Address similarity model training method, address similarity determination method and device

Abstract

The invention discloses an address similarity model training method, an address similarity determination method and an address similarity determination device, which are used for accurately evaluating the similarity between addresses. The address similarity model training method comprises the steps of determining an address text of an address in a sample address pair, and splicing the address text and geographical location information of the address to obtain a spliced text; predicting the similarity of the sample address pairs based on an address similarity model and the spliced text; and based on the similarity and the similarity label of the sample address pair, performing optimization training on the address similarity model.

Description

Technical Field

[0001] The present application relates to the field of information processing technology, and in particular to a method for training an address similarity model, and a method and device for determining address similarity. Background Art

[0002] With the continuous development of Internet technology, the application of address similarity calculation technology is becoming more and more extensive. For example, in Internet financial business scenarios, address similarity calculation technology is used to identify abnormal users; in online e-commerce business scenarios, address similarity calculation technology is used to identify abnormal transaction behaviors, and so on. Summary of the invention

[0003] The purpose of the embodiments of the present application is to provide a method for training an address similarity model, a method and device for determining address similarity, so as to accurately evaluate the similarity between addresses.

[0004] In order to achieve the above objectives, the present application embodiment adopts the following technical solutions:

[0005] In a first aspect, an embodiment of the present application provides a method for training an address similarity model, including:

[0006] Determine the address text of the address in the sample address pair, and concatenate the address text with the geographical location information of the address to obtain a concatenated text;

[0007] Predicting the similarity of the sample address pairs based on the address similarity model and the concatenated text;

[0008] The address similarity model is optimized and trained based on the similarity and the similarity labels of the sample address pairs.

[0009] In a second aspect, an embodiment of the present application provides a method for determining address similarity, including:

[0010] Determine the address text of the address in the target address pair, and concatenate the address text with the geographic location information of the address to obtain a concatenated text;

[0011] Based on the address similarity model and the concatenated text, the similarity of the target address pair is predicted.

[0012] In a third aspect, an embodiment of the present application provides a training device for an address similarity model, including:

[0013] A splicing module, used for determining the address text of the address in the sample address pair, and splicing the address text with the geographical location information of the address to obtain a spliced ​​text;

[0014] A prediction module, used for predicting the similarity of the sample address pair based on the address similarity model and the concatenated text;

[0015] A training module is used to optimize and train the address similarity model based on the similarity and the similarity labels of the sample address pairs.

[0016] In a fourth aspect, an embodiment of the present application provides an address similarity determination device, including:

[0017] A splicing module, used for determining the address text of the address in the target address pair, and splicing the address text with the geographical location information of the address to obtain a spliced ​​text;

[0018] A prediction module is used to predict the similarity of the target address pair based on the address similarity model and the concatenated text.

[0019] In a fifth aspect, an embodiment of the present application provides an electronic device, including:

[0020] processor;

[0021] a memory for storing instructions executable by the processor;

[0022] The processor is configured to execute the instructions to implement the training method of the address similarity model provided in the first aspect; or, the processor is configured to execute the instructions to implement the address similarity determination method provided in the second aspect.

[0023] In a sixth aspect, an embodiment of the present application provides a computer-readable storage medium, which, when instructions in the storage medium are executed by a processor of an electronic device, enables the electronic device to execute the training method of the address similarity model provided in the first aspect; or, when instructions in the storage medium are executed by a processor of an electronic device, enables the electronic device to execute the address similarity determination method provided in the second aspect.

[0024] In the seventh aspect, an embodiment of the present application provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to cause a computer to execute part or all of the steps in the training method of the address similarity model provided in the first aspect or the address similarity determination method provided in the second aspect.

[0025] At least one of the above technical solutions adopted in the embodiments of the present application can achieve the following beneficial effects:

[0026] The address text and geographic location information of the address are used as prior information to train the address similarity model, so that the address similarity model can fully learn the similarities and differences between different addresses from the prior information, so as to accurately evaluate the similarity between different addresses, especially for the case where two addresses are similar in address text but have large differences in actual geographic location. During the training process, by splicing the address text and geographic location information of the address in the sample address pair, the resulting spliced ​​text can more objectively and accurately reflect the differences in geographic location and address text of the sample address pair; further, based on the address similarity model and the spliced ​​text, the similarity of the sample address pair is predicted, and based on the predicted similarity of the sample address pair and the similar pair label, the address similarity model is optimized and trained, which can effectively guide the address similarity model to fully learn the differences in geographic location and address text of different addresses, thereby improving the model training effect and accurately evaluating the similarity between different addresses. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0028] Figure 1 A flowchart of a method for training an address similarity model provided for one embodiment of the present application;

[0029] Figure 2 A schematic diagram of the structure of an address similarity model provided for an embodiment of the present application;

[0030] Figure 3 A flowchart of a method for training an address similarity model provided for another embodiment of the present application;

[0031] Figure 4 A flowchart of a method for determining address similarity provided for an embodiment of the present application;

[0032] Figure 5 A schematic diagram of the structure of a training device for an address similarity model provided in one embodiment of the present application;

[0033] Figure 6 A schematic diagram of the structure of an address similarity determination device provided for an embodiment of the present application;

[0034] Figure 7 A schematic diagram of the structure of an electronic device provided for one embodiment of the present application. DETAILED DESCRIPTION

[0035] In order to make the purpose, technical solution and advantages of the present application clearer, the technical solution of the present application will be clearly and completely described below in combination with the specific embodiments of the present application and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present application.

[0036] The terms "first", "second", etc. in this specification and claims are used to distinguish similar objects, and are not used to describe a particular order or precedence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application can be implemented in an order other than those illustrated or described herein. In addition, "and / or" in this specification and claims means at least one of the connected objects, and the character " / " generally means that the objects associated with each other are in an "or" relationship.

[0037] The applicant has studied the existing address similarity calculation methods and found that these methods determine the similarity of different addresses by evaluating the text similarity of different addresses. The defect of this is that two addresses with similar text but far apart actual geographical locations will be judged as similar addresses, resulting in inaccurate similarity calculation results, which in turn affects the accuracy of subsequent business processing results. For example, the similarity between the address text "Datun Road, Beijing" and the address text "Xiaotun Road, Beijing" is relatively high. If the similarity of the two addresses is evaluated only based on the similarity of the address text, the two addresses will be judged as similar addresses, but in fact they belong to two different addresses.

[0038] To this end, the technical solution proposed in the embodiment of the present application not only considers the address text of the address, but also the geographical location information of the address in the process of evaluating the address similarity. Specifically, the address text and geographical location information of the address are used as prior information to train the address similarity model, so that the address similarity model can fully learn the similarities and differences between different addresses from the prior information, so as to accurately evaluate the similarity between different addresses.

[0039] During the training process, by splicing the address text and geographical location information of the addresses in the sample address pairs, the resulting spliced ​​text can more objectively and accurately reflect the differences in geographical location and address text of the sample address pairs; further, based on the address similarity model and the spliced ​​text, the similarity of the sample address pairs is predicted, and based on the predicted similarity of the sample address pairs and the similar pair labels, the address similarity model is optimized and trained, which can effectively guide the address similarity model to fully learn the differences in geographical location and address text of different addresses, thereby improving the model training effect and accurately evaluating the similarity between different addresses.

[0040] Afterwards, using the trained address similarity model, the similarity between addresses can be accurately evaluated based on the address text and geographic location information, which is conducive to improving the accuracy of subsequent business processing results.

[0041] It should be understood that the training method of the address similarity model and the address similarity determination method provided in the embodiments of the present application can be performed by an electronic device. The so-called electronic device here can include terminal devices, such as smart phones, tablet computers, laptops, desktop computers, intelligent voice interaction devices, smart home appliances, smart watches, vehicle terminals, aircraft, etc.; or, the electronic device can also include a server, such as an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides cloud computing services.

[0042] The technical solutions provided by various embodiments of the present application are described in detail below in conjunction with the accompanying drawings.

[0043] Please refer to Figure 1 , is a flow chart of a method for training an address similarity model provided in an embodiment of the present application, the method comprising the following steps:

[0044] S102, determining the address text of the address in the sample address pair, and concatenating the address text with the geographical location information of the address to obtain a concatenated text.

[0045] The sample address pair includes two addresses to be evaluated. Each address has address text and geographic location information. The address text of an address refers to the text used to describe the address, such as "Province A, City B, District C". The geographic location information of an address refers to the information used to describe the geographic location represented by the address, such as but not limited to the longitude and latitude of the address in the geographic coordinate system, the location coordinates of the address on the map, etc. The geographic location information of an address can be obtained in various ways, such as but not limited to a combination of one or more of the following ways: obtained by manual marking, obtained by positioning with a positioning device, read from historical business logs, obtained by crawling with web crawler technology, etc.

[0046] The sample address pairs are marked with similarity labels. The similarity labels indicate the similarity of the sample address pairs. As an example, if the two addresses in the sample address pair are similar, the similarity label of the sample address pair is 1; if the two addresses in the sample address pair are not similar, the similarity label of the sample address pair is 0.

[0047] The sample address pair and its similarity label can be constructed in various ways, which are not limited in the embodiments of the present application. As an example, two addresses are randomly selected from the sample address set as a sample address pair. If the two addresses are similar, the similarity label of the sample address pair is marked as a first identifier, such as 1; if the two addresses are not similar, the similarity label of the sample address pair is marked as a second identifier, such as 0. The sample address set includes multiple addresses.

[0048] In practical applications, multiple sample address pairs can be constructed, and the address text and geographic location information of the addresses in the multiple sample address pairs can be used to train the address similarity model to improve the model training effect, thereby improving the prediction accuracy of the address similarity model.

[0049] In the embodiment of the present application, the above S102 can be implemented in various ways, which is not limited in the embodiment of the present application.

[0050] In one embodiment, the address text and geographic location information of each address in the sample address pair can be directly concatenated in order to obtain a concatenated text. For example, the sample address pair includes a first address and a second address, the address text of the first address is addr1_text, the geographic location information of the first address is geo_1, the address text of the second address is addr2_text, and the geographic location information of the second address is geo_2, and the concatenated text is obtained by concatenating in order:

[0051] [CLS]addr1_text[SEP]geo_1[SEP]addr2_text[SEP]geo_2[SEP].

[0052] Among them, [CLS] is the start identifier, indicating the beginning of the concatenated text, and [SEP] is the segmentation identifier.

[0053] In another embodiment, the spliced ​​text includes geographic splicing information and address text of each address. In this case, the above S102 includes the following steps: splicing the geographic location information of the two addresses in the sample address pair to obtain address splicing information, the address text of each address, and the address splicing information to obtain the spliced ​​text. For example, still taking the sample address pair including the first address and the second address as an example, the geographic location information of the two addresses is combined and identified with a segmentation identifier to obtain spliced ​​geographic information: geo_1[SEP]geo_2; then, based on the address text addr1_text of the first address, the address text addr2_text of the second address and the spliced ​​geographic information, three independent texts are obtained:

[0054] [CLS]addr1_text[SEP]

[0055] [CLS]addr2_text[SEP]

[0056] [CLS]geo_1[SEP]geo_2[SEP]

[0057] In the above implementation, by splicing the geographic location information of the two addresses in the sample address pair, the spliced ​​geographic information can objectively and accurately reflect the difference in the geographic location of the two addresses; at the same time, the address texts of the two addresses can objectively and accurately reflect the difference in the address texts of the two addresses. These two differences can guide the address similarity model to more fully learn the differences between different addresses, thereby more accurately predicting address similarity.

[0058] In another embodiment, the concatenated text includes a first text, a second text, and a third text. The above S102 includes the following steps: concatenating the address text and the geographic location information of the first address to obtain the first text; concatenating the address text and the geographic location information of the second address to obtain the second text; and concatenating the geographic location information of the first address and the geographic location information of the second address to obtain the third text.

[0059] For example, Figure 2 As shown, the address text addr1_text of the first address and the geographic location information geo_1 of the first address are concatenated to obtain a first text [CLS]addr1_text[SEP]geo_1[SEP], recorded as text1_geo1; the address text addr2_text of the second address and the geographic location information geo_2 of the second address are concatenated to obtain a second text [CLS]addr2_text[SEP]geo_2[SEP], recorded as text2_geo2; the geographic location information geo_1 of the first address and the geographic location information geo_2 of the second address are concatenated to obtain a third text [CLS]geo_1[SEP]geo_2[SEP], recorded as geo.

[0060] The first text spliced ​​together in the above manner can intuitively and accurately reflect the correlation between the address text of the first address and its geographic location information, the second text can intuitively and accurately reflect the correlation between the address text of the second address and its geographic location information, and the third text can intuitively and accurately reflect the difference in geographic location information between the first address and the second address. Therefore, the address similarity model can accurately obtain the difference between the first address and the second address in address text and geographic location information based on the first text, the second text and the third text, and thus accurately evaluate the similarity between the two addresses.

[0061] The embodiment of the present application shows some implementation methods of the above S102. Of course, it should be understood that the above S102 can also be implemented in other ways, and the embodiment of the present application does not limit this.

[0062] S104, predicting the similarity of the sample address pairs based on the address similarity model and the concatenated text.

[0063] The address similarity model may obtain the predicted similarity of the sample address pairs in various ways, which are not limited in the embodiments of the present application.

[0064] In one embodiment, the above S104 includes the following steps:

[0065] S141, encoding the first text, the second text, and the third text based on the address similarity model to obtain a first text vector, a second text vector, and a third text vector.

[0066] The first text vector is a text vector obtained by encoding the first text, and the dimension is n. The second text vector is a text vector obtained by encoding the second text, and the dimension is n. The third text vector is a text vector obtained by encoding the third text, and the dimension is n.

[0067] Encoding each text can be implemented in various ways, which is not limited in the embodiments of the present application. As an example, for each text in the first text, the second text, and the third text, the text is embedded (Embedding) to obtain a representation vector of a specified dimension; then, the representation vector is pooled (Pooling) to obtain the text vector of the text. Among them, the pooling process can adopt average pooling, that is, the average value of all token output vectors in the text is calculated as the text vector of the entire text.

[0068] S142, concatenate the first text vector, the second text vector, and the third text vector to obtain a concatenated vector.

[0069] By concatenating the first text vector, the second text vector and the third text vector, the address text and geographic location information of the address in the sample address pair can be more fully integrated, thereby guiding the address similarity model to make more accurate address similarity predictions.

[0070] The above S142 can be implemented in various ways, which are not limited in the embodiments of the present application. As an example, the first text vector, the second text vector and the third text vector can be sequentially spliced ​​together to obtain a spliced ​​vector. For example, the first text vector is vec1, the second text vector is vec2, and the third text vector is geo_vec. The three are sequentially spliced ​​together to obtain a spliced ​​vector (vec1, vec2, geo_vec).

[0071] As another example, a difference vector between the first text vector and the second text vector is determined; the first text vector, the second text vector, the difference vector and the third text vector are concatenated to obtain a concatenated vector.

[0072] For example, the first text vector is vec1, the second text vector is vec2, and the third text vector is geo_vec. First, determine that the difference vector between the first text vector and the second text vector is |vec1-vec2|. Then, concatenate the first text vector vec1, the second text vector vec2, the difference vector |vec1-vec2|, and the third text vector geo_vec in sequence to obtain a concatenated vector (vec1, vec2, |vec1-vec2|, geo_vec).

[0073] By concatenating the three text vectors in the above manner, the differences between the sample address pairs in address text and geographic location information can be better integrated into the concatenated vector, thereby more effectively guiding the address similarity model to perform address similarity prediction.

[0074] S143, predicting the similarity of the sample address pairs based on the address similarity model and the concatenation vector.

[0075] Specifically, the concatenated vector is combined with the weight matrix w of the address similarity model t ∈R 4n×k After multiplication and classification, the similarity of the sample address pairs can be obtained. Among them, the weight matrix is ​​the parameter of the address similarity model, which can be continuously updated during the training process of the address similarity model.

[0076] In practical applications, the address similarity model may have various appropriate structures, which may be set according to actual needs, and the present application embodiment does not limit this. As an example, Figure 2 As shown, the address similarity model includes a coding network and a classifier. The coding network can adopt various networks with coding capabilities, such as language models such as BERT, which is not limited in the embodiments of the present application.

[0077] The above S141 is implemented by a coding network, and the above S143 is implemented by a classifier. Figure 2 As shown, after the first text, the second text and the third text are input into the address similarity model, the three texts are embedded in the encoding network and then pooled to obtain the first text vector vec1, the second text vector vec2 and the third text vector geo_vec; then, the first text vector, the second text vector and the third text vector are concatenated to obtain the concatenated vector; finally, the classifier performs classification prediction based on the concatenated vector to obtain the similarity of the sample address pair.

[0078] Since the concatenated text can objectively and accurately reflect the differences in address text and geographical location of sample address pairs, compared with a single address text, it increases the prior information of the model input. Therefore, the concatenated text can better guide the address similarity model to fully learn such differences, thereby more accurately evaluating the similarity between addresses.

[0079] The embodiment of the present application shows some implementation methods of the above S104. Of course, the above S104 can also be implemented in other ways, which are not limited in the embodiment of the present application.

[0080] S106: Optimize and train the address similarity model based on the similarity and similarity labels of the sample address pairs.

[0081] Specifically, based on the similarity and similarity labels of the sample address pairs, the prediction loss of the address similarity model is determined; then, based on the prediction loss, the parameters of the address similarity model are optimized using one or more optimization algorithms such as stochastic gradient descent algorithm, AdaGrad, Adam, etc. The above S102 to S106 are repeated multiple times until the optimization training is stopped when the preset stop condition is met, and the address similarity model at this time is used to predict the similarity of any two addresses in the application stage.

[0082] In another embodiment of the present application, considering that the accuracy of the address text of the address in the sample address pair will have a significant impact on the training effect of the address similarity model, in order to further enhance the model training effect and improve the prediction accuracy of the address similarity model, before the above S102, the original address text of the address in the sample address pair is also corrected to obtain the address text of the address.

[0083] Please refer to Figure 3 , is a flow chart of a method for training an address similarity model provided in another embodiment of the present application, the method comprising the following steps:

[0084] S302, performing address element recognition on the original address text of the address in the sample address pair to obtain element information of the n-level address element.

[0085] Address elements are the components of an address. An address usually includes multiple levels of address elements. As an example, according to administrative divisions, an address includes at least three levels of address elements. The first-level address element refers to the first-level administrative region, namely the provincial-level administrative region, specifically including provinces, autonomous regions, municipalities directly under the central government, and special administrative regions; the second-level address element refers to the second-level administrative region, namely the prefecture-level administrative region, specifically including prefecture-level cities, regions, autonomous prefectures, leagues, etc.; the third-level address element refers to the third-level administrative region, namely the county-level administrative region, specifically including urban districts, county-level cities, counties, autonomous counties, banners, autonomous banners, forest areas, special zones, etc. For example, the address "Province A, City B, District C, Road D, Building E" includes five levels of address elements, namely Province A, City B, District C, Road D, and Building E.

[0086] For each level of address elements, its element information may include, for example, but is not limited to, at least one of the following information: element text, pinyin string of element text, etc. For example, the original address text of the address "Qixia District, Nanjing City, Jiangsu Province" includes three levels of address elements, the element text of the first level address element is Jiangsu Province, and its pinyin string is "jiang susheng", the element text of the second level address element is Nanjing City, and its pinyin string is "nan jing shi", and the element text of the third level address element is Qixia District, and its pinyin string is "qi xia qu".

[0087] Address element recognition of the original address text of the address can be implemented in various ways, which are not limited in the embodiments of the present application. As an example, keywords used to distinguish address elements, such as "province", "city", "district", "county", "road", etc., can be matched with the original address text of the address, and based on the matching results, the element texts of the address elements at all levels are extracted from the original address text, thereby obtaining the pinyin string of the element text.

[0088] S304, obtaining reference element information of each level of address elements from the reference address library.

[0089] The reference address database includes reference element information of commonly used address elements at all levels. The reference element information of address elements at all levels may include, but is not limited to, at least one of the following information: reference element text of address elements at all levels, pinyin character string of reference element text, etc.

[0090] S306: Based on the element information and reference element information of the n-level address element, the original address text of the address is modified to obtain a candidate address text of the address.

[0091] Specifically, for each level of address elements, the original address text of the address is corrected based on the element information and reference element information of the address element, which can ensure the accuracy of the element information of each level of address elements.

[0092] In one embodiment, the above S306 includes the following steps: Step A1, for each level of address element, comparing the element text of the address element with the reference element text of the address element to obtain the text comparison result of the address element; Step A2, comparing the pinyin character string of the element text with the pinyin character string of the reference element text of the address element to obtain the pinyin comparison result of the address element; Step A3, if the text comparison result is inconsistent and the pinyin comparison result is consistent, then correcting the element text of the address element based on the reference element text of the address element to obtain the candidate address text of the address.

[0093] For example, the original address text of the address is "Xia District, Nanjing City, Jiangsu Province". By performing address element recognition on the original address text, the element information of the third-level address element is obtained as follows:

[0094] Element information of the first-level address element: Jiangsu Province jiang su sheng

[0095] Element information of the second-level address element: Nanjing nan jing shi

[0096] Element information of the third-level address element: district qi xia qu

[0097] The reference element information of the above-mentioned address elements at all levels is obtained from the reference address database as follows:

[0098] Reference element information for the first-level address element: Jiangsu Province jiang su sheng

[0099] Reference element information of the second-level address element: Nanjing nan jing shi

[0100] Reference element information of the third-level address element: Qixia District

[0101] By comparing the element information of the above-mentioned address elements at all levels and their reference element information, it is determined that the element text of the third-level address element is inconsistent with the reference element text, and the pinyin string of the element text of the third-level address element is consistent with the pinyin string of the reference element text. The reference element text of the third-level address element is used to replace the element text of the third-level address element to obtain the candidate address text of "Qixia District, Nanjing City, Jiangsu Province".

[0102] Through the above implementation, it is possible to correct polyphonetic errors in the original address text to obtain an accurate address text, thereby improving the prediction accuracy of the address similarity model.

[0103] In another embodiment, the above S306 includes the following steps: step B1, for each level of address element, comparing the element text of the address element with the reference element text of the address element to obtain the text comparison result of the address element; step B2, determining the text distance between the element text of the address element and the reference element text of the address element; step B3, if the text comparison result is inconsistent and the text distance is less than the first preset distance threshold, then correcting the element text of the address element based on the reference element text of the address element to obtain the candidate address text of the address.

[0104] The text distance is used to represent the similarity between the element text of the address element and the reference element text. The text distance can be obtained by various text similarity calculation methods, such as but not limited to at least one of the following methods: calculating the edit distance between two texts, calculating the Jaccard similarity coefficient between two texts, and calculating the longest common subsequence.

[0105] For example, the original address text of the address is "A Province B City C District D Road Baochen Times". By performing address element recognition on the original address text, the element text of the 5-level address element is obtained as follows:

[0106] Element information of the first-level address element: Province A

[0107] Element information of the second-level address element: City B

[0108] Element information of level 3 address element: Area C

[0109] Element information of the fourth-level address element: D Road

[0110] Element information of level 5 address element: Baochen Era

[0111] The reference element texts of the above-mentioned address elements at all levels are obtained from the reference address library as follows:

[0112] Element information of the first-level address element: Province A

[0113] Element information of the second-level address element: City B

[0114] Element information of level 3 address element: Area C

[0115] Element information of the fourth-level address element: D Road

[0116] Element information of level 5 address elements: Baozhan Times

[0117] By comparing the element information of the above-mentioned address elements at all levels and their reference element information, it is determined that the element text of the 5th-level address element is inconsistent with the reference element text, but the text distance between the two is less than the first preset distance threshold. The reference element text of the 5th-level address element is used to replace the element text of the 5th-level address element to obtain the candidate address text of the address "A Province B City C District D Road Baozhan Times".

[0118] Through the above implementation, it is possible to correct the errors of similar characters in the original address text to obtain an accurate address text, thereby improving the prediction accuracy of the address similarity model.

[0119] In another embodiment, the above S306 includes the following steps: for each level of address element, if the element information of the address element is a null value, the element information of the address element is filled based on the reference element information of the address element to obtain the candidate address text of the address element.

[0120] For example, the original address text of the address is "Qixia District". By performing address element recognition on the original address text, the element text of the third-level address element is obtained as follows:

[0121] Element information of level 1 address element: null value

[0122] Element information of the second-level address element: null value

[0123] Element information of the third-level address element: Qixia District

[0124] The reference element texts of the above-mentioned address elements at all levels are obtained from the reference address library as follows:

[0125] Reference element information for level 1 address elements: Jiangsu Province

[0126] Reference element information for the second-level address element: Nanjing City

[0127] Reference element information of the third-level address element: Qixia District

[0128] Furthermore, the element text of the first-level address element is filled with the reference element text of the first-level address element, and the element text of the second-level address element is filled with the reference element text of the second-level address element, to obtain the candidate address text of "Qixia District, Nanjing City, Jiangsu Province".

[0129] Through the above implementation, it is possible to supplement the missing address element text in the original address text to obtain an accurate address text, thereby improving the prediction accuracy of the address similarity model.

[0130] The embodiment of the present application shows some implementation methods of the above S306. Of course, it should be understood that the above S306 can also be implemented in other ways, and the embodiment of the present application does not limit this.

[0131] S308: Determine the address text of the address based on the verification result of the candidate address text.

[0132] By verifying the candidate address text of the address, it can be determined whether the candidate address text is accurate; if the candidate address text is accurate, the candidate address text can be used as the address text of the address; if the candidate address text is inaccurate, the original address text continues to be used as the address text of the address.

[0133] Verification of the candidate address text may be implemented in various ways, which are not limited in the embodiments of the present application. In one embodiment, the above S308 includes the following steps: determining candidate geographic location information that matches the candidate address text; if the geographic distance between the candidate geographic location information and the geographic location information of the address does not exceed a second preset distance threshold, determining that the candidate address text is accurate, and then using the candidate address text as the address text of the address; if the distance between the candidate geographic location information and the geographic location information of the address is greater than the second preset distance threshold, determining that the candidate address text is inaccurate, and then using the address text of the address as the address text of the address.

[0134] The geographical location information matched by the candidate address text is used to describe the geographical location represented by the candidate address text. In practical applications, the geographical location information matched by the candidate address text can be represented by the geographical location represented by the first three levels of address elements in the candidate address text.

[0135] For example, both District C of City B of Province A and District C' of City B of Province A have Road D, the original address text of the address is Road D of District C of City B of Province A, and the candidate address text of the address is Road D of District C' of City B of Province A. If the distance between the geographic location 1 matched by the original address text and the geographic location 2 matched by the candidate address text is greater than a second preset distance threshold, it is determined that the candidate address text is inaccurate, and the modification of the original address text is abandoned, and the original address text continues to be used as the address text of the address.

[0136] Through the above implementation, the accuracy of the address text can be further improved, thereby more effectively guiding the model training process and improving the prediction accuracy of the address similarity model.

[0137] The embodiment of the present application shows some implementation methods of the above S308. Of course, it should be understood that the above S308 can also be implemented in other ways, and the embodiment of the present application does not limit this.

[0138] S310, concatenating the address text and geographic location information of the address in the sample address pair to obtain a concatenated text.

[0139] The specific implementation of the above S310 is Figure 1 The specific implementation of S102 in the illustrated embodiment is similar and will not be described in detail.

[0140] S312, predicting the similarity of the sample address pairs based on the address similarity model and the concatenated text.

[0141] The specific implementation of S312 is the same as Figure 1 The specific implementation of S104 in the illustrated embodiment is similar and will not be described in detail.

[0142] S314: Optimize and train the address similarity model based on the similarity and similarity labels of the sample address pairs.

[0143] The specific implementation of S314 is the same as Figure 1 The specific implementation of S106 in the illustrated embodiment is similar and will not be described in detail.

[0144] One or more embodiments of the present application provide a method for training an address similarity model, which uses the address text and geographic location information of the address as prior information to train the address similarity model, so that the address similarity model can fully learn the similarities and differences between different addresses from the prior information, thereby accurately evaluating the similarity between different addresses, especially for the situation where two addresses are similar in address text but have large differences in actual geographic locations. During the training process, by splicing the address text and geographic location information of the address in the sample address pair, the resulting spliced ​​text can more objectively and accurately reflect the differences in geographic location and address text of the sample address pair; further, based on the address similarity model and the spliced ​​text, the sample address pair is predicted for similarity, and based on the predicted similarity of the sample address pair and the similar pair label, the address similarity model is optimized and trained, which can effectively guide the address similarity model to fully learn the differences in geographic location and address text of different addresses, thereby improving the model training effect and accurately evaluating the similarity between different addresses.

[0145] Based on the address similarity model trained by the above method, the present application embodiment also provides a method for determining address similarity. Figure 4 , is a flow chart of a method for determining address similarity provided by an embodiment of the present application, the method comprising the following steps:

[0146] S402, determining the address text of the address in the target address pair, and concatenating the address text with the geographical location information of the address to obtain a concatenated text.

[0147] The target address pair includes two addresses to be compared. Figure 1 The specific implementation of S102 in the illustrated embodiment is similar and will not be described in detail.

[0148] S404: predicting the similarity of the target address pair based on the address similarity model and the concatenated text.

[0149] The specific implementation of the above S404 is the same as the above Figure 1 The specific implementation of S104 in the illustrated embodiment is similar and will not be described in detail.

[0150] The method for determining address similarity provided in the embodiment of the present application is applicable to a scenario where the number of addresses to be compared is two or more. In the case where the number of addresses to be compared is multiple, these addresses can be combined to obtain multiple target address pairs, each target address pair including two addresses to be compared. Further, for each target address pair, the above S402 to S404 are executed to obtain the similarity of the target address pair.

[0151] For example, the addresses to be compared include address 1, address 2, address 3, and address 4. By combining these addresses, the following 6 address pairs can be obtained: address pair 1 (address 1 and address 2), address pair 1 (address 1 and address 3), address pair 1 (address 1 and address 4), address pair 1 (address 2 and address 4), address pair 1 (address 2 and address 4), address pair 1 (address 3 and address 4). By executing the above S402 to S404 for each address pair, the similarity between these addresses can be obtained.

[0152] One or more embodiments of the present application provide a method for determining address similarity. In the process of evaluating address similarity, not only the address text of the address but also the geographic location information of the address is considered. By splicing the address text of the address in the target address pair and the geographic location information of the address, the resulting spliced ​​text can more objectively and accurately reflect the differences in the geographic location and the differences in the address text of the target address pair, thereby guiding the address similarity model to more accurately predict the similarity of the target address pair.

[0153] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0154] Based on the same inventive concept, the present application also provides a training device for an address similarity model. Figure 5 , is a structural diagram of a training device 500 for an address similarity model provided in an embodiment of the present application, the device comprising: a splicing module 510, a prediction module 520 and a training module 530.

[0155] The splicing module 510 is used to determine the address text of the address in the sample address pair, and splice the address text of the address with the geographical location information of the address to obtain the spliced ​​text.

[0156] The prediction module 520 is used to predict the similarity of the sample address pairs based on the address similarity model and the concatenated text.

[0157] The training module 530 is used to optimize and train the address similarity model based on the similarity and the similarity labels of the sample address pairs.

[0158] In another embodiment, the sample address pair includes a first address and a second address; the concatenated text includes a first text, a second text and a third text;

[0159] The splicing module is used for:

[0160] concatenating the address text of the first address and the geographic location information of the first address to obtain the first text;

[0161] concatenating the address text of the second address and the geographical location information of the second address to obtain the second text;

[0162] The geographic location information of the first address and the geographic location information of the second address are concatenated to obtain the third text.

[0163] In another embodiment, the prediction module:

[0164] Based on the address similarity model, encoding the first text, the second text and the third text to obtain a first text vector, a second text vector and a third text vector;

[0165] splicing the first text vector, the second text vector and the third text vector to obtain a spliced ​​vector;

[0166] Based on the address similarity model and the concatenation vector, the similarity of the sample address pairs is predicted.

[0167] In another embodiment, when the prediction module concatenates the first text vector, the second text vector and the third text vector to obtain a concatenated vector, the prediction module performs the following steps:

[0168] determining a difference vector between the first text vector and the second text vector;

[0169] The first text vector, the second text vector, the difference vector and the third text vector are concatenated to obtain the concatenated vector.

[0170] In another embodiment, the apparatus 500 further includes:

[0171] An identification module, used for performing address element identification on the original address text of the address in the sample address pair to obtain element information of the n-level address element;

[0172] An acquisition module, used for acquiring reference element information of each level of address elements from a reference address library;

[0173] A correction module, used for correcting the original address text of the address based on the element information of the n-level address element and the reference element information to obtain a candidate address text of the address;

[0174] A determination module is used to determine the address text of the address based on the verification result of the candidate address text.

[0175] In another embodiment, the element information of the address element includes the element text of the address element and the pinyin string of the element text, and the reference element information includes the reference element text and the pinyin string of the reference element text;

[0176] The correction module is used for:

[0177] For each level of address elements, the element text of the address element is compared with the reference element text of the address element to obtain a text comparison result of the address element;

[0178] Compare the pinyin character string of the element text with the pinyin character string of the reference element text to obtain the pinyin comparison result of the address element;

[0179] If the text comparison result is inconsistent and the pinyin comparison result is consistent, the element text is corrected based on the reference element text to obtain a candidate address text for the address.

[0180] In another embodiment, the element information of the address element includes element text of the address element, and the reference element information includes reference element text;

[0181] The correction module is used for:

[0182] For each level of address elements, the element text of the address element is compared with the reference element text of the address element to obtain a text comparison result of the address element;

[0183] determining a text distance between an element text of the address element and a reference element text of the address element;

[0184] If the text comparison result is inconsistent and the text distance is less than a first preset distance threshold, the element text is corrected based on the reference element text to obtain a candidate address text for the address.

[0185] In another embodiment, the correction module is used to:

[0186] For each level of address elements, if the element information of the address element is a null value, the element information of the address element is filled based on the reference element information of the address element to obtain the candidate address text of the address element.

[0187] In another embodiment, the determining module is used to:

[0188] Determining candidate geographic location information matching the candidate address text;

[0189] If the geographical distance between the candidate geographical location information and the geographical location information of the address does not exceed a second preset distance threshold, using the candidate address text as the address text of the address;

[0190] If the geographical distance between the candidate geographical location information and the geographical location information of the address is greater than the second preset distance threshold, the original address text of the address is used as the address text of the address.

[0191] Obviously, the training device 500 for the address similarity model provided in the embodiment of the present application can be used as the above Figure 1 The execution body of the training method of the address similarity model shown in FIG. Figure 1 The functions realized are as follows. Since the principles are the same, they will not be described in detail.

[0192] Based on the same inventive concept, the present application embodiment also provides an address similarity determination device. Figure 6 , is a structural diagram of an address similarity determination device 600 provided in an embodiment of the present application, the device comprising: a splicing module 610 and a prediction module 620.

[0193] The splicing module 610 is used to determine the address text of the address in the target address pair, and splice the address text of the address with the geographical location information of the address to obtain a spliced ​​text.

[0194] The prediction module 620 is used to predict the similarity of the target address pair based on the address similarity model and the concatenated text.

[0195] Obviously, the address similarity determination device 600 provided in the embodiment of the present application can be used as the above Figure 4 The address similarity determination method shown in the figure is implemented by the address similarity determination device. Figure 4 The functions realized are as follows. Since the principles are the same, they will not be described in detail.

[0196] Figure 7 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present application. Figure 7 At the hardware level, the electronic device includes a processor, and optionally also includes an internal bus, a network interface, and a memory. The memory may include a memory, such as a high-speed random access memory (RAM), and may also include a non-volatile memory (non-volatile memory), such as at least one disk storage. Of course, the electronic device may also include hardware required for other services.

[0197] The processor, network interface and memory can be interconnected through an internal bus, which can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 7 Only one bidirectional arrow is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0198] The memory is used to store the program. Specifically, the program may include a program code, and the program code includes a computer operation instruction. The memory may include a memory and a non-volatile memory, and provides instructions and data to the processor.

[0199] The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it, forming a training device for the address similarity model at the logical level. The processor executes the program stored in the memory and is specifically used to perform the following operations:

[0200] Determine the address text of the address in the sample address pair, and concatenate the address text of the address with the geographic location information of the address to obtain a concatenated text;

[0201] Predicting the similarity of the sample address pairs based on the address similarity model and the concatenated text;

[0202] The address similarity model is optimized and trained based on the similarity and the similarity labels of the sample address pairs.

[0203] Alternatively, the processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it, forming an address similarity determination device at the logical level. The processor executes the program stored in the memory and is specifically used to perform the following operations:

[0204] Determine the address text of the address in the target address pair, and concatenate the address text of the address with the geographic location information of the address to obtain a concatenated text;

[0205] Based on the address similarity model and the concatenated text, the similarity of the target address pair is predicted.

[0206] The above application Figure 1 The method performed by the training device of the address similarity model disclosed in the embodiment shown, or the above-mentioned method of the present application Figure 4The method performed by the address similarity determination device disclosed in the illustrated embodiment can be applied to a processor or implemented by a processor. The processor may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in the processor or an instruction in the form of software. The above processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present application can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the embodiments of the present application can be directly embodied as a hardware decoding processor for execution, or a combination of hardware and software modules in the decoding processor for execution. The software module can be located in a storage medium mature in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0207] The electronic device may also perform Figure 1 The method is used to implement the training device of the address similarity model. Figure 1 , Figure 2 , Figure 4 The functions of the embodiments shown, or the electronic device may also perform Figure 4 The method and the device for determining the address similarity are implemented in Figure 4 The functions of the illustrated embodiment will not be described in detail in the embodiments of the present application.

[0208] Of course, in addition to software implementation methods, the electronic device of the present application does not exclude other implementation methods, such as logic devices or a combination of software and hardware, etc. That is to say, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0209] The embodiment of the present application also provides a computer-readable storage medium, which stores one or more programs, wherein the one or more programs include instructions, which, when executed by a portable electronic device including a plurality of application programs, enable the portable electronic device to execute Figure 1 The method of the embodiment shown is specifically used to perform the following operations:

[0210] Determine the address text of the address in the sample address pair, and concatenate the address text of the address with the geographic location information of the address to obtain a concatenated text;

[0211] Predicting the similarity of the sample address pairs based on the address similarity model and the concatenated text;

[0212] The address similarity model is optimized and trained based on the similarity and the similarity labels of the sample address pairs.

[0213] Alternatively, when the instruction is executed by a portable electronic device including a plurality of application programs, the portable electronic device can execute Figure 4 The method of the embodiment shown is specifically used to perform the following operations:

[0214] Determine the address text of the address in the target address pair, and concatenate the address text of the address with the geographic location information of the address to obtain a concatenated text;

[0215] Based on the address similarity model and the concatenated text, the similarity of the target address pair is predicted.

[0216] An embodiment of the present application also provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to cause a computer to execute part or all of the steps in the training method of the address similarity model provided in the embodiment of the present application or the address similarity determination method provided in the embodiment of the present application.

[0217] In short, the above description is only a preferred embodiment of the present application and is not intended to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0218] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0219] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0220] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0221] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

Claims

1. A method for training an address similarity model, characterized in that: include: Determine the address text of the address in the sample address pair, and concatenate the address text with the geographical location information of the address to obtain a concatenated text; Predicting the similarity of the sample address pairs based on the address similarity model and the concatenated text; The address similarity model is optimized and trained based on the similarity and the similarity labels of the sample address pairs.

2. The method according to claim 1, characterized in that: The sample address pair includes a first address and a second address; the concatenated text includes a first text, a second text and a third text; The step of splicing the address text and the geographical location information of the address to obtain a spliced ​​text includes: concatenating the address text of the first address and the geographic location information of the first address to obtain the first text; concatenating the address text of the second address and the geographical location information of the second address to obtain the second text; The geographic location information of the first address and the geographic location information of the second address are concatenated to obtain the third text.

3. The method according to claim 2, characterized in that The predicting the similarity of the sample address pair based on the address similarity model and the concatenated text includes: Based on the address similarity model, encoding the first text, the second text and the third text to obtain a first text vector, a second text vector and a third text vector; splicing the first text vector, the second text vector and the third text vector to obtain a spliced ​​vector; Based on the address similarity model and the concatenation vector, the similarity of the sample address pairs is predicted.

4. The method according to claim 3, characterized in that The step of concatenating the first text vector, the second text vector and the third text vector to obtain a concatenated vector includes: determining a difference vector between the first text vector and the second text vector; The first text vector, the second text vector, the difference vector and the third text vector are concatenated to obtain the concatenated vector.

5. The method according to any one of claims 1 to 4, characterized in that The step of determining the address text of the address in the sample address pair comprises: Performing address element recognition on the original address text of the address in the sample address pair to obtain element information of the n-level address element; Obtain reference element information of each level of address elements from the reference address database; Based on the element information of the n-level address element and the reference element information, the original address text of the address is modified to obtain a candidate address text of the address; Based on the verification result of the candidate address text, the address text of the address is determined.

6. The method according to claim 5, characterized in that The element information of the address element includes the element text of the address element and the pinyin string of the element text, and the reference element information includes the reference element text and the pinyin string of the reference element text; The step of correcting the original address text of the address based on the element information of the n-level address element and the reference element information to obtain the candidate address text of the address includes: For each level of address elements, the element text of the address element is compared with the reference element text of the address element to obtain a text comparison result of the address element; Compare the pinyin character string of the element text with the pinyin character string of the reference element text to obtain the pinyin comparison result of the address element; If the text comparison result is inconsistent and the pinyin comparison result is consistent, the element text is corrected based on the reference element text to obtain a candidate address text for the address.

7. The method according to claim 5, characterized in that The element information of the address element includes the element text of the address element, and the reference element information includes the reference element text; The step of correcting the original address text of the address based on the element information of the n-level address element and the reference element information to obtain the candidate address text of the address includes: For each level of address elements, the element text of the address element is compared with the reference element text of the address element to obtain a text comparison result of the address element; determining a text distance between an element text of the address element and a reference element text of the address element; If the text comparison result is inconsistent and the text distance is less than a first preset distance threshold, the element text is corrected based on the reference element text to obtain a candidate address text for the address.

8. The method according to claim 5, characterized in that The step of correcting the original address text of the address based on the element information of the n-level address element and the reference element information to obtain the candidate address text of the address includes: For each level of address elements, if the element information of the address element is a null value, the element information of the address element is filled based on the reference element information of the address element to obtain the candidate address text of the address element.

9. The method according to claim 5, characterized in that The step of determining the address text of the address based on the verification result of the candidate address text comprises: Determining candidate geographic location information matching the candidate address text; If the geographical distance between the candidate geographical location information and the geographical location information of the address does not exceed a second preset distance threshold, using the candidate address text as the address text of the address; If the geographical distance between the candidate geographical location information and the geographical location information of the address is greater than the second preset distance threshold, the original address text of the address is used as the address text of the address.

10. A method for determining address similarity, characterized in that: include: Determine the address text of the address in the target address pair, and concatenate the address text with the geographic location information of the address to obtain a concatenated text; Based on an address similarity model and the concatenated text, the similarity of the target address pair is predicted, wherein the address similarity model is trained according to the method according to any one of claims 1-9.

11. A training device for an address similarity model, characterized in that: include: A splicing module, used for determining the address text of the address in the sample address pair, and splicing the address text of the address with the geographical location information of the address to obtain a spliced ​​text; A prediction module, used for predicting the similarity of the sample address pair based on the address similarity model and the concatenated text; A training module is used to optimize and train the address similarity model based on the similarity and the similarity labels of the sample address pairs.

12. An address similarity determination device, characterized in that: include: A splicing module, used for determining the address text of the address in the target address pair, and splicing the address text with the geographical location information of the address to obtain a spliced ​​text; A prediction module is used to predict the similarity of the target address pair based on an address similarity model and the concatenated text, wherein the address similarity model is trained according to the method according to any one of claims 1-9.

13. An electronic device, characterized in that: include: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the training method of the address similarity model as described in any one of claims 1 to 9; or, the processor is configured to execute the instructions to implement the address similarity determination method as described in claim 10.

14. A computer-readable storage medium, characterized in that: When the instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the training method of the address similarity model as described in any one of claims 1 to 9; or, when the instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the address similarity determination method as described in claim 10.

15. A computer program product, characterized in that The computer program product includes a non-transitory computer-readable storage medium storing a computer program, wherein the computer program is operable to cause a computer to execute part or all of the steps in the method for training an address similarity model as described in any one of claims 1 to 9 or the method for determining address similarity as described in claim 10.

Images