Similarity determination method, similarity prediction model training method and device

Abstract

This disclosure provides a method for determining similarity, a method for training a similarity prediction model, and an apparatus. This disclosure relates to the field of internet technology, and particularly to artificial intelligence technologies such as intelligent search, intelligent matching, intelligent layout, and intelligent recommendation. The specific implementation scheme is as follows: Obtain a first network record and a second network record; determine a first feature of the first network record and a second feature of the second network record, wherein both the first and second features include a first type of feature and a second type of feature; determine the target weights corresponding to the first type of feature and the second type of feature; input the target weights corresponding to the first type of feature and the second type of feature, as well as the first and second features, into a similarity prediction model, and obtain the similarity between the first network record and the second network record output by the similarity prediction model. According to the scheme of this disclosure, the accuracy of similarity prediction can be improved.

Description

Technical Field

[0001] This disclosure relates to the field of Internet technology, and in particular to the field of artificial intelligence technology, such as intelligent search, intelligent matching, intelligent layout, and intelligent recommendation. Background Technology

[0002] With the rapid development of internet technology, the fast-moving consumer goods (FMCG) industry has gradually emerged. Within this industry, there are numerous outlets and sales points. These outlets are often very similar in name and address, leading to a low degree of matching between the actual information of some outlets and the information in the database. Summary of the Invention

[0003] This disclosure provides a method for determining similarity, a method for training a similarity prediction model, and an apparatus.

[0004] According to a first aspect of this disclosure, a similarity determination method is provided, comprising: acquiring a first network node record and a second network node record; determining a first feature of the first network node record and a second feature of the second network node record, wherein the first feature and the second feature each include a first type of feature and a second type of feature, wherein the first type of feature is a feature with adjustable weights and the second type of feature is a feature with non-adjustable weights; determining target weights corresponding to the first type of feature and the second type of feature respectively; inputting the target weights corresponding to the first type of feature and the second type of feature, as well as the first feature and the second feature, into a similarity prediction model to obtain the similarity between the first network node record and the second network node record output by the similarity prediction model.

[0005] According to a second aspect of this disclosure, a method for training a similarity prediction model is provided, comprising: acquiring training data, the training data including a first feature of a first dot record sample and a second feature of a second dot record sample, and ground truth similarity values ​​of the first dot record sample and the second dot record sample; inputting the first feature of the first dot record sample and the second feature of the second dot record sample into the model to be trained, thereby obtaining a similarity prediction value of the first dot record sample and the second dot record sample output by the model to be trained; and training the model to be trained based on the similarity prediction value of the first dot record sample and the second dot record sample, and the ground truth similarity values ​​of the first dot record sample and the second dot record sample, thereby obtaining a similarity prediction model.

[0006] According to a third aspect of this disclosure, a similarity determination apparatus is provided, comprising: a first acquisition module for acquiring a first network node record and a second network node record; a first determination module for determining a first feature of the first network node record and a second feature of the second network node record, wherein both the first feature and the second feature include a first type of feature and a second type of feature, wherein the first type of feature is a feature with adjustable weights and the second type of feature is a feature with non-adjustable weights; a second determination module for determining target weights corresponding to the first type of feature and the second type of feature respectively; and a second acquisition module for inputting the target weights corresponding to the first type of feature and the second type of feature, as well as the first feature and the second feature, into a similarity prediction model to acquire the similarity between the first network node record and the second network node record output by the similarity prediction model.

[0007] According to a fourth aspect of this disclosure, a training apparatus for a similarity prediction model is provided, comprising: a fourth acquisition module for acquiring training data, the training data including a first feature of a first dot record sample and a second feature of a second dot record sample, and ground truth similarity values ​​of the first dot record sample and the second dot record sample; a prediction module for inputting the first feature of the first dot record sample and the second feature of the second dot record sample into the training model to obtain a similarity prediction value of the first dot record sample and the second dot record sample output by the training model; and a training module for training the training model based on the similarity prediction value of the first dot record sample and the second dot record sample, and the ground truth similarity values ​​of the first dot record sample and the second dot record sample, to obtain a similarity prediction model.

[0008] According to a fifth aspect of this disclosure, an electronic device is provided, comprising: at least one processor; a memory communicatively connected to the at least one processor; the memory storing instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the similarity determination method provided in the first aspect and / or the similarity prediction model training method provided in the second aspect.

[0009] According to a sixth aspect of this disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to cause a computer to perform the similarity determination method provided in the first aspect and / or the similarity prediction model training method provided in the second aspect.

[0010] According to a seventh aspect of this disclosure, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the similarity determination method provided in the first aspect and / or the similarity prediction model training method provided in the second aspect.

[0011] According to the technical solution disclosed herein, the accuracy of similarity prediction can be improved.

[0012] The above overview is for illustrative purposes only and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments, and features described above, further aspects, embodiments, and features of this application will become readily apparent from the accompanying drawings and the following detailed description. Attached Figure Description

[0013] In the accompanying drawings, unless otherwise specified, the same reference numerals throughout the various drawings denote the same or similar parts or elements. These drawings are not necessarily drawn to scale. It should be understood that these drawings depict only some embodiments disclosed in this application and should not be construed as limiting the scope of this application.

[0014] Figure 1 This is a flowchart illustrating a similarity determination method according to an embodiment of the present disclosure;

[0015] Figure 2 This is a schematic diagram of the process for mining blank network points based on the number of target network points according to an embodiment of this disclosure;

[0016] Figure 3 This is a schematic diagram illustrating the determination of the weights of each feature based on the feature parameter information of the similarity prediction model according to an embodiment of this disclosure;

[0017] Figure 4 This is a flowchart illustrating a similarity prediction model training method according to an embodiment of the present disclosure;

[0018] Figure 5 This is a schematic diagram illustrating the feature mining of network record samples using a dictionary method according to an embodiment of this disclosure;

[0019] Figure 6 This is a schematic diagram of the architecture for training a similarity prediction model according to an embodiment of this disclosure;

[0020] Figure 7 This is a schematic diagram of the overall process of training a similarity prediction model according to an embodiment of the present disclosure;

[0021] Figure 8 This is a schematic diagram of the structure of a similarity determination device according to an embodiment of the present disclosure;

[0022] Figure 9 This is a schematic diagram of the structure of a similarity prediction model training device according to an embodiment of the present disclosure;

[0023] Figure 10 This is a schematic diagram of a scenario based on the similarity determination method according to an embodiment of this disclosure;

[0024] Figure 11 This is a schematic diagram of a scenario for a similarity prediction model training method according to an embodiment of the present disclosure;

[0025] Figure 12 This is a schematic diagram of the structure of an electronic device used to implement the similarity determination method and / or similarity prediction model training method of the embodiments of this disclosure. Detailed Implementation

[0026] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0027] The terms "first," "second," and "third," etc., used in the embodiments, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, such as including a series of steps or units. A method, system, product, or apparatus is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products, or apparatuses.

[0028] Taking convenience store scenarios as an example, users typically upload a first location record, which includes the location's name and address. A second location record is then obtained by querying a Point of Interest (POI) database based on this first location. Related technologies determine their similarity level by measuring the correlation between the first and second location records and manually setting thresholds; alternatively, they compare individual address elements and determine the matching level based on the address element hierarchy and the store name hierarchy. First, while these methods effectively match addresses to some extent, they often fail if the address records are non-standard, such as missing address or store name elements, excessive redundant information, or a lack of natural separators between the store name and address elements. Second, these methods also fail to identify the matching relationship when two identical addresses have minimal text overlap. Third, when Chinese and English texts differ and there are no clear word segmentation boundaries, using only word segmentation tools may lead to errors in downstream model recognition due to segmentation mistakes. In addition, existing prediction models can only learn the optimal weights through the model itself, lacking human intervention.

[0029] Clearly, the above processing method has significant uncertainties and cannot accurately predict the similarity between the first and second location records, resulting in a low matching rate between the user's location records and the map's location records. Furthermore, this method cannot clean and verify existing locations, nor can it deduplicate existing locations or uncover blank locations.

[0030] In order to at least partially solve one or more of the above-mentioned problems and other potential problems, this disclosure proposes a similarity prediction scheme that can accurately predict the similarity between the first network point record and the second network point record, improve the matching accuracy between the first network point record and the second network point record, and can both clean and verify existing network points, as well as deduplicate existing network points and discover blank network points.

[0031] This disclosure provides a similarity determination method. Figure 1 This is a flowchart illustrating a similarity determination method according to an embodiment of the present disclosure. This similarity determination method can be applied to a similarity determination device. The similarity determination device is located in an electronic device. The electronic device includes, but is not limited to, fixed devices and / or mobile devices. For example, fixed devices include, but are not limited to, servers, which can be cloud servers or ordinary servers. Mobile devices include, but are not limited to, mobile phones, tablets, and vehicle-mounted terminals. In some possible implementations, the similarity determination method can also be implemented by a processor calling computer-readable instructions stored in memory. Figure 1 As shown, the similarity determination method includes:

[0032] S101: Obtain the records of the first and second branch locations;

[0033] S102: Determine the first feature of the first network record and the second feature of the second network record. The first feature and the second feature both include a first type of feature and a second type of feature. The first type of feature is a feature with adjustable weights, and the second type of feature is a feature with non-adjustable weights.

[0034] S103: Determine the target weights corresponding to the first type of feature and the second type of feature respectively;

[0035] S104: Input the target weights corresponding to the first type of feature and the second type of feature, as well as the first feature and the second feature, into the similarity prediction model to obtain the similarity between the first network record and the second network record output by the similarity prediction model.

[0036] In this embodiment, the methods for obtaining the first and second branch site records include, but are not limited to, user-uploaded records and retrieval from a POI database. For example, the first branch site record may be a user-uploaded record; the second branch site record may also be a user-uploaded record. Another example is that the first branch site record may be user-uploaded information, and the second branch site record may be information retrieved from a POI database. Yet another example is that the first branch site record may be user-uploaded information, and the second branch site record may be information retrieved from a POI database based on the first record. The above are merely illustrative examples and are not intended to limit all possible methods of obtaining the first and second branch site records; they are simply not exhaustive.

[0037] In this embodiment of the disclosure, both the first and second outlet records include the name and address of the outlet. For example, the first outlet record is 7-11 Convenience Store, Wenxin Fifth Road, Nanshan District, Shenzhen. Another example is Yonghui Supermarket, No. 5-1, Lane 7, Xinghua First Road, Bao'an District, Shenzhen. Yet another example is Bianlifeng Convenience Store, Central Business Plaza, Haide Third Road, Nanshan District, Shenzhen, Guangdong Province. The above are merely illustrative examples and are not intended to limit the relevant information included in the first and second outlet records; they are simply not exhaustive.

[0038] In this embodiment of the disclosure, the first feature recorded by the first outlet and the second feature recorded by the second outlet may include the following features: location features, address text features, structured address features, outlet name features, semantic features, and channel features. The first and second features include at least one of the features mentioned above. The above is merely illustrative and is not intended to limit the scope of all possible features included in the first and second features; an exhaustive list is not provided here.

[0039] In this embodiment, the first feature and the second feature are obtained based on a feature acquisition model. This feature acquisition model is a pre-trained model, such as a Bidirectional Long Short-Term Memory (BiLSTM) network model, and can be a component of a similarity prediction model. This disclosure does not limit the specific training method of the feature acquisition model. The specific training method for the similarity prediction model will be described in detail later and will not be repeated here.

[0040] In this embodiment of the disclosure, the first feature recorded by the first outlet may include only the first type of feature or the second type of feature, or it may include both the first type of feature and the second type of feature. Similarly, the first feature recorded by the second outlet may include only the first type of feature or the second type of feature, or it may include both the first type of feature and the second type of feature.

[0041] In this embodiment, the first type of feature is one whose weights can be manually adjusted; this type of feature includes location features and structured address features. The second type of feature is one whose weights cannot be manually adjusted; this type of feature includes address text features, store name features, semantic features, and channel features. The weights of the second type of feature are fixed weights learned by the learning model. In this embodiment, the learning model can be a pre-trained eXtreme Gradient Boosting (XGBoost) model, and the learning model can be a component of the similarity prediction model. How to train the similarity prediction model will be described in detail later and will not be repeated here.

[0042] For example, the learning model learns that the sum of the weights of the first type of features is 0.3, where the weight of the location feature is 0.11 and the weight of the structured address feature is 0.19; the sum of the weights of the second type of features is 0.7, where the weight of the address text feature is 0.56, the weight of the outlet / store name feature is 0.03, the weight of the semantic feature is 0.09, and the weight of the channel feature is 0.02. Since the weights of the first type of features can be manually adjusted, the weight of the location feature can be adjusted to 0.16, and the weight of the structured address feature can be adjusted to 0.14. Alternatively, the location feature can be adjusted to 0.09, and the weight of the structured address feature can be adjusted to 0.21. Since the weights of the second type of features cannot be manually adjusted, the weight of the address text feature should be 0.56, the weight of the outlet / store name feature should be 0.03, the weight of the semantic feature should be 0.09, and the weight of the channel feature should be 0.02. The above is merely an illustrative example and is not intended to limit the specific numerical values ​​of the weights included in the first and second categories of features; it is simply not an exhaustive list.

[0043] In this embodiment, the user uploads N 7-Eleven store addresses, i.e., the first store record; based on the first store record, the user queries the POI database for N 7-Eleven store address records related to the N first store records, i.e., the second store record. Here, N is an integer greater than or equal to 1. The first and second store records are preprocessed to remove noise and other redundant information, resulting in a 7-Eleven store store dataset text. The 7-Eleven store store dataset text is then labeled. The 7-Eleven store store dataset text is input into a BiLSTM model, which parses the 7-Eleven store store dataset text to obtain store names, address elements, and vectors. The target weights of the first and second types of features for the first and second store records are determined. The target weights of the first and second types of features, along with the first and second features, are input into a similarity prediction model. If the similarity between the first and second store records output by the similarity prediction model is 0.9, then the first and second store records are considered to be the same store.

[0044] The technical solution of this disclosure determines a first feature of a first network point record and a second feature of a second network point record, as well as target weights corresponding to the first and second types of features. The target weights corresponding to the first and second types of features, along with the first and second features, are input into a similarity prediction model to obtain the similarity between the first and second network point records output by the similarity prediction model. This improves the accuracy of similarity prediction, enabling accurate prediction of the similarity between the first and second network point records, thus improving the matching accuracy between them and enabling effective cleaning and verification of existing network points. Furthermore, the ability to manually adjust the weights of some features further enhances the matching accuracy between the first and second network point records.

[0045] In some embodiments, the similarity determination method may further include:

[0046] S105: If the similarity between the first node record and the second node record exceeds the preset similarity threshold, normalize the first node record and the second node record.

[0047] In this embodiment, the preset similarity threshold can be set or adjusted according to requirements such as accuracy or speed. For example, if the preset similarity value is 0.5, and the similarity between the first and second halftone records is 0.7, then the first and second halftone records will be normalized. Alternatively, if the similarity is 0.4, then it can be determined that the first and second halftone records are not the same halftone. The above are merely illustrative examples and are not intended to limit the preset similarity threshold to all possibilities; an exhaustive list is not provided here.

[0048] In this embodiment of the disclosure, the normalization process includes: deduplicating two identical nodes and retaining only one of them. The normalization process may further include: merging two identical nodes into a single node.

[0049] In this embodiment, the user uploads P addresses of Yonghui Supermarket stores, forming the first store record; the user then uploads Q addresses of Yonghui Supermarket stores, forming the second store record. The first and second store records are preprocessed to remove noise and other redundant information, resulting in a text dataset of Yonghui Supermarket store locations. This text dataset is then labeled. The text dataset is input into a BiLSTM model, which parses it to obtain store names, address elements, and vectors. The target weights for the first and second types of features in both the first and second store records are then determined. The target weights corresponding to the first and second types of features, along with the first and second features themselves, are input into the similarity prediction model. If the similarity between the i-th address among the P network addresses output by the similarity prediction model and any network address among the Q network addresses is lower than a preset similarity threshold (assumed to be 0.8), then the i-th network address among the P network addresses is not normalized. If the similarity between the j-th network address among the P network addresses and the k-th network address among the Q network addresses is higher than the preset similarity threshold (assumed to be 0.8), then the j-th network address among the P network addresses and the k-th network address among the Q network addresses are normalized and merged into one network address. Here, 1≤i≤P, 1≤j≤Q, 1≤k≤Q. By identifying different address elements and store name elements through the BiLSTM model, the matching accuracy between the first and second network address records is improved when non-standard addresses are input, thereby achieving the purpose of cleaning and verifying existing network addresses. In addition, by adjusting the weights of the XGBoost output, manual intervention is possible, thereby improving the matching accuracy between the first and second network point records.

[0050] The technical solution of this disclosure determines the relationship between the first and second site records based on the similarity between the first site record and the second site record. This provides more accurate data support for matching the first and second site records, thereby helping to improve the accuracy of the matching results between the first and second site records.

[0051] In some embodiments, the similarity determination method may further include:

[0052] S106: Obtain the number of target outlets;

[0053] S107: Determine the number and address of the target network points that have been deployed based on the similarity between the first target network point included in the first network point record and the second target network point included in the second network point record;

[0054] S108: Based on the number and address of the already laid target points, determine the number and address of blank points that can be laid within the preset area of ​​the map.

[0055] In this embodiment of the disclosure, the target outlet is any key account (KA) outlet. Here, the target outlet can be a chain store. For example, 7-Eleven convenience store. Another example is Yonghui Supermarket. Yet another example is Bianlifeng. The above are merely illustrative examples and are not intended to limit all possible target outlets; they are simply not exhaustive.

[0056] In this embodiment of the disclosure, the target number of outlets refers to the planned number of outlets within a certain geographical area. For example, the planned number of 7-Eleven convenience stores in Shenzhen is 30. Another example is the planned number of Yonghui Supermarket outlets in Guangdong Province, which is 200. Yet another example is the planned number of Bianlifeng outlets in Nanshan District, which is 100. These are merely illustrative examples and are not intended to limit the target number of outlets; they are simply not exhaustive.

[0057] Figure 2 This diagram illustrates the process of identifying blank network points based on the target number of network points. Figure 2 As shown, based on the similarity between the first target network point included in the first network point record and the second target network point included in the second network point record, the number and address of the already laid target network points are determined; based on the number and address of the already laid target network points, the number and address of the blank network points that can be laid within the preset area of ​​the map are determined.

[0058] In this embodiment of the disclosure, the method for obtaining the number of target service points may include: obtaining the number of target service points uploaded by the user. The method for obtaining the number of target service points may also include: retrieving the names and addresses of target service points in a map to obtain the existing number of target service points; and evaluating the number of target service points based on the retrieval results. The above is merely an illustrative example and is not intended to limit all possible methods for obtaining the number of target service points; it is simply not exhaustive.

[0059] In some implementations, a user uploads the addresses of 100 7-Eleven stores, known as the first store record. Based on the first store record, S address records of 7-Eleven stores are obtained from the POI database, known as the second store record. The first and second store records are preprocessed to remove noise and other redundant information, resulting in a text dataset of 7-Eleven store locations. This text dataset is then labeled. The text dataset is input into a BiLSTM model, which parses it to obtain store names, address elements, and vectors. If, based on the 100 7-Eleven store addresses uploaded by the user, the number of target stores already established is determined to be 60, and the addresses of these 60 target stores are determined, then the number of unestablished stores that can be established within a preset area on the map, and their addresses, can be determined.

[0060] The technical solution of this disclosure embodiment can determine the number and address of the laid target points based on the similarity between the first target point and the second target point, and then determine the number and address of the blank points that can be laid within the preset area of ​​the map, thereby helping to achieve deduplication of existing points and discovery of blank points.

[0061] In some embodiments, the first type of features may include location features and structured address features. The second type of features may include address text features, outlet name features, semantic features, and channel features.

[0062] In some embodiments, S103 may include:

[0063] S1031: Obtain feature parameter information of the similarity prediction model;

[0064] S1032: Determine the default weights corresponding to the first type of feature and the second type of feature based on the feature parameter information, and determine the default weights corresponding to the second type of feature as the target weights of the second type of feature;

[0065] S1033: Obtain weight allocation information, which is used to indicate the weight ratio of each feature included in the first type of feature;

[0066] S1034: Determine the target weight of the first type of feature based on the weight allocation information and the default weight corresponding to the first type of feature.

[0067] In this embodiment of the disclosure, the target weight of the first type of feature includes: the weight value of each feature included in the first type of feature. The target weight of the second type of feature includes: the weight value of each feature included in the second type of feature.

[0068] In this embodiment, the feature parameter information is the weight parameter data corresponding to each feature parameter obtained by the similarity prediction model based on the training data. How the similarity prediction model is trained will be described in detail later and will not be repeated here.

[0069] In this embodiment of the disclosure, the weight allocation information is used to indicate the weight percentage of each feature included in the first type of features. The weight allocation information can be determined based on the weight allocation rules of the first type of features. The weight allocation rules can be obtained by: the user formulating weight allocation rules based on actual conditions. The weight allocation rules can also be obtained by: automatically generating weight allocation rules based on the first and second network point records input by the user. The weight allocation rules can also be obtained by: a default weight allocation rule obtained based on a similarity prediction model.

[0070] Figure 3 This diagram illustrates how the weights of each feature are determined based on the feature parameter information of a similarity prediction model. Figure 3 As shown, the feature parameter information of the similarity prediction model is obtained; based on the feature parameter information, the default weights corresponding to the first and second types of features are determined. The similarity prediction model determines the target weights corresponding to the first and second types of features based on the data corresponding to each feature parameter obtained from the training data.

[0071] For example, if the similarity prediction model yields a sum of weights of 0.4 for the first type of features and 0.6 for the second type of features, with the weight of the location feature in the first type of features being 0.21 and the weight of the structured address feature being 0.19, and the sum of weights of the second type of features being 0.7, where the weight of the address text feature is 0.46, the weight of the outlet / store name feature is 0.03, the weight of the semantic feature is 0.09, and the weight of the channel feature is 0.02, then since the weights of the first type of features can be manually adjusted, the weight of the location feature can be adjusted to 0.26, and the weight of the structured address feature can be adjusted to 0.14. Alternatively, the location feature can be adjusted to 0.19, and the structured address feature can be adjusted to 0.21. Since the weights of the second type of features cannot be manually adjusted, the weight of the address text feature should be 0.46, the weight of the outlet / store name feature should be 0.03, the weight of the semantic feature should be 0.09, and the weight of the channel feature should be 0.02. The above is merely an illustrative example and is not intended to limit the specific numerical values ​​of the weights included in the first and second categories of features; it is simply not an exhaustive list.

[0072] The technical solution of this disclosure, based on weight allocation information, determines the weight of the first type of feature, which improves the flexibility of similarity determination and thus improves the accuracy of similarity determination between the first network record and the second network record.

[0073] In some embodiments, S104 may further include:

[0074] S1041: Based on the address text features in the first feature of the first network point record sample and the address text features in the second feature of the second network point record sample, determine the predicted address text similarity value between the first network point record sample and the second network point record sample.

[0075] S1042: Based on the structured address features in the first feature of the first network point record sample and the structured address features in the second feature of the second network point record sample, determine the predicted value of the structured address similarity between the first network point record sample and the second network point record sample.

[0076] S1043: Based on the store name feature in the first feature of the first store record sample and the store name feature in the second feature of the second store record sample, determine the predicted store name similarity value between the first store record sample and the second store record sample.

[0077] S1044: Based on the semantic features in the first feature of the first network point record sample and the semantic features in the second feature of the second network point record sample, determine the predicted semantic similarity value between the first network point record sample and the second network point record sample.

[0078] S1045: Based on the channel features in the first feature of the first branch record sample and the channel features in the second feature of the second branch record sample, determine the predicted channel similarity value between the first branch record sample and the second branch record sample.

[0079] S1046: Based on the positional features in the first feature of the first network point record sample and the positional features in the second feature of the second network point record sample, determine the predicted spatial distance similarity value between the first network point record sample and the second network point record sample.

[0080] S1047: Based on the predicted values ​​of address text similarity, structured address similarity, store name similarity, semantic similarity, channel similarity, and spatial distance similarity, determine the predicted similarity values ​​of the first store record sample and the second store record sample.

[0081] In this embodiment of the disclosure, the address text features are determined by comparing the address texts of the first and second branch records and based on the correlation between the two address texts. For example, the address text features are obtained by comparing the number of characters in the address texts of the first and second branch records. Another example is by comparing the overlapping words in the address texts of the first and second branch records. The above are merely illustrative examples and are not intended to limit all possible methods of obtaining address text features; they are simply not exhaustive.

[0082] In this embodiment of the disclosure, a first outlet record and a second outlet record are obtained; the address text feature of the first outlet record is No. 2118 Jintian Road, Futian District, Shenzhen, Guangdong Province, near the Modern International Building, and the address text feature of the second outlet record is 102A, Modern International Building, Jintian Road; by comparing the number of repeated characters between the first outlet record and the second outlet record, which is 8, the address text similarity prediction value between the first outlet record and the second outlet record can be obtained.

[0083] In this embodiment, the structured address features are determined by comparing the address text of the first and second network node records and based on their address text structures. For example, the structured address features are obtained by comparing whether the text addresses of the first and second network node records are symmetrical. Another example is by comparing the overlapping text portions of the text addresses of the first and second network node records. These are merely illustrative examples and are not intended to limit all possible methods of obtaining structured address features; they are simply not exhaustive.

[0084] In this embodiment of the disclosure, a first location record and a second location record are obtained; the address text of the first location record is No. 2118 Jintian Road, Futian District, Shenzhen, Guangdong Province, near the Modern International Building, and the address text of the second location record is 102A, Modern International Building, Jintian Road; by comparing the structurally overlapping text "Modern International Building, Jintian Road" between the first location record and the second location record, the structured address feature similarity prediction value between the first location record and the second location record can be obtained.

[0085] In this embodiment of the disclosure, the store name feature is obtained by comparing the store name text of the first store record and the second store record. For example, the store name feature can be obtained by comparing the number of characters in the store name text of the first store record and the second store record. Another example is by comparing the overlapping words in the store name text of the first store record and the second store record. The above are merely illustrative examples and are not intended to limit all possible methods of obtaining store name features; they are simply not exhaustive.

[0086] In this embodiment, a first outlet record and a second outlet record are obtained; the store name text feature of the first outlet record is Huayou Convenience Store (Modern Building Store), and the store name text feature of the second outlet record is Huayou Convenience Modern International Store; by comparing the number of repeated characters between the first outlet record and the second outlet record, which is 7, the address text similarity prediction value between the first outlet record and the second outlet record can be obtained.

[0087] In this embodiment of the disclosure, the semantic features are obtained by analyzing the semantics of the first and second outlet records. For example, semantic features are obtained by analyzing the meaning of the store name text in the first and second outlet records. As another example, semantic features are obtained by analyzing the meaning of the address text in the first and second outlet records. The above are merely illustrative examples and are not intended to limit all possible ways of obtaining semantic features; they are simply not exhaustive.

[0088] In this embodiment of the disclosure, a first outlet record and a second outlet record are obtained; the store name text feature of the first outlet record is Huayou Convenience Store (Modern Building Store), and the store name text feature of the second outlet record is Huayou Convenience Modern International Store; by comparing the semantic features between the first outlet record and the second outlet record, the address text similarity prediction value between the first outlet record and the second outlet record can be obtained.

[0089] In this embodiment of the disclosure, the channel characteristics are obtained by analyzing the channels of the first outlet record and the second outlet record. For example, the channel characteristics are obtained by analyzing the channel of the store name text in the first outlet record and the second outlet record. Another example is that the channel characteristics are obtained by analyzing the channel of the address text in the first outlet record and the second outlet record. The above are merely illustrative examples and are not intended to limit all possible ways of obtaining channel characteristics; they are simply not exhaustive.

[0090] In this embodiment of the disclosure, a first outlet record and a second outlet record are obtained; the store name text feature of the first outlet record is Huayou Convenience Store (Modern Building Store), and the store name text feature of the second outlet record is Huayou Hair Salon Modern International Store; by comparing the channel features between the first outlet record and the second outlet record, it is clear that one is a convenience store and the other is a hair salon, and the channels are different, so the address text similarity prediction value between the first outlet record and the second outlet record can be obtained.

[0091] In this embodiment of the disclosure, the location feature is obtained by analyzing the spatial distance between the first outlet record and the second outlet record. For example, the location feature is obtained by searching the distance between the text of the store names of the first outlet record and the second outlet record on a map. Another example is that the location feature is obtained by searching the distance between the text of the addresses of the first outlet record and the second outlet record on a map. The above are merely illustrative examples and are not intended to limit all possible methods of obtaining location features; they are simply not exhaustive.

[0092] In this embodiment of the disclosure, a first outlet record and a second outlet record are obtained; the store name text feature of the first outlet record is Huayou Convenience Store (Modern Building Store), and the store name text feature of the second outlet record is Huayou Hair Salon Modern International Store; by comparing the spatial distance between the first outlet record and the second outlet record on the map, the location features of the two are obtained, and the address text similarity prediction value of the first outlet record and the second outlet record can be obtained.

[0093] If the predicted value is 0, it means that the first network record and the second network record are not the same network; if the predicted value is 1, it means that the first network record and the second network record are the same network.

[0094] This disclosure does not restrict the execution order of S1041, S1042, S1043, S1044, S1045 and S1046.

[0095] The technical solution of this disclosure, based on address text similarity prediction value, structured address similarity prediction value, outlet name similarity prediction value, semantic similarity prediction value, channel similarity prediction value, and spatial distance similarity prediction value, determines the similarity prediction value of the first outlet record sample and the second outlet record sample. This can provide comprehensive data for confirming the similarity between the first outlet record and the second outlet record, thereby helping to improve the accuracy of confirming the similarity between the first outlet record and the second outlet record.

[0096] It should be understood that Figure 2 and Figure 3 The schematic diagrams shown are merely illustrative and not limiting, and are scalable; those skilled in the art can use them as a basis. Figure 2 and Figure 3 Even with various obvious changes and / or substitutions to the examples, the resulting technical solutions still fall within the scope of this disclosure.

[0097] This disclosure provides a method for training a similarity prediction model, which can be applied to electronic devices. The following will be combined with... Figure 4 The flowchart shown illustrates a similarity prediction model training method provided in this disclosure. It should be noted that although the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order.

[0098] S401: Obtain the training data, which includes the first feature of the first dot record sample and the second feature of the second dot record sample, as well as the ground truth of the similarity between the first dot record sample and the second dot record sample.

[0099] S402: Input the first feature of the first dot record sample and the second feature of the second dot record sample into the model to be trained, and obtain the similarity prediction value of the first dot record sample and the second dot record sample output by the model to be trained.

[0100] S403: Based on the similarity prediction values ​​of the first and second network point record samples, and the ground truth similarity values ​​of the first and second network point record samples, train the model to be trained to obtain a similarity prediction model.

[0101] The training data includes first site record samples and second site record samples, as well as first features of the first site record samples and second features of the second site record samples, and may also include first-class features and second-class features.

[0102] In this embodiment of the disclosure, the first and second node record samples can be preprocessed. For example, the first and second node record samples can be converted from simplified to traditional Chinese characters. Another example is the conversion between full-width and half-width characters. Yet another example is the normalization of symbols and numbers in the first and second node record samples. The above are merely illustrative examples and are not intended to limit all possible preprocessing methods; they are simply not exhaustive.

[0103] In this embodiment of the disclosure, the ground truth value for the similarity between the first network point record sample and the second network point record sample can be a manually labeled ground truth value or a reliable ground truth value obtained from a data source. It should be noted that this disclosure does not limit the source of the ground truth value.

[0104] In this embodiment of the disclosure, the similarity prediction model is used to output a similarity prediction result based on the input data. The similarity prediction result may include the predicted similarity values ​​of the first network point record sample and the second network point record sample.

[0105] The model to be trained is used to obtain a first feature of a first-type network record sample and a second feature of a second-type network record sample based on the training data, and outputs a similarity prediction value between the first-type network record sample and the second-type network record sample. Based on this, it can be understood that in this embodiment, the model to be trained may include: at least one learning model, used at least to obtain the weights corresponding to the first-type feature and the second feature based on the training data; at least one feature acquisition model, used at least to acquire the first feature of the first-type network record sample and the second feature of the second-type network record sample; at least one prediction model, used to predict the predicted values ​​of the first-type network record sample and the second-type network record sample; the similarity prediction model is obtained by training the model to be trained using the training data, and therefore has the same model structure as the model to be trained, the difference being that the model parameters are updated after training.

[0106] Table 1 shows examples of positive and negative samples in the training data. Positive samples can be collected text records from a large number of dots, or they can be user-inputted historical records. Negative samples can be random samples of non-positive samples from the user-input data. The positive and negative samples are input into the model to be trained to obtain labels. If the label is 0, it indicates that the first dot record and the second dot record are dissimilar; if the label is 1, it indicates that the first dot record and the second dot record are similar.

[0107]

[0108] Table 1

[0109] Therefore, training the model using the predicted similarity values ​​of the first and second netpoint record samples, as well as the ground truth similarity values, can improve the accuracy of the trained similarity prediction model, thereby enhancing the accuracy of the model's output of the predicted similarity values ​​for the first and second netpoint record samples. Furthermore, based on the predicted similarity values ​​of the first and second netpoint record samples, it is possible to further deduplicate existing netpoints and identify blank netpoints.

[0110] In some embodiments, the similarity prediction model training method may further include:

[0111] S404: Retrieve multiple branch record texts from the branch record database;

[0112] S405: Use a feature acquisition model to parse multiple network point record texts to obtain the first feature of the first network point record sample and the second feature of the second network point record sample.

[0113] In this embodiment of the disclosure, pre-training data is obtained from a site record database, which includes site record text collected from the Internet; it also includes historical site records input by the user; and may also include site record samples generated based on map retrieval.

[0114] In this embodiment, the first feature and the second feature are obtained based on a feature acquisition model. The feature acquisition model is a pre-trained BiLSTM model, which can be a component of a similarity prediction model. The BiLSTM model uses the feature acquisition model to parse multiple dotted record texts to obtain the first feature of the first dotted record sample and the second feature of the second dotted record sample.

[0115] In this way, the first and second features can be better identified, which can improve the training speed of the similarity prediction model and further improve the prediction accuracy of the first and second network records.

[0116] In some embodiments, the similarity prediction model training method may further include:

[0117] S406: Obtain the store name features corresponding to the first and second store record samples based on the store name domain term dictionary;

[0118] S407: Obtain the semantic features corresponding to the first and second dot record samples based on the normalized word dictionary;

[0119] S408: Obtain the channel features corresponding to the first and second outlet record samples based on the channel matching dictionary.

[0120] In this embodiment, the domain dictionary for store names is a pre-set dictionary of domains related to store names. For example, it might include details of XXX chain convenience stores, YYY chain supermarkets, and ZZZ chain hotels. The content in this domain dictionary can be added to or removed as needed.

[0121] In this embodiment of the disclosure, the normalized word dictionary is a pre-set dictionary of different representations of the same store. For example, "711" and "seven eleven" are normalized to "711". Here, the contents of this normalized word dictionary can be added to or deleted from as needed.

[0122] In this embodiment, the channel matching dictionary is a pre-set dictionary about the business scope of the outlet. Channels include, but are not limited to, business scopes such as cosmetics, retail, hairdressing, and clothing. For example, CC Hair Salon cannot be matched as CC Hotel. The content in this channel matching dictionary can be added to or deleted as needed.

[0123] Figure 5 This diagram illustrates the process of mining features from dot record samples using a dictionary approach, such as... Figure 5 As shown, the network store name features corresponding to the first and second network store record samples are obtained based on the network store name domain word dictionary; the semantic features corresponding to the first and second network store record samples are obtained based on the normalized word dictionary; and the channel features corresponding to the first and second network store record samples are obtained based on the channel matching dictionary.

[0124] This disclosure does not restrict the execution order of S406, S407 and S408.

[0125] In this way, we can better determine the characteristics of store names, semantic features, and channel features, improve the training speed of similarity prediction models, and further improve the prediction accuracy of the first and second store records.

[0126] In similarity prediction model training methods, both the first and second features include one or more of the following features: location features; address text features; structured address features; store name features; semantic features; and channel features. For example... Figure 6As shown, the similarity prediction model's process includes a feature acquisition stage, a weight adjustment stage, and a model prediction stage. In the feature acquisition stage, a massive amount of network point record text is collected, and the elements related to store names and addresses are labeled. A BiLSTM sequence labeling model is trained. A domain-specific dictionary for network point names, a normalized dictionary, and a channel matching dictionary are established. The text is parsed to obtain the following levels of relevant features from the XGBoost model: spatial distance, address text similarity level, structured address matching level, store name similarity level, semantic similarity level, and channel matching level. In the weight adjustment stage, the weights are manually adjusted based on the input information. In the model prediction stage, the optimal XGBoost model and weights are read for prediction, obtaining the confidence score, i.e., the similarity, between the first and second network point record samples.

[0127] Thus, by combining location features, address text features, structured address features, outlet name features, semantic features, and channel features to obtain the predicted values ​​of the first outlet record and the second outlet record, accurate data support can be provided for model training, improving the accuracy of the similarity prediction model and enhancing the accuracy of model training.

[0128] In some embodiments, S402 includes:

[0129] S4021: Based on the address text features in the first feature of the first branch record sample and the address text features in the second feature of the second branch record sample, determine the predicted address text similarity value between the first branch record sample and the second branch record sample.

[0130] S4022: Based on the structured address features in the first feature of the first network point record sample and the structured address features in the second feature of the second network point record sample, determine the predicted value of the structured address similarity between the first network point record sample and the second network point record sample.

[0131] S4023: Based on the store name feature in the first feature of the first store record sample and the store name feature in the second feature of the second store record sample, determine the predicted store name similarity value between the first store record sample and the second store record sample.

[0132] S4024: Based on the semantic features in the first feature of the first network point record sample and the semantic features in the second feature of the second network point record sample, determine the predicted semantic similarity value between the first network point record sample and the second network point record sample.

[0133] S4025: Based on the channel features in the first feature of the first branch record sample and the channel features in the second feature of the second branch record sample, determine the predicted channel similarity value between the first branch record sample and the second branch record sample.

[0134] S4026: Based on the positional features in the first feature of the first network point record sample and the positional features in the second feature of the second network point record sample, determine the predicted spatial distance similarity value between the first network point record sample and the second network point record sample;

[0135] S4027: Based on the predicted values ​​of address text similarity, structured address similarity, store name similarity, semantic similarity, channel similarity, and spatial distance similarity, determine the predicted similarity values ​​of the first store record sample and the second store record sample.

[0136] Thus, based on the predicted values ​​of address text similarity, structured address similarity, store name similarity, semantic similarity, channel similarity, and spatial distance similarity, the predicted similarity values ​​of the first and second store record samples are determined. This provides more accurate data support for predicting the similarity between the first and second store records, thereby improving the training and optimization effect of the similarity prediction model and further enhancing the accuracy of the similarity prediction model in predicting the similarity between the first and second store records.

[0137] Figure 7 The diagram illustrates the overall process of training a similarity prediction model, as follows: Figure 7 As shown, the overall process is divided into two main stages: obtaining vector representations of network point parsing elements and training a similarity prediction model. Specifically, network point record text is obtained from a FMCG network point database. This database is preprocessed, and a trained BiLSTM sequence labeling model is used to parse the preprocessed network point records, obtaining network point names, address elements, and vectors. The preprocessed network point record text is then manually or intelligently labeled to obtain the network point dataset text. Based on the labeled network point dataset text and the network point names, address elements, and vectors obtained from the BiLSTM sequence labeling model, relevant features of the network point dataset text are obtained, such as spatial distance features, address text features, structured address features, store name features, semantic features, and channel features. The labeled network point dataset is divided into training, debugging, and testing sets. Based on the training and debugging sets, multiple similarity prediction models with an XGBoost structure are trained. Based on the testing set, the optimal similarity prediction model is evaluated from among the multiple similarity prediction models.

[0138] By identifying different address and store name elements using the BiLSTM model and performing structured matching based on these elements, non-standard addresses can be effectively identified. The BiLSTM model is used for sequence labeling tasks on massive datasets, exhibiting strong domain characteristics and effectively mitigating segmentation errors caused by general word segmentation tools. The weights output by XGBoost can be adjusted to allow for human intervention.

[0139] This disclosure proposes a highly efficient similarity prediction scheme, which predicts the similarity between the first and second network point records and determines the deployment status of the target network point. This enables the deduplication of existing network points and the discovery of blank network points. Furthermore, the improved accuracy of similarity prediction not only increases the accuracy of the comparison between the first and second network point records but also saves manpower and resources, thereby maximizing resource utilization.

[0140] This disclosure provides a similarity determination device, such as... Figure 8 As shown, the similarity determination device may include: a first acquisition module 801, used to acquire a first network node record and a second network node record; a first determination module 802, used to determine a first feature of the first network node record and a second feature of the second network node record, wherein both the first feature and the second feature include a first type of feature and a second type of feature, the first type of feature being a feature with adjustable weights and the second type of feature being a feature with non-adjustable weights; a second determination module 803, used to determine the target weights corresponding to the first type of feature and the second type of feature respectively; and a second acquisition module 804, used to input the target weights corresponding to the first type of feature and the second type of feature, the first feature and the second feature into a similarity prediction model, and obtain the similarity between the first network node record and the second network node record output by the similarity prediction model.

[0141] In some embodiments, the similarity determination apparatus may further include: a processing module 805. Figure 8 (not shown in the image) is used to normalize the first and second node records when the similarity between the first and second node records exceeds a preset similarity threshold.

[0142] In some embodiments, the similarity determination device may further include: a third acquisition module 806. Figure 8 (Not shown in the image), used to obtain the number of target outlets; the third determining module 807 ( Figure 8 (Not shown in the image), used to determine the number and address of the deployed target network points based on the similarity between the first target network point included in the first network point record and the second target network point included in the second network point record; the fourth determining module 808 ( Figure 8 (Not shown in the image), used to determine the number and address of blank points that can be laid within a preset area of ​​the map based on the number and address of the already laid target points.

[0143] In some embodiments, the second determining module 803 includes: a first acquiring submodule, configured to acquire feature parameter information of a similarity prediction model; a first determining submodule, configured to determine the default weights corresponding to a first type of feature and a second type of feature based on the feature parameter information, and to determine the default weights corresponding to the second type of feature as the target weights of the second type of feature; a second acquiring submodule, configured to acquire weight allocation information, the weight allocation information being used to indicate the weight ratios of each feature included in the first type of feature; and a second determining submodule, configured to determine the target weights of the first type of feature based on the weight allocation information and the default weights corresponding to the first type of feature.

[0144] Those skilled in the art should understand that the functions of each processing module in the similarity determination device of this disclosure can be understood with reference to the relevant description of the foregoing similarity determination method. Each processing module in the similarity determination device of this disclosure can be implemented by an analog circuit that implements the functions described in the embodiments of this disclosure, or by running software that performs the functions described in the embodiments of this disclosure on an electronic device.

[0145] The similarity determination device of this disclosure can predict the similarity between the queried network points and the network points in the POI database, thereby realizing the deduplication of existing network points and the discovery of blank network points.

[0146] This disclosure provides a similarity prediction model training device, such as... Figure 9 As shown, the similarity prediction model training device includes: a fourth acquisition module 901, used to acquire training data, which includes a first feature of a first dot record sample and a second feature of a second dot record sample, as well as the ground truth similarity between the first dot record sample and the second dot record sample; a prediction module 902, used to input the first feature of the first dot record sample and the second feature of the second dot record sample into the model to be trained, and obtain the similarity prediction value of the first dot record sample and the second dot record sample output by the model to be trained; and a training module 903, used to train the model to be trained based on the similarity prediction value of the first dot record sample and the second dot record sample, as well as the ground truth similarity value of the first dot record sample and the second dot record sample, to obtain a similarity prediction model.

[0147] In this embodiment of the disclosure, the similarity prediction model training device may further include: a fifth acquisition module ( Figure 9 (Not shown in the image), used to retrieve multiple branch record texts from the branch record database; parsing module ( Figure 9 (not shown in the image) is used to parse multiple network point record texts using a feature acquisition model to obtain the first feature of the first network point record sample and the second feature of the second network point record sample.

[0148] In this embodiment of the disclosure, the similarity prediction model training device may further include: a sixth acquisition module ( Figure 9 (Not shown in the image), used to obtain the store name features corresponding to the first and second store record samples based on the store name domain term dictionary; the seventh acquisition module ( Figure 9 (not shown in the image), used to obtain the semantic features corresponding to the first and second dot record samples respectively based on the normalized word dictionary; the eighth acquisition module ( Figure 9 (not shown in the text) is used to obtain the channel features corresponding to the first and second outlet record samples based on the channel matching dictionary.

[0149] In this embodiment of the disclosure, the first feature and the second feature in the similarity prediction model training device each include one or more of the following features: location feature; address text feature; structured address feature; outlet name feature; semantic feature; channel feature.

[0150] In this embodiment of the disclosure, the prediction module 902 includes: a first prediction submodule, used to determine the address text similarity prediction value between the first and second outlet record samples based on the address text features in the first feature of the first outlet record sample and the address text features in the second feature of the second outlet record sample; a second prediction submodule, used to determine the structured address similarity prediction value between the first and second outlet record samples based on the structured address features in the first feature of the first outlet record sample and the structured address features in the second feature of the second outlet record sample; a third prediction submodule, used to determine the outlet name similarity prediction value between the first and second outlet record samples based on the outlet name features in the first feature of the first outlet record sample and the outlet name features in the second feature of the second outlet record sample; and a fourth prediction submodule, used to determine the outlet name similarity prediction value between the first and second outlet record samples based on the semantic features in the first feature of the first outlet record sample. The first and second branch record samples are used to determine the semantic similarity prediction values ​​of the first and second branch record samples based on the semantic features in the second feature of the first and second branch record samples; the fifth prediction submodule is used to determine the channel similarity prediction values ​​of the first and second branch record samples based on the channel features in the first feature of the first and second branch record samples; the sixth prediction submodule is used to determine the spatial distance similarity prediction values ​​of the first and second branch record samples based on the location features in the first feature of the first and second branch record samples; the seventh prediction submodule is used to determine the similarity prediction values ​​of the first and second branch record samples based on the address text similarity prediction value, structured address similarity prediction value, branch store name similarity prediction value, semantic similarity prediction value, channel similarity prediction value, and spatial distance similarity prediction value.

[0151] Those skilled in the art should understand that the functions of each processing module in the similarity prediction model training device of this disclosure can be understood with reference to the relevant description of the similarity prediction model training method described above. Each processing module in the similarity prediction model training device of this disclosure can be implemented by an analog circuit that implements the functions described in the embodiments of this disclosure, or it can be implemented by running software that performs the functions described in the embodiments of this disclosure on an electronic device.

[0152] The similarity prediction model training apparatus of this disclosure can improve the accuracy of the trained similarity prediction model, thereby improving the accuracy of the similarity prediction values ​​output by the similarity prediction model.

[0153] This disclosure provides a scenario illustration of a similarity determination method, such as... Figure 10As shown. As previously described, the similarity determination method provided in this disclosure is applied to an electronic device. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The electronic device performs the following functions: acquiring a first site record and a second site record; determining a first feature of the first site record and a second feature of the second site record, wherein both the first feature and the second feature include a first type of feature and a second type of feature, the first type of feature being a feature with adjustable weights, and the second type of feature being a feature with non-adjustable weights; determining the target weights corresponding to the first type of feature and the second type of feature respectively; inputting the target weights corresponding to the first type of feature and the second type of feature, the first feature, and the second feature into a similarity prediction model, and obtaining the similarity between the first site record and the second site record output by the similarity prediction model.

[0154] The first and second site records, as well as the first and second features, can be obtained from the data source. The data source can be various forms of data storage devices, such as laptops, desktop computers, workstations, personal digital assistants (PDAs), servers, blade servers, mainframes, and other suitable computers. The data source can also represent various forms of mobile devices, such as PDAs, cellular phones, smartphones, wearable devices, and other similar computing devices. Furthermore, the data source and the user terminal can be the same device.

[0155] It should be understood that Figure 10 The scene diagrams shown are merely illustrative and not restrictive; those skilled in the art can interpret them based on... Figure 10 Even with various obvious changes and / or substitutions to the examples, the resulting technical solutions still fall within the scope of this disclosure.

[0156] This disclosure also provides a scenario illustration of a similarity prediction model training method, such as... Figure 11As shown above, the similarity prediction model training method provided in this disclosure is applied to an electronic device. The electronic device is intended to represent various forms of digital computers, such as laptops, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The electronic device performs the following functions: acquiring training data, which includes a first feature of a first dot record sample and a second feature of a second dot record sample, as well as the ground truth similarity between the first dot record sample and the second dot record sample; inputting the first feature of the first dot record sample and the second feature of the second dot record sample into the model to be trained, obtaining the predicted similarity value of the first dot record sample and the second dot record sample output by the model to be trained; and training the model to be trained based on the predicted similarity value of the first dot record sample and the second dot record sample, and the ground truth similarity value of the first dot record sample and the second dot record sample, to obtain a similarity prediction model.

[0157] The training data can be obtained from a data source. The data source can be various forms of data storage devices, such as laptops, desktop computers, workstations, personal digital assistants (PDAs), servers, blade servers, mainframes, and other suitable computers. The data source can also represent various forms of mobile devices, such as PDAs, cellular phones, smartphones, wearable devices, and other similar computing devices. Furthermore, the data source and the user terminal can be the same device.

[0158] It should be understood that Figure 11 The scene diagrams shown are merely illustrative and not restrictive; those skilled in the art can interpret them based on... Figure 11 Even with various obvious changes and / or substitutions to the examples, the resulting technical solutions still fall within the scope of this disclosure.

[0159] The acquisition, storage, and application of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0160] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0161] Figure 12A schematic block diagram of an example electronic device 1200 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0162] like Figure 12 As shown, device 1200 includes a computing unit 1201, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 1202 or a computer program loaded from storage unit 1208 into random access memory (RAM) 1203. The RAM 1203 may also store various programs and data required for the operation of device 1200. The computing unit 1201, ROM 1202, and RAM 1203 are interconnected via bus 1204. An input / output (I / O) interface 1205 is also connected to bus 1204.

[0163] Multiple components in device 1200 are connected to I / O interface 1205, including: input unit 1206, such as keyboard, mouse, etc.; output unit 1207, such as various types of monitors, speakers, etc.; storage unit 1208, such as disk, optical disk, etc.; and communication unit 1209, such as network card, modem, wireless transceiver, etc. Communication unit 1209 allows device 1200 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0164] The computing unit 1201 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 1201 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. The computing unit 1201 performs the various methods and processes described above, such as similarity determination methods / similarity prediction model training methods. For example, in some embodiments, the similarity determination method / similarity prediction model training method can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 1208. In some embodiments, part or all of the computer program can be loaded and / or installed on device 1200 via ROM 1202 and / or communication unit 1209. When the computer program is loaded into RAM 1203 and executed by computing unit 1201, one or more steps of the similarity determination method / similarity prediction model training method described above can be performed. Alternatively, in other embodiments, computing unit 1201 can be configured to perform the similarity determination method / similarity prediction model training method by any other suitable means (e.g., by means of firmware).

[0165] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0166] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable similarity determination device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0167] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory, read-only memory, erasable programmable read-only memory (EPROM), flash memory, optical fiber, compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0168] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a cathode ray tube (CRT) or liquid crystal display (LCD) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0169] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication (e.g., a communication network) of any form or medium. Examples of communication networks include Local Area Networks (LANs), Wide Area Networks (WANs), and the Internet.

[0170] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0171] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0172] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A similarity determination method, comprising: Retrieve the records of the first and second branch locations; The first feature of the first outlet record and the second feature of the second outlet record are determined. Both the first feature and the second feature include a first type of feature and a second type of feature. The first type of feature includes location features and structured address features. The second type of feature includes address text features, outlet name features, semantic features, and channel features. The first type of feature is a feature with adjustable weights, and the second type of feature is a feature with non-adjustable weights. Determine the target weights corresponding to the first type of feature and the second type of feature respectively; wherein, determining the target weights corresponding to the first type of feature and the second type of feature respectively includes: obtaining feature parameter information of a similarity prediction model; determining the default weights corresponding to the first type of feature and the second type of feature respectively based on the feature parameter information; determining the default weights corresponding to the second type of feature as the target weights of the second type of feature; obtaining weight allocation information, the weight allocation information being used to indicate the weight ratios of each feature included in the first type of feature; and determining the target weights of the first type of feature based on the weight allocation information and the default weights corresponding to the first type of feature; The target weights corresponding to the first type of feature and the second type of feature, as well as the first feature and the second feature, are input into the similarity prediction model to obtain the similarity between the first network point record and the second network point record output by the similarity prediction model.

2. The method according to claim 1, further comprising: If the similarity between the first network record and the second network record exceeds a preset similarity threshold, the first network record and the second network record are normalized.

3. The method according to claim 1, further comprising: Obtain the number of target outlets; Based on the similarity between the first target network point included in the first network point record and the second target network point included in the second network point record, the number and address of the deployed target network points are determined. Based on the number and address of the already deployed target points, determine the number and address of any blank points that can be deployed within the preset area of ​​the map.

4. A method for training a similarity prediction model, comprising: Acquire training data, which includes a first feature of a first dot record sample and a second feature of a second dot record sample, as well as the ground truth similarity between the first dot record sample and the second dot record sample; wherein, the first feature and the second feature both include a first type of feature and a second type of feature; The first feature of the first network point record sample and the second feature of the second network point record sample are input into the model to be trained to obtain the similarity prediction value of the first network point record sample and the second network point record sample output by the model to be trained. Based on the similarity prediction values ​​of the first and second network point record samples, and the ground truth similarity values ​​of the first and second network point record samples, the model to be trained is trained to obtain a similarity prediction model; wherein, the training is configured such that the trained similarity prediction model has internal weight parameters associated with the first type of feature and the second type of feature respectively; the internal weight parameters are used as the basis for determining the default weights corresponding to the first type of feature and the second type of feature respectively when applied.

5. The method according to claim 4, further comprising: Retrieve multiple branch record texts from the branch record database; The feature acquisition model is used to parse the multiple network point record texts to obtain the first feature of the first network point record sample and the second feature of the second network point record sample.

6. The method according to claim 4, wherein, Both the first feature and the second feature include outlet name features, semantic features, and channel features; The method further includes: Based on the domain term dictionary of outlet names, obtain the outlet name features corresponding to the first outlet record sample and the second outlet record sample respectively; Semantic features corresponding to the first and second network point record samples are obtained based on a normalized word dictionary; The channel features corresponding to the first and second outlet record samples are obtained based on the channel matching dictionary.

7. The method according to claim 4, wherein, The step of inputting the first feature of the first dot record sample and the second feature of the second dot record sample into the model to be trained, and obtaining the similarity prediction value of the first dot record sample and the second dot record sample output by the model to be trained, includes: Based on the address text features in the first feature of the first branch record sample and the address text features in the second feature of the second branch record sample, the predicted address text similarity value between the first branch record sample and the second branch record sample is determined. Based on the structured address features in the first feature of the first branch record sample and the structured address features in the second feature of the second branch record sample, the predicted value of the structured address similarity between the first branch record sample and the second branch record sample is determined. Based on the store name feature in the first feature of the first store record sample and the store name feature in the second feature of the second store record sample, the predicted store name similarity value between the first store record sample and the second store record sample is determined. Based on the semantic features in the first feature of the first branch record sample and the semantic features in the second feature of the second branch record sample, the semantic similarity prediction value between the first branch record sample and the second branch record sample is determined. Based on the channel features in the first feature of the first branch record sample and the channel features in the second feature of the second branch record sample, the predicted channel similarity value between the first branch record sample and the second branch record sample is determined. Based on the location features in the first feature of the first network point record sample and the location features in the second feature of the second network point record sample, the spatial distance similarity prediction value between the first network point record sample and the second network point record sample is determined. Based on at least one of the address text similarity prediction value, the structured address similarity prediction value, the outlet name similarity prediction value, the semantic similarity prediction value, the channel similarity prediction value, and the spatial distance similarity prediction value, the similarity prediction value between the first outlet record sample and the second outlet record sample is determined.

8. A similarity determination device, comprising: The first acquisition module is used to acquire the first branch record and the second branch record; The first determining module is used to determine the first feature of the first outlet record and the second feature of the second outlet record. The first feature and the second feature both include a first type of feature and a second type of feature. The first type of feature includes location features and structured address features, and the second type of feature includes address text features, outlet name features, semantic features, and channel features. The first type of feature is a feature with adjustable weight, and the second type of feature is a feature with non-adjustable weight. The second determining module is used to determine the target weights corresponding to the first type of feature and the second type of feature, respectively. The second determining module includes: a first acquiring submodule, used to acquire feature parameter information of a similarity prediction model; the first determining submodule is used to determine the default weights corresponding to the first type of feature and the second type of feature based on the feature parameter information, and to determine the default weights corresponding to the second type of feature as the target weights of the second type of feature; a second acquiring submodule, used to acquire weight allocation information, the weight allocation information indicating the weight ratio of each feature included in the first type of feature; and a second determining submodule, used to determine the target weights of the first type of feature based on the weight allocation information and the default weights corresponding to the first type of feature. The second acquisition module is used to input the target weights corresponding to the first type of feature and the second type of feature, as well as the first feature and the second feature, into the similarity prediction model to obtain the similarity between the first network record and the second network record output by the similarity prediction model.

9. The apparatus according to claim 8, further comprising: The processing module is used to normalize the first network record and the second network record when the similarity between the first network record and the second network record exceeds a preset similarity threshold.

10. The apparatus according to claim 8, further comprising: The third acquisition module is used to acquire the number of target outlets; The third determining module is used to determine the number and address of the deployed target sites based on the similarity between the first target sites included in the first site record and the second target sites included in the second site record. The fourth determination module is used to determine the number and address of blank network points that can be laid within the preset area of ​​the map, based on the number and address of the already laid target network points.

11. A similarity prediction model training device, comprising: The fourth acquisition module is used to acquire training data, which includes a first feature of a first dot record sample and a second feature of a second dot record sample, as well as the ground truth similarity between the first dot record sample and the second dot record sample; wherein, the first feature and the second feature both include a first type of feature and a second type of feature; The prediction module is used to input the first feature of the first network point record sample and the second feature of the second network point record sample into the model to be trained, and obtain the similarity prediction value of the first network point record sample and the second network point record sample output by the model to be trained. A training module is used to train the model to be trained based on the similarity prediction values ​​of the first and second network point record samples, and the ground truth similarity values ​​of the first and second network point record samples, to obtain a similarity prediction model; wherein the training is configured such that the trained similarity prediction model has internal weight parameters associated with the first type of feature and the second type of feature respectively; the internal weight parameters are used as the basis for determining the default weights corresponding to the first type of feature and the second type of feature respectively during application.

12. The apparatus of claim 11, further comprising: The fifth acquisition module is used to retrieve multiple branch record texts from the branch record database; The parsing module is used to parse the multiple network point record texts using a feature acquisition model to obtain the first feature of the first network point record sample and the second feature of the second network point record sample.

13. The apparatus of claim 11, further comprising: The sixth acquisition module is used to acquire the store name features corresponding to the first store record sample and the second store record sample respectively based on the store name domain word dictionary; The seventh acquisition module is used to acquire the semantic features corresponding to the first network point record sample and the second network point record sample respectively based on the normalized word dictionary; The eighth acquisition module is used to acquire the channel features corresponding to the first outlet record sample and the second outlet record sample respectively based on the channel matching dictionary.

14. The apparatus according to claim 11, wherein, The prediction module includes: The first prediction submodule is used to determine the predicted address text similarity value between the first branch record sample and the second branch record sample based on the address text features in the first feature of the first branch record sample and the address text features in the second feature of the second branch record sample. The second prediction submodule is used to determine the predicted value of the structured address similarity between the first network point record sample and the second network point record sample based on the structured address features in the first feature of the first network point record sample and the structured address features in the second feature of the second network point record sample. The third prediction submodule is used to determine the predicted similarity value of the store names of the first store record sample and the second store record sample based on the store name features in the first feature of the first store record sample and the store name features in the second feature of the second store record sample. The fourth prediction submodule is used to determine the semantic similarity prediction value between the first network point record sample and the second network point record sample based on the semantic features in the first feature of the first network point record sample and the semantic features in the second feature of the second network point record sample. The fifth prediction submodule is used to determine the predicted channel similarity value between the first network point record sample and the second network point record sample based on the channel feature in the first feature of the first network point record sample and the channel feature in the second feature of the second network point record sample. The sixth prediction submodule is used to determine the spatial distance similarity prediction value between the first network point record sample and the second network point record sample based on the positional features in the first feature of the first network point record sample and the positional features in the second feature of the second network point record sample. The seventh prediction submodule is used to determine the similarity prediction value between the first outlet record sample and the second outlet record sample based on at least one of the address text similarity prediction value, the structured address similarity prediction value, the outlet name similarity prediction value, the semantic similarity prediction value, the channel similarity prediction value, and the spatial distance similarity prediction value.

15. An electronic device comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.

16. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-7.

17. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1-7.

Images