Geographical functional area identification method, device, equipment and storage medium

Abstract

The present application discloses a method, device, equipment and storage medium for identifying geographic functional areas. The method includes: obtaining road layer data and POI data of the area to be identified, dividing the area to be identified into multiple sub-areas based on the road layer data, determining POI data pairs that meet the contextual relationship based on the spatial distance between the POI data, generating a vector group based on the feature information of the POI data and using it as the input of the word vector training model, obtaining the word vector of the POI data, weightedly calculating the word vector to obtain the partition vector of the sub-area, and performing cluster analysis on the partition vector to determine the functional area category to which the sub-area belongs. The POI data pairs involved in this solution are selected in accordance with the laws of social and economic activities and statistical data. The word vectors obtained by training the POI data pairs can better mine semantic information such as the spatial distribution characteristics of the POI data, and provide support for the subsequent identification of the functional areas of the sub-areas.

Description

Technical Field

[0001] The present invention relates to the field of geographical area planning, design and research, and in particular to a method, device, equipment and storage medium for identifying geographical functional areas. Background Art

[0002] Functional zones, as the basic building blocks of geographic regions, are an essential component of regional planning and management. Their identification is crucial for revealing the physical and social characteristics of a region. Points of Interest (POIs) primarily refer to geographical entities that are closely related to people's lives and can be abstracted as points, such as government offices, schools, hospitals, and shopping malls. These POIs offer advantages such as large data volumes, diverse categories, wide coverage, and ease of access. They integrate geographic location and attribute information, and represent a comprehensive reflection of the interactions among relevant elements of a geographic functional zone, including population, land, economy, and society.

[0003] Related technologies for identifying geographic functional areas using POI data often determine the area's function solely by calculating the frequency density of POIs, categorizing functional areas as either single-function or mixed-function areas. This results in functional area identification results that favor a larger number of common POI types (e.g., restaurants, shopping, and businesses), failing to effectively represent the area's functions. Furthermore, traditional semantic models such as the Latent Dirichlet Allocation (LDA) model often only consider the latent semantic information of POI data and fail to incorporate the richer information in POI data to accurately identify geographic functional areas. Summary of the Invention

[0004] In view of this, embodiments of the present application provide a method, apparatus, device, and storage medium for identifying geographic functional areas, aiming to accurately identify geographic functional areas.

[0005] The technical solution of the embodiment of the present invention is achieved as follows:

[0006] An embodiment of the present invention provides a method for identifying geographic functional areas, including:

[0007] Obtaining road layer data of the area to be identified and point of interest (POI) data within the area to be identified;

[0008] Dividing the area to be identified into at least two sub-areas based on the road layer data;

[0009] Determining POI data pairs that meet the contextual relationship based on the spatial distance and the distance threshold between the POI data;

[0010] Generating a vector group of the POI data pair based on feature information of the POI data, the vector group including vectors corresponding to the two POI data in the POI data pair;

[0011] Inputting each of the vector groups into a word vector training model to learn and obtain the word vector of each of the POI data;

[0012] Based on the correspondence between the POI data and the sub-regions, weightedly calculating the word vector of the POI data in each sub-region to obtain a partition vector of each sub-region;

[0013] The partition vectors are clustered and analyzed based on a set number of clusters, and the functional area category to which each sub-area belongs is determined according to the cluster clusters obtained from the cluster analysis.

[0014] An embodiment of the present invention further provides a device for identifying a geographical functional area, comprising:

[0015] A data acquisition module is used to acquire road layer data of the area to be identified and point of interest (POI) data within the area to be identified;

[0016] A sub-area division module, configured to divide the area to be identified into at least two sub-areas based on the road layer data;

[0017] A data pair determination module, configured to determine a POI data pair that meets the contextual relationship based on the spatial distance and the distance threshold between the POI data;

[0018] a vector generation module, configured to generate a vector group of the POI data pair based on feature information of the POI data, wherein the vector group includes vectors corresponding to the two POI data in the POI data pair;

[0019] A word vector training module, configured to input each of the vector groups into a word vector training model to learn and obtain the word vectors of each of the POI data;

[0020] A partition vector calculation module is used to calculate the word vector of the POI data in each sub-region based on the corresponding relationship between the POI data and the sub-region, so as to obtain the partition vector of each sub-region;

[0021] The clustering and partitioning module is used to perform cluster analysis on the partitioning vector based on a set number of clusters, and determine the functional area category to which each sub-area belongs according to the cluster clusters obtained from the cluster analysis.

[0022] An embodiment of the present invention further provides a geographic functional area identification device, comprising: a processor and a memory for storing a computer program that can be run on the processor, wherein the processor is used to execute the steps of the geographic functional area identification method described in the embodiment of the present invention when running the computer program.

[0023] An embodiment of the present invention further provides a storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the geographic functional area identification method described in the embodiment of the present invention are implemented.

[0024] The technical solution provided by the embodiment of the present invention determines that the two POI data meet the contextual relationship when the distance between the two POI data is within the distance threshold. The setting of the distance threshold is closely related to the laws of social and economic activities and the statistical range in the region. When the two POI data that meet the contextual relationship are determined as a POI data pair, and the corresponding vector group of the POI data pair is used as the input of the word vector training model, the probability distribution of the POI data pair can be supervised and learned, so as to better mine the semantic relevance of the geographic space distribution of the POI data through the trained word vector. The word vector containing the semantic relevance of the POI data obtained by model training is weighted and calculated to obtain the sub-region partition vector, and finally the identification of the geographic functional area is achieved through cluster analysis of the partition vector. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This is a flow chart of a method for identifying geographic functional areas according to an embodiment of the present application;

[0026] Figure 2 A flow chart of parameter adjustment for a word vector training model according to an embodiment of the present application;

[0027] Figure 3 This is a schematic diagram of the structure of the device for identifying geographic functional areas according to an embodiment of the present application;

[0028] Figure 4 This is a schematic diagram of the structure of the geographic functional area identification device according to an embodiment of the present application;

[0029] Figure 5 This is an example diagram of the POI distribution characteristic diagram of an embodiment of the present application;

[0030] Figure 6 This is an example diagram of the visualization display of semantic spaces of different dimensions of POI types in the embodiment of the present application;

[0031] Figure 7 This is an example diagram of the average precision (MAP) curve of the word vector training model in the embodiment of the present application;

[0032] Figure 8 This is an example diagram of the loss curve of the word vector training model in the embodiment of this application. DETAILED DESCRIPTION

[0033] The present application will be described in further detail below with reference to the accompanying drawings and embodiments.

[0034] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application pertains. The terms used herein in the specification of this application are for the purpose of describing specific embodiments only and are not intended to limit this application.

[0035] Before further explaining the embodiments of the present application in detail, the nouns and terms involved in the embodiments of the present application are explained. The nouns and terms involved in the embodiments of the present application are subject to the following interpretations:

[0036] In a geographic information system, a point of interest (POI) can be a house, a shop, a mailbox, a bus stop, and so on. POIs can be categorized, for example, into primary, secondary, and tertiary levels. Each category has a corresponding industry code and name, making it easier to record and distinguish information during collection.

[0037] In natural language processing tasks, the first thing to consider is how words are represented in a computer. Generally, there are two representation methods: discrete representation (one-hot representation) and distributed representation (distribution representation). Discrete representation represents each word as a long vector whose dimensions are the size of the vocabulary. Only one dimension in the vector has a value of 1, and the rest are 0. This dimension represents the current word. Distributed representation represents a word as a continuous, dense vector of a fixed length.

[0038] Word embedding converts words into distributed representations, also known as word vectors. It's a general term for language models and representation learning techniques in natural language processing (NLP). Conceptually, it involves embedding a high-dimensional space (equal to the number of words) into a much lower-dimensional continuous vector space, where each word or phrase is mapped to a vector in the real domain.

[0039] Word vectors have good semantic properties and are a common way to represent word features. The value of each dimension of a word vector represents a feature with a certain semantic and grammatical interpretation. Therefore, each dimension of a word vector can be called a word feature.

[0040] Language models generate word embeddings by training a neural network language model (NNLM). Word embeddings are a byproduct of the language model. The basic idea behind NNLMs is to predict words based on their contextual appearance. This contextual prediction is essentially a form of learning co-occurrence statistics. Well-known methods for generating word embeddings using NNLMs include the continuous bag-of-words (CBOW) model and the skip-gram model. The core idea of ​​the CBOW model is to predict a word based on its context. The skip-gram model, in contrast to the CBOW model, uses "word occurrences" to predict "context." These models are shallow, two-layer neural networks trained to reconstruct linguistic text. After training, the trained word embedding model can be used to map each word to a word vector, representing the relationships between words.

[0041] A weighted average multiplies each value by its corresponding weight, then adds the total to get the overall value, which is then divided by the total number of units. The weighted average depends not only on the value of each unit in the population (the variable value), but also on the number of times each value appears (the frequency). Because the number of times each value appears weighs its influence on the average, it is called a weight. Because the weighted average is calculated based on different weights, it is also called a weighted average.

[0042] In statistics, the Kendall correlation coefficient is named after Maurice Kendall and is often represented by the Greek letter τ (tau). Kendall's coefficient of consistency is a statistical indicator used to reflect the closeness of the relationship between multiple groups of variables. A Kendall test is a nonparametric hypothesis test that uses a calculated correlation coefficient to test the statistical dependence of two random variables. The Kendall correlation coefficient ranges from -1 to 1. When τ is 1, it indicates that the two random variables have the same rank correlation; when τ is -1, it indicates that the two random variables have completely opposite rank correlations; and when τ is 0, it indicates that the two random variables are independent.

[0043] Clustering is the process of organizing similar data members into classes or clusters. This allows objects within the same cluster to be highly similar, while objects in different clusters are highly different. Cluster analysis is an exploratory process that eliminates the need for predefined criteria. Cluster analysis can automatically classify data based on sample data.

[0044] Clustering algorithms are also called "unsupervised classification". Their purpose is to divide data into meaningful or useful groups (or clusters). The k-means clustering algorithm is an iterative clustering analysis algorithm.

[0045] The Silhouette Coefficient (SCC) is a measure of clustering effectiveness that combines cohesion and separation. It can be used to evaluate the impact of different algorithms, or different algorithm execution modes, on clustering results based on the same raw data.

[0046] The goal of clustering is usually expressed as an objective function that depends on the proximity of points to each other or to the centroid of the cluster; for example, minimizing the square of the distance from each point to the nearest centroid. The sum of squared errors (SSE, also known as spread) is used as an objective function to measure the quality of clustering. The error of each data point is calculated as its Euclidean distance to the nearest centroid, and then the sum of the squares of the errors is calculated.

[0047] The Hopkins statistic is a spatial statistic that tests the spatial randomness of spatially distributed variables. If the entire sample space is uniform and has no clustering tendency (no obvious clustering), the Hopkins statistic is approximately 0.5. Conversely, if the space has a clustering tendency (obvious clustering), the Hopkins statistic is close to 1.

[0048] Related technologies that use POI data for geographic functional area identification simply determine the function of an area by calculating the frequency density of POIs, which fails to effectively represent the region's function. Furthermore, traditional semantic models, such as the LDA model, often only consider the latent semantic information of POI data and fail to incorporate the rich information in POI data to accurately identify geographic functional areas.

[0049] Based on this, in various embodiments of the present application, by setting selection rules for POI data pairs, a vector group is generated for the POI data pairs based on the feature information of the POI data and used as the input of the word vector training model to learn the probability distribution of the POI data pairs. The trained word vector contains the semantic information of the POI data in the geographic space. On this basis, the word vector is weighted and calculated to obtain the partition vector of the sub-region. The partition vector of the sub-region is clustered and analyzed, and then the functional area category to which the sub-region belongs is determined based on the clustering results.

[0050] The present application embodiment provides a method for identifying geographical functional areas, such as Figure 1 As shown, the geographic functional area identification method includes:

[0051] Step 101: Obtain road layer data of the area to be identified and POI data within the area to be identified.

[0052] Here, the area to be identified is a geographical area for which the category of the geographical functional area needs to be determined, which can be an urban area or a township area.

[0053] Here, POI data is point data of a specific location within the area to be identified. The social function attributes of the specific spatial location can be determined based on the attribute content of the POI data.

[0054] Here, we counted and ranked the number of POI data of each category within the identified area. The statistical results show that POI data follows a power-law distribution: a small number of POI data types predominate, corresponding to high-frequency words in natural language, while the majority of POI data types correspond to low-frequency words in natural language. This indicates that POI data in geographic space corresponds to words in natural language. By analogizing POI data with words in natural language, we can obtain word vectors containing POI semantic information, and then explore the spatial distribution probability and regional attribute information of POI data within the region.

[0055] In practical applications, we can also use the natural language word embedding method to build a corpus containing POI data of each sub-region.

[0056] Step 102: Divide the area to be identified into at least two sub-areas based on the road layer data.

[0057] Here, the road layer data may be a dataset in a geographic information system that includes all roads in the area to be identified. The area to be identified can be divided into multiple sub-areas according to road boundaries based on multiple road data.

[0058] In practical applications, the area to be identified can be divided into multiple sub-areas based on urban main roads, or based on non-urban main roads. The area to be identified can be divided into multiple sub-areas using road layer data alone, or the road layer data can be combined with the raster network to divide the area to be identified into multiple sub-areas.

[0059] In practical applications, the area to be identified is divided into at least two sub-areas based on road layer data. The division process may include: first, obtaining urban water surface data, and using software tools or plug-ins to remove the water area of ​​the urban area under study (rivers and lakes are not included in the calculation); then, obtaining road data of the area to be identified, using software tools or plug-ins to generate road layer data, and dividing the area to be identified into at least two sub-areas. The sub-areas can be traffic analysis zones (TAZs) or areas divided based on other specifications and scales. For example, sub-areas with an area smaller than a set threshold (for example, 1,000 square meters) can also be removed, thereby removing the debris areas generated during the division process and obtaining valid sub-areas.

[0060] Step 103: determining POI data pairs that meet the contextual relationship based on the spatial distance and distance threshold between the POI data;

[0061] Here, the distance threshold may be determined based on the statistical number of POI data within the distance threshold range, or may be determined based on the socio-economic activity patterns of the region.

[0062] In practical applications, a distance threshold can be set to define POI data pairs such as "central POI data and contextual POI data." The relationship between the proportion of POI data with a certain amount of contextual POI data and the distance threshold can be statistically analyzed. For example, when the distance threshold is increased to 500 meters, the proportion of POIs with adjacent POI data exceeding the set interval (20, 40, 60, 80, and 100) reaches a high level, indicating that a sufficient number of POI data pairs can be captured with a distance threshold of 500 meters.

[0063] In some embodiments, the POI data includes geographic location information, and determining POI data pairs that meet the contextual relationship based on the spatial distance and distance threshold between the POI data includes:

[0064] Calculate the spatial distance between any two POI data based on the geographic location information of each POI data;

[0065] Pairs of POI data with a spatial distance less than or equal to a distance threshold are counted as POI data pairs that meet the contextual relationship.

[0066] In one application example, the geographic location information contained in the POI data is longitude and latitude. The distance between two POI data can be calculated directly based on the longitude and latitude, or the longitude and latitude can be converted into coordinates of a plane projection coordinate system before calculating the spatial distance between the two POI data.

[0067] In an application example, the geographic location information included in the POI data is a projection coordinate, and the spatial distance between two POI data can be calculated based on the projection coordinate.

[0068] In actual applications, depending on the size of the sub-area to be identified and / or the recognition accuracy requirements, you can choose to directly use longitude and latitude or projection coordinates to calculate the spatial distance between two POI data, or you can convert longitude and latitude into projection coordinates and then calculate the spatial distance between two POI data.

[0069] In practical applications, the spatial distance between two POIs can be calculated based on the longitude and latitude of the data. This can be done by using the conversion relationship between longitude and latitude and length, or by using the existing formula for calculating distance based on longitude and latitude. External tool software or built-in plug-ins can also be used to convert longitude and latitude information into spatial distance.

[0070] In practical applications, the distance threshold can be set to a range of 200 meters to 500 meters as needed.

[0071] Step 104 : generating a vector group of the POI data pair based on the feature information of the POI data, where the vector group includes vectors corresponding to the two POI data in the POI data pair.

[0072] Here, the characteristic information of the POI data can distinguish each POI data from other POI data. A vector is a vector representation of each POI data obtained by digitizing the characteristic information of the POI data. The dimension of this vector is equal to the total number of POI data in the area to be identified. Therefore, this vector dimension is usually very high and sparse, which cannot be used to calculate semantic relevance between them.

[0073] Here, the vector group includes two vectors. For example, the kth POI data pair includes the i-th POI data and the j-th POI data that meet the contextual relationship. The i-th POI data is vectorized as vector v i , the j-th POI data is vectorized and represented as vector v j , then the vector group v k (v i , v j )Two vectors.

[0074] In some embodiments, generating a vector group of POI data pairs based on feature information of POI data includes:

[0075] Encode the characteristic information of the POI data to generate a vector corresponding to the POI data, wherein the characteristic information includes at least: geographic location information and POI multi-level categories;

[0076] Based on the correspondence between POI data and vectors, and the contextual relationship between POI data pairs, a vector group of POI data pairs is generated.

[0077] Here, the POI data may include: name, address name, longitude and latitude, region, POI multi-level category and other information, and the characteristic information of the POI data includes at least geographic location information and POI multi-level category.

[0078] Here, multi-level POI classification refers to a multi-level POI classification with progressively smaller subdivisions. POI classification is a tree-like hierarchical structure. Lower-level POI types contain richer and more specific semantic information, making them more conducive to geographic functional area analysis based on POI type semantics. For example, an internet cafe named "Internet Cafe A" would be classified as "Sports and Leisure Services" at the first level, "Entertainment Venues" at the second level, and "Internet Cafe" at the third level.

[0079] It should be noted that the early POI multi-level classification may have only two categories, and in order to better classify POI data, POI classification may have more than three levels. The POI multi-level classification is not limited to the above example content.

[0080] Here, one-hot vectors can be used to encode the feature information of POI data.

[0081] Step 105: Input each vector group into a word vector training model to learn and obtain the word vector of each POI data.

[0082] Here, the word vector training model can be used to embed each sparse high-dimensional input vector into a lower-dimensional weight matrix, learn the probability distribution of each POI data pair, and obtain the word vector of each POI data based on the trained word vector model.

[0083] Here, the word vector training model can be a skip-gram model or other neural network language models suitable for semantic recognition of POI data.

[0084] In some embodiments, each vector group is input into a word vector training model to learn and obtain the word vector of each POI data, including:

[0085] Each vector group is used as a single training sample to generate a training sample set containing each vector group, and the training sample set is used as the input of the word vector training model;

[0086] Each vector is trained through a word vector training model to learn word embedding and obtain the word vector of each POI data.

[0087] Here, the frequency and order of words in a document reflect the specific context and semantics of the document. The region to be identified is analogized to a corpus, and the categories of POI data in geographic space are considered words. The word embeddings of the surrounding POI data are predicted based on the central POI data in the geographic space. The embedding of the POI data type is learned by predicting the probability distribution of the contextual (i.e., neighboring) POI data types given the central word POI data type. Each POI data in the training set is considered the "central POI data," and its neighboring POI data are considered the "contextual POI data." The underlying idea is that "places can be classified by their neighbors," which is similar to the assumption in linguistics that "you should recognize a word by its surrounding words." The contextual POI data of each central POI data can be defined by their spatial proximity in geographic space. Vectors of all "central POI data, contextual POI data" pairs in the geographic space are then extracted as the input to the word embedding training model.

[0088] Here, we consider regions as documents, land use types within geographic functional zones as topics, and POI data within the identified region as base words. The spatial distribution of POI data within an identified region can be considered a word sequence within a document. Therefore, the relationship between POI data sequences and geographic functional zone types can be quantified using a continuous spatial language model. By mining the latent relationships within contextual relationships, we can better explore the information within POI data. Using spatial distribution metrics for POI data, we can derive the latent distribution characteristics of its final-level categories, thereby detecting and extracting the first-level categories as the basis for determining geographic functional zone categories.

[0089] In one application example, the word vectors of the trained POI data can be used to calculate POI-class semantic similarity. The calculated results can then be used to determine whether the POI-class semantic similarity has been captured. After model training, the resulting POI-class semantic similarity sequence can be used to verify the effectiveness of the model training. Table 1 shows the POI-class semantic similarity sequence.

[0090] Table 1 Sequence of semantic similarity of POI categories

[0091]

[0092]

[0093] Here, each POI data in the data set consisting of all POI data in the area to be identified can be regarded as the central POI data, and the POI data within a certain distance from the central POI data is selected as the corresponding context POI data. The surrounding information can be fully utilized to determine the semantic features of the central POI data, which is conducive to capturing the potential semantics of POI data closer to the central POI data.

[0094] In one application example, the word vector training model can be a skip-word model including an input layer and an output layer. Since the input vector group corresponds to the vectors of two POI data, and the two POI data are each other's central words and background words, when using the skip-word model to train the word vectors of POI data, it can be considered that the central words and background words are both certain. At this time, the input of the training sample can be used to train the neural network of the model. By traversing all the training samples in the training sample set, the model parameters are continuously iterated and updated, and finally a more accurate word vector is obtained from the trained word vector training model.

[0095] Step 106 : Based on the correspondence between the POI data and the sub-regions, a weighted word vector of the POI data in each sub-region is calculated to obtain a partition vector for each sub-region.

[0096] Here, the correspondence between POI data and sub-regions can be determined based on the word vectors obtained by the word vector training model and the geographic location information contained in the POI data. The word vectors of the POI data in the sub-region are weighted and calculated based on the correspondence to obtain the partition vector of the sub-region.

[0097] Here, the sub-region is regarded as a document, the word vector of the POI data is regarded as a word, and the weighted average of the word vector is used to represent the functional characteristics of the sub-region. The partition vector representing the functional characteristics of the sub-region is calculated by the following formula:

[0098]

[0099] Among them, type(v i,k ) represents the word vector of the k-th POI data in the i-th sub-region (SR), and N represents the total number of word vectors in the sub-region.

[0100] Here, by calculating the weighted average of the word vectors in the sub-region, the sub-region partition vector can become a semantic vector that is endowed with specific meaning.

[0101] Step 107 : performing cluster analysis on the partition vector based on the set number of clusters, and determining the functional area category to which each sub-area belongs according to the cluster clusters obtained from the cluster analysis.

[0102] Here, the K-means algorithm can be used to cluster the sub-region partition vectors. The number of clusters K is set from 2 to 20, and its silhouette coefficient (Silhouette Coefficient) and cluster error variance (SSE) are calculated respectively to determine the optimal number of clusters for clustering the sub-region partition vectors.

[0103] Here, the number of clusters can be set according to the accuracy requirement of geographic functional area identification, and the number of clusters can be the same as or positively correlated with the total number of geographic functional area categories.

[0104] Here, we can determine whether the sub-region partition vectors are spatially non-uniformly distributed, that is, whether they are clustered. We can calculate the Hopkins statistic corresponding to all sub-region partition vectors. The higher the value, the more obvious the non-uniform distribution of the sub-region feature partition vectors, and the better the clustering trend.

[0105] Here, based on the cluster size after clustering the partition vectors of each sub-region, the cluster formed by the largest number of final-level category POI data can be determined as the cluster representing the function of the sub-region, the first-level category to which the final-level category belongs can be used as the functional category of the cluster, and the geographical functional area category of the sub-region can be confirmed in combination with the functional category of the cluster.

[0106] In some embodiments, the POI data includes: at least two levels of categories that are subdivided level by level, wherein each level of categories has multiple levels; the method further includes:

[0107] Based on the clusters of cluster analysis, the density data of POI data of each first-level category in each sub-area is calculated. The density value is used to reflect the density weight value of POI data of each first-level category in the cluster;

[0108] Based on the clusters of cluster analysis, the enrichment factor of POI data of each first-level category in the sub-region is calculated. The enrichment factor is used to reflect the enrichment degree of POI data of each first-level category in the cluster;

[0109] Accordingly, the functional area category to which each sub-region belongs is determined based on the clusters obtained from the cluster analysis, including:

[0110] For each sub-region, the number of POI data in the cluster and the density weight value and / or enrichment factor of the POI data are weighted and summed, and the category of the cluster with the largest sum is selected as the functional area category to which the corresponding sub-region belongs.

[0111] Here, the density of each first-class POI data in each sub-region can be calculated to reflect the weight of each first-class POI data in each sub-region cluster. The calculation formula is as follows:

[0112]

[0113] Among them, d i q represents the density of POI data of category q in sub-region cluster i, N i q represents the number of POI data in category q, A i represents the total area of ​​the sub-regions of sub-region cluster i.

[0114] Here, we can also calculate the enrichment factor of the POI data of each first-class category in the sub-region. The enrichment factor reflects the enrichment degree of the POI data of each first-class category in the cluster. It can be used to reduce the impact of high-density distribution of first-class POI data such as catering and shopping on sub-region identification. The formula is as follows:

[0115]

[0116] Among them, EF i q N represents the enrichment factor of the POI data of the first category q in the cluster of the i-th sub-region, i q N represents the number of POI data with the first category q in the i-th sub-region cluster, i Represents the total number of POI data in cluster i, N q represents the total number of POI data of the first-level category q, and N is the total number of POI data in the area to be identified.

[0117] In some embodiments, based on the cluster clusters of cluster analysis, the density data of POI data of each first-level category in each sub-area can be calculated. The density value is used to reflect the density weight value of POI data of each first-level category in the cluster cluster. For each sub-area, the number of POI data in the cluster cluster and the density weight value of the POI data are weightedly summed, and the category of the cluster cluster with the largest sum value is selected as the functional area category to which the corresponding sub-area belongs.

[0118] In some embodiments, based on the cluster analysis, an enrichment factor index can be introduced to weight the POI data of each first-level category with high density distribution in the cluster. For each sub-area, the number of POI data in the cluster and the enrichment factor of the POI data are weighted and summed, and the category of the cluster with the largest sum value is selected as the functional area category to which the corresponding sub-area belongs.

[0119] In some embodiments, based on the cluster clusters of cluster analysis, the density data of the POI data of each first-level category in each sub-area can be calculated, and the density value is used to reflect the density weight value of the POI data of each first-level category in the cluster cluster. The enrichment factor index can also be introduced to weight the POI data of each first-level category with high density distribution in the cluster cluster. For each sub-area, the number of POI data in the cluster cluster and the density weight value and enrichment factor of the POI data are weighted and summed, and the category of the cluster cluster with the largest sum value is selected as the functional area category to which the corresponding sub-area belongs.

[0120] In one application example, the density of POI data doesn't fully reflect the characteristics of each sub-region cluster. This is because certain primary POI types, such as "catering services" and "life services," may have a high density across the entire area to be identified. This high density across the entire area interferes with determining the specific functions to which the sub-region clusters belong. In this case, an enrichment factor can be added to the density weight to reduce the impact of certain high-density primary POI types on the functional area identification results.

[0121] In one application example, the density and enrichment factor of each first-level category POI data in each sub-region cluster can be calculated based on the clustering results, a distribution characteristic map of the POI categories in the sub-region cluster can be drawn, and the geographical functional areas to which the sub-region clusters belong can be analyzed and identified. The following table shows the density weight values ​​and enrichment factors of the POI categories in the sub-region clusters. The zoning orientation of each cluster can be intuitively analyzed based on the content of Table 2.

[0122] Table 2 Sub-region cluster density weights and enrichment factors

[0123]

[0124] By inputting the data in Table 2 into professional mapping and data analysis software, a proportion-oriented POI distribution characteristic map can be drawn. The distribution characteristic map can be Figure 5 shown in the form.

[0125] In some embodiments, the method further comprises:

[0126] Each sub-area is displayed in a partitioned manner based on the functional area category to which it belongs.

[0127] Here, based on the results of cluster analysis and functional area identification, a POI category distribution feature map in the sub-region cluster is drawn, and the specific geographical functional area to which the sub-region cluster belongs is identified and displayed.

[0128] In practical applications, a single color can be used to render sub-regions belonging to the same geographic functional area to obtain the zoning display results of the entire study area. Clusters representing the main functions and other clusters of secondary functions in each sub-region can also be displayed in three dimensions, and clusters of the same POI category can be marked with the same color or the same elements.

[0129] In some embodiments, the method further comprises:

[0130] The word vectors obtained by the word vector training model are mapped into a two-dimensional semantic space, and POI types belonging to the same top-level category are rendered with the same color to intuitively display the relationship between POI classes.

[0131] In practical applications, we can use the dimensionality reduction algorithm to map the word vectors of all POI types trained with semantic vector dimensions of 40, 60, 80, and 100 into a two-dimensional semantic space, and obtain the semantic space of POI types in different dimensions. The semantic space can be Figure 6 The visual display is shown as follows.

[0132] In an application example, the functional area recognition result and the sampling recognition result can also be compared and verified, and the accuracy of the functional area recognition result based on the technical solution of the present invention can be evaluated according to the degree of consistency between the two.

[0133] Compared with the huge corpus used for natural semantic recognition, the corpus constructed based on all the POI data obtained from the area to be identified has the disadvantage of being small in scale, which is not conducive to the mining and utilization of the semantic relevance of POI data. Therefore, this technical solution also adopts technical means to improve the accuracy of POI data word vector training.

[0134] Based on this, in some embodiments, the POI data includes: at least two levels of categories that are subdivided level by level, wherein each level of categories is multiple; Figure 2 As shown, the method further includes:

[0135] Step 201: using a semantic similarity measurement method, calculate the semantic similarity between at least two final-level categories to obtain a first POI category semantic similarity sequence;

[0136] Here, the semantic similarity measurement methods include distribution-based semantic similarity measurement and information content-based semantic similarity measurement:

[0137] The calculation expression of the first distribution-based semantic similarity measurement method is as follows:

[0138]

[0139] t icsDefined as the nearest common parent class of class t1 and class t2. N1 is the distance from t1 to t ics The shortest path N2 is from t2 to t ics The shortest path of t ics The shortest path to the root node.

[0140] The calculation expression of the second distribution-based semantic similarity measurement method is as follows:

[0141]

[0142] Where D is the maximum depth of the classification hierarchy and N is the shortest path between types t1 and t2.

[0143] The calculation expression of the first semantic similarity measurement method based on information content is as follows:

[0144]

[0145] The second semantic similarity measurement method based on information content is calculated as follows:

[0146]

[0147] Among them, IC is the information content corresponding to each node type, t ics Defined as the nearest common parent of classes t1 and t2.

[0148] Here, based on the two metrics and their corresponding two calculation formulas, four evaluation methods are used. Therefore, these four evaluation methods can be combined to evaluate the original hierarchical structure derived from the POI classification system. Specifically, at least two POI categories belonging to the same final category are selected and their semantic similarity is calculated to obtain four sets of results. The Kendall consistency coefficient is used to evaluate the consistency of the rankings of the four metrics. When the average Kendall coefficient is high, it indicates that the similarity rankings obtained by these metrics are almost completely consistent. Based on the four sets of calculation results after evaluation, the semantic similarity sequence of the first POI category is obtained.

[0149] Step 202: Based on the training results of the word vector training model, the semantic similarity between the POI data pairs that meet the contextual relationship is calculated to obtain a second POI category semantic similarity sequence;

[0150] Here, through the word vector training model, we can learn the word vectors of all POI data. Their semantic relationship and semantic similarity are reflected in the relationship between their word embeddings. The semantic correlation between the word vectors of POI data can be calculated using cosine similarity. The cosine similarity formula is as follows:

[0151]

[0152] Here, Sim(x,y) represents the semantic similarity between the word vectors x and y of the POI data, with the value of semantic similarity ranging from 1 to -1. For each target POI data type, we can also conveniently calculate its semantic similarity with other POI data types using the cosine similarity formula. Based on the semantic similarity, we can also obtain the semantic similarity sequence with the second POI type.

[0153] Step 203: obtaining a first semantic similarity by weighted calculation based on the first POI semantic similarity sequence;

[0154] Step 204: Obtain a second semantic similarity by weighted calculation based on the second POI semantic similarity sequence;

[0155] Here, for each target POI data category, a hierarchical semantic similarity measurement method can be used to obtain the most similar first POI category semantic similarity sequence. The semantic similarity between the word vectors of the POI data can also be used to obtain the second POI category semantic similarity sequence. The data in these two sequences are weighted and summed to obtain the first and second semantic similarities.

[0156] Step 205: Based on the comparison result of the first semantic similarity and the second semantic similarity, adjust the training parameters of the word vector training model, where the training parameters include at least one of the following: vector dimension and number of training rounds.

[0157] Here, the goal of word vector training can be to make the second semantic similarity calculated based on the word vector after model training closer to the first semantic similarity, and the settings of the vector dimension and the number of training rounds in the word vector training model are positively correlated with the accuracy of the word vector obtained by training to a certain extent. Therefore, based on the comparison results of the first semantic similarity and the second semantic similarity, adjusting the training parameters of the word vector training model can obtain more accurate word vectors.

[0158] Here, the vector dimension of the word vector training model is set in the range of 40 to 100, the step size is 10, and the number of model training rounds is 0 to 200. During model training, you can set the target detection evaluation index and draw it as follows Figure 7 The MAP curve shown in the figure shows the optimal word embedding dimension for the entire training process. The MAP score increases with the number of model training rounds. In the later stages of training, the MAP growth trend slows, indicating that the accuracy of the word embeddings is gradually improving and stabilizing.

[0159] Here, in the early stage of training, the model converges quickly and then gradually stabilizes. When the model training loss curve finally stabilizes, the model training is completed and the model training effect is good. The loss curve is as follows Figure 8 shown.

[0160] After the word embedding model is trained, the word embedding ranking can be obtained to test the effectiveness of the model training. The semantic similarity of the POI types is calculated based on the word embeddings obtained. The calculation results are used to determine whether the semantic similarity of the POI types is captured.

[0161] In order to implement the method of the embodiment of the present application, the embodiment of the present application also provides a geographic functional area identification device, which corresponds to the above-mentioned geographic functional area identification method, and each step in the above-mentioned geographic functional area identification method embodiment is also fully applicable to the embodiment of the present geographic functional area identification device.

[0162] like Figure 3 The device shown includes: a data acquisition module 301, a sub-region division module 302, a data pair determination module 303, a vector generation module 304, a word vector training module 305, a partition vector calculation module 306 and a cluster partition module 307, wherein the data acquisition module 301 is used to obtain the road layer data of the area to be identified and the point of interest POI data in the area to be identified; the sub-region division module 302 is used to divide the area to be identified into at least two sub-regions based on the road layer data; the data pair determination module 303 is used to determine the POI data pairs that meet the contextual relationship based on the spatial distance and distance threshold between each POI data. ; The vector generation module 304 is used to generate a vector group of POI data pairs based on the feature information of the POI data, and the vector group includes the vectors corresponding to the two POI data in the POI data pair; the word vector training module 305 is used to input each vector group into the word vector training model to learn and obtain the word vector of each POI data; the partition vector calculation module 306 is used to weightedly calculate the word vector of the POI data in each sub-region based on the correspondence between the POI data and the sub-region, and obtain the partition vector of each sub-region; the cluster partitioning module 307 is used to perform cluster analysis on the partition vector based on the set number of clusters to determine the functional area category to which each sub-region belongs.

[0163] In some embodiments, the POI data includes geographic location information. Based on the spatial distance and distance threshold between the POI data, POI data pairs that meet the contextual relationship are determined. The data pair determination module 303 is specifically configured to:

[0164] Calculate the spatial distance between any two POI data based on the geographic location information of each POI data;

[0165] Pairs of POI data with a spatial distance less than or equal to a distance threshold are counted as POI data pairs that meet the contextual relationship.

[0166] In some embodiments, a vector group of POI data pairs is generated based on feature information of the POI data. The vector generation module 304 is specifically configured to:

[0167] Encode the characteristic information of the POI data to generate a vector corresponding to the POI data, wherein the characteristic information includes at least: geographic location information and POI multi-level categories;

[0168] Based on the correspondence between POI data and vectors, and the contextual relationship between POI data pairs, a vector group of POI data pairs is generated.

[0169] In some embodiments, the word vector training module 305 is further configured to:

[0170] The word vectors obtained by the word vector training model are mapped into a two-dimensional semantic space, and POI types belonging to the same top-level category are rendered with the same color to intuitively display the relationship between POI classes.

[0171] In some embodiments, the POI data includes at least two levels of categories that are subdivided level by level, wherein each level of categories has multiple levels, and the clustering and partitioning module 307 is specifically configured to:

[0172] Based on the clusters of cluster analysis, the density data of POI data of each first-level category in each sub-area is calculated. The density value is used to reflect the weight value of POI data of each first-level category in the cluster;

[0173] Based on the clusters of cluster analysis, the enrichment factor of POI data of each first-level category in the sub-region is calculated. The enrichment factor is used to reflect the enrichment degree of POI data of each first-level category in the cluster;

[0174] Accordingly, the functional area category to which each sub-area belongs is determined based on the clusters obtained through cluster analysis. The cluster partitioning module 307 is further configured to:

[0175] For each sub-region, the number of POI data in the cluster and the density weight value and / or enrichment factor of the POI data are weighted and summed, and the category of the cluster with the largest sum is selected as the functional area category to which the corresponding sub-region belongs.

[0176] In some embodiments, the cluster partitioning module 307 is further configured to:

[0177] Each sub-area is displayed in a partitioned manner based on the functional area category to which it belongs.

[0178] In actual application, the data acquisition module 301, sub-region division module 302, data pair determination module 303, vector generation module 304, word vector training module 305, partition vector calculation module 306, and cluster partition module 307 can be implemented by a processor in the geographic functional area identification device. Of course, the processor needs to run a computer program in the memory to implement its functions.

[0179] It should be noted that the geographic functional area identification device provided in the above embodiment only uses the division of the above-mentioned program modules as an example to illustrate the geographic functional area identification. In actual applications, the above-mentioned processing can be assigned to different program modules as needed, that is, the internal structure of the device can be divided into different program modules to complete all or part of the above-mentioned processing. In addition, the geographic functional area identification device provided in the above embodiment and the geographic functional area identification method embodiment are based on the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0180] Based on the hardware implementation of the above program modules, and in order to implement the method of the embodiment of the present application, the embodiment of the present application also provides a geographic functional area identification device. Figure 4 Only the exemplary structure of the geographic functional area identification device is shown, not the entire structure, and can be implemented as needed. Figure 4 Partial or complete structure shown.

[0181] like Figure 4 As shown, the geographic functional area identification device 400 provided in the embodiment of the present application includes: at least one processor 401, a memory 402, a user interface 403 and at least one network interface 404. The various components in the geographic functional area identification device 400 are coupled together through a bus system 405. It can be understood that the bus system 405 is used to achieve connection and communication between these components. In addition to including a data bus, the bus system 405 also includes a power bus, a control bus and a status signal bus. However, for the sake of clarity, Figure 4 Various buses are labeled as bus system 405 .

[0182] The user interface 403 may include a display, a keyboard, a mouse, a trackball, a click wheel, keys, buttons, a touch pad or a touch screen.

[0183] The memory 402 in the embodiment of the present application is used to store various types of data to support the operation of the geographic functional area recognition device. Examples of such data include: any computer program used to operate on the geographic functional area recognition device.

[0184] The geographic functional area identification method disclosed in the embodiments of the present application can be applied to or implemented by processor 401. Processor 401 may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the geographic functional area identification method can be completed by hardware integrated logic circuits in processor 401 or by software instructions. The aforementioned processor 401 can be a general-purpose processor, a digital signal processor (DSP), or other programmable logic device, discrete gate or transistor logic device, discrete hardware component, etc. Processor 401 can implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of the present application. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the embodiments of the present application can be directly implemented and executed by a hardware decoding processor, or by a combination of hardware and software modules in the decoding processor. The software module can be located in a storage medium located in memory 402. Processor 401 reads the information in memory 402 and, in conjunction with its hardware, completes the steps of the geographic functional area identification method provided in the embodiments of the present application.

[0185] In an exemplary embodiment, the geographic functional area identification device can be implemented by one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontrollers (MCUs), microprocessors, or other electronic components to execute the aforementioned method.

[0186] It is understood that the memory 402 can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memories. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a magnetic random access memory (FRAM), a flash memory, a magnetic surface memory, an optical disk, or a compact disc read-only memory (CD-ROM); the magnetic surface memory can be a magnetic disk or a magnetic tape. The volatile memory can be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static random access memory (SRAM), synchronous static random access memory (SSRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct rambus random access memory (DRRAM).The memories described in the embodiments of this application are intended to include, but are not limited to, these and any other suitable types of memories.

[0187] In an exemplary embodiment, the present application also provides a storage medium, namely, a computer storage medium, which may be a computer-readable storage medium. For example, the storage medium 402 may store a computer program. The computer program may be executed by the processor 401 of the geographic functional area identification device to complete the steps of the method of the present application. The computer-readable storage medium may be a memory such as a ROM, PROM, EPROM, EEPROM, Flash Memory, magnetic surface storage, optical disk, or CD-ROM.

[0188] It should be noted that: "first", "second", etc. are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0189] In addition, the technical solutions described in the embodiments of the present application can be arbitrarily combined without conflict.

[0190] The above are only specific embodiments of the present application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by any person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A method for identifying geographical functional areas, characterized in that: include: Acquire road layer data of the area to be identified and point of interest (POI) data within the area to be identified; Dividing the area to be identified into at least two sub-areas based on the road layer data; Determining POI data pairs that meet the contextual relationship based on the spatial distance and the distance threshold between the POI data; Generating a vector group of the POI data pair based on feature information of the POI data, the vector group including vectors corresponding to the two POI data in the POI data pair; Inputting each of the vector groups into a word vector training model to learn and obtain the word vector of each of the POI data; Based on the correspondence between the POI data and the sub-regions, weightedly calculating the word vector of the POI data in each sub-region to obtain a partition vector of each sub-region; Performing cluster analysis on the partition vector based on a set number of clusters, and determining the functional area category to which each sub-area belongs according to the cluster clusters obtained from the cluster analysis; The POI data includes: at least two levels of categories that are subdivided level by level, wherein each level of categories has multiple levels; the method further includes: Using a semantic similarity measurement method, the semantic similarity between at least two final-level categories is calculated to obtain a first POI category semantic similarity sequence; Based on the training results of the word vector training model, the semantic similarity between the POI data pairs that meet the contextual relationship is calculated to obtain a second POI category semantic similarity sequence; A first semantic similarity is obtained by weighted calculation based on the first POI category semantic similarity sequence; A second semantic similarity is obtained by weighted calculation based on the second POI category semantic similarity sequence; Adjusting training parameters of the word vector training model based on a comparison result between the first semantic similarity and the second semantic similarity, the training parameters including at least one of the following: vector dimension and number of training rounds; The POI data includes: at least two levels of categories that are subdivided level by level, wherein each level of categories has multiple levels; the method further includes: Calculating density data of the POI data of each first-level category in each of the sub-areas based on the clusters obtained by the cluster analysis, wherein the density data is used to reflect the density weight value of the POI data of each first-level category in the cluster; Calculating, based on the clusters obtained from the cluster analysis, an enrichment factor of the POI data of each first-level category in the sub-region, wherein the enrichment factor is used to reflect the degree of enrichment of the POI data of each first-level category in the cluster; Accordingly, determining the functional area category to which each sub-area belongs based on the cluster clusters obtained by the cluster analysis includes: For each sub-region, a weighted sum is taken of the number of POI data in the cluster and the density weight value and / or enrichment factor of the POI data, and the category of the cluster with the largest sum is selected as the functional area category to which the corresponding sub-region belongs.

2. The method according to claim 1, characterized in that The POI data includes: geographic location information, and determining POI data pairs that meet the contextual relationship based on the spatial distance and distance threshold between the POI data includes: Calculate the spatial distance between any two POI data based on the geographic location information of each POI data; Pairs of POI data whose spatial distance is less than or equal to a distance threshold are counted as POI data pairs that meet the contextual relationship.

3. The method according to claim 1, characterized in that Generating the vector group of the POI data pair based on the feature information of the POI data includes: Encoding characteristic information of the POI data to generate a vector corresponding to the POI data, wherein the characteristic information includes at least geographic location information and POI multi-level categories; Based on the correspondence between the POI data and the vectors, and the contextual relationship of the POI data pairs, a vector group of the POI data pairs is generated.

4. The method according to claim 1, wherein Inputting each of the vector groups into a word vector training model to learn and obtain the word vectors of each of the POI data includes: Taking each of the vector groups as a single training sample, generating a training sample set including each of the vector groups, and taking the training sample set as input of the word vector training model; Each of the vectors is embedded in the word through the word vector training model to obtain the word vector of each of the POI data.

5. The method according to claim 1, further comprising: Each sub-region is displayed in a partitioned manner based on the functional area category to which each sub-region belongs.

6. A device for identifying geographical functional areas, characterized in that: include: A data acquisition module is used to acquire road layer data of the area to be identified and point of interest (POI) data in the area to be identified; A sub-area division module, configured to divide the area to be identified into at least two sub-areas based on the road layer data; A data pair determination module, configured to determine a POI data pair that meets the contextual relationship based on the spatial distance and the distance threshold between the POI data; a vector generation module, configured to generate a vector group of the POI data pair based on feature information of the POI data, wherein the vector group includes vectors corresponding to the two POI data in the POI data pair; A word vector training module, configured to input each of the vector groups into a word vector training model to learn and obtain the word vectors of each of the POI data; A partition vector calculation module is used to calculate the word vector of the POI data in each sub-region based on the corresponding relationship between the POI data and the sub-region, so as to obtain the partition vector of each sub-region; A clustering and partitioning module, configured to perform cluster analysis on the partitioning vectors based on a set number of clusters, and determine the functional area category to which each sub-area belongs; The POI data includes: at least two levels of categories that are subdivided step by step, wherein each level of categories has multiple levels; the word vector training module is further used to: A semantic similarity measurement method is used to calculate the semantic similarity between at least two final-level categories to obtain a first POI-class semantic similarity sequence; based on the training results of the word vector training model, the semantic similarity between the POI data pairs that meet the contextual relationship is calculated to obtain a second POI-class semantic similarity sequence; a first semantic similarity is obtained by weighted calculation based on the first POI-class semantic similarity sequence; a second semantic similarity is obtained by weighted calculation based on the second POI-class semantic similarity sequence; based on the comparison result of the first semantic similarity and the second semantic similarity, the training parameters of the word vector training model are adjusted, the training parameters including at least one of the following: vector dimension and number of training rounds; The POI data includes: at least two levels of categories that are subdivided level by level, wherein each level of categories has multiple levels; the clustering and partitioning module is specifically used to: Based on the clusters of the cluster analysis, calculating density data of the POI data of each first-level category in each of the sub-areas, the density data being used to reflect the density weight value of the POI data of each first-level category in the cluster; based on the clusters of the cluster analysis, calculating enrichment factors of the POI data of each first-level category in the sub-areas, the enrichment factors being used to reflect the enrichment degree of the POI data of each first-level category in the cluster; Correspondingly, the cluster partitioning module is also used to: for each of the sub-areas, perform weighted summation on the number of POI data in the cluster cluster and the density weight value and / or enrichment factor of the POI data, and select the category to which the cluster cluster with the largest sum value belongs as the functional area category to which the corresponding sub-area belongs.

7. A geographical functional area identification device, characterized in that: include: A processor and a memory for storing a computer program capable of being executed on the processor, wherein: The processor is configured to execute the steps of the method according to any one of claims 1 to 5 when running a computer program.

8. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images