Cross-city point of interest distribution data migration prediction method and system

By gridding and clustering of urban POI data, selecting source cities with high similarity for transfer learning, combining TrAdaBoost and CART decision tree models, the problem of similarity neglect in urban transfer learning is solved, and cross-city interest points are improved Prediction accuracy of distributed data.

CN115422993BActive Publication Date: 2024-11-08BEIJING JIAOTONG UNIV
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202210928411.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-08-03
Publication Date
2024-11-08
Estimated Expiration
2042-08-03

AI Technical Summary

Technical Problem

In the study of urban transfer learning, the existing technology focuses on the problem of knowledge differences, but ignores the consideration of urban similarity, resulting in inaccurate prediction results.

Method used

By obtaining the city POI distribution data, performing grid processing and K-means clustering, selecting the source city with similarity to the target city for transfer learning, using the TrAdaBoost model for prediction, setting thresholds to filter noise data, and using the CART decision tree model for Multi-categorical prediction.

Benefits of technology

Considering urban similarity from the source, the accuracy of data migration prediction of cross-city point-of-interest distribution data is improved and the stability of the model is obtained, and more accurate prediction results are obtained.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115422993B_ABST
    Figure CN115422993B_ABST
Patent Text Reader

Abstract

The present invention provides a method and system for predicting the migration of point of interest distribution data across cities, which belongs to the field of target recognition technology, and obtains city POI distribution data; grids the obtained city POI distribution data, takes the number of POI dimensions as the feature of the grid, and forms a city grid POI data set, that is, a feature matrix of the city; based on the city grid POI data set, clusters using the K-means clustering method to obtain the city POI distribution clustering result; based on the city POI distribution clustering result, selects a city in the same category as the target city as the source city, and uses a pre-trained migration prediction model to perform migration learning prediction to obtain the point of interest distribution result of the target city. The present invention uses city clustering to solve the problem from the source from the perspective of similarity discrimination, and cooperates with sample-based migration technology to obtain more accurate prediction results.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of target recognition, and in particular to a method and system for predicting the migration of inter-city point of interest distribution data. Background Art

[0002] In recent years, scholars have conducted a lot of research on non-time series data transfer learning such as cross-city distribution, and have achieved some results. Wei et al. (2016) proposed a FLORAL transfer method, which realizes knowledge transfer between domains based on urban multimodal data. Wang et al. (2018) first applied urban transfer learning to urban spatiotemporal data prediction, and used passenger flow data as an example to verify the accuracy of the prediction results. Guo (2018) et al., based on the recommendation algorithm, combined the data of the target hotel and all hotels to recommend the location of the target hotel in the target city. Wang et al. (2019) proposed the CoTrans model, which transfers the knowledge learned from taxi trajectories. Li et al. (2020) solved the problem of uneven Poisson distribution of data through resampling and deep neural networks. For the problem of regional distribution prediction of shared bicycles, there is a prediction model CoFa-GeoConv based on joint factor decomposition and convolutional neural network. The model mainly includes two core steps: 1) The factor decomposition algorithm is used to act on the feature space of the original city and the target city at the same time to reduce the feature dimension and obtain the latent variable expression of the original feature; 2) The latent feature expression of the city grid and the grid neighbors is used as the input of the convolutional neural network. The model is trained based on the known bicycle distribution data in the source city. When applied in the target city, the number of bicycles in each area of ​​the target city is predicted in turn, and then the overall bicycle distribution of the target city is calculated.

[0003] The above-mentioned studies on urban non-time series data have different focuses, but there are also problems that need to be solved urgently: when conducting research on urban transfer learning, they all focus on solving the problem of knowledge differences, but ignore the source problem - the consideration of urban similarities. Summary of the invention

[0004] The object of the present invention is to provide a method and system for predicting the migration of cross-city POI distribution data, so as to solve at least one technical problem existing in the above-mentioned background technology.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] In one aspect, the present invention provides a method for predicting cross-city point of interest distribution data migration, comprising:

[0007] Get city POI distribution data;

[0008] The obtained city POI distribution data is gridded, and the number of each POI dimension is used as the feature of the grid to form a city grid POI dataset, that is, the city's feature matrix;

[0009] Based on the urban grid POI dataset, the K-means clustering method is used to cluster and obtain the urban POI distribution clustering results;

[0010] Based on the city POI distribution clustering results, cities in the same category as the target city are selected as source cities, and the pre-trained migration prediction model is used to perform transfer learning prediction to obtain the POI distribution results of the target city.

[0011] Preferably, when the migration prediction model is trained, the target city POI data is used for training, and the source city POI data is used for prediction to obtain the source city POI data prediction results; a threshold is set, and according to the source city POI data prediction results, a subset of samples with weighted residuals less than the set threshold is selected as the final source city data set, and then the TrAdaBoost algorithm process is performed.

[0012] Preferably, the TrAdaBoost model uses a CART decision tree model as a base learner.

[0013] Preferably, the loss function used in the migration prediction model training is:

[0014] ;

[0015] in, Indicates the number of target city data, Indicates the target city i The predicted value of the group data, Indicates the target city i The true value of the group data, Indicates the number of data heat classification categories, Represents the weight of the data sample.

[0016] In a second aspect, the present invention provides a cross-city point of interest distribution data migration prediction system, comprising:

[0017] Acquisition module, used to obtain city POI distribution data;

[0018] A processing module is used to perform grid processing on the acquired city POI distribution data, taking the number of each POI dimension as the feature of the grid, and forming a city grid POI data set, that is, a feature matrix of the city;

[0019] The clustering module is used to cluster the urban grid POI dataset using the K-means clustering method to obtain the urban POI distribution clustering results;

[0020] The prediction module is used to select cities in the same category as the target city as source cities based on the city POI distribution clustering results, and use the pre-trained migration prediction model to perform transfer learning prediction to obtain the POI distribution results of the target city.

[0021] In a third aspect, the present invention provides a non-transitory computer-readable storage medium, which is used to store computer instructions. When the computer instructions are executed by a processor, the cross-city point of interest distribution data migration prediction method as described above is implemented.

[0022] In a fourth aspect, the present invention provides a computer program product, including a computer program, which, when executed on one or more processors, is used to implement the cross-city point of interest distribution data migration prediction method as described above.

[0023] In a fifth aspect, the present invention provides an electronic device, comprising: a processor, a memory and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory so that the electronic device executes instructions for implementing the cross-city point of interest distribution data migration prediction method as described above.

[0024] Terminology explanation:

[0025] POI: It is the abbreviation of "Point of Interest", which can be translated into Chinese as "point of interest". In a geographic information system, a POI can be a house, a shop, a mailbox, a bus stop, etc.

[0026] The beneficial effects of the present invention are as follows: by using city clustering, the problem is solved from the perspective of similarity judgment, that is, from the source, and by combining with sample-based migration technology, a more accurate prediction result is obtained.

[0027] Additional advantages of the present invention will be more clearly given in the following description or learned through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0029] Figure 1 This is a schematic diagram of the clustering result of the city POI data according to an embodiment of the present invention.

[0030] Figure 2 This is an overall structural diagram of the Two-step TrAdaBoost-Multi model described in an embodiment of the present invention.

[0031] Figure 3 This is a structural diagram of the cross-city POI distribution data migration prediction framework described in an embodiment of the present invention.

[0032] Figure 4 Schematic diagram of the relationship between half entropy, Gini index and classification error rate in two-category classification described in an embodiment of the present invention.

[0033] Figure 5 A schematic diagram of the basic structure of the integrated learning described in an embodiment of the present invention.

[0034] Figure 6 The figure is a schematic diagram of the basic structure of the AdaBoost algorithm described in an embodiment of the present invention. DETAILED DESCRIPTION

[0035] The embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements with the same or similar functions. The embodiments described below by the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be interpreted as limiting the present invention.

[0036] It should be understood by those skilled in the art that unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those skilled in the art to which this invention belongs.

[0037] It should also be understood that terms, such as those defined in commonly used dictionaries, should be understood to have a meaning consistent with that in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless as defined herein.

[0038] Those skilled in the art will appreciate that, unless otherwise stated, the singular forms "a", "an", "said" and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements and / or groups thereof.

[0039] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification and the features of different embodiments or examples, unless they are contradictory.

[0040] To facilitate understanding of the present invention, the present invention is further explained below with reference to specific embodiments in conjunction with the accompanying drawings, and the specific embodiments do not constitute a limitation on the embodiments of the present invention.

[0041] Those skilled in the art should understand that the drawings are merely schematic diagrams of embodiments, and the components in the drawings are not necessarily necessary for implementing the present invention.

[0042] Example 1

[0043] This embodiment 1 provides a cross-city point of interest distribution data migration prediction system, including:

[0044] Acquisition module, used to obtain city POI distribution data;

[0045] A processing module is used to perform grid processing on the acquired city POI distribution data, taking the number of each POI dimension as the feature of the grid, and forming a city grid POI data set, that is, a feature matrix of the city;

[0046] The clustering module is used to cluster the urban grid POI dataset using the K-means clustering method to obtain the urban POI distribution clustering results;

[0047] The prediction module is used to select cities in the same category as the target city as source cities based on the city POI distribution clustering results, and use the pre-trained migration prediction model to perform transfer learning prediction to obtain the POI distribution results of the target city.

[0048] In this embodiment 1, a method for predicting the migration of cross-city POI distribution data is implemented using the above system, including:

[0049] Get city POI distribution data;

[0050] The obtained city POI distribution data is gridded, and the number of each POI dimension is used as the feature of the grid to form a city grid POI dataset, that is, the city's feature matrix;

[0051] Based on the urban grid POI dataset, the K-means clustering method is used to cluster and obtain the urban POI distribution clustering results;

[0052] Based on the city POI distribution clustering results, cities in the same category as the target city are selected as source cities, and the pre-trained migration prediction model is used to perform transfer learning prediction to obtain the POI distribution results of the target city.

[0053] When the migration prediction model is trained, the target city POI data is used for training, and the source city POI data is used for prediction to obtain the source city POI data prediction results; a threshold is set, and according to the source city POI data prediction results, a subset of samples with weighted residuals less than the set threshold is selected as the final source city data set, and then the TrAdaBoost algorithm process is performed.

[0054] The TrAdaBoost model uses the CART decision tree model as the base learner.

[0055] The loss function used in the migration prediction model training is:

[0056] ;

[0057] in, Indicates the number of target city data, Indicates the target city i The predicted value of the group data, Indicates the target city i The true value of the group data, Indicates the number of data heat classification categories, Represents the weight of the data sample.

[0058] Example 2

[0059] like Figures 1 to 6 As shown, this embodiment 2 provides a method for predicting the migration of inter-city POI distribution data, the method comprising the following steps:

[0060] Step S1: Gridding of city POI data

[0061] In this embodiment 2, POI data of 60 cities including Beijing, Shanghai, Xiamen, Baoding, etc. are obtained through the Baidu Map Open Platform API, and POI types include 19 dimensions including food, education, medical care, nature, shopping, beauty, finance, media, hotels, etc. In order to better quantify the city POI data and construct the city POI features, the city POI distribution data is gridded, and the city is divided into 30*30 square grids with a distance of 2km.

[0062] After dividing each city into grids, the data in each grid is counted, and the number of POI dimensions is used as the feature of the grid to form a city grid POI dataset, that is, the city's feature matrix. After that, we will carry out clustering, prediction and other work based on the POI matrix generated for each city.

[0063] Step S2: Clustering of city distribution characteristics

[0064] In order to measure the similarity between city distributions, the city grid POI matrix is ​​used as the feature matrix of each city, and the K-means clustering method is used to cluster the cities. The clusters are defined as , Represents the mean vector of the cluster, then the desired result is to minimize the square error SE, and its mathematical formula is:

[0065] ;

[0066] Among them, the distance measurement method of K-means uses Euclidean distance, which is expressed as follows:

[0067] ;

[0068] DBI is used as an indicator to measure the K-means algorithm. DBI represents the ratio of the inner distance to the outer distance of a cluster. Therefore, the larger the DBI value, the higher the degree of dispersion and the worse the clustering effect. On the contrary, the better the clustering effect. Its mathematical expression is as follows:

[0069] ;

[0070] Finally, the k value is set to 3 to form the city POI distribution clustering result, as shown in Figure 1 As shown in the figure. Based on the clustering results, it is believed that cities of the same category are similar in the overall distribution of POIs, which has a positive impact on the migration effect. Therefore, when performing transfer learning, cities of the same category as the target city will be selected as source cities for transfer learning training, rather than randomly selecting them.

[0071] Step S3: Two-stepTrAdaBoost-Multi migration prediction model

[0072] Through the investigation of the existing transfer learning model, it is found that its migration ideas are mainly divided into two types: migration for sample features and migration for models. The migration of data is to achieve migration between data by learning the similarity of the data itself. City clustering provides more similar data distribution for transfer learning, making it easier for the model to learn the similarities between the two data, reducing the overhead of model fitting, and effectively improving the transfer learning model. Therefore, in response to this problem, the sample-based model is finally selected in this embodiment 2 to carry out research, so as to effectively combine the similarity clustering of urban data with the transfer learning model to form a more complete framework.

[0073] As a method of extracting instances from source data, the TrAdaBoost algorithm aims to extract usable parts from richer source domain data to assist the insufficient target domain data, so that the combination of the two has better results than simply using insufficient target domain data to train the model.

[0074] First, the weight of each data is normalized to make it identically distributed. Then the weak classifier is called. The source domain and target domain data are used as the overall training data, and the process is the same as AdaBoost training weak classifier. Here, the source domain data participates in the training and plays its role. When calculating the error rate, only the error rate of the target domain data is calculated, because the target domain is the real purpose. After that, the weight adjustment rate of each domain data is calculated separately, and the weight of the data is updated. In the target domain, if the data is not correctly classified and it is considered that the data is difficult to distinguish, its weight is increased and its features are further learned, which is consistent with the traditional AdaBoost algorithm. However, when the source domain data is misclassified, it means that it is very different from our target domain data and is useless data, so the weight value is reduced. At the final output, the voting results of the weak classifiers in the second half (i.e., N / 2~N) shall prevail.

[0075] When the source city data and the target city data are more similar, it is obvious that the TrAdaBoost model can use more source city data, and finally achieve more sufficient training and obtain better migration effect. However, when the distribution of city data is very different, the utilization rate of source city data in the training process will be greatly reduced, which also leads to poor training effect. The results of city clustering help us to judge and select data in advance. Therefore, city clustering lays a good foundation for the TrAdaBoost model, and the TrAdaBoost model further extracts the characteristics of the source city and the target city to achieve better prediction effect. It can be said that the two form a good combination effect.

[0076] However, the TrAdaBoost model still has some problems. If there are many invalid samples in the source city data when training begins, and the number of iterations is not well controlled, it will lead to failure to obtain a good model effect. In addition, the TrAdaBoost model only supports binary classification problems, not multi-classification problems. In view of the above problems, in this embodiment 2, the TrAdaBoost model is improved and a Two-stepTrAdaBoost-Multi model is proposed.

[0077] In order to avoid the problem of high sample noise in the source city data, a Two-step process is proposed before the formal TrAdaBoost process. The algorithm pseudo code is shown in Table 1 below:

[0078] Table 1

[0079] ;

[0080] The two steps of Two-step are: the first step is to train the model using the target city data, and use the model to predict the source city data to obtain the prediction results; the second step is to set a threshold, and according to the prediction results of the source city, select the subset whose sample weighted residual is less than the set threshold as the final source city data set, and then perform the TrAdaBoost process. By setting the threshold, the Two-step process can effectively filter out some noise data with large differences in the source domain and improve the model fitting efficiency. In addition, compared with the TrAdaBoost model that can only use the second half of the base learner, the Two-step process also improves the fitting effect of the first half of the base learner and increases the available proportion of the overall base learner.

[0081] In addition, the base learner used in the TrAdaBoost model is a support vector machine model (SVM), which uses quadratic programming when solving support vectors, so the SVM model cannot cope with a large number of samples. Moreover, the traditional SVM only supports binary classification problems, and its effect on solving multi-classification problems is not ideal. Therefore, in this embodiment 2, the CART decision tree model is used to replace the SVM model, and its internal output is adjusted, and a softmax layer is added to support multi-classification. The loss function L of the model is set to:

[0082] ;

[0083] in, Indicates the number of target city data, Indicates the target city i The predicted value of the group data, Indicates the target city i The true value of the group data, Indicates the number of data heat classification categories, Represents the weight of the data sample.

[0084] Since the goal of the model is to have good performance in the target city, the model only focuses on the loss of the target city data. Considering the size relationship between the distribution heat categories, the loss of a single data is set as the difference between the predicted label and the true label divided by the total number of categories to represent its prediction gap. When calculating the total loss, the weight of each data is taken into account, so the loss of a single data is multiplied by its weight, and the loss results of the entire data set are averaged to obtain the total loss of the model. Finally, the Two-step TrAdaBoost-Multi model is formed, and its overall structure is as follows: Figure 2 shown.

[0085] The CART decision tree uses a binary tree instead of a multi-branch tree, and uses the GINI coefficient to reduce the amount of calculation, which simplifies the model and maintains the main advantages of the entropy model. Due to its formation process, the decision tree itself has unstable characteristics, that is, when the sample changes, even if the change is subtle, the decision process of the tree will no longer adapt. In addition, the problem of determining the optimal solution of the decision tree is also NP-hard, because the problem of local optimality is very easy to occur in the determination process. The above problems are complementary to the idea of ​​ensemble learning and are improved by making the tree form a "forest". Therefore, TrAdaBoost and decision trees can form an organic combination.

[0086] Finally, the above three steps are combined to form the entire cross-city POI distribution data migration prediction framework, and its overall structure is as follows: Figure 3 shown.

[0087] In this embodiment 2, the CART decision tree algorithm is widely used in many fields. It is improved based on the structure of ID3. In addition, compared with the defects of C4.5, such as the multi-tree structure of the model and the entropy measurement method, which will consume a lot of calculations, the CART algorithm has also optimized it: the entropy measurement is changed to the Gini coefficient measurement, and the multi-tree is simplified to a binary tree. In addition, we can also see from the name that CART supports both classification and regression. Here, only the classification problem is considered in this embodiment 2.

[0088] The most important step of a decision tree is splitting, which is the process of generating subtrees downward. The goal of this process is to further purify the data so that the output of the decision tree is closer to reality. For CART, the GINI coefficient and sample variance are used to measure the degree of node purification in classification and regression problems respectively. As the degree of node purification gradually increases, the node will also obtain a better fitting effect.

[0089] The calculation formula of GINI value is:

[0090] ;

[0091] in, represents the number of categories, It corresponds to the possibility of the i-th category.

[0092] ;

[0093] In the formula, represents the number of items belonging to this category, and |S| represents the total amount of data. Assume that the dataset S is divided into two parts based on feature D. The GINI value of the divided dataset is:

[0094] ;

[0095] Since CART generates a binary tree, there are only two possible eigenvalues ​​when selecting features, thus ensuring that the number of child nodes does not exceed two. Using the GINI coefficient to replace information entropy greatly reduces the computational complexity, but there will also be some differences. Figure 4 It can be seen that the two curves are very close, with only a significant difference at the 45° angle.

[0096] Information gain involves a large number of logarithmic operations, and its complexity is much higher than that of multiplication and division operations. Therefore, using the GINI coefficient as a substitute can effectively reduce the computational complexity and greatly simplify the algorithm's operation process, while not completely losing the advantages of the original algorithm. Therefore, its advantages outweigh its disadvantages.

[0097] When pruning a tree, CART uses a post-pruning method. After generating the tree, all pruning results are obtained, and then a set of validation data is used to cross-validate the results, and finally the pruning result with the smallest loss is selected. The loss function of this process is: Loss(T) = CART(T) + α|T|; where T represents the subtree, CART(T) represents the prediction error, and α represents the regularization parameter. When no regularization operation is performed, that is, α is 0, the original CART tree is our optimal result. As α increases, the stronger the pruning operation, the smaller the optimal decision tree obtained will be compared to the initial state. For each α, there is a subtree that minimizes the Loss, which corresponds to the optimal result.

[0098] In this embodiment 2, the AdaBoost algorithm is an ensemble learning algorithm. Its concept is to integrate multiple weakly supervised models to form a strong supervised model with better overall performance. That is to say, when multiple decision makers make decisions together, even if a few decision makers make wrong decisions, the majority of other decision makers can correct them without causing errors in the final result. Figure 5 The following is the basic structure diagram of integrated learning.

[0099] Boosting algorithm is a form of ensemble learning, which means "enhancement" in Chinese. It achieves overall optimization through gradual optimization. AdaBoost is one of the most classic algorithms.

[0100] Compared with Boosting, AdaBoost adds Adaptive, i.e., adaptive. The basic structure of AdaBoost is as follows Figure 6 As shown in the figure, its characteristics are: for a sample data, if the current weak classifier does not correctly distinguish the category of the sample, then in the next round of classification training of the new weak classifier, the weight corresponding to the sample will be enhanced. Each round of training of the new weak classifier is the overall sample. After the training is completed, the new sample weight will be output, and the weight belonging to the weak classifier will also be output. The weight corresponds to the weak classifier's right to speak when participating in the overall decision. Finally, the model will iterate until the termination requirement is met.

[0101] The mathematical representation of the ensemble learner is shown below, where G(x) represents the supervised model. Represents the importance of the weakly supervised model, that is, the weight:

[0102] .

[0103] In this embodiment 2, the TrAdaBoost algorithm is evolved from the AdaBoost algorithm. The AdaBoost algorithm is based on the traditional machine learning application scenario, that is, its basic assumption is that all data involved in model training and prediction have the same data distribution. However, this is completely different from the scenario of transfer learning. Transfer learning technology is precisely to solve the adaptation problem when the distribution of data is different, because data with different distributions is equivalent to large data noise, which will greatly affect the model's fitting of data features, making it difficult for the model's prediction effect to meet expectations.

[0104] Therefore, the central idea of ​​the TrAdaBoost algorithm is to use weights to influence each data, and further adjust its influence on the training process of the entire model by adjusting the size of the data weight, so that the model can achieve a satisfactory fitting effect.

[0105] When training a new base learner in each iteration, if the base learner does not correctly distinguish a source domain data, it is considered that the data may have a large gap with the target data, so the weight corresponding to the data needs to be adjusted to make it smaller. Multiply the sample by a positive weight less than 1, and adjust the weight coefficient to achieve a lower impact of the sample on the classification model in subsequent iterations. After a series of iterations, the weight of the source domain that is more similar to the target domain data and is helpful will be higher than other data, while the weight of other data will be relatively lower. The main steps of the algorithm are as follows:

[0106] If the number of source domain and target domain data is a and b respectively, when initializing the data weight, the weight is often set to:

[0107] ;

[0108] At the same time, a parameter value θ is set to update the weight. If the number of base learners is E, the formula is:

[0109] ;

[0110] After training, we get a base learner ( ), we need to calculate the error rate of the base learner on the target domain:

[0111] ;

[0112] The value of θ is further updated by the error rate:

[0113] ;

[0114] The set θ value will be used to update the weight:

[0115] ;

[0116] Based on the above algorithm process, the process of adjusting the data weight is to multiply the data by , where θ is a positive value less than 1. After several iterations, the source domain data will obtain a more appropriate weight according to the similarity with the target domain data, that is, the degree of help to the target data, and finally achieve a positive migration effect.

[0117] Example 3

[0118] Embodiment 3 of the present invention provides a non-transitory computer-readable storage medium, wherein the non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, a method for predicting the migration of cross-city point of interest distribution data is implemented. The method includes:

[0119] Get city POI distribution data;

[0120] The obtained city POI distribution data is gridded, and the number of each POI dimension is used as the feature of the grid to form a city grid POI dataset, that is, the city's feature matrix;

[0121] Based on the urban grid POI dataset, the K-means clustering method is used to cluster and obtain the urban POI distribution clustering results;

[0122] Based on the city POI distribution clustering results, cities in the same category as the target city are selected as source cities, and the pre-trained migration prediction model is used to perform transfer learning prediction to obtain the POI distribution results of the target city.

[0123] Example 4

[0124] Embodiment 4 of the present invention provides a computer program (product), including a computer program, wherein when the computer program is executed on one or more processors, the computer program is used to implement a method for predicting the migration of distribution data of points of interest across cities, the method comprising:

[0125] Get city POI distribution data;

[0126] The obtained city POI distribution data is gridded, and the number of each POI dimension is used as the feature of the grid to form a city grid POI dataset, that is, the city's feature matrix;

[0127] Based on the urban grid POI dataset, the K-means clustering method is used to cluster and obtain the urban POI distribution clustering results;

[0128] Based on the city POI distribution clustering results, cities in the same category as the target city are selected as source cities, and the pre-trained migration prediction model is used to perform transfer learning prediction to obtain the POI distribution results of the target city.

[0129] Example 5

[0130] Embodiment 5 of the present invention provides an electronic device, including: a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to execute instructions for implementing a method for predicting the migration of distribution data of points of interest across cities, the method including:

[0131] Get city POI distribution data;

[0132] The obtained city POI distribution data is gridded, and the number of each POI dimension is used as the feature of the grid to form a city grid POI dataset, that is, the city's feature matrix;

[0133] Based on the urban grid POI dataset, the K-means clustering method is used to cluster and obtain the urban POI distribution clustering results;

[0134] Based on the city POI distribution clustering results, cities in the same category as the target city are selected as source cities, and the pre-trained migration prediction model is used to perform transfer learning prediction to obtain the POI distribution results of the target city.

[0135] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may 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 code.

[0136] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0137] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0138] These computer program instructions can also be loaded onto a computer or other programmable data processing device, and a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0139] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative work on the basis of the technical solution disclosed in the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for predicting the migration of inter-city POI distribution data, characterized in that: include: Get city POI distribution data; The obtained city POI distribution data is gridded, and the number of each POI dimension is used as the feature of the grid to form a city grid POI dataset, that is, the city's feature matrix; Based on the urban grid POI dataset, the K-means clustering method is used to cluster and obtain the urban POI distribution clustering results; Based on the city POI distribution clustering results, select cities in the same category as the target city as the source city, and use the pre-trained migration prediction model to perform transfer learning prediction to obtain the distribution results of the POIs of the target city. The migration prediction model is a Two-step TrAdaBoost-Multi model obtained by improving the TrAdaBoost model. When the migration prediction model is trained, the target city POI data is used for training, and the source city POI data is used for prediction to obtain the source city POI data prediction result; A threshold is set, and according to the prediction results of the source city POI data, a subset of samples with weighted residuals less than the set threshold is selected as the final source city data set, and then the TrAdaBoost algorithm process is performed; The base learner support vector machine model SVM in the TrAdaBoost model is adjusted to use the CART decision tree model as the base learner, and the internal output of the CART decision tree model is adjusted to add a softmax layer to support multi-classification, which is used to solve the problem of large number of samples and multi-classification; The steps of TrAdaBoost algorithm are as follows: If the number of source domain and target domain data is a and b respectively, when initializing the data weight, the weight is often set to: At the same time, a parameter value θ is set to update the weight. If the number of base learners is E, the formula is: After training, we get a base learner ( ), calculate the error rate of the base learner on the target domain: The value of θ is further updated by the error rate: The set θ value will be used to update the weight: Where θ is a positive value less than 1. After several iterations, the source domain data will be weighted according to the degree of similarity with the target domain data and the degree of help to the target data, thus achieving a positive migration effect. The loss function used in the migration prediction model training is: in, Indicates the number of target city data, Indicates the target city i The predicted value of the group data, Indicates the target city i The true value of the group data, Represents the weight of the data sample.

2. A cross-city point of interest distribution data migration prediction system, characterized in that: The method for implementing the cross-city point of interest distribution data migration prediction method described in claim 1 comprises: Acquisition module, used to obtain city POI distribution data; A processing module is used to perform grid processing on the acquired city POI distribution data, taking the number of each POI dimension as the feature of the grid, and forming a city grid POI data set, that is, a feature matrix of the city; The clustering module is used to cluster the urban grid POI dataset using the K-means clustering method to obtain the urban POI distribution clustering results; The prediction module is used to select cities in the same category as the target city as source cities based on the city POI distribution clustering results, and use the pre-trained migration prediction model to perform transfer learning prediction to obtain the POI distribution results of the target city.

3. A non-transitory computer-readable storage medium, characterized in that: The non-transitory computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by the processor, the cross-city point of interest distribution data migration prediction method described in claim 1 is implemented.

4. A computer program product, characterized in that It comprises a computer program, which, when running on one or more processors, is used to implement the cross-city point of interest distribution data migration prediction method as claimed in claim 1.

5. An electronic device, characterized in that: include: A processor, a memory and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to execute instructions for implementing the cross-city point of interest distribution data migration prediction method described in claim 1.

Citation Information

Patent Citations

  • Data verification method based on multi-source transfer learning

    CN108549907A

  • Cross-city traffic flow joint prediction method based on deep migration learning

    CN110148296A

  • Cross-modal data prediction method based on transfer learning

    CN112863182A

  • Heterogeneous graph-based multi-mode traffic operation situation association rule mining method

    CN114238491A