text data, time series data, and geographic information data; inputting the training sample data into an initial model including an encoding module and a decoding module for iterative training, and the training process is as follows: the attribute data of each geographic object is classified according to the above The classification categories are respectively input into the corresponding encoding modules for encoding to generate data vectors; the data vectors output by each encoding module are respectively input into the sub-decoders corresponding to each encoding module, and the data vectors output by each encoding module are fused and input into the global decoder; the evaluation results are respectively obtained by decoding through each sub-decoder, and the residuals between each evaluation result and the target result are calculated, and the residuals are fed back to the corresponding encoders to adjust the parameters; the evaluation results are obtained by decoding through the global decoder, and the total residuals between the evaluation results and the target result are calculated, and the total residuals are respectively fed back to each encoder to adjust the parameters; the encoding and decoding process is iteratively performed until the total residual meets the set value.

[0015] Optionally, other geographic object data in the relationship between the geographic object location and the region in which it is located includes: geographic object relationship data and region relationship data; The geographic object relationship data is used to represent the relationship data between a selected geographic object and surrounding associated geographic objects, wherein the surrounding associated geographic objects are geographic objects whose distance from the selected geographic object is within a first preset distance range and have an associated influence on the user in the selected geographic object; The area relationship data is used to represent the relationship data between an area where a selected geographical object is located and a neighboring area, and the neighboring area is an area whose distance from the area where the selected geographical object is located is within a second preset distance range.

[0016] Optionally, the relationship data between the selected geographic object and surrounding associated geographic objects is obtained in the following manner: Acquire the position data of the selected geographic object as the selected geographic object position vector; Determine surrounding associated geographical objects of the selected geographical object; For each of the surrounding associated geographic objects, determine its position data as a position vector of each surrounding associated geographic object; Calculate the difference between the position vector of each of the surrounding associated geographic objects and the position vector of the selected geographic object as a position difference vector; After linear processing, each position difference vector is added to the corresponding position vector of the surrounding associated geographic object to obtain a geocoded vector corresponding to each surrounding associated geographic object; All geocoded vectors are fused through a multi-head attention mechanism to obtain the relationship data between the selected geographic object and the surrounding associated geographic objects.

[0017] Optionally, the relationship data between the area where the selected geographical object is located and the neighboring areas are obtained in the following manner: Obtaining characteristic data of other geographical objects and time series data of ownership transfer in the area where the selected geographical object is located as the selected area data vector; Determine the neighborhood area; Obtaining geographic object feature data and ownership transfer time series data of the neighboring area as a neighboring area data vector; Relationship data between the selected target region and the neighbor region is obtained according to the selected region data vector and the neighbor region data vector.

[0018] Optionally, the step of obtaining characteristic data of other geographical objects and time series data of ownership transfer in the area where the selected geographical object is located as the selected area data vector includes: Classify the characteristic data of other geographical objects in the area where the selected geographical object is located and the time series data of ownership transfer into text data and time series data, and for each geographical object in the other geographical objects, concatenate the text data and the time series data as a sub-vector, and combine the sub-vectors of all other geographical objects as the selected area data vector; The step of obtaining the geographic object feature data and ownership transfer time series data of the neighboring area as a neighboring area data vector includes: The characteristic data of the geographical objects in the neighboring area and the time series data of the ownership transfer are classified into text data and time series data. For each geographical object in the neighboring area, the text data and the time series data are spliced ​​as a sub-vector, and the sub-vectors of all geographical objects are combined as the data vector of the neighboring area.

[0019] Optionally, determining the neighbor area includes: Among the text data and time series data classified from the characteristic data of other geographical objects and the time series data of ownership transfer in the area where the selected geographical object is located, as well as the text data and time series data classified from the characteristic data of geographical objects and the time series data of ownership transfer in the neighboring area, the text data and the time series data are encoded using a factorization machine encoder and a temporal convolutional network encoder respectively to generate data vectors for the selected area and the neighboring area.

[0020] The embodiment of the present application also provides a geographic object data processing device, including: a target geographic object determination unit to be evaluated, used to determine the target geographic object to be evaluated; an attribute data acquisition unit, used to acquire attribute data of the target geographic object, the attribute data including: characteristic data of the target geographic object, characteristic data of the region where the target geographic object is located, time series data of object ownership transfer in the region where the target geographic object is located, and other geographic object data in the relationship between the target geographic object position and the region where the target geographic object is located; an evaluation unit, used to input the attribute data into a pre-trained geographic object value evaluation model to obtain a value evaluation result for the target geographic object; wherein the geographic object value evaluation model is trained by the following method: obtaining training sample data, the training sample data including attribute data of multiple geographic objects, the attribute data of each geographic object including: characteristic data of the geographic object, characteristic data of the region where the geographic object is located, time series data of object ownership transfer in the region where the geographic object is located, and other geographic object data in the relationship between the geographic object position and the region where the geographic object is located The invention discloses a method for extracting geographic object data from a geographic object; classifying the training sample data into numerical data, text data, time series data and geographic information data; inputting the training sample data into an initial model including an encoding module and a decoding module for iterative training, wherein the training process is as follows: inputting the attribute data of each geographic object into the corresponding encoding module according to the above classification category for encoding to generate a data vector; inputting the data vectors output by each encoding module into the sub-decoders corresponding to each encoding module, and merging the data vectors output by each encoding module and inputting them into the global decoder; decoding by each sub-decoder to obtain evaluation results, calculating each residual between each evaluation result and the target result, and feeding back each residual to the corresponding encoder to adjust the parameters; decoding by the global decoder to obtain evaluation results, calculating the total residual between the evaluation result and the target result, and feeding back the total residual to each encoder to adjust the parameters; iteratively executing the encoding and decoding process until the total residual meets the set value.

[0021] The embodiment of the present application also provides a training device for a geographic object value assessment model, including: a training sample data acquisition unit, used to acquire training sample data, the training sample data includes attribute data of multiple geographic objects, and the attribute data of each geographic object includes: geographic object feature data, feature data of the region where the geographic object is located, time series data of object ownership transfer in the region where the geographic object is located, and other geographic object data in the relationship between the geographic object position and the region where it is located; a classification unit, used to classify the training sample data into numerical data, text data, time series data and geographic information data; an iterative training unit, used to input the training sample data into an initial model including an encoding module and a decoding module for iterative training, and the training process is as follows :The attribute data of each geographic object is input into the corresponding encoding module according to the above classification category for encoding to generate data vectors; the data vectors output by each encoding module are input into the sub-decoders corresponding to each encoding module, and the data vectors output by each encoding module are fused and input into the global decoder; the evaluation results are obtained by decoding through each sub-decoder, and the residuals between each evaluation result and the target result are calculated, and the residuals are fed back to the corresponding encoders to adjust the parameters; the evaluation results are obtained by decoding through the global decoder, and the total residuals between the evaluation results and the target result are calculated, and the total residuals are fed back to each encoder to adjust the parameters; the encoding and decoding process is iteratively performed until the total residual meets the set value.

[0022] The present application also provides an electronic device, which includes a processor and a memory; a computer program is stored in the memory, and the processor executes the above method after running the computer program.

[0023] The present application also provides a computer storage medium, wherein the computer storage medium stores a computer program, and after the computer program is run by a processor, the above method is executed.

[0024] Compared with the prior art, the embodiments of the present application have the following advantages: The embodiment of the present application provides a method for processing geographic object data, including: determining a target geographic object to be evaluated; obtaining attribute data of the target geographic object, wherein the attribute data includes: characteristic data of the target geographic object, characteristic data of the region where the target geographic object is located, time series data of object ownership transfer in the region where the target geographic object is located, and other geographic object data in the relationship between the target geographic object position and the region where the target geographic object is located; inputting the attribute data into a pre-trained geographic object value evaluation model to obtain a value evaluation result for the target geographic object; wherein the geographic object value evaluation model is trained using the following method: obtaining training sample data, wherein the training sample data includes attribute data of multiple geographic objects, wherein the attribute data of each geographic object includes: characteristic data of the geographic object, characteristic data of the region where the geographic object is located, time series data of object ownership transfer in the region where the geographic object is located, and other geographic object data in the relationship between the geographic object position and the region where the geographic object is located; dividing the training sample data into The classes are numerical data, text data, time series data and geographic information data; the training sample data is input into an initial model including an encoding module and a decoding module for iterative training, and the training process is as follows: the attribute data of each geographic object is input into the corresponding encoding module according to the above classification category for encoding to generate a data vector; the data vectors output by each encoding module are input into the sub-decoders corresponding to each encoding module respectively, and the data vectors output by each encoding module are fused and input into the global decoder; the evaluation results are obtained by decoding through each sub-decoder respectively, and the residuals between each evaluation result and the target result are calculated, and the residuals are fed back to the corresponding encoders to adjust the parameters; the evaluation results are obtained by decoding through the global decoder, and the total residuals between the evaluation result and the target result are calculated, and the total residuals are fed back to the encoders to adjust the parameters; the encoding and decoding processes are iteratively performed until the total residual meets the set value.

[0025] The geographic object data processing method described in the embodiment of the present application comprehensively considers multiple attribute data such as the characteristic data of the target geographic object, the characteristic data of the area where the target geographic object is located, the time series data of the object ownership transfer in the area where the target geographic object is located, and other geographic object data in the relationship between the target geographic object position and the area where it is located when obtaining the value assessment result of the target geographic object. By inputting the above attribute data into a pre-trained geographic object value assessment model, the value assessment result of the target geographic object is finally obtained. The geographic object data processing method improves the accuracy of the value assessment of the geographic object to be assessed, and the data required for the assessment calculation is ordered and definite, which reduces the consumption of computing resources and improves the efficiency of the assessment calculation and the stability of the calculation results.

[0026] The embodiment of the present application also provides a training method for a geographic object value assessment model, including: obtaining training sample data, the training sample data including attribute data of multiple geographic objects, the attribute data of each geographic object including: geographic object feature data, feature data of the region where the geographic object is located, time series data of object ownership transfer in the region where the geographic object is located, and other geographic object data in the relationship between the geographic object position and the region where it is located; classifying the training sample data into numerical data, text data, time series data, and geographic information data; inputting the training sample data into an initial model including an encoding module and a decoding module for iterative training, and the training process is as follows: the attribute data of each geographic object is classified according to the above The classification categories are respectively input into the corresponding encoding modules for encoding to generate data vectors; the data vectors output by each encoding module are respectively input into the sub-decoders corresponding to each encoding module, and the data vectors output by each encoding module are fused and input into the global decoder; the evaluation results are respectively obtained by decoding through each sub-decoder, and the residuals between each evaluation result and the target result are calculated, and the residuals are fed back to the corresponding encoders to adjust the parameters; the evaluation results are obtained by decoding through the global decoder, and the total residuals between the evaluation results and the target result are calculated, and the total residuals are respectively fed back to each encoder to adjust the parameters; the encoding and decoding process is iteratively performed until the total residual meets the set value.

[0027] The geographic object value assessment model training method described in the embodiment of the present application can classify the training sample data into numerical data, text data, time series data and geographic information data, and input them into the corresponding encoding modules for encoding according to the above classification categories, generate data vectors, input the data vectors output by each encoding module into the sub-decoders corresponding to each encoding module, and input the data vectors output by each encoding module into the global decoder after fusion. On the one hand, the residuals between each evaluation result obtained after decoding by each sub-decoder and the target result can be fed back to the corresponding encoders, so as to adjust the parameters of each sub-encoder; on the other hand, the total residuals between the evaluation result obtained after decoding by the global decoder and the target result can be fed back to each encoder, so as to adjust the parameters of each sub-encoder, through multiple encoding and decoding iterative training processes, until the total residual meets the set value, the geographic object value assessment model training method improves the accuracy of the value assessment of the geographic object to be assessed, and the data required for the assessment calculation is orderly and determined, which reduces the consumption of computing resources, improves the efficiency of the assessment calculation and the stability of the calculation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 It is a schematic diagram of a first application scenario of a geographic object data processing method provided in the first embodiment of the present application.

[0029] Figure 2 It is a schematic diagram of a second application scenario of a geographic object data processing method provided in the first embodiment of the present application.

[0030] Figure 3 It is a schematic diagram of a third application scenario of a geographic object data processing method provided in the first embodiment of the present application.

[0031] Figure 4 It is a flowchart of a geographic object data processing method provided in the first embodiment of the present application.

[0032] Figure 5 It is a schematic diagram of the interface relationship of the Geo Transformer module provided in the first embodiment of the present application.

[0033] Figure 6 It is a flowchart of a method for training a geographic object value assessment model provided in the second embodiment of the present application.

[0034] Figure 7 It is a schematic diagram of a geographic object data processing device provided in the third embodiment of the present application.

[0035] Figure 8 It is a schematic diagram of a training device for a geographic object value assessment model provided in the fourth embodiment of the present application.

[0036] Fig. 9 This is a schematic diagram of an electronic device provided in the fifth embodiment of the present application. DETAILED DESCRIPTION

[0037] Many specific details are described in the following description to facilitate a full understanding of the present application. However, the present application can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the connotation of the present application, so the present application is not limited by the specific implementation disclosed below.

[0038] First, in order to enable those skilled in the art to better understand the present application, a specific application scenario of an embodiment of the present application is described in detail below based on a geographic object data processing method provided in the present application.

[0039] Before describing in detail the specific application scenarios of the embodiments of the present application, the background of the solution is first described.

[0040] With the development of the economy, transactions of geographic objects (such as real estate, garages, etc.) are becoming more and more frequent. Geographic object transactions mainly refer to the transfer of ownership of geographic objects. Taking real estate transactions as an example, real estate valuation is particularly important in the process of real estate transactions. It is a key technology for realizing asset digitization, and the accuracy of real estate valuation is of great significance to real estate transactions. It is also an important prerequisite for the digitization of asset transactions. In the increasingly complex real estate evaluation, accurately evaluating the value of real estate has become particularly challenging. When the ownership of geographic objects is transferred, such as house sales, house rentals, etc., it is usually necessary to evaluate the value of geographic objects. In the scenario of evaluating the value of geographic objects, one is to rely on manual calculation to evaluate the value of geographic objects. Manual calculation valuation is not only long and costly, but also has large evaluation differences due to the different experience and preferences of appraisers. Another is to use a geographic object evaluation model to evaluate the value of geographic objects and estimate the value of geographic objects in a digital form. Although it can improve the accuracy of online geographic object value evaluation compared with manual calculation, when using the existing real estate valuation model to estimate the value of real estate, taking the geographic object as real estate as an example, one is to use a rule-based model, also known as a repeat model, to predict the house price. This model simply uses the most recent transaction price in the same residential community, only the latest price, and does not fully consider various factors of real estate price evaluation. One is to use traditional machine learning methods to estimate the value of real estate. This method can implement linear regression, support vector regression and LGBM (Light Gradient Multivariate Model). BoostingMachine) model, which is a machine learning algorithm based on the gradient boosting framework, is specifically used to solve classification and regression problems. However, this method has high requirements for feature processing and cannot fully learn multimodal data. Another model is to evaluate the value of real estate based on a neural network model. The model includes a MugRep multi-task model based on a hierarchical community module and a ReGram model that integrates the relationship between target transactions and surrounding neighbors through an attention mechanism. This type of model does not integrate the geographic information module and the time series module well. Based on the above problems, when using the existing geographic object valuation model to evaluate the value of geographic objects, it is necessary to collect a large amount of disordered related data, and the collected data will vary from user to user. This results in the method not only occupying a large amount of computing resources, time-consuming and labor-intensive, but also there will be large differences between the valuation results of geographic objects and the actual transaction results, and between the results of use by different users, resulting in the situation where the geographic object transaction cycle is too long or the transaction cannot be completed. Therefore, how to improve the efficiency of geographic object value evaluation while reducing computing resources is an urgent problem to be solved.

[0041] Based on this, an embodiment of the present application provides a geographic object data processing method. When obtaining the value assessment result of the target geographic object, the characteristic data of the target geographic object, the characteristic data of the area where the target geographic object is located, the time series data of the object ownership transfer in the area where the target geographic object is located, and the target geographic object position and other geographic object data in the relationship between the area are comprehensively considered. By inputting the above-mentioned attribute data into a pre-trained geographic object value assessment model, the value assessment result of the target geographic object is finally obtained. The geographic object data processing method improves the accuracy of the value assessment of the geographic object to be assessed, and the data required for the assessment calculation is ordered and definite, which reduces the consumption of computing resources, improves the efficiency of the assessment calculation and the stability of the calculation results.

[0042] The following is an explanation of the technical terms involved in the embodiments of the present application: Multi-modal: Multi-modal refers to a mode that combines multiple types of data and information sources for processing and analysis. For example, in the fields of machine learning and natural language processing, multimodal learning may use text, image, and audio data at the same time. By combining different types of data, multimodal systems can improve the accuracy of understanding and prediction. FM: Factorization Machines (FM) is a machine learning algorithm based on matrix decomposition, which mainly solves the problem of feature interaction and parameter estimation under high-dimensional sparse data. FM improves model performance by introducing feature combinations and matrix decomposition of latent vectors. It is particularly suitable for dealing with data sparsity and feature interaction complexity in scenarios such as recommendation systems. FM can be used for classification and regression tasks. It is an extension of linear models and can efficiently capture the interaction between features. The core of FM is to represent feature interactions through the inner product of low-dimensional vectors, so that the number of its parameters grows linearly with the dimension, thereby reducing the computational complexity.

[0043] RoBERTa: RoBERTa (A Robustly Optimized BERT Pretraining Approach) is an improved version of the BERT (Bidirectional Encoder Representation from Transformer) model. RoBERTa aims to improve the performance of natural language processing tasks through pre-training on larger datasets, longer training time, and improved training strategies.

[0044] Temporal CNN: Temporal CNN is a convolutional neural network architecture for processing time series data. It captures features in the time dimension by extending the traditional convolution operation and can process input sequences of arbitrary length. Temporal CNN is often used in tasks such as video analysis and rhythm signal processing.

[0045] Geo Transformer: Geo Transformer is a model that combines geographic information and deep learning frameworks such as Transformer. It is often used for analysis and prediction of geospatial data and can effectively process geographic location information and other types of data, such as remote sensing images, map data, etc. Geo Transformer improves the analysis capabilities of spatial data by modeling spatial relationships and features.

[0046] Multi-head Attention: Multi-head Attention is an extension of the attention mechanism in the Transformer model, allowing the model to independently calculate attention on different "heads". Each head can learn different aspects of the input sequence, thereby extracting diverse information. Multi-head attention enables the model to better focus on different parts of the input and improves the ability to understand context.

[0047] The following combination Figure 1-Figure 3 The application scenarios of the geographic object data processing method provided in the embodiments of the present application are described in detail.

[0048] See also Figure 1 , Figure 1 This is a schematic diagram of the first application scenario of a geographic object data processing method provided in the first embodiment of the present application. In the embodiment of the present application, the geographic object is house A, and the geographic object value assessment result is the valuation result of house A as an example for description. Figure 1 The property valuation interface is shown in , in which the user can select the property to be evaluated, such as Property A, and then, in the property data item, complete the property data of Property A, for example, Property A has 22 floors, faces south, has an area of ​​150 square meters, and was built in 2000, etc. In the surrounding supporting entity item, complete the important entities around Property A, for example, the important entities near Property A are schools, subways, and parks, but there are no hospitals. In the nearby community property transaction data, complete the historical transaction prices of properties in the community near Property A, for example, the transaction price of Property B is 5.6 million, the transaction price of Property C is 5.3 million, the transaction price of Property D is 6 million, the transaction price of Property E is 6.2 million, etc. After the user clicks OK, he will enter the property valuation result interface, please refer to Figure 2 , Figure 2 This is a schematic diagram of a second application scenario of a geographic object data processing method provided in the first embodiment of the present application. For example, the final valuation result of property A is 5.8 million.

[0049] It should be noted that the above process of obtaining the valuation result of a property based on the property data, the supporting entity data around the property, and the property transaction data of the neighboring community is actually obtained by inputting the property data, the supporting entity data around the property, and the property transaction data of the neighboring community into the multimodal model. The multimodal model is the geographic object value assessment model. For the specific implementation process of the multimodal model, please refer to Figure 3 ,like Figure 3 As shown, it is a schematic diagram of the third application scenario of a geographic object data processing method provided by the first embodiment of the present application. The multimodal model includes a multimodal information input module, a multimodal encoding module and a multimodal fusion module. Among them, the multimodal information input module is used to input the multimodal information of the geographic object, and the multimodal input information can specifically be a variety of attribute data of the geographic object, including: geographic object feature data, feature data of the region where the geographic object is located, time series data of the object ownership transfer in the region where the geographic object is located, and other geographic object data in the relationship between the location of the geographic object and the region where the geographic object is located. The model classifies the above attribute data into numerical information, text information, time series information and geographic information. Take the geographic object as a house source as an example, wherein the numerical information can specifically be the relevant attribute information of the house source and the community where the house source is located. The house source information can specifically include the area, floor, hall, room and bathroom of the house source, and the information of the community where the house source is located can specifically include the greening rate, volume ratio and water, electricity and fuel costs of the community. The text information mainly includes the description information of the house source, and the description information of the house source can be information such as the house source introduction, community introduction, surrounding facilities, and traffic conditions. Time series information mainly refers to the historical transaction sequence information of the community, administrative district, and city where the property is located. Geographic information mainly includes supporting entity information and related information of neighboring communities and properties. Supporting entity information refers to surrounding geographical factors that have a significant impact on the value of the property, such as schools, subways, hospitals, or other important entities.

[0050] After the multimodal information input module obtains the above information, it transmits the above information to the multimodal encoding module, which is used to encode the above multimodal information and generate the encoded vector representation information. In the specific implementation, when encoding information of different modes, the corresponding encoders are also different. When encoding numerical information, FM (factor decomposition machine) is used as the encoder, and after encoding, the first vector is generated; when encoding text information, RoBERTa is used as the encoder, and after encoding, the second vector is generated; when encoding time series information, Temporal CNN (temporal convolutional network) is used as the encoder, and after encoding, the third vector is generated; when encoding geographic information, Geo Transformer is used as the encoder, and the geographic information includes supporting entity information and neighboring community information. After encoding the supporting entity information using the self-attention mechanism of Transformer, the fourth vector is generated, and after encoding the neighboring community information, the fifth vector is generated.

[0051] After the multimodal encoding module encodes the multimodal information, it transmits the generated vector to the multimodal fusion module, which is used to fuse the vectors. In the specific implementation, the Multi-headAttention mechanism is used to input the first vector, the second vector, the third vector, the fourth vector and the fifth vector into the preset multi-head self-attention mechanism model for fusion processing to generate the fusion vector of the geographic object. Afterwards, the fusion vector of the geographic object is input into the preset global decoder for decoding to obtain the estimated price of the geographic object. At the same time, the actual price y of the geographic object is obtained, and the estimated price of the geographic object is Subtract the actual price y of the geographic object to obtain the residual data, that is, the total residual. For example, the actual price y of the geographic object is 6 million, and the estimated price of the geographic object is is 5.8 million, then the residual data is 200,000. Then, the residual data is transmitted to the global decoder D, and at the same time, the residual data is also transmitted to the total encoder, and the parameters of the total encoder (FM encoder, RoBERTa encoder, Temporal CNN encoder, Temporal CNN encoder) are adjusted inversely based on the residual data. Then, the vector is output again as the updated vector, and the updated vector is input again into the preset multi-head self-attention mechanism model for fusion processing to generate the updated fusion vector of the geographic object. After that, the updated fusion vector of the geographic object is input into the preset global decoder for decoding, and the estimated price of the geographic object can be obtained again, until the difference between the estimated price data finally output by the encoder after adjusting the parameters and the actual price data is close to 0.

[0052] Of course, in the specific implementation, while the data vectors output by each encoder are fused and input into the global decoder, the data vectors output by each encoder are also input into the sub-decoders corresponding to each encoder, that is, the first vector is input into the first decoder to obtain the first estimated price, the second vector is input into the second decoder to obtain the second estimated price, the third vector is input into the third decoder to obtain the third estimated price, the fourth vector is input into the fourth decoder to obtain the fourth estimated price, the fifth vector is input into the fifth decoder to obtain the fifth estimated price, and the first estimated price, the second estimated price, the third estimated price, the fourth estimated price, and the fifth estimated price are respectively subtracted from the actual price to obtain each residual data, and each residual data is respectively fed back to the corresponding encoder to adjust the parameters of each encoder, that is, each residual data is reversely transmitted to the sub-decoder under each mode, and the sub-decoder is then transmitted to the corresponding sub-encoder, and the encoder parameters under each of the above modes are adjusted according to each residual data, that is, the FM encoder, RoBERTa encoder, Temporal CNN encoder, Temporal The parameters of each encoder in the CNN encoder are improved and optimized to make the final estimated price closer to the actual price. It should be noted that when training the geographic object value assessment model, batches of geographic object sample data are input into the model for training. Correspondingly, batches of residual data are obtained. In order to prevent the randomness of a single sample data, the final residual data is the average or median of the batch residual data.

[0053] The above is the full process analysis of the geographic object data processing method described in the embodiment of the present application. The method adopts a multi-modal approach and combines the multimodal geographic data of the geographic object, including the numerical information of the target geographic object, text information, historical transaction sequence information of the area where the geographic object is located, supporting entity information around the geographic object, and related information of neighboring communities and houses. Through a multi-level, geographic information-driven multimodal method, it is possible to better handle complex spatial information and diverse perspectives, enhance the model's ability to understand and express different scenarios, thereby improving the overall prediction ability of the model and making the final prediction value of the geographic object more accurate.

[0054] The present application is described in detail below through multiple embodiments and drawings. First embodiment

[0055] The first embodiment of the present application provides a method for processing geographic object data. Figure 4 The geographic object data processing method is described in detail.

[0056] Step S401: Determine the target geographic object to be evaluated.

[0057] This step is used to determine the geographic objects that need to be evaluated as the target geographic objects to be evaluated.

[0058] Step S402: Acquire the attribute data of the target geographic object, the attribute data including: characteristic data of the target geographic object, characteristic data of the region where the target geographic object is located, time series data of object ownership transfer in the region where the target geographic object is located, and other geographic object data in the relationship between the target geographic object location and the region where it is located.

[0059] This step is used to further obtain the attribute data of the target geographic object after determining the target geographic object to be evaluated. The attribute data of the target geographic object includes: characteristic data of the target geographic object, characteristic data of the region where the target geographic object is located, time series data of the transfer of ownership of the target geographic object in the region where the target geographic object is located, and other geographic object data in the relationship between the location of the target geographic object and the region where it is located.

[0060] Step S403: Input the attribute data into a pre-trained geographic object value assessment model to obtain a value assessment result for the target geographic object; wherein the geographic object value assessment model is trained by the following method: obtain training sample data, the training sample data includes attribute data of multiple geographic objects, and the attribute data of each geographic object includes: geographic object feature data, feature data of the region where the geographic object is located, time series data of object ownership transfer in the region where the geographic object is located, and other geographic object data in the relationship between the geographic object position and the region where it is located; classify the training sample data into numerical data, text data, time series data, and geographic information data; input the training sample data into an initial model including an encoding module and a decoding module for iterative training; The training process is as follows: the attribute data of each geographic object is input into the corresponding encoding module according to the above classification category for encoding to generate data vectors; the data vectors output by each encoding module are input into the sub-decoders corresponding to each encoding module, and the data vectors output by each encoding module are fused and input into the global decoder; the evaluation results are obtained by decoding through each sub-decoder, and the residuals between each evaluation result and the target result are calculated, and the residuals are fed back to the corresponding encoders to adjust the parameters; the evaluation results are obtained by decoding through the global decoder, and the total residuals between the evaluation results and the target result are calculated, and the total residuals are fed back to each encoder to adjust the parameters; the encoding and decoding process is iteratively performed until the total residual meets the set value.

[0061] This step is used to input the attribute data of the target geographic object into the pre-trained geographic object value assessment model, and finally obtain the value assessment result for the target geographic object.

[0062] In the embodiment of the present application, the value of the target geographic object is mainly evaluated by using a geographic object value assessment model. The geographic object value assessment model is obtained through training. During training, first, training sample data is obtained, and the training sample data includes attribute data of multiple geographic objects. The attribute data of each geographic object includes: geographic object feature data, feature data of the area where the geographic object is located, time series data of object ownership transfer in the area where the geographic object is located, and other geographic object data in the relationship between the geographic object location and the area where the geographic object is located; then, the training sample data is classified into numerical data, text data, time series data and geographic information data; finally, the training sample data is input into an initial model including an encoding module and a decoding module for iterative training.

[0063] During specific implementation, the characteristic data of the geographic object may be the numerical information of the geographic object or the textual information of the geographic object. In the embodiment of the present application, the description is made by taking the geographic object as a housing source as an example, and the geographic object may be various types of housing sources, such as commercial housing, residences, apartments, villas, houses, garages, warehouses, etc. The numerical information may specifically be the relevant attribute information of the housing source and the community where the housing source is located. The housing source information may specifically include the construction year, area, floor, hall, room, bathroom, orientation and other information of the housing source. The information of the community where the housing source is located may specifically include the greening rate of the community, the plot ratio (PlotRatio / Floor Area Ratio / Volume Fraction, which refers to the ratio of the total ground building area to the net land area of ​​a community, also known as the gross density of the building area) and water, electricity and fuel costs and other information. The textual information mainly includes the description information of the housing source, and the description information of the housing source may be information such as the introduction of the housing source, the introduction of the community, the surrounding facilities, and the traffic conditions.

[0064] It should be noted that the building year, area, floor and room type of the house can directly reflect the basic situation of the house and is an important factor in evaluating the value of the house. Whether the supporting facilities around the community are complete, whether the transportation around the community is convenient, and whether the community is in a prosperous commercial area are all important factors affecting the valuation results of the house.

[0065] In the embodiment of the present application, the characteristic data of the area where the geographical object is located refers to the characteristic data of the area where the geographical object is located. For example, it can be the characteristic data of the community where the house is located. The characteristic data of the area where the geographical object is located can also include quantifiable numerical information and descriptive text information. The specific content of the numerical information and text information is similar to the above description. Please refer to the above description for details, and will not be described in detail here. The area where the geographical object is located can refer to the community where the house is located, or it can refer to the district information to which the house belongs, or it can refer to the city information to which the house belongs. It should be noted that the area where the house is located has an important impact on the valuation results of the house, for example, whether the city level where the house is located is a first-tier city, a second-tier city or a third-tier city, whether the supporting facilities around the location of the house include subways, parks, schools, hospitals, etc., whether the transportation around the house is convenient, and whether the house is in a prosperous commercial area.

[0066] In the embodiment of the present application, the time series data of the transfer of ownership of the object in the area where the geographic object is located can specifically be the historical transaction price data of other houses in the community where the house is located. Among them, the transfer of ownership of the geographic object can specifically be the transfer of house ownership, that is, house sale, or the transfer of house use rights, that is, house leasing. The time series data mainly refers to the historical transaction sequence data of the community, zoning information, and city where the house is located.

[0067] In an embodiment of the present application, in the training of the geographic object value assessment model, other geographic object data in the relationship between the geographic object location and the region where it is located include: geographic object relationship data and region relationship data. The geographic object relationship data is used to represent the relationship data between a selected geographic object and surrounding associated geographic objects, and the surrounding associated geographic objects are geographic objects whose distance from the selected geographic object is within a first preset distance range and have an associated influence on the users in the selected geographic object; the region relationship data is used to represent the relationship data between the region where a selected geographic object is located and the neighboring region, and the neighboring region is an area whose distance from the region where the selected geographic object is located is within a second preset distance range. It should be noted that the relationship data between the selected geographic object and the surrounding associated geographic objects can refer to the data spliced ​​together between the housing source and the surrounding subway stations, hospitals, etc., or it can be the convenience data of the supporting entities to the housing source obtained through calculation.

[0068] In the embodiment of the present application, the surrounding associated geographic objects are geographic objects whose distance from the selected geographic object is within a first preset distance range and have an associated influence on the user in the selected geographic object. The surrounding associated geographic objects can specifically be supporting entity information, and the supporting entity information refers to surrounding geographical factors that have a significant impact on the value of real estate, such as schools, subways, hospitals, parks, supermarkets, shopping malls or other important entities. In specific implementation, an entity list can be pre-set to determine the influence of the entities in the entity list on the housing source, or the housing source data and entity data can be input into a preset model, and the model is used to predict the influence of the entity on the housing source, and then determine whether the entity is an associated entity of the housing source.

[0069] It should be noted that the surrounding geographical objects refer to the surrounding geographical objects of the house. The surrounding geographical objects of the house can be entities outside the community where the house is located, or entities inside the community where the house is located, such as convenience stores and schools inside the community, convenience stores and fast food restaurants in commercial buildings, etc. The surroundings not only refer to the outside of the community or the outside of the commercial building, but also refer to the inside of the community or the inside of the commercial building.

[0070] As an embodiment, the relationship data between the selected geographical object and the surrounding associated geographical objects can be obtained in the following manner: obtain the vector representation data of the selected geographical object as the selected geographical object vector; determine the surrounding associated geographical objects; obtain the vector representation data of the surrounding associated geographical objects as the surrounding associated geographical object vector; splice the selected geographical object vector with the surrounding associated geographical object vector to obtain a geographical object splicing vector as the relationship data between the selected geographical object and the surrounding associated geographical objects.

[0071] As another embodiment, the relationship data between the selected geographical object and the surrounding associated geographical objects can also be obtained in the following manner: obtaining vector representation data of the selected geographical object as the selected geographical object vector; determining the surrounding associated geographical objects; obtaining vector representation data of the surrounding associated geographical objects as the surrounding associated geographical object vector; obtaining a model based on the selected geographical object vector, the surrounding associated geographical object vector and the geographical object relationship data, and obtaining the relationship data between the selected geographical object and the surrounding associated geographical objects, wherein the relationship data between the selected geographical object and the surrounding associated geographical objects obtained according to the geographical object relationship data obtaining model is used to represent the influence of the surrounding associated geographical objects on the users in the selected geographical object.

[0072] It should be noted that when obtaining the relationship data between the selected geographic object and the surrounding associated geographic objects, one way is to first obtain the vector representation data of the selected geographic object and the vector representation data of the surrounding associated geographic objects, and then splice the vector representation data of the selected geographic object and the vector representation data of the surrounding associated geographic objects, and use the spliced ​​vector as the relationship data between the selected geographic object and the surrounding associated geographic objects; another way is to first pre-train the geographic object relationship data to obtain a model, and the relationship data between the selected geographic object and the surrounding associated geographic objects obtained by the model is used to represent the influence of the surrounding associated geographic object on the users in the selected geographic object, such as the influence of surrounding schools on users in the housing source, the influence of surrounding hospitals on users in the housing source, etc. Then, the vector representation data of the selected geographic object and the vector representation data of the surrounding associated geographic objects are obtained. Finally, according to the selected geographic object vector, the surrounding associated geographic object vector and the geographic object relationship data acquisition model, the relationship data between the selected geographic object and the surrounding associated geographic objects are obtained.

[0073] As an embodiment, when obtaining the relationship data between the selected geographic object and the surrounding associated geographic objects, the portrait data of the demander of the selected geographic object can also be obtained; the characteristic data of the surrounding associated geographic objects can be obtained; a model is obtained according to the portrait data of the demander of the selected geographic object, the characteristic data of the surrounding associated geographic objects and the geographic object relationship data, and the relationship data between the selected geographic object and the surrounding associated geographic objects is obtained, wherein the relationship data between the selected geographic object and the surrounding associated geographic objects obtained according to the geographic object relationship data acquisition model is used to indicate the influence of the surrounding associated geographic objects on the users in the selected geographic object. It should be noted that the portrait data of the demander of the selected geographic object is used to indicate the degree of demand of the selected target object demander for the surrounding associated geographic objects. For example, for users who need to go to school, whether there is a school around the community is more important; for users who need to take care of the elderly, whether there is a hospital around the community is more important. For another example, in the housing auction scenario, the portrait data of the buyer user is obtained with the authorization of the buyer user. According to the portrait data of the buyer user, the behavior habits of the buyer user can be obtained. If the buyer user usually relies on the subway to travel more, the buyer user is more inclined to the housing source with a subway nearby.

[0074] As an embodiment, the relationship data between the selected geographic object and the surrounding associated geographic objects are obtained in the following manner: obtaining the position data of the selected geographic object as the selected geographic object position vector; determining the surrounding associated geographic objects and their position data as the surrounding associated geographic object data vector; determining the relationship data between the selected geographic object and the surrounding associated geographic objects according to the selected geographic object position vector and the surrounding associated geographic object data vector. Wherein, the position data includes longitude and latitude. When determining the relationship data between the selected geographic object and the surrounding associated geographic objects according to the selected geographic object position vector and the surrounding associated geographic object data vector, the selected geographic object position vector and the surrounding associated geographic object data vector may be specifically concatenated.

[0075] As an embodiment, the relationship data between the selected geographic object and the surrounding associated geographic objects are obtained in the following manner: obtaining the position data of the selected geographic object as the selected geographic object position vector; determining the surrounding associated geographic objects of the selected geographic object; for each of the surrounding associated geographic objects, determining its position data as each surrounding associated geographic object position vector; calculating the difference between the position vector of each surrounding associated geographic object and the position vector of the selected geographic object as a position difference vector; linearly processing each position difference vector, adding it to the corresponding surrounding associated geographic object position vector to obtain the geocoded vector corresponding to each surrounding associated geographic object; fusing all geocoded vectors through a multi-head attention mechanism to obtain the relationship data between the selected geographic object and the surrounding associated geographic objects. Wherein, the position data includes longitude and latitude; the position difference vector includes longitude difference and latitude difference.

[0076] For specific implementation, see Figure 5 , Figure 5 It is a schematic diagram of the interface relationship of the Geo Transformer module provided in the first embodiment of the present application. Figure 5 In the table, lat and lon represent latitude and longitude respectively, p j refers to the jth property, i.e. the selected geographical object, o i refers to the i-th entity around the property, i.e., the surrounding associated geographic object, W g refers to the learnable parameters. Indicates parameter W g Multiply it with the vector composed of longitude and latitude, and q1 is the result. Add q and o respectively to get the geocoded vector. Finally, multiple q are fused through multi-head attention to get the relationship data between the selected geographic object and the surrounding related geographic objects.

[0077] In the embodiment of the present application, the calculation of the difference between the position vector of each of the surrounding associated geographic objects and the position vector of the selected geographic object as the position difference vector includes: calculating the longitude difference and latitude difference between each of the surrounding associated geographic objects and the selected geographic object; normalizing the longitude difference and latitude difference, and mapping them to a coded value of a set dimension to obtain the position difference vector. It should be noted that after the longitude difference and latitude difference are normalized, the data obtained is more suitable for model processing, making the data processing speed of the model faster, thereby improving the data processing efficiency of the model.

[0078] The normalizing the longitude difference and the latitude difference and mapping them to a code of a set dimension to obtain the position difference vector includes: normalizing the longitude difference and the latitude difference by taking the sine and cosine of the longitude difference and the latitude difference respectively; multiplying the longitude difference and the latitude difference by 10 respectively, and taking 2i / d as the base; model The dimension is expanded by an exponential factor, where i is the index of the dimension and d model is the set dimension; the longitude difference and latitude difference are multiplied by 10 respectively, and 2i / d model The data obtained after the exponential factorization are subjected to sine and cosine transformation respectively, and all the generated data are used as elements of the position difference vector to obtain the position difference vector.

[0079] It should be noted that the use of multi-head attention mechanism for fusion enables the model to consider not only the individual position of the geographic entity, but also the distance and direction of the geographic entity relative to a certain attribute. For the i-th entity, its relative position relative to the j-th attribute is calculated as follows:

[0080] The normalization method of the relative position is similar to the position encoding used for the absolute position. Specifically, the relative position may be represented by sinusoidal relative encoding. The formula of sinusoidal relative encoding is as follows:

[0081] in, and Respectively represent the latitude and longitude differences between two geographic entities. For example, suppose there are two geographic entities with coordinates (lat1, lon1) = (30, 120) and (lat2, lon2) = (35, 125), then the relative positions are: =35-30=5, =125-120=5, for 8-dimensional encoding (d model =8), the relative position encoding is:

[0082] It should be noted that the above example only shows the relative position encoding result of the latitude difference. The relative position encoding of the longitude difference is similar to the latitude difference and is not shown here. By expanding the dimensions of the longitude difference and the latitude difference, more information can be extracted, thereby improving the estimation accuracy of the model.

[0083] In an embodiment of the present application, the geographic object value assessment model inputs the training sample data into an initial model including a coding module and a decoding module for iterative training, and the training process is as follows: the attribute data of each geographic object is input into the corresponding coding module according to the above classification category for encoding to generate a data vector; the data vectors output by each coding module are input into the sub-decoders corresponding to each coding module, and the data vectors output by each coding module are fused and input into the global decoder; the evaluation results are obtained by decoding through each sub-decoder, and the residuals between each evaluation result and the target result are calculated, and the residuals are fed back to the corresponding encoders to adjust the parameters; the evaluation results are obtained by decoding through the global decoder, and the total residuals between the evaluation result and the target result are calculated, and the total residuals are fed back to the encoders to adjust the parameters; the encoding and decoding processes are iteratively performed until the total residual meets the set value.

[0084] In specific implementation, when encoding the attribute data of geographic objects to generate data vectors, a preset encoder can be used to encode the attribute data of geographic objects. For numerical data in the attribute data of geographic objects, an FM encoder can be used. The FM encoder is widely used to process tabular data. The use of the FM encoder can capture the interactive relationship between numerical attributes; for text data in the attribute data of geographic objects, RoBERTa can be used as an encoder. The RoBERTa encoder is an advanced transformer-based pre-trained language model to encode text inputs. RoBERTa's deep contextual understanding ability enables it to effectively capture subtle meanings and relationships, thereby converting text descriptions into rich speech codes.

[0085] In the embodiment of the present application, the relationship data between the selected geographical object and the surrounding associated geographical objects are obtained in the following manner: obtaining the position data of the selected geographical object as the selected geographical object position vector; determining the surrounding associated geographical object and its position data as the surrounding associated geographical object data vector; determining the relationship data between the selected geographical object and the surrounding associated geographical object according to the selected geographical object position vector and the surrounding associated geographical object data vector. In specific implementation, the determination of the surrounding associated geographical object and its position data includes: determining the surrounding associated geographical object and its position data according to the selected geographical object position data, the positional relationship between the surrounding geographical object and the selected geographical object, and the degree of influence of the surrounding geographical object on the value assessment of the selected geographical object. It should be noted that the surrounding associated geographical object can specifically be a supporting geographical entity such as a school, a hospital, a subway, a park, etc. The position data of the surrounding associated geographical object includes the numerical data and text data of the supporting geographical entity, and also includes the geographical coordinate data of the surrounding associated geographical object, that is, the longitude coordinate data and latitude coordinate data of the geographical entity. When determining the surrounding associated geographic objects and their location data, full consideration is given to the location data of the selected geographic object, the location relationship between the surrounding geographic objects and the selected geographic object, and the degree of influence of the surrounding geographic objects on the value assessment of the selected geographic object, thereby further improving the accuracy of the model prediction results.

[0086] The determining of the surrounding associated geographical objects includes: obtaining a preset list of surrounding associated geographical objects corresponding to the target geographical object, and determining the surrounding associated geographical objects from the list of surrounding associated geographical objects corresponding to the target geographical object; or obtaining a preset list of surrounding associated geographical objects corresponding to the target area, and determining the surrounding associated geographical objects from the list of surrounding associated geographical objects corresponding to the target area; or obtaining a preset list of surrounding associated geographical objects corresponding to the building where the target geographical object is located, and determining the surrounding associated geographical objects from the list of surrounding associated geographical objects corresponding to the building where the target geographical object is located; or obtaining a preset list of all geographical objects; filtering out a list of geographical objects whose distances to the target geographical object are within the first preset distance range from the list of all geographical objects as a list of candidate surrounding associated geographical objects; obtaining feature data of each candidate surrounding associated geographical object in the list of candidate surrounding associated geographical objects; and filtering out the surrounding associated geographical objects from the list of candidate surrounding associated geographical objects according to the feature data of the target geographical object, the feature data of each candidate surrounding associated geographical object, and a surrounding associated geographical object screening model.

[0087] It should be noted that there can be multiple surrounding associated geographical objects. When determining the surrounding associated geographical objects, the surrounding associated geographical objects can be determined specifically from the surrounding associated geographical object list corresponding to the preset target geographical object. For example, the surrounding associated geographical object list corresponding to property A shows that the associated geographical objects of property A are schools, hospitals, and shopping malls; or, the surrounding associated geographical objects can be determined from the surrounding associated geographical object list corresponding to the preset target area. For example, the geographical objects associated with the area where property A is located are subways, parks, and schools; or, the surrounding associated geographical objects can be determined from the surrounding associated geographical object list corresponding to the preset building where the target geographical object is located. For example, the geographical object associated with the community where property A is located is a school. ; Alternatively, the candidate surrounding associated geographic objects can be screened out from the full geographic object list, and then the surrounding associated geographic objects can be screened out from the candidate surrounding associated geographic object list based on the target geographic object feature data, the feature data of each candidate surrounding associated geographic object and the surrounding associated geographic object screening model. For example, the full geographic object list shows schools, hospitals, shopping malls, parks, subways, supermarkets, etc., and the candidate surrounding associated geographic objects within the first preset distance range of house A include schools, hospitals, and parks. Then, based on the feature data of house A, the feature data of the school, the feature data of the hospital, the feature data of the park and the surrounding associated geographic object screening model, the surrounding associated geographic objects of house A are finally screened out as schools and hospitals.

[0088] As an embodiment, the relationship data between the area where the selected geographical object is located and the neighboring area is obtained in the following manner: obtaining vector representation data of the area where the selected geographical object is located as the vector of the area where the selected geographical object is located; determining the neighboring area; obtaining vector representation data of the neighboring area as the neighboring area vector; splicing the vector of the area where the selected geographical object is located with the neighboring area vector to obtain a region splicing vector as the relationship data between the area where the selected geographical object is located and the neighboring area.

[0089] As another embodiment, the relationship data between the area where the selected geographical object is located and the neighboring areas are obtained in the following manner: obtaining vector representation data of the area where the selected geographical object is located as the vector of the area where the selected geographical object is located; determining the neighboring areas; obtaining vector representation data of the neighboring areas as the neighboring areas vector; obtaining a model based on the vector of the area where the selected geographical object is located, the neighboring areas vector and the area relationship data, and obtaining the relationship data between the area where the selected geographical object is located and the neighboring areas, wherein the relationship data between the area where the selected geographical object is located and the neighboring areas obtained by the model based on the area relationship data is used to represent the influence of the neighboring areas on the area where the selected geographical object is located.

[0090] It should be noted that when obtaining the relationship data between the area where the selected geographical object is located and the neighboring area, one way is to splice the characteristic data of the community and the characteristic data and historical transaction data of its neighboring communities together as relationship data; another way is to input the characteristic data of the community and the characteristic data and historical transaction data of its neighboring communities into a preset model to obtain the relationship data of the community and each surrounding community. The transaction situation of the neighboring community with large relationship data has a greater impact on the price of housing in the community.

[0091] In a specific implementation, the method of obtaining the vector representation data of the area where the selected geographical object is located includes: encoding the characteristic data of the area where the selected geographical object is located to obtain the characteristic vector of the area where the selected geographical object is located; encoding the time series data of object ownership transfer in the area where the selected geographical object is located to obtain the time series vector of object ownership transfer in the area where the selected geographical object is located; and splicing the characteristic vector of the area where the selected geographical object is located and the time series vector of object ownership transfer in the area where the selected geographical object is located to obtain the vector representation data of the area where the selected geographical object is located.

[0092] The method of obtaining the vector representation data of the neighboring area includes: obtaining the characteristic data of the neighboring area as the neighboring area characteristic data; encoding the neighboring area characteristic data to obtain the characteristic vector of the neighboring area; obtaining the time series data of geographical object ownership transfer of the neighboring area as the time series data of geographical object ownership transfer of the neighboring area; encoding the time series data of geographical object ownership transfer of the neighboring area to obtain the time series vector of geographical object ownership transfer of the neighboring area; and concatenating the characteristic vector of the neighboring area and the time series vector of geographical object ownership transfer of the neighboring area to obtain the vector representation data of the neighboring area.

[0093] It should be noted that when encoding the characteristic data of the area where the selected geographic object is located and the characteristic data of the neighboring area, the FM encoder can be used for numerical data, and the RoBERTa encoder can be used for text data; when encoding the time series data of the transfer of ownership of objects in the target area and the time series data of the transfer of ownership of geographic objects in the neighboring area, Temporal CNN (temporal convolutional network) can be used. This network is very effective in modeling sequence data and extracting time dependencies. Temporal CNN's ability to handle long-range dependencies enables it to recognize and utilize patterns in each transaction sequence.

[0094] As an embodiment, the relationship data between the area where the selected geographical object is located and the neighboring area is obtained in the following manner: obtaining the feature data of other geographical objects in the area where the selected geographical object is located and the time series data of ownership transfer as the selected area data vector; determining the neighboring area; obtaining the feature data of geographical objects in the neighboring area and the time series data of ownership transfer The shift time series data is used as a neighbor area data vector; and the relationship data between the area where the selected geographic object is located and the neighbor area is obtained according to the selected area data vector and the neighbor area data vector.

[0095] The step of obtaining the characteristic data of other geographical objects and the time series data of ownership transfer in the area where the selected geographical object is located as the selected area data vector comprises: classifying the characteristic data of other geographical objects and the time series data of ownership transfer in the area where the selected geographical object is located into text data and time series data, splicing the text data and the time series data for each geographical object in the other geographical objects as a sub-vector, and combining the sub-vectors of all other geographical objects as the selected area data vector; the step of obtaining the characteristic data of geographical objects and the time series data of ownership transfer in the neighboring area as the neighboring area data vector comprises: classifying the characteristic data of geographical objects and the time series data of ownership transfer in the neighboring area as text data and time series data, splicing the text data and the time series data for each geographical object in the neighboring area as a sub-vector, and combining the sub-vectors of all geographical objects as the data vector of the neighboring area.

[0096] It should be noted that when obtaining the characteristic data of other geographical objects and the time series data of ownership transfer in the area where the selected geographical object is located as the selected area data vector, the characteristic data of other geographical objects and the time series data of ownership transfer in the area where the selected geographical object is located are first classified into text data and time series data, and for each geographical object among the other geographical objects, the text data and the time series data are spliced ​​as a sub-vector; then, the sub-vectors of all other geographical objects are combined as the selected area data vector.

[0097] When obtaining the geographic object feature data and ownership transfer time series data of the neighboring area as the neighboring area data vector, similarly, first classify the geographic object feature data and ownership transfer time series data of the neighboring area into text data and time series data, and for each geographic object in the neighboring area geographic object, splice the text data and the time series data as a sub-vector; then, combine the sub-vectors of all geographic objects as the data vector of the neighboring area.

[0098] In a specific implementation, the determining of the neighbor area includes: obtaining a preset neighbor area list corresponding to the selected area, and determining the neighbor area from the corresponding neighbor area list; or, obtaining a preset full area list; and filtering out, from the full geographic object list, neighbor areas whose distance from the selected area is within the second preset distance range.

[0099] In the embodiment of the present application, the valuation result of the target geographic object is mainly used when the target geographic object has a transfer of ownership event. The valuation result of the target geographic object having a transfer of ownership event can specifically be the funds required to complete the sale of a house, or the funds required to complete the rental of a house. During model training, the training sample data is classified into numerical data, text data, time series data, and geographic information data; the training sample data is input into an initial model including an encoding module and a decoding module for iterative training, which can make the final value assessment result of the target geographic object more accurate.

[0100] In specific implementation, the attribute data of each geographic object is input into the corresponding encoding module according to the above classification category for encoding to generate a data vector. Specifically, it can be: respectively encode the geographic object feature data, the feature data of the area where the geographic object is located, the time series data of the object ownership transfer in the area where the geographic object is located, the geographic object relationship data and the regional relationship data to determine a plurality of encoded vector representation data; fuse the plurality of encoded vector representation data to determine the fused vector representation data as the geographic object fused vector data; based on the geographic object fused vector data, obtain the value assessment result of the geographic object.

[0101] The method of respectively encoding the geographic object feature data, the feature data of the region where the geographic object is located, the time series data of object ownership transfer in the region where the geographic object is located, the geographic object relationship data and the region relationship data to determine a plurality of encoded vector representation data includes: encoding the geographic object feature data to obtain feature vector representation data of the geographic object as the geographic object feature vector; encoding the feature data of the region where the geographic object is located to obtain feature vector representation data of the region where the geographic object is located as the feature vector of the region where the geographic object is located; encoding the time series data of object ownership transfer in the region where the geographic object is located to obtain the time series vector of object ownership transfer in the region where the geographic object is located; encoding the geographic object relationship data to obtain feature vector representation data of the geographic object relationship as the geographic object relationship feature vector; encoding the region relationship data to obtain feature vector representation data of the region relationship as the region relationship feature vector.

[0102] In specific implementation, when encoding the feature data of geographic objects and the feature data of the region where the geographic objects are located, different encoders can be selected according to the data type. When encoding the numerical information in the feature data of geographic objects and the feature data of the region where the geographic objects are located, FM (factor decomposition machine) can be used as the encoder; when encoding the text information in the feature data of geographic objects and the feature data of the region where the geographic objects are located, RoBERTa can be used as the encoder; when encoding the time series data of the transfer of ownership of the objects in the region where the geographic objects are located, TemporalCNN (temporal convolutional network) can be used as the encoder; when encoding the geographic object relationship data and the regional relationship data, Geo Transformer can be used as the encoder. Geo Transformer uses the self-attention mechanism of Transformer to effectively capture the interactions and relative importance between various geographic entities in the geographic object relationship data, and in order to integrate the latitude and longitude coordinates of the geographic entities and the houses to be evaluated, Geo Transformer can include relative spatial relationships. Please continue to refer to the interface relationship of the GeoTransformer module. Figure 5 , which will not be described in detail here. In addition, considering the relationship between the house and the neighboring communities, each community is characterized by numerical attributes and historical and transaction time series data. The numerical attributes are encoded using the FM encoder. The time series data is encoded using Temporal CNN, and the encoded features are spliced ​​together. Using longitude and latitude coordinates, Geo Transformer can be used to provide an assessment of how neighboring communities affect the value of the target house. This method can capture complex spatial and temporal dynamics and improve the accuracy of the target house value estimation results.

[0103] The above process is to respectively encode the feature data of the geographic object, the feature data of the area where the geographic object is located, the time series data of the object ownership transfer in the area where the geographic object is located, the geographic object relationship data and the area relationship data, and determine a plurality of encoded vector representation data. After determining the geographic object vector data, it is also necessary to fuse the plurality of encoded vector representation data. The following describes the process of fusing the plurality of encoded vector representation data to obtain the geographic object fusion vector and the process of obtaining the value assessment result of the geographic object based on the geographic object fusion vector.

[0104] The step of fusing the plurality of encoded vector representation data and determining the fused vector representation data as the fused vector data of the geographic object includes: fusing the feature vector of the geographic object, the feature vector of the region where the geographic object is located, the time series vector of the object ownership transfer in the region where the geographic object is located, the feature vector of the geographic object relationship, and the feature vector of the region relationship to determine the fused vector representation data as the fused vector data of the geographic object. The fused vector data of the geographic object is input into the global decoder, and the evaluation result is obtained by decoding the data through the global decoder, and the total residual between the evaluation result and the target result is calculated, and the total residual is fed back to each encoder to adjust the parameters; and the encoding and decoding process is iteratively performed until the total residual meets the set value.

[0105] In specific implementation, you can use the Multi-head Attention mechanism, please continue to refer to Figure 3 In order to effectively integrate the above multimodal data (E1, E2, E3, E4, E5), a fusion mechanism based on the residual teacher-student strategy can be adopted. For each modality Ei, a multi-head self-attention mechanism is used to calculate its contextual representation in the context of other modalities. . Mi represents the contextual enhanced representation of the modality. The multi-head self-attention mechanism enables the model to focus on diverse features in different modalities and effectively capture complex interactions between modalities. Once the contextual representations of each modality are obtained, these representations will be passed through a global decoder. The decoder adopts the Transformer decoder style, aggregates the contextual information from each modality, and outputs the final valuation prediction : . The prediction from the global decoder is used as the “teacher” prediction, and the residual between the teacher prediction and the actual label y is defined as: For example, the actual price y of the geographic object is 6 million, and the estimated price of the geographic object is The residual data is 5.8 million, and the residual data is 200,000. The residual data is transmitted to the global decoder, and the global decoder transmits it to the corresponding global encoder. The parameters of the global encoder are adjusted inversely based on the residual data until the difference between the estimated price data and the actual price data finally output by the encoder after adjusting the parameters is close to 0.

[0106] In the embodiment of the present application, the parameters of the encoder under each of the above modes can also be adjusted according to the residual data, so as to improve and optimize each sub-encoder so that the final estimated price is closer to the actual price. In specific implementation, the residual data can be transmitted to the FM encoder via the first decoder (decoder D1 corresponding to the FM encoder), and the FM encoder generates an updated first vector after re-encoding. The residual data is transmitted to the RoBERTa encoder via the second decoder (decoder D2 corresponding to the RoBERTa encoder), and the RoBERTa encoder generates an updated second vector after re-encoding. The residual data is transmitted to the Temporal CNN encoder via the third decoder (decoder D3 corresponding to the Temporal CNN encoder), and the Temporal CNN encoder generates an updated third vector after re-encoding. The residual data is transmitted to the Geo Transformer encoder via the fourth decoder (decoder D4 corresponding to the Geo Transformer encoder), and the Geo Transformer encoder generates an updated fourth vector after re-encoding. The residual data is transmitted to the Geo Transformer encoder via the fifth decoder (decoder D5 corresponding to the Geo Transformer encoder), and the Geo Transformer encoder generates an updated fifth vector after re-encoding. Then, the updated first vector, the updated second vector, the updated third vector, the updated fourth vector and the updated fifth vector are input into the multi-head self-attention model to generate an updated fusion vector, and the updated fusion vector is decoded by the decoder to obtain the estimated price data again, until the difference between the estimated price data finally output by the sub-encoder after adjusting the parameters and the actual price data is close to 0.

[0107] Of course, in the specific implementation, while the data vectors output by each encoder are fused and input into the global decoder, the data vectors output by each encoder are also input into the sub-decoders corresponding to each encoder, that is, the first vector is input into the first decoder to obtain the first estimated price, the second vector is input into the second decoder to obtain the second estimated price, the third vector is input into the third decoder to obtain the third estimated price, the fourth vector is input into the fourth decoder to obtain the fourth estimated price, the fifth vector is input into the fifth decoder to obtain the fifth estimated price, and the first estimated price, the second estimated price, the third estimated price, the fourth estimated price, and the fifth estimated price are respectively subtracted from the actual price to obtain each residual data, and each residual data is respectively fed back to the corresponding encoder to adjust the parameters of each encoder, that is, each residual data is reversely transmitted to the sub-decoder under each mode, and the sub-decoder is then transmitted to the corresponding sub-encoder, and the encoder parameters under each of the above modes are adjusted according to each residual data, that is, the FM encoder, RoBERTa encoder, Temporal CNN encoder, Temporal The parameters of each encoder in the CNN encoder are improved and optimized to make the final estimated price closer to the actual price. It should be noted that when training the geographic object value assessment model, batches of geographic object sample data are input into the model for training. Correspondingly, batches of residual data are obtained. In order to prevent the randomness of a single sample data, the final residual data is the average or median of the batch residual data.

[0108] In practice, the residual between the teacher’s prediction and the actual label y is defined as: , this residual guides the training of the “student” decoder for each individual modality. For each sub-modality decoder, the residual is back-propagated: . This method enables the sub-modal decoders to improve their respective accuracy by learning the optimized output of residual adaptation. The teacher-student enhances the model capability by iteratively utilizing the global prediction (teacher) to improve and optimize each sub-modal decoder (student). Based on the residual teacher-student framework, this modality fusion method not only improves the overall prediction capability of the model, but also provides a stronger learning signal for each modality, achieving two-way optimization from fusion to each modality level.

[0109] The embodiment of the present application provides a method for processing geographic object data, including: determining a target geographic object to be evaluated; obtaining attribute data of the target geographic object, wherein the attribute data includes: characteristic data of the target geographic object, characteristic data of the region where the target geographic object is located, time series data of object ownership transfer in the region where the target geographic object is located, and other geographic object data in the relationship between the target geographic object position and the region where the target geographic object is located; inputting the attribute data into a pre-trained geographic object value evaluation model to obtain a value evaluation result for the target geographic object; wherein the geographic object value evaluation model is trained using the following method: obtaining training sample data, wherein the training sample data includes attribute data of multiple geographic objects, wherein the attribute data of each geographic object includes: characteristic data of the geographic object, characteristic data of the region where the geographic object is located, time series data of object ownership transfer in the region where the geographic object is located, and other geographic object data in the relationship between the geographic object position and the region where the geographic object is located; dividing the training sample data into The classes are numerical data, text data, time series data and geographic information data; the training sample data is input into an initial model including an encoding module and a decoding module for iterative training, and the training process is as follows: the attribute data of each geographic object is input into the corresponding encoding module according to the above classification category for encoding to generate a data vector; the data vectors output by each encoding module are input into the sub-decoders corresponding to each encoding module respectively, and the data vectors output by each encoding module are fused and input into the global decoder; the evaluation results are obtained by decoding through each sub-decoder respectively, and the residuals between each evaluation result and the target result are calculated, and the residuals are fed back to the corresponding encoders to adjust the parameters; the evaluation results are obtained by decoding through the global decoder, and the total residuals between the evaluation result and the target result are calculated, and the total residuals are fed back to the encoders to adjust the parameters; the encoding and decoding processes are iteratively performed until the total residual meets the set value.

[0110] The geographic object data processing method described in the embodiment of the present application comprehensively considers multiple attribute data such as the characteristic data of the target geographic object, the characteristic data of the area where the target geographic object is located, the time series data of the object ownership transfer in the area where the target geographic object is located, and other geographic object data in the relationship between the target geographic object position and the area where it is located when obtaining the value assessment result of the target geographic object. By inputting the above attribute data into a pre-trained geographic object value assessment model, the value assessment result of the target geographic object is finally obtained. The geographic object data processing method improves the accuracy of the value assessment of the geographic object to be assessed, and the data required for the assessment calculation is ordered and definite, which reduces the consumption of computing resources and improves the efficiency of the assessment calculation and the stability of the calculation results. Second embodiment

[0111] In the above-mentioned first embodiment, a geographic object data processing method is provided. Correspondingly, the second embodiment of the present application provides a method for training a geographic object value assessment model. The parts of this embodiment that are the same as the first embodiment will not be repeated here. Please refer to the corresponding parts in the first embodiment.

[0112] Please refer to Figure 6 , which is a flowchart of a method for training a geographic object value assessment model provided in the second embodiment of the present application.

[0113] Step S601: Acquire training sample data, the training sample data includes attribute data of multiple geographic objects, and the attribute data of each geographic object includes: feature data of the geographic object, feature data of the region where the geographic object is located, time series data of object ownership transfer in the region where the geographic object is located, and other geographic object data in the relationship between the location of the geographic object and the region where it is located; Step S602: classifying the training sample data into numerical data, text data, time series data and geographic information data; Step S603: input the training sample data into an initial model including an encoding module and a decoding module for iterative training. The training process is as follows: Input the attribute data of each geographic object into the corresponding encoding module according to the above classification categories for encoding to generate a data vector; The data vectors output by each encoding module are respectively input into the sub-decoders corresponding to each encoding module, and the data vectors output by each encoding module are fused and input into the global decoder; Decoding is performed through each subdecoder to obtain evaluation results respectively, each residual between each evaluation result and the target result is calculated, and each residual is fed back to the corresponding encoder to adjust the parameters; Decoding by a global decoder to obtain an evaluation result, calculating a total residual between the evaluation result and a target result, and feeding the total residual back to each encoder to adjust parameters; The encoding and decoding process is iterated until the total residual satisfies a set value.

[0114] Optionally, other geographic object data in the relationship between the geographic object location and the region in which it is located includes: geographic object relationship data and region relationship data; The geographic object relationship data is used to represent the relationship data between a selected geographic object and surrounding associated geographic objects, wherein the surrounding associated geographic objects are geographic objects whose distance from the selected geographic object is within a first preset distance range and have an associated influence on the user in the selected geographic object; The area relationship data is used to represent the relationship data between an area where a selected geographical object is located and a neighboring area, and the neighboring area is an area whose distance from the area where the selected geographical object is located is within a second preset distance range.

[0115] Optionally, the relationship data between the selected geographic object and surrounding associated geographic objects is obtained in the following manner: Obtaining the position data of the selected geographic object as a selected geographic object position vector; Determine the surrounding associated geographic objects and their location data as surrounding associated geographic object data vectors; According to the selected geographic object position vector and the surrounding associated geographic object data vector, relationship data between the selected geographic object and the surrounding associated geographic objects is determined.

[0116] Optionally, the relationship data between the selected geographic object and surrounding associated geographic objects is obtained in the following manner: Acquire the position data of the selected geographic object as the selected geographic object position vector; Determine surrounding associated geographical objects of the selected geographical object; For each of the surrounding associated geographic objects, determine its position data as a position vector of each surrounding associated geographic object; Calculate the difference between the position vector of each of the surrounding associated geographic objects and the position vector of the selected geographic object as a position difference vector; After linear processing, each position difference vector is added to the corresponding position vector of the surrounding associated geographic object to obtain a geocoded vector corresponding to each surrounding associated geographic object; All geocoded vectors are fused through a multi-head attention mechanism to obtain the relationship data between the selected geographic object and the surrounding associated geographic objects.

[0117] Optionally, the location data includes longitude and latitude; The position difference vector includes a longitude difference and a latitude difference.

[0118] Optionally, the location data includes longitude and latitude; The step of calculating the difference between the position vector of each of the surrounding associated geographic objects and the position vector of the selected geographic object as a position difference vector comprises: Calculating the longitude difference and latitude difference between each of the surrounding associated geographic objects and the selected geographic object; The longitude difference and the latitude difference are normalized and mapped into a code value of a set dimension to obtain the position difference vector.

[0119] Optionally, the normalizing the longitude difference and the latitude difference, mapping them into codes of a set dimension, and using the encoded data as the position difference vector includes: Normalization is performed by taking sine and cosine of the longitude difference and latitude difference respectively; By multiplying the longitude difference and latitude difference by 10, we can get 2i / d model The dimension is expanded by an exponential factor, where i is the index of the dimension and d model is the dimension of the setting; The longitude difference and latitude difference are multiplied by 10 respectively, and 2i / d model The data obtained after the exponential factorization are subjected to sine and cosine transformation respectively, and all the generated data are used as elements of the position difference vector to obtain the position difference vector.

[0120] Optionally, the determining the surrounding associated geographical objects includes: Obtaining a preset list of surrounding associated geographical objects corresponding to the selected geographical object, and determining the surrounding associated geographical object from the list of surrounding associated geographical objects corresponding to the selected geographical object; Alternatively, a preset list of surrounding associated geographical objects corresponding to the area where the selected geographical object is located is obtained, and the surrounding associated geographical object is determined from the list of surrounding associated geographical objects corresponding to the area where the selected geographical object is located; Alternatively, a preset list of surrounding associated geographical objects corresponding to the building where the selected geographical object is located is obtained, and the surrounding associated geographical object is determined from the list of surrounding associated geographical objects corresponding to the building where the selected geographical object is located; Alternatively, a preset list of all geographic objects is obtained; a list of geographic objects whose distances to the selected geographic object are within the first preset distance range is filtered out from the full list of geographic objects as a list of candidate surrounding associated geographic objects; feature data of each candidate surrounding associated geographic object in the list of candidate surrounding associated geographic objects is obtained; and the surrounding associated geographic objects are filtered out from the list of candidate surrounding associated geographic objects based on the selected geographic object feature data, the feature data of each candidate surrounding associated geographic object, and a surrounding associated geographic object screening model.

[0121] Optionally, the determining the surrounding associated geographical objects and their location data includes: The surrounding associated geographical objects and their position data are determined according to the position data of the selected geographical object, the positional relationship between the surrounding geographical objects and the selected geographical object, and the influence of the surrounding geographical objects on the value assessment of the selected geographical object.

[0122] Optionally, the relationship data between the area where the selected geographical object is located and the neighboring areas are obtained in the following manner: Obtaining characteristic data of other geographical objects and time series data of ownership transfer in the area where the selected geographical object is located as the selected area data vector; Determine the neighborhood area; Obtaining geographic object feature data and ownership transfer time series data of the neighboring area as a neighboring area data vector; Relationship data between the selected target region and the neighbor region is obtained according to the selected region data vector and the neighbor region data vector.

[0123] Optionally, the step of obtaining characteristic data of other geographical objects and time series data of ownership transfer in the area where the selected geographical object is located as the selected area data vector includes: Classify the characteristic data of other geographical objects in the area where the selected geographical object is located and the time series data of ownership transfer into text data and time series data, and for each geographical object in the other geographical objects, concatenate the text data and the time series data as a sub-vector, and combine the sub-vectors of all other geographical objects as the selected area data vector; The step of obtaining the geographic object feature data and ownership transfer time series data of the neighboring area as a neighboring area data vector includes: The characteristic data of the geographical objects in the neighboring area and the time series data of the ownership transfer are classified into text data and time series data. For each geographical object in the neighboring area, the text data and the time series data are spliced ​​as a sub-vector, and the sub-vectors of all geographical objects are combined as the data vector of the neighboring area.

[0124] Optionally, determining the neighbor area includes: Among the text data and time series data classified from the characteristic data of other geographical objects and the time series data of ownership transfer in the area where the selected geographical object is located, as well as the text data and time series data classified from the characteristic data of geographical objects and the time series data of ownership transfer in the neighboring area, the text data and the time series data are encoded using a factorization machine encoder and a temporal convolutional network encoder respectively to generate data vectors of the selected area and the neighboring area.

[0125] Optionally, a multi-head attention mechanism is applied to fuse the data vectors output by each encoding module.

[0126] The training method of the geographic object value assessment model described in the embodiment of the present application includes: obtaining training sample data, the training sample data includes attribute data of multiple geographic objects, and the attribute data of each geographic object includes: geographic object feature data, feature data of the region where the geographic object is located, time series data of object ownership transfer in the region where the geographic object is located, and other geographic object data in the relationship between the geographic object position and the region where it is located; classifying the training sample data into numerical data, text data, time series data and geographic information data; inputting the training sample data into an initial model including an encoding module and a decoding module for iterative training, and the training process is as follows: the attribute data of each geographic object is classified according to the above The classification categories are respectively input into the corresponding encoding modules for encoding to generate data vectors; the data vectors output by each encoding module are respectively input into the sub-decoders corresponding to each encoding module, and the data vectors output by each encoding module are fused and input into the global decoder; the evaluation results are respectively obtained by decoding through each sub-decoder, and the residuals between each evaluation result and the target result are calculated, and the residuals are fed back to the corresponding encoders to adjust the parameters; the evaluation results are obtained by decoding through the global decoder, and the total residuals between the evaluation results and the target result are calculated, and the total residuals are respectively fed back to each encoder to adjust the parameters; the encoding and decoding process is iteratively performed until the total residual meets the set value.

[0127] The geographic object value assessment model training method described in the embodiment of the present application can classify the training sample data into numerical data, text data, time series data and geographic information data, and input them into the corresponding encoding modules for encoding according to the above classification categories, generate data vectors, input the data vectors output by each encoding module into the sub-decoders corresponding to each encoding module, and input the data vectors output by each encoding module into the global decoder after fusion. On the one hand, each residual between each evaluation result obtained after decoding by each sub-decoder and the target result can be fed back to the corresponding encoder, so as to adjust the parameters of each sub-encoder; on the other hand, the total residual between the evaluation result obtained after decoding by the global decoder and the target result can be fed back to each encoder, so as to adjust the parameters of each sub-encoder, through multiple encoding and decoding iterative training processes, until the total residual meets the set value, the geographic object value assessment model training method improves the accuracy of the estimated results of the geographic object value assessment model. Third embodiment

[0128] In the above-mentioned first embodiment, a geographic object data processing method is provided. Correspondingly, the third embodiment of the present application provides a geographic object data processing device. Since the device embodiment is basically similar to the first method embodiment, the description is relatively simple. For the relevant parts, please refer to the partial description of the method embodiment. The device embodiment described below is only illustrative.

[0129] Please refer to Figure 7 , which is a schematic diagram of a geographic object data processing device provided in the third embodiment of the present application.

[0130] The geographic object data processing device 700 comprises: A target geographic object to be evaluated determining unit 701, used to determine a target geographic object to be evaluated; The attribute data acquisition unit 702 is used to acquire the attribute data of the target geographic object, wherein the attribute data includes: characteristic data of the target geographic object, characteristic data of the region where the target geographic object is located, time series data of object ownership transfer in the region where the target geographic object is located, and other geographic object data in the relationship between the position of the target geographic object and the region where the target geographic object is located; An evaluation unit 703 is used to input the attribute data into a pre-trained geographic object value evaluation model to obtain a value evaluation result for the target geographic object; The geographic object value assessment model is trained using the following method: Acquire training sample data, the training sample data including attribute data of multiple geographic objects, the attribute data of each geographic object including: feature data of the geographic object, feature data of the region where the geographic object is located, time series data of object ownership transfer in the region where the geographic object is located, and other geographic object data in the relationship between the location of the geographic object and the region where the geographic object is located; Classifying the training sample data into numerical data, text data, time series data and geographic information data; The training sample data is input into an initial model including an encoding module and a decoding module for iterative training. The training process is as follows: Input the attribute data of each geographic object into the corresponding encoding module according to the above classification categories for encoding to generate a data vector; The data vectors output by each encoding module are respectively input into the sub-decoders corresponding to each encoding module, and the data vectors output by each encoding module are fused and input into the global decoder; Decoding is performed through each subdecoder to obtain evaluation results respectively, each residual between each evaluation result and the target result is calculated, and each residual is fed back to the corresponding encoder to adjust the parameters; Decoding by a global decoder to obtain an evaluation result, calculating a total residual between the evaluation result and a target result, and feeding the total residual back to each encoder to adjust parameters; The encoding and decoding process is iterated until the total residual satisfies a set value. Fourth embodiment

[0131] In the above second embodiment, a method for training a geographic object value assessment model is provided. Correspondingly, the fourth embodiment of the present application provides a device for training a geographic object value assessment model. Since the device embodiment is basically similar to the second method embodiment, the description is relatively simple. For relevant parts, please refer to the partial description of the method embodiment. The device embodiment described below is only illustrative.

[0132] Please refer to Figure 8 , which is a schematic diagram of a training device for a geographic object value assessment model provided in the fourth embodiment of the present application.

[0133] The training device 800 of the geographic object value assessment model includes: The training sample data acquisition unit 801 is used to acquire training sample data, wherein the training sample data includes attribute data of multiple geographic objects, and the attribute data of each geographic object includes: feature data of the geographic object, feature data of the region where the geographic object is located, time series data of object ownership transfer in the region where the geographic object is located, and other geographic object data in the relationship between the location of the geographic object and the region where the geographic object is located; A classification unit 802 is used to classify the training sample data into numerical data, text data, time series data and geographic information data; The iterative training unit 803 is used to input the training sample data into the initial model including the encoding module and the decoding module for iterative training. The training process is as follows: Input the attribute data of each geographic object into the corresponding encoding module according to the above classification categories for encoding to generate a data vector; The data vectors output by each encoding module are respectively input into the sub-decoders corresponding to each encoding module, and the data vectors output by each encoding module are fused and input into the global decoder; Decoding is performed through each subdecoder to obtain evaluation results respectively, each residual between each evaluation result and the target result is calculated, and each residual is fed back to the corresponding encoder to adjust the parameters; Decoding by a global decoder to obtain an evaluation result, calculating a total residual between the evaluation result and a target result, and feeding the total residual back to each encoder to adjust parameters; The encoding and decoding process is iterated until the total residual satisfies a set value. Fifth embodiment

[0134] Corresponding to the above method embodiment of the present application, the fifth embodiment of the present application further provides an electronic device. Fig. 9 As shown, Fig. 9 This is a schematic diagram of an electronic device provided in the fifth embodiment of the present application. The electronic device includes: at least one processor 901, at least one communication interface 902, at least one memory 903 and at least one communication bus 904; optionally, the communication interface 902 may be an interface of a communication module, such as an interface of a GSM module; the processor 901 may be a processor CPU, or an application specific integrated circuit ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of the present invention. The memory 903 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk storage. Among them, the memory 903 stores a program, and the processor 901 calls the program stored in the memory 903 to execute the method provided in the above embodiment of the present application. Sixth embodiment

[0135] Corresponding to the above method of the present application, the sixth embodiment of the present application further provides a computer storage medium. The computer storage medium stores a computer program, which is executed by a processor to execute the method provided in the above embodiment of the present application.

[0136] Although the present application is disclosed as above in the form of a preferred embodiment, it is not intended to limit the present application. Any technical personnel in this field may make possible changes and modifications without departing from the spirit and scope of the present application. Therefore, the scope of protection of the present application shall be based on the scope defined by the claims of the present application.

[0137] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0138] Memory may include non-permanent storage in a computer-readable medium, in the form of random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0139] 1. Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be used to store information by any method or technology. Information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include non-transitory media such as modulated data signals and carrier waves.

[0140] 2. Those skilled in the art should understand that the embodiments of the present application can be provided as methods, systems or computer program products. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0141] It should be noted that the embodiments of the present application may involve the use of user data. In actual applications, user-specific personal data can be used in the scheme described in this article within the scope permitted by applicable laws and regulations, subject to the requirements of applicable laws and regulations of the country where the user is located (for example, with the user's explicit consent, effective notification to the user, etc.).

[0142] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

Claims

1. A geographic object data processing method, characterized in that: include: Identify the target geographic objects to be evaluated; Acquire attribute data of the target geographic object, the attribute data including: characteristic data of the target geographic object, characteristic data of the region where the target geographic object is located, time series data of object ownership transfer in the region where the target geographic object is located, and other geographic object data in the relationship between the position of the target geographic object and the region where it is located; Inputting the attribute data into a pre-trained geographic object value assessment model to obtain a value assessment result for the target geographic object; The geographic object value assessment model is trained using the following method: Acquire training sample data, the training sample data including attribute data of multiple geographic objects, the attribute data of each geographic object including: feature data of the geographic object, feature data of the region where the geographic object is located, time series data of object ownership transfer in the region where the geographic object is located, and other geographic object data in the relationship between the location of the geographic object and the region where the geographic object is located; Classifying the training sample data into numerical data, text data, time series data and geographic information data; The training sample data is input into an initial model including an encoding module and a decoding module for iterative training. The training process is as follows: Input the attribute data of each geographic object into the corresponding encoding module according to the above classification categories for encoding to generate a data vector; The data vectors output by each encoding module are respectively input into the sub-decoders corresponding to each encoding module, and the data vectors output by each encoding module are fused and input into the global decoder; Decoding is performed through each subdecoder to obtain evaluation results respectively, each residual between each evaluation result and the target result is calculated, and each residual is fed back to the corresponding encoder to adjust the parameters; Decoding by a global decoder to obtain an evaluation result, calculating a total residual between the evaluation result and a target result, and feeding the total residual back to each encoder to adjust parameters; The encoding and decoding process is iterated until the total residual satisfies a set value.

2. The geographic object data processing method according to claim 1, characterized in that: In the training of the geographic object value assessment model, other geographic object data in the relationship between the geographic object location and the region in which it is located includes: geographic object relationship data and region relationship data; The geographic object relationship data is used to represent the relationship data between a selected geographic object and surrounding associated geographic objects, wherein the surrounding associated geographic objects are geographic objects whose distance from the selected geographic object is within a first preset distance range and have an associated influence on the user in the selected geographic object; The area relationship data is used to represent the relationship data between an area where a selected geographical object is located and a neighboring area, and the neighboring area is an area whose distance from the area where the selected geographical object is located is within a second preset distance range.

3. The geographic object data processing method according to claim 2, characterized in that: The relationship data between the selected geographic object and the surrounding associated geographic objects is obtained in the following manner: Obtaining the position data of the selected geographic object as a selected geographic object position vector; Determine the surrounding associated geographic objects and their location data as surrounding associated geographic object data vectors; According to the selected geographic object position vector and the surrounding associated geographic object data vector, relationship data between the selected geographic object and the surrounding associated geographic objects is determined.

4. The geographic object data processing method according to claim 2, characterized in that: The relationship data between the selected geographic object and the surrounding associated geographic objects is obtained in the following manner: Acquire the position data of the selected geographic object as the selected geographic object position vector; Determine surrounding associated geographical objects of the selected geographical object; For each of the surrounding associated geographic objects, determine its position data as a position vector of each surrounding associated geographic object; Calculate the difference between the position vector of each of the surrounding associated geographic objects and the position vector of the selected geographic object as a position difference vector; After linear processing, each position difference vector is added to the corresponding position vector of the surrounding associated geographic object to obtain a geocoded vector corresponding to each surrounding associated geographic object; All geocoded vectors are fused through a multi-head attention mechanism to obtain the relationship data between the selected geographic object and the surrounding associated geographic objects.

5. The geographic object data processing method according to claim 4, characterized in that: The location data includes longitude and latitude; The position difference vector includes a longitude difference and a latitude difference.

6. The geographic object data processing method according to claim 4, characterized in that: The location data includes longitude and latitude; The step of calculating the difference between the position vector of each of the surrounding associated geographic objects and the position vector of the selected geographic object as a position difference vector comprises: Calculating the longitude difference and latitude difference between each of the surrounding associated geographic objects and the selected geographic object; The longitude difference and the latitude difference are normalized and mapped into a code value of a set dimension to obtain the position difference vector.

7. The geographic object data processing method according to claim 2, characterized in that: The relationship data between the area where the selected geographical object is located and the neighboring areas are obtained in the following manner: Obtaining characteristic data of other geographical objects and time series data of ownership transfer in the area where the selected geographical object is located as the selected area data vector; Determine the neighborhood area; Obtaining geographic object feature data and ownership transfer time series data of the neighboring area as a neighboring area data vector; According to the selected area data vector and the neighbor area data vector, the relationship data between the area where the selected geographic object is located and the neighbor area is obtained.

8. The geographic object data processing method according to claim 7, characterized in that: The step of obtaining characteristic data of other geographical objects and time series data of ownership transfer in the area where the selected geographical object is located as the selected area data vector includes: Classify the characteristic data of other geographical objects in the area where the selected geographical object is located and the time series data of ownership transfer into text data and time series data, and for each geographical object in the other geographical objects, concatenate the text data and the time series data as a sub-vector, and combine the sub-vectors of all other geographical objects as the selected area data vector; The step of obtaining the geographic object feature data and ownership transfer time series data of the neighboring area as a neighboring area data vector includes: The characteristic data of the geographical objects in the neighboring area and the time series data of the ownership transfer are classified into text data and time series data. For each geographical object in the neighboring area, the text data and the time series data are spliced ​​as a sub-vector, and the sub-vectors of all geographical objects are combined as the data vector of the neighboring area.

9. A method for training a geographic object value assessment model, characterized in that: include: Acquire training sample data, the training sample data including attribute data of multiple geographic objects, the attribute data of each geographic object including: feature data of the geographic object, feature data of the region where the geographic object is located, time series data of object ownership transfer in the region where the geographic object is located, and other geographic object data in the relationship between the location of the geographic object and the region where the geographic object is located; Classifying the training sample data into numerical data, text data, time series data and geographic information data; The training sample data is input into an initial model including an encoding module and a decoding module for iterative training. The training process is as follows: Input the attribute data of each geographic object into the corresponding encoding module according to the above classification categories for encoding to generate a data vector; The data vectors output by each encoding module are respectively input into the sub-decoders corresponding to each encoding module, and the data vectors output by each encoding module are fused and input into the global decoder; Decoding is performed through each subdecoder to obtain evaluation results respectively, each residual between each evaluation result and the target result is calculated, and each residual is fed back to the corresponding encoder to adjust the parameters; Decoding by a global decoder to obtain an evaluation result, calculating a total residual between the evaluation result and a target result, and feeding the total residual back to each encoder to adjust parameters; The encoding and decoding process is iterated until the total residual satisfies a set value.

10. The method for training a geographic object value assessment model according to claim 9, characterized in that: Other geographic object data in the relationship between the location of the geographic object and the region in which it is located include: geographic object relationship data and region relationship data; The geographic object relationship data is used to represent the relationship data between a selected geographic object and surrounding associated geographic objects, wherein the surrounding associated geographic objects are geographic objects whose distance from the selected geographic object is within a first preset distance range and have an associated influence on the user in the selected geographic object; The area relationship data is used to represent the relationship data between an area where a selected geographical object is located and a neighboring area, and the neighboring area is an area whose distance from the area where the selected geographical object is located is within a second preset distance range.

11. The method for training a geographic object value assessment model according to claim 10, characterized in that: The relationship data between the selected geographic object and the surrounding associated geographic objects is obtained in the following manner: Acquire the position data of the selected geographic object as the selected geographic object position vector; Determine surrounding associated geographical objects of the selected geographical object; For each of the surrounding associated geographic objects, determine its position data as a position vector of each surrounding associated geographic object; Calculate the difference between the position vector of each of the surrounding associated geographic objects and the position vector of the selected geographic object as a position difference vector; After linear processing, each position difference vector is added to the corresponding position vector of the surrounding associated geographic object to obtain a geocoded vector corresponding to each surrounding associated geographic object; All geocoded vectors are fused through a multi-head attention mechanism to obtain the relationship data between the selected geographic object and the surrounding associated geographic objects.

12. The method for training a geographic object value assessment model according to claim 10, characterized in that: The relationship data between the area where the selected geographical object is located and the neighboring areas are obtained in the following manner: Obtaining characteristic data of other geographical objects and time series data of ownership transfer in the area where the selected geographical object is located as the selected area data vector; Determine the neighborhood area; Obtaining geographic object feature data and ownership transfer time series data of the neighboring area as a neighboring area data vector; Relationship data between the selected target region and the neighbor region is obtained according to the selected region data vector and the neighbor region data vector.

13. The method for training a geographic object value assessment model according to claim 12, characterized in that: The step of obtaining characteristic data of other geographical objects and time series data of ownership transfer in the area where the selected geographical object is located as the selected area data vector includes: Classify the characteristic data of other geographical objects in the area where the selected geographical object is located and the time series data of ownership transfer into text data and time series data, and for each geographical object in the other geographical objects, concatenate the text data and the time series data as a sub-vector, and combine the sub-vectors of all other geographical objects as the selected area data vector; The step of obtaining the geographic object feature data and ownership transfer time series data of the neighboring area as a neighboring area data vector includes: The characteristic data of the geographical objects in the neighboring area and the time series data of the ownership transfer are classified into text data and time series data. For each geographical object in the neighboring area, the text data and the time series data are spliced ​​as a sub-vector, and the sub-vectors of all geographical objects are combined as the data vector of the neighboring area.

14. The method for training a geographic object value assessment model according to claim 12 or 13, characterized in that: The determining of the neighbor area includes: Among the text data and time series data classified from the characteristic data of other geographical objects and the time series data of ownership transfer in the area where the selected geographical object is located, as well as the text data and time series data classified from the characteristic data of geographical objects and the time series data of ownership transfer in the neighboring area, the text data and the time series data are encoded using a factorization machine encoder and a temporal convolutional network encoder respectively to generate data vectors for the selected area and the neighboring area.

15. A geographic object data processing device, characterized in that: include: A target geographic object to be evaluated determining unit, used to determine the target geographic object to be evaluated; An attribute data acquisition unit is used to acquire the attribute data of the target geographic object, wherein the attribute data includes: characteristic data of the target geographic object, characteristic data of the region where the target geographic object is located, time series data of object ownership transfer in the region where the target geographic object is located, and other geographic object data in the relationship between the position of the target geographic object and the region where it is located; An evaluation unit, used for inputting the attribute data into a pre-trained geographic object value evaluation model to obtain a value evaluation result for the target geographic object; The geographic object value assessment model is trained using the following method: Acquire training sample data, the training sample data including attribute data of multiple geographic objects, the attribute data of each geographic object including: feature data of the geographic object, feature data of the region where the geographic object is located, time series data of object ownership transfer in the region where the geographic object is located, and other geographic object data in the relationship between the location of the geographic object and the region where the geographic object is located; Classifying the training sample data into numerical data, text data, time series data and geographic information data; The training sample data is input into an initial model including an encoding module and a decoding module for iterative training. The training process is as follows: Input the attribute data of each geographic object into the corresponding encoding module according to the above classification categories for encoding to generate a data vector; The data vectors output by each encoding module are respectively input into the sub-decoders corresponding to each encoding module, and the data vectors output by each encoding module are fused and input into the global decoder; Decoding is performed through each subdecoder to obtain evaluation results respectively, each residual between each evaluation result and the target result is calculated, and each residual is fed back to the corresponding encoder to adjust the parameters; Decoding by a global decoder to obtain an evaluation result, calculating a total residual between the evaluation result and a target result, and feeding the total residual back to each encoder to adjust parameters; The encoding and decoding process is iterated until the total residual satisfies a set value.

16. An electronic device, characterized in that: The electronic device comprises a processor and a memory; The memory stores a computer program, and after the processor runs the computer program, the method described in any one of claims 1-8 or 9-14 is executed.

17. A computer storage medium, characterized in that: The computer storage medium stores a computer program, and after the computer program is run by the processor, the method described in any one of claims 1-8 or 9-14 is executed.

Images