Map data fusion method, device, electronic device and storage medium

Abstract

The embodiment of the present application discloses a map data fusion method, device, electronic device and storage medium, the method comprising: obtaining a POI within a preset distance range from a POI to be compared as a recall POI; determining the statistical features of the inherent attributes of the POI to be compared and the recall POI; determining the semantic similarity between the text attributes of the POI to be compared and the text attributes of the recall POI through a deep semantic model; determining the implicit features between the semantic similarity and the inherent attributes of the POI to be compared and the recall POI through a deep learning model; inputting the statistical features and the implicit features into a logistic regression model to obtain a fusion score of the POI to be compared and the recall POI; and determining the fusion result of the POI to be compared and the recall POI based on the fusion score. The embodiment of the present application combines a logistic regression model and a deep model, having both the memory ability of a traditional model and the generalization ability of a deep model, thereby improving the accuracy of data fusion.

Description

Technical Field

[0001] The embodiments of the present application relate to the technical field of electronic maps, and in particular to a map data fusion method, device, electronic device and storage medium. Background Art

[0002] Due to map collection qualifications or cost issues, many map databases are collected through map capture and data fusion. First, the map data of other manufacturers is captured through technical means, for example, the map data of companies A and B are captured now; then, the map data of companies A and B are merged into one. The most critical part of this process is data fusion, that is, how to identify the POI (Point of Information) data shared by the map data of companies A and B.

[0003] In the prior art, data fusion solutions are mainly rule-based, using artificially set feature rules to determine whether two POIs point to the same POI. The first step is to recall data, which is to select POI data that is suspected to be duplicated with the POI to be compared from the map data set; then the POI points to be compared and the recalled POI points are compared one by one, and the relevant features of each set of POI data are calculated. Then, a score in the range of 0-1 is given through rules or traditional machine learning models, and based on the score, it is determined whether another data set contains the same POI as the POI to be compared.

[0004] The models used in the existing technology are all based on statistics, but the real world is complex. Models using statistical principles may have overfitting problems, which may lead to misjudgment. The data fusion accuracy is not high, and there will still be a lot of duplicate data in the fused map data. Summary of the invention

[0005] The embodiments of the present application provide a map data fusion method, device, electronic device and storage medium, which are helpful to improve the accuracy of data fusion.

[0006] In order to solve the above problems, in a first aspect, an embodiment of the present application provides a map data fusion method, comprising:

[0007] Obtaining POIs within a preset distance range from the POI to be compared as recalled POIs;

[0008] Determining statistical features of inherent attributes of the POI to be compared and the recalled POI;

[0009] Determining the semantic similarity between the text attributes of the POI to be compared and the text attributes of the recalled POI through a deep semantic model;

[0010] Determine the implicit features between the semantic similarity and inherent attributes of the compared POI and the recalled POI through a deep learning model;

[0011] Inputting the statistical features and latent features into a logistic regression model to obtain a fusion score of the POI to be compared and the recalled POI;

[0012] According to the fusion score, a fusion result of the POI to be compared and the recalled POI is determined.

[0013] In a second aspect, an embodiment of the present application provides a map data fusion device, including:

[0014] A recalled POI acquisition module is used to acquire a POI within a preset distance range from the POI to be compared as a recalled POI;

[0015] A statistical feature determination module, used to determine the statistical features of the inherent attributes of the POI to be compared and the recalled POI;

[0016] A semantic similarity determination module, used to determine the semantic similarity between the text attributes of the POI to be compared and the text attributes of the recalled POI through a deep semantic model;

[0017] An implicit feature determination module, used for determining implicit features between semantic similarity and inherent attributes of the POI to be compared and the recalled POI through a deep learning model;

[0018] A fusion module, used for inputting the statistical features and latent features into a logistic regression model to obtain a fusion score of the POI to be compared and the recalled POI;

[0019] The fusion result determination module is used to determine the fusion result of the POI to be compared and the recalled POI according to the fusion score.

[0020] In a third aspect, an embodiment of the present application further provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the map data fusion method described in the embodiment of the present application when executing the computer program.

[0021] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, and when the program is executed by a processor, the steps of the map data fusion method disclosed in the embodiment of the present application are performed.

[0022] The map data fusion method, device, electronic device and storage medium provided in the embodiments of the present application determine the statistical features of the inherent attributes of the POI to be compared and the recalled POI after obtaining the recalled POI corresponding to the POI to be compared, determine the semantic similarity between the text attributes of the POI to be compared and the recalled POI through a deep semantic model, determine the implicit features between the semantic similarity and the inherent attributes of the POI to be compared and the recalled POI through a deep learning model, input the statistical features and the implicit features into a logistic regression model to obtain a fusion score of the POI to be compared and the recalled POI, and determine the fusion result of the POI to be compared and the recalled POI according to the fusion score. Since the statistical features and the implicit features output by the deep semantic model and the deep learning model are used simultaneously in the process of map data fusion, the traditional logistic regression model and the deep model are combined, and both the memory ability of the traditional model and the generalization ability of the deep model are possessed, thereby avoiding the problem of poor generalization ability of the traditional machine learning model and improving the accuracy of data fusion. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0024] Figure 1 is a flow chart of a map data fusion method according to Embodiment 1 of the present application;

[0025] Figure 2 is a structural diagram of a map data fusion device according to Embodiment 2 of the present application;

[0026] Figure 3 It is a schematic diagram of the structure of the electronic device of the third embodiment of the present application. DETAILED DESCRIPTION

[0027] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0028] Embodiment 1

[0029] This embodiment provides a map data fusion method, such as Figure 1 As shown, the method includes: step 110 to step 160.

[0030] Step 110 : obtaining POIs within a preset distance range from the POI to be compared as recalled POIs.

[0031] The POI to be compared is a POI in one data set, and is a POI used for data fusion with a POI in another data set. The inherent attributes and text attributes of the POI to be compared are known, and the inherent attributes and text attributes of each POI in the other data set are also known. The inherent attributes include coordinates, so that according to the POI to be compared and the preset distance range, the POI within the preset distance range of the POI to be compared can be obtained in the other data set, and the obtained POI is used as the recalled POI.

[0032] In one embodiment of the present application, the step of obtaining a POI within a preset distance range from the POI to be compared as a recalled POI includes:

[0033] Obtaining a POI within a preset distance range from the POI to be compared as an initial POI;

[0034] Initial POIs that are not related to the POI name of the POI to be compared are filtered out, and the remaining initial POIs are used as the recalled POIs.

[0035] The POI names being irrelevant means that the POI to be compared and the initial POI do not contain the same characters in their POI names. The preset distance range generally means that the distance from the POI to be compared does not exceed the preset distance.

[0036] With the coordinates of the POI to be compared as the center and a preset distance (for example, 200 meters) as the radius, a circle is determined, and the circle is used as the preset distance range. The POIs within the preset distance range are obtained, and the obtained POIs are used as initial POIs. For example, there are 400 initial POIs, and then strategy filtering is performed, and the POI names of the 400 initial POIs are compared with the POI names of the POI to be compared respectively to determine whether the initial POI and the POI names of the POI to be compared contain the same characters, and the initial POIs with irrelevant POI names are filtered out, and the remaining initial POIs are used as recalled POIs. For example, 100 initial POIs are filtered out, and 300 initial POIs remain. The remaining 300 initial POIs are used as recalled POIs, that is, there are 300 recalled POIs. By filtering the acquired initial POIs within a preset distance range from the POI to be compared and using the remaining initial POIs as recalled POIs, the processing speed of map data fusion can be improved, and data fusion processing of POIs that are completely unrelated to the POI to be compared can be avoided.

[0037] Step 120: Determine the statistical features of the inherent attributes of the POI to be compared and the recalled POI.

[0038] The inherent attributes may include coordinates, telephone numbers, POI types, house numbers, etc. The coordinates may be longitude and latitude coordinates, and the POI types may include gas stations, restaurants, barber shops, etc.

[0039] For each recalled POI, the inherent attributes of the POI to be compared are compared with the corresponding inherent attributes of the recalled POI to obtain statistical features.

[0040] In one embodiment of the present application, the determining of the statistical features of the inherent attributes of the POI to be compared and the recalled POI includes:

[0041] Determining the distance between the POI to be compared and the recalled POI according to the coordinates of the POI to be compared and the coordinates of the recalled POI;

[0042] Comparing whether the telephone number of the POI to be compared is the same as the telephone number of the recalled POI, and obtaining a telephone comparison result;

[0043] Comparing whether the POI type of the POI to be compared is the same as the POI type of the recalled POI, and obtaining a type comparison result;

[0044] Comparing the house number of the POI to be compared with the house number of the recalled POI to determine whether they are the same, and obtaining a house number comparison result;

[0045] At least one of the distance, the telephone comparison result, the type comparison result and the house number comparison result is used as the statistical feature.

[0046] The statistical features may include at least one of the distance between the POI to be compared and the recalled POI, a telephone comparison result, a POI type comparison result, and a house number comparison result.

[0047] According to the coordinates of the POI to be compared and the coordinates of the recalled POI, the distance between the POI to be compared and the recalled POI can be calculated; each POI has a corresponding telephone number, and the telephone number of the POI to be compared is compared with the telephone number of the recalled POI to obtain a telephone comparison result. For example, if the telephone number of the POI to be compared is the same as the telephone number of the recalled POI, it can be said that the POI to be compared is the same as the recalled POI. If the telephone number of the POI to be compared is different from the telephone number of the recalled POI, it cannot be determined whether the POI to be compared is the same as the recalled POI, because the same POI can also have multiple telephone numbers; each POI corresponds to a POI type, and the POI type of the POI to be compared is compared with the recalled POI to obtain a POI type comparison result; each POI has a corresponding house number, and the house number of the POI to be compared is compared with the house number of the recalled POI to obtain a house number comparison result; at least one of the distance, telephone comparison result, POI type comparison result and house number comparison result obtained above is used as a statistical feature of the inherent attributes of the POI to be compared and the recalled POI. When determining the statistical features of the inherent attributes of the POI to be compared and the recalled POIs, for each recalled POI, the statistical features of the inherent attributes of the POI to be compared are determined respectively, so as to obtain the statistical features of the inherent attributes of each recalled POI and the POI to be compared.

[0048] Step 130 : Determine the semantic similarity between the text attributes of the POI to be compared and the text attributes of the recalled POI by using a deep semantic model.

[0049] The text attributes may include a POI name and / or address.

[0050] The deep semantic model can perform semantic extraction on the text attributes of the POI to be compared and the text attributes of the recalled POI, and calculate the similarity of the extracted semantic features to obtain the semantic similarity between the text attributes of the POI to be compared and the text attributes of the recalled POI. For each recalled POI, the semantic similarity with the text attributes of the POI to be compared is determined respectively, and the semantic similarity between the text attributes of each recalled POI and the POI to be compared is obtained.

[0051] In one embodiment of the present application, determining the semantic similarity between the text attributes of the POI to be compared and the text attributes of the recalled POI by using a deep semantic model includes:

[0052] Preprocessing the text attributes of the POI to be compared and the text attributes of the recalled POI respectively by using a deep semantic model to obtain the preprocessed text attributes of the POI to be compared and the text attributes of the recalled POI;

[0053] Semantically extracting the preprocessed text attributes of the POI to be compared and the text attributes of the recalled POI respectively through the deep semantic model to obtain semantic features of the POI to be compared and the semantic features of the recalled POI;

[0054] The similarity between the semantic features of the POI to be compared and the semantic features of the recalled POI is calculated by using the deep semantic model to obtain the semantic similarity between the text attributes of the POI to be compared and the text attributes of the recalled POI.

[0055] The text attributes of the POI to be compared and the text attributes of a recalled POI are input into the deep semantic model together. The deep semantic model first preprocesses the text attributes of the POI to be compared and the text attributes of the recalled POI respectively, so as to unify the capitalization of letters, convert traditional Chinese characters into simplified Chinese characters, remove spaces, etc., and obtain the preprocessed text attributes of the POI to be compared and the text attributes of the recalled POI. After the text attributes are preprocessed, the preprocessed text attributes of the POI to be compared and the text attributes of the recalled POI are respectively input into the semantic extraction layer of the deep semantic model. The preprocessed text attributes of the POI to be compared and the text attributes of the recalled POI are respectively semantically extracted to obtain the semantic features of the POI to be compared and the semantic features of the recalled POI; the similarity between the semantic features of the POI to be compared and the semantic features of the recalled POI is calculated by the deep semantic model to obtain the semantic similarity between the text attributes of the POI to be compared and the text attributes of the recalled POI, and the distance between the semantic features of the POI to be compared and the semantic features of the recalled POI can be calculated, and the distance is used as the semantic similarity, for example, the distance can be the Manhattan distance, etc. The text attributes of the POI to be compared and the recalled POI are first preprocessed by the deep semantic model, and then semantic extraction is performed, and then the semantic similarity is calculated, which can improve the accuracy of the calculated semantic similarity, and then the relevance of the POI name of the POI to be compared and the recalled POI can be clearly determined according to the semantic similarity.

[0056] For example, when the text attribute is a POI name, the deep semantic model can output the semantic similarity between the POI name of the POI to be compared and the POI name of the recalled POI. For example, the POI name of the POI to be compared is "Quanjude Roast Duck" and the POI name of the recalled POI is "Carrefour Supermarket". When these two POI names are input into the deep semantic model, the semantic similarity between the two can be obtained to be 0.0013; the POI name of the POI to be compared is "Hua Cai Building" and the POI name of the recalled POI is "Hua Cai Center". When these two POI names are input into the deep semantic model, the semantic similarity between the two can be obtained to be 0.941. The higher the semantic similarity, the higher the correlation between the two POI names, and the lower the semantic similarity, the lower the correlation between the two POI names.

[0057] In one embodiment of the present application, the semantic extraction of the preprocessed text attributes of the POI to be compared and the text attributes of the recalled POI by the deep semantic model to obtain the semantic features of the POI to be compared and the semantic features of the recalled POI includes:

[0058] The preprocessed text attributes of the POI to be compared and the text attributes of the recalled POI are semantically extracted respectively through the attention mechanism in the deep semantic model to obtain the semantic features of the POI to be compared and the semantic features of the recalled POI.

[0059] The semantic extraction layer in the deep semantic model can be realized by the attention mechanism, that is, the semantic extraction layer is an attention mechanism structure. The attention mechanism first calculates the attention weights of the preprocessed text attributes of the POI to be compared and the recalled POI, and obtains the attention weights of each attribute value of the text attribute of the POI to be compared and the attention weights of each attribute value of the text attribute of the recalled POI. Then, the attribute values ​​are weighted according to the attention weights of each attribute value of the text attribute of the POI to be compared, and the semantic features of the POI to be compared are obtained. The semantic features of the recalled POI are weighted according to the attention weights of each attribute value of the text attribute of the recalled POI. The accuracy of semantic extraction can be improved by performing semantic extraction on the text attributes of the POI to be compared and the text attributes of the recalled POI through the attention mechanism.

[0060] Step 140 : determining implicit features between the semantic similarity and inherent attributes of the POI to be compared and the recalled POI through a deep learning model.

[0061] The deep learning model may include multiple fully connected layers, so that the semantic similarity and inherent attributes of the POIs to be compared are input into the deep learning model. Through the multiple fully connected layers, implicit features between the semantic similarity and the inherent attributes can be obtained.

[0062] In one embodiment of the present application, determining the implicit features between the semantic similarity and inherent attributes of the POI to be compared and the recalled POI by using a deep learning model includes:

[0063] The semantic similarity between the POI to be compared and the recalled POI, the distance in the inherent attributes, and the POI type are input into the deep learning model to obtain the implicit features between the semantic similarity and the inherent attributes between the POI to be compared and the recalled POI.

[0064] The inherent attributes may include coordinates, telephone numbers, POI types, house numbers, etc. The distance between the POI to be compared and the recalled POI can be obtained based on the coordinates of the POI to be compared and the coordinates of the recalled POI, and the distance can also be used as an inherent attribute. Since telephone numbers and house numbers are relatively certain features, there is no need to learn implicit features. When determining implicit features, the semantic similarity, distance and POI type of the POI to be compared and the recalled POI are input into the deep learning model, and the semantic similarity, distance and POI type are fully connected through multiple fully connected layers in the deep learning model to obtain the implicit features between the semantic similarity, distance and POI type. For example, when the distance is 100 meters, the meanings represented by two small supermarkets and two parks are different. The implicit features obtained by the deep learning model can clearly distinguish different POIs.

[0065] Step 150: Input the statistical features and latent features into a logistic regression model to obtain a fusion score of the POI to be compared and the recalled POI.

[0066] Among them, the logistic regression model can be a traditional machine learning model. In an embodiment of the present application, the input features of the logistic regression model include not only the statistical features, but also the implicit features output by the deep learning model.

[0067] The statistical features and the latent features are input into a logistic regression model, and logistic regression processing is performed on the statistical features and the latent features through the logistic regression model to obtain a fusion score of the POI to be compared and the recalled POI.

[0068] When multiple recalled POIs are included, for each recalled POI, the fusion score of each recalled POI and the POI to be compared is calculated according to steps 120 to 140 to obtain the fusion score of the POI to be compared and each recalled POI.

[0069] Step 160: Determine a fusion result of the POI to be compared and the recalled POI according to the fusion score.

[0070] The fusion score of the POI to be compared and each recalled POI is compared with a preset threshold to determine the fusion result of the POI to be compared and the recalled POI. The preset threshold may include a first threshold and a second threshold, the first threshold is greater than the second threshold, if there is a recalled POI with a fusion score greater than the first threshold, it is determined that the POI to be compared and the recalled POI are the same POI, if all the fusion scores are less than the second threshold, it is determined that the POI to be compared and the recalled POI are different.

[0071] In one embodiment of the present application, determining the fusion result of the POI to be compared and the recalled POI according to the fusion score includes:

[0072] Determine the highest fusion score among the fusion scores of the POI to be compared and each recalled POI;

[0073] If the highest fusion score is greater than or equal to a first threshold, determining that the fusion result is that the POI to be compared and the recalled POI corresponding to the highest fusion score are the same POI;

[0074] If the highest fusion score is less than the first threshold and greater than or equal to the second threshold, determining that the fusion result is uncertain;

[0075] If the highest fusion score is less than the second threshold, determining that the fusion result is that the POI to be compared is different from the recalled POI;

[0076] The first threshold is greater than the second threshold.

[0077] When determining the fusion result of the POI to be compared and the recalled POI, since it may include multiple recalled POIs, the highest fusion score among the fusion scores of the POI to be compared and each recalled POI can be determined first, and then the relationship between the highest fusion score and the first threshold and the second threshold can be compared. If the highest fusion score is greater than or equal to the first threshold, the fusion result is determined to be that the POI to be compared and the recalled POI corresponding to the highest fusion score are the same POI. If the highest fusion score is less than the first threshold and greater than or equal to the second threshold, the fusion result is determined to be uncertain. If the highest fusion score is less than the second threshold, the fusion result is determined to be that the POI identical to the POI to be compared is not found in another data set, that is, the POI to be compared and the recalled POI are different. The fusion score is a value between 0 and 1, and the first threshold and the second threshold are both greater than 0 and less than 1.

[0078] For example, 300 POIs are recalled, that is, there are 300 recalled POIs. Through steps 120 to 140, the fusion scores of the 300 recalled POIs and the POI to be compared are obtained, that is, 300 fusion scores are obtained. The 300 fusion scores are compared to determine the highest fusion score. Assuming that the first threshold is 0.8 and the second threshold is 0.2, if the highest fusion score is greater than 0.8, it can be determined that the POI to be compared and the recalled POI corresponding to the highest fusion score are the same POI, that is, the same POI. If the highest fusion score is less than 0.2, it can be determined that the POI to be compared does not have the same POI within the preset distance range. If the highest fusion score is less than 0.8 and greater than 0.2, the fusion result is uncertain, that is, it cannot be determined whether the POI to be compared is the same as the recalled POI.

[0079] It should be noted that the above-mentioned deep semantic model, deep learning model and logistic regression model together constitute the map data fusion model in the embodiment of the present application in accordance with the above-mentioned processing order. Before using the map data fusion model for map data fusion, the map data fusion model must be trained. During training, the map data fusion model is uniformly trained using manually annotated sample data. After the training is completed, the map data fusion model is used for map data fusion processing.

[0080] The map data fusion method provided in the embodiment of the present application determines the statistical features of the inherent attributes of the POI to be compared and the recalled POI after obtaining the recalled POI corresponding to the POI to be compared, determines the semantic similarity between the text attributes of the POI to be compared and the recalled POI through a deep semantic model, determines the implicit features between the semantic similarity and the inherent attributes of the POI to be compared and the recalled POI through a deep learning model, inputs the statistical features and the implicit features into a logistic regression model to obtain a fusion score of the POI to be compared and the recalled POI, and determines the fusion result of the POI to be compared and the recalled POI according to the fusion score. Since the statistical features and the implicit features output by the deep semantic model and the deep learning model are used simultaneously in the process of map data fusion, the traditional logistic regression model and the deep model are combined, and both the memory ability of the traditional model and the generalization ability of the deep model are possessed, thereby avoiding the problem of poor generalization ability of the traditional machine learning model and improving the accuracy of data fusion.

[0081] Embodiment 2

[0082] This embodiment provides a map data fusion device, such as Figure 2 As shown, the map data fusion device 200 includes:

[0083] The recalled POI acquisition module 210 is used to acquire a POI within a preset distance range from the POI to be compared as a recalled POI;

[0084] A statistical feature determination module 220, configured to determine the statistical features of the inherent attributes of the POI to be compared and the recalled POI;

[0085] A semantic similarity determination module 230, configured to determine the semantic similarity between the text attributes of the POI to be compared and the text attributes of the recalled POI through a deep semantic model;

[0086] An implicit feature determination module 240 is used to determine implicit features between the semantic similarity and inherent attributes of the POI to be compared and the recalled POI through a deep learning model;

[0087] A fusion module 250 is used to input the statistical features and implicit features into a logistic regression model to obtain a fusion score of the POI to be compared and the recalled POI;

[0088] The fusion result determination module 260 is used to determine the fusion result of the POI to be compared and the recalled POI according to the fusion score.

[0089] Optionally, the semantic similarity determination module includes:

[0090] A preprocessing unit, used to preprocess the text attributes of the POI to be compared and the text attributes of the recalled POI respectively through a deep semantic model to obtain the preprocessed text attributes of the POI to be compared and the text attributes of the recalled POI;

[0091] A semantic extraction unit, configured to perform semantic extraction on the preprocessed text attributes of the POI to be compared and the text attributes of the recalled POI respectively through the deep semantic model to obtain semantic features of the POI to be compared and the semantic features of the recalled POI;

[0092] The semantic similarity calculation unit is used to calculate the similarity between the semantic features of the POI to be compared and the semantic features of the recalled POI through the deep semantic model, so as to obtain the semantic similarity between the text attributes of the POI to be compared and the text attributes of the recalled POI.

[0093] Optionally, the semantic extraction unit is specifically used for:

[0094] The preprocessed text attributes of the POI to be compared and the text attributes of the recalled POI are semantically extracted respectively through the attention mechanism in the deep semantic model to obtain the semantic features of the POI to be compared and the semantic features of the recalled POI.

[0095] Optionally, the text attribute includes a POI name and / or address.

[0096] Optionally, the statistical feature determination module includes:

[0097] a distance determination unit, configured to determine the distance between the POI to be compared and the recalled POI according to the coordinates of the POI to be compared and the coordinates of the recalled POI;

[0098] A telephone comparison unit, used to compare whether the telephone number of the POI to be compared is the same as the telephone number of the recalled POI, and obtain a telephone comparison result;

[0099] A POI type comparison unit, used to compare whether the POI type of the POI to be compared is the same as the POI type of the recalled POI, and obtain a POI type comparison result;

[0100] A house number comparison unit, used to compare whether the house number of the POI to be compared is the same as the house number of the recalled POI, and obtain a house number comparison result;

[0101] The statistical feature determination unit is used to use at least one of the distance, the telephone comparison result, the POI type comparison result and the house number comparison result as the statistical feature.

[0102] Optionally, the fusion result determination module is specifically used for:

[0103] Determine the highest fusion score among the fusion scores of the POI to be compared and each recalled POI;

[0104] If the highest fusion score is greater than or equal to a first threshold, determining that the fusion result is that the POI to be compared and the recalled POI corresponding to the highest fusion score are the same POI;

[0105] If the highest fusion score is less than the first threshold and greater than or equal to the second threshold, determining that the fusion result is uncertain;

[0106] If the highest fusion score is less than the second threshold, determining that the fusion result is that the POI to be compared is different from the recalled POI;

[0107] The first threshold is greater than the second threshold.

[0108] Optionally, the implicit feature determination module is specifically used to:

[0109] The semantic similarity between the POI to be compared and the recalled POI, the distance in the inherent attributes, and the POI type are input into the deep learning model to obtain the implicit features between the semantic similarity and the inherent attributes between the POI to be compared and the recalled POI.

[0110] Optionally, the recalled POI acquisition module includes:

[0111] A POI recall unit, used to obtain a POI within a preset distance range from the POI to be compared as an initial POI;

[0112] The POI filtering unit is used to filter out initial POIs that are irrelevant to the POI name of the POI to be compared, and use the remaining initial POIs as the recalled POIs.

[0113] The map data fusion device provided in the embodiment of the present application is used to implement the various steps of the map data fusion method described in the first embodiment of the present application. The specific implementation methods of each module of the device refer to the corresponding steps and will not be repeated here.

[0114] The map data fusion device provided by the embodiment of the present application obtains POIs within a preset distance range from the POI to be compared as the recalled POI through the recalled POI acquisition module, and the statistical feature determination module determines the statistical features of the inherent attributes of the POI to be compared and the recalled POI, the semantic similarity determination module determines the semantic similarity between the text attributes of the POI to be compared and the recalled POI through the deep semantic model, the implicit feature determination module determines the implicit features between the semantic similarity and the inherent attributes of the POI to be compared and the recalled POI through the deep learning model, the fusion module inputs the statistical features and the implicit features into the logistic regression model to obtain the fusion score of the POI to be compared and the recalled POI, and the fusion result determination module determines the fusion result of the POI to be compared and the recalled POI according to the fusion score. Since the statistical features and the implicit features output by the deep semantic model and the deep learning model are used simultaneously in the process of map data fusion, the traditional logistic regression model and the deep model are combined, and the memory ability of the traditional model and the generalization ability of the deep model are both possessed, the problem of poor generalization ability of the traditional machine learning model is avoided, and the accuracy of data fusion is improved.

[0115] Embodiment 3

[0116] The present application also provides an electronic device, such as Figure 3 As shown, the electronic device 300 may include one or more processors 310 and one or more memories 320 connected to the processor 310. The electronic device 300 may also include an input interface 330 and an output interface 340 for communicating with another device or system. The program code executed by the processor 310 may be stored in the memory 320.

[0117] The processor 310 in the electronic device 300 calls the program code stored in the memory 320 to execute the map data fusion method in the above embodiment.

[0118] The above elements in the above electronic device may be connected to each other via a bus, such as one or any combination of a data bus, an address bus, a control bus, an expansion bus and a local bus.

[0119] The embodiment of the present application also provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of the map data fusion method as described in the first embodiment of the present application are implemented.

[0120] Each embodiment in this specification is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0121] The above is a detailed introduction to a map data fusion method, device, electronic device and storage medium provided in the embodiments of the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea; at the same time, for general technical personnel in this field, according to the ideas of the present application, there will be changes in the specific implementation methods and application scopes. In summary, the content of this specification should not be understood as a limitation on the present application.

[0122] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

Claims

1. A map data fusion method, characterized in that: include: Obtaining POIs within a preset distance range from the POI to be compared as recalled POIs; Determining statistical features of inherent attributes of the POI to be compared and the recalled POI; Determine the semantic similarity between the text attributes of the POI to be compared and the text attributes of the recalled POI through a deep semantic model; determine the implicit features between the semantic similarity and the inherent attributes of the POI to be compared and the recalled POI through a deep learning model; input the statistical features and the implicit features into a logistic regression model to obtain a fusion score of the POI to be compared and the recalled POI; determine the fusion result of the POI to be compared and the recalled POI according to the fusion score; The method of determining the semantic similarity between the text attributes of the POI to be compared and the text attributes of the recalled POI by using the deep semantic model comprises: preprocessing the text attributes of the POI to be compared and the text attributes of the recalled POI respectively by using the deep semantic model to obtain the preprocessed text attributes of the POI to be compared and the text attributes of the recalled POI; performing semantic extraction on the preprocessed text attributes of the POI to be compared and the text attributes of the recalled POI respectively by using the deep semantic model to obtain the semantic features of the POI to be compared and the semantic features of the recalled POI; calculating the similarity between the semantic features of the POI to be compared and the semantic features of the recalled POI by using the deep semantic model to obtain the semantic similarity between the text attributes of the POI to be compared and the text attributes of the recalled POI; The method of determining the implicit features between the semantic similarity and inherent attributes of the POI to be compared and the recalled POI by using a deep learning model includes: inputting the semantic similarity, the distance in the inherent attributes, and the POI type of the POI to be compared and the recalled POI into the deep learning model to obtain the implicit features between the semantic similarity and inherent attributes of the POI to be compared and the recalled POI.

2. The method according to claim 1, characterized in that The method of performing semantic extraction on the preprocessed text attributes of the POI to be compared and the text attributes of the recalled POI respectively through the deep semantic model to obtain the semantic features of the POI to be compared and the semantic features of the recalled POI includes: performing semantic extraction on the preprocessed text attributes of the POI to be compared and the text attributes of the recalled POI respectively through the attention mechanism in the deep semantic model to obtain the semantic features of the POI to be compared and the semantic features of the recalled POI.

3. The method according to claim 1, characterized in that The text attributes include the POI name and / or address.

4. The method according to any one of claims 1 to 2, characterized in that: The statistical features of determining the inherent attributes of the POI to be compared and the recalled POI include: determining the distance between the POI to be compared and the recalled POI according to the coordinates of the POI to be compared and the coordinates of the recalled POI; comparing whether the telephone number of the POI to be compared is the same as the telephone number of the recalled POI to obtain a telephone comparison result; comparing whether the POI type of the POI to be compared is the same as the POI type of the recalled POI to obtain a POI type comparison result; comparing whether the house number of the POI to be compared is the same as the house number of the recalled POI to obtain a house number comparison result; and using at least one of the distance, the telephone comparison result, the POI type comparison result and the house number comparison result as the statistical feature.

5. The method according to any one of claims 1-2, characterized in that: Determining the fusion result of the POI to be compared and the recalled POI according to the fusion score includes: determining the highest fusion score among the fusion scores of the POI to be compared and each recalled POI; if the highest fusion score is greater than or equal to a first threshold, determining that the fusion result is that the POI to be compared and the recalled POI corresponding to the highest fusion score are the same POI; if the highest fusion score is less than the first threshold and greater than or equal to a second threshold, determining that the fusion result is uncertain; if the highest fusion score is less than the second threshold, determining that the fusion result is that the POI to be compared is different from the recalled POI; wherein the first threshold is greater than the second threshold.

6. The method according to any one of claims 1 to 3, characterized in that: The obtaining of POIs within a preset distance range from the POI to be compared as recalled POIs includes: obtaining POIs within a preset distance range from the POI to be compared as initial POIs; filtering out initial POIs that are irrelevant to the POI name of the POI to be compared, and using the remaining initial POIs as the recalled POIs.

7. A map data fusion device, characterized in that: include: A recalled POI acquisition module is used to acquire a POI within a preset distance range from the POI to be compared as a recalled POI; A statistical feature determination module, used to determine the statistical features of the inherent attributes of the POI to be compared and the recalled POI; The semantic similarity determination module is used to determine the semantic similarity between the text attributes of the POI to be compared and the text attributes of the recalled POI through a deep semantic model, wherein the semantic similarity between the text attributes of the POI to be compared and the text attributes of the recalled POI through a deep semantic model comprises: preprocessing the text attributes of the POI to be compared and the text attributes of the recalled POI respectively through a deep semantic model to obtain the preprocessed text attributes of the POI to be compared and the text attributes of the recalled POI; performing semantic extraction on the preprocessed text attributes of the POI to be compared and the text attributes of the recalled POI respectively through the deep semantic model to obtain the semantic features of the POI to be compared and the semantic features of the recalled POI; calculating the similarity between the semantic features of the POI to be compared and the semantic features of the recalled POI through the deep semantic model to obtain the semantic features of the POI to be compared; The method comprises the following steps: determining the semantic similarity between the text attributes of the compared POI and the recalled POI and determining the implicit features between the semantic similarity and the inherent attributes of the compared POI and the recalled POI through a deep learning model, wherein the method comprises: inputting the semantic similarity between the compared POI and the recalled POI, the distance in the inherent attributes and the POI type into the deep learning model to obtain the implicit features between the semantic similarity and the inherent attributes of the compared POI and the recalled POI; a fusion module, which is used to input the statistical features and the implicit features into a logistic regression model to obtain the fusion score of the compared POI and the recalled POI; and a fusion result determination module, which is used to determine the fusion result of the compared POI and the recalled POI according to the fusion score.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the map data fusion method according to any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the map data fusion method described in any one of claims 1 to 6 are implemented.

Images