Address data processing method, system, apparatus, device, and medium

Abstract

The application relates to the technical field of artificial intelligence, and discloses an address data processing method, device and equipment and a storage medium, which are used for quickly obtaining suitable training text data and improving model training precision. The method part comprises the following steps: when a target neural network model needs to be trained, a plurality of target address data are sequentially called from a pre-constructed address database according to weight sizes, and calling address data are obtained, wherein each target address data is given a corresponding weight according to a required degree; a plurality of address templates are called from a pre-constructed address template library, and calling address templates are obtained; a plurality of training text data are generated through the calling address data and the calling address templates, wherein the greater the weight of the calling address data is, the more times the calling address data are filled into each calling address template to generate the training text data; and the training text data are delivered to the target neural network model.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, and particularly relates to an address data processing method, system and device, computer equipment and a storage medium. BACKGROUND

[0002] At present, the demand for address-related in business is gradually increasing, and a large amount of address data is needed for training in the field of natural language processing (NLP) or in the field of automatic speech recognition technology (ASR), for example, in the field of NLP, address data is needed to train the model, so as to identify address information such as province, city, district and county and put it out for feedback.

[0003] In the traditional address data generation scheme, address perturbation data enhancement is usually performed on the searched address data to obtain a large amount of address data for training data. However, due to the randomness of data enhancement, the actual enhanced training text data may contain invalid addresses, which can easily reduce the accuracy of the model. SUMMARY

[0004] The present application provides an address data processing method, system, device, computer equipment and storage medium to solve the technical problem that the traditional method easily reduces the accuracy of the model.

[0005] In a first aspect, the present application provides an address data processing method, which comprises:

[0006] When the target neural network model needs to be trained, a plurality of target address data are called from the pre-constructed address database in order of weight size, and the called address data are obtained, wherein each target address data is assigned with a corresponding weight according to the required degree;

[0007] A plurality of address templates are called from the pre-constructed address template library, and the called address templates are obtained;

[0008] A plurality of training text data are generated by calling the address data and the address templates, wherein the greater the weight of the called address data, the more times the called address data are filled into each called address template to generate the training text data;

[0009] The training text data are input to the target neural network model.

[0010] Optionally, the plurality of training text data are generated by calling the address data and the address templates, comprising:

[0011] The each calling address data is filled into each calling address template in sequence until a plurality of training text data meeting the training quantity requirement is obtained.

[0012] Optionally, a plurality of address templates are called from a pre-constructed address template library to obtain the calling address templates, including:

[0013] According to the training quantity requirement and the weight of each calling address data, the template calling quantity is determined.

[0014] A plurality of address templates corresponding to the template calling quantity are randomly called from the pre-constructed address template library to obtain the calling address templates.

[0015] Optionally, the address database is constructed by the following method:

[0016] The original address data is obtained, wherein the original address data is real address data searched out.

[0017] The original address data is taken as the target address data.

[0018] According to the degree of meeting the requirement of the target address data, the corresponding weight and address storage time stamp are given to the target address data.

[0019] The plurality of target address data to which the weight and the address storage time stamp are given is stored in the address database.

[0020] Optionally, the original address data is taken as the target address data, including:

[0021] According to a preset segmentation rule, the address information of the original address data is segmented to obtain the target address data including a plurality of levels of address information, wherein each target address data includes the same number of levels of address information.

[0022] Optionally, the address template library is constructed by the following method:

[0023] The business data provided by a business application is obtained, and the address templates are parsed from the business data as original address templates; and / or, the address templates are generated by an address template generation model as original address templates.

[0024] The plurality of original address templates are subjected to semantic analysis to obtain the semantic fluency of each original address template.

[0025] From the plurality of original address templates, a target original address template with a semantic fluency greater than a preset fluency is selected.

[0026] The target original address template that passes the artificial inspection is stored in the address template library.

[0027] In a second aspect, the present application provides an address data processing system, comprising an address data processing device, an address database and an address template library, wherein the address data processing device is configured to:

[0028] When the target neural network model needs to be trained, a plurality of target address data are sequentially called from the address database according to the weight size, and the called address data are obtained, wherein each target address data is assigned with a corresponding weight according to the degree of compliance;

[0029] A plurality of address templates are called from the address template library, and the called address templates are obtained.

[0030] The plurality of training text data are generated by calling the address data and the called address templates, wherein the greater the weight of the called address data, the more times the called address data are filled into each called address template to generate the training text data.

[0031] The training text data are delivered to the target neural network model.

[0032] In a third aspect, the present application provides an address data processing device, comprising:

[0033] An address data calling module, configured to sequentially call a plurality of target address data from a pre-constructed address database according to the weight size when the target neural network model needs to be trained, and obtain called address data, wherein each target address data is assigned with a corresponding weight according to the degree of compliance.

[0034] An address template calling module, configured to call a plurality of address templates from a pre-constructed address template library, and obtain called address templates.

[0035] A generating module, configured to generate a plurality of training text data by calling the address data and the called address templates, wherein the greater the weight of the called address data, the more times the called address data are filled into each called address template to generate the training text data.

[0036] A sending module, configured to deliver the training text data to the target neural network model.

[0037] In a fourth aspect, the present application provides a computer device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any one of the address data processing methods when executing the computer program.

[0038] In a fifth aspect, the present application provides a computer readable storage medium, which stores a computer program, wherein the computer program is executed by a processor to implement the steps of any one of the address data processing methods.

[0039] In one of the aforementioned solutions, target address data is first retrieved from the address database and combined with address templates from the address template library to quickly generate training text data for training the target neural network. Since the greater the weight of the retrieved address data, the more times it is used to populate each address template to generate training text data, each target address data is assigned a corresponding weight based on its suitability. Therefore, based on the weights, suitable target address data can be quickly retrieved and effective training text data can be generated rapidly during model training. This facilitates quickly obtaining the required training data and improves the efficiency of address data acquisition. Compared to simply using the model to perform text augmentation such as address perturbation to generate training text data, this method of applying address templates from the template library better ensures the cascading relationship and correctness of addresses, improves the accuracy of model training, and is also easier to maintain and generate. Attached Figure Description

[0040] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 This is a schematic diagram of an application environment for an address data processing method according to an embodiment of this application;

[0042] Figure 2 This is a flowchart illustrating an address data processing method according to one embodiment of this application;

[0043] Figure 3 yes Figure 2 A flowchart illustrating a specific implementation of step S20;

[0044] Figure 4 This is a schematic diagram of the address database construction process in one embodiment of this application;

[0045] Figure 5 This is a schematic diagram of the construction process of the address template library in one embodiment of this application;

[0046] Figure 6 This is a schematic diagram of the structure of an address data processing system according to an embodiment of this application;

[0047] Figure 7 This is a schematic diagram of the structure of an address data processing device according to an embodiment of this application;

[0048] Figure 8 This is a schematic diagram of the structure of a computer device according to one embodiment of this application. Detailed Implementation

[0049] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0050] This application provides an address data processing method that can be applied to applications such as... Figure 1 In the application environment, the client communicates with the server. The server implements the address data processing method in this application. When training the target neural network model, it sequentially retrieves multiple target address data from a pre-built address database according to their weights to obtain call address data, wherein each target address data is assigned a corresponding weight based on its compliance with requirements; it retrieves multiple address templates from a pre-built address template library to obtain call address templates; it generates multiple training text data using the call address data and call address templates, wherein the greater the weight of the call address data, the more times the call address data is filled into each call address template to generate training text data; and it sends the training text data to the target neural network model. In this application, target address data is first retrieved from the address database and combined with address templates from the address template library to quickly generate training text data for training the target neural network. Since the greater the weight of the retrieved address data, the more times it is used to populate each address template to generate training text data, each target address data is assigned a corresponding weight based on its suitability. Therefore, based on the weights, suitable target address data can be quickly retrieved and effective training text data can be generated quickly during model training. This facilitates obtaining the required training data quickly and improves the efficiency of address data acquisition. Compared to simply using the model to perform text augmentation such as address perturbation to generate training text data, this method of applying address templates from the template library can better ensure the cascading relationship and correctness of addresses, improve the accuracy of model training, and is also easier to maintain and generate.

[0051] The client can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented using a standalone server or a server cluster consisting of multiple servers.

[0052] It should be noted that the demand for address-related in the business is gradually increasing, whether in the field of natural language processing (NLP) or in the field of automatic speech recognition technology (ASR), a large amount of address data is needed for training, for example, in the field of NLP, address data is needed to train the model, so as to recognize the address information such as province, city, district and county and put it out for feedback; If there is address data in the ASR field, the recognized address by speech recognition can be verified, thereby improving the accuracy of the model. The training text data obtained in the embodiment of the present application can be used in the model training in the above-mentioned NLP field or ASR field, which is not limited. For example, it can be applied to the NER address extraction model.

[0053] In an embodiment, as shown in Figure 2 , an address data processing method is provided, which is applied to the server in Figure 1 for example, and includes the following steps:

[0054] S10: When the target neural network model needs to be trained, a plurality of target address data are called from the pre-constructed address database in order according to the weight size, and the called address data are obtained, wherein each target address data is assigned with a corresponding weight according to the degree of compliance.

[0055] Before step S10 is performed, an address database needs to be constructed and uniquely address the database containing the incoming address data, the target address data, for example, the address database can be a SQL database, not limited, convenient for the target address data in the address database for retrieval, query and maintenance operation, can use SQL query statement, can quickly filter the required address. And each address data is assigned a corresponding weight according to its own compliance degree. That is, the target address data in the address database corresponds to a weight. Wherein, whether the target address data meets the requirements, including whether the address information is detailed and / or whether the address information has wrong characters, and / or whether the address information is obviously different from the real address, the more detailed the address information, the fewer the wrong characters, the more real the address information, the higher the corresponding weight. For example, the weight of some target address data can be set to 0, which can be defaulted as invalid and cannot be called, but is still saved in the address database, and the weight of some target address data can be set to greater than 0, which means that the target address data is valid and can be called to generate training text data. The size of the weight value is related to the number of times the target address data is called or used to generate training text data combined with the address template in the same training, for example, the weight of a certain target address data is set to 1, and the target address data will be repeated n times, which is positively related to the weight.

[0056] It should be noted that the above examples of weight values are only exemplary and do not limit the present application. It is worth noting that the invalid target address data in the address database is still maintained, which can facilitate subsequent reuse of the target address data. When these invalid target address data is needed, it can be directly used without repeated acquisition, reducing the data extraction process, and in many cases, it may not be convenient to reacquire the same data. Through the present application, it can be easily acquired.

[0057] After the above address database is constructed, when the target neural network model needs to be trained, multiple target address data is called from the pre-constructed address database according to the weight size to obtain the called address data. For example, in this example, the larger the weight, the more times the target address data is repeated, so the multiple target address data is called from the pre-constructed address database in order of weight from large to small, and the specific number of calls depends on the training text data to be called, which will be described in detail below. In addition, it should be noted that the target neural network model includes but is not limited to NLP model or ASR model, etc., which is not limited.

[0058] S20: Call multiple address templates from the pre-constructed address template library to obtain called address templates.

[0059] In this embodiment, when the target neural network model needs to be trained, a plurality of address templates are also called from the pre-constructed address template library to obtain called address templates. The address template library includes a large number of pre-constructed address templates. It should be noted that the address template is a template constructed in the present application by a preset method, and address information in the address data is directly filled into the corresponding position of the address template to obtain a new text data containing addresses, i.e., training text data. The construction of the address template library is described in detail in the embodiments below.

[0060] S30: Generate a plurality of training text data by calling address data and calling address templates. The greater the weight of the called address data, the more times the called address data is filled into each calling address template to generate training text data.

[0061] S40: Deliver the training text data to the target neural network model.

[0062] In this embodiment, after obtaining a plurality of called address data and a plurality of called address templates, the corresponding positions in the called address templates can be filled with the called address data to generate a plurality of training text data. The difference between the two steps and the traditional scheme includes that in the process of training text data by calling address data and calling address templates, the greater the weight of the called address data, the more times the called address data is filled into each calling address template to generate training text data. That is, the called address templates are filled according to the weight of the target address data, and the target address data is differentiated, so that the obtained training text data meets the requirements. Finally, the generated training text data is delivered to the target neural network model for model training.

[0063] In summary, the address data processing method provided in the embodiments of the present application can quickly retrieve appropriate target address data and quickly generate effective training text data during model training according to the weight, which is beneficial to quickly obtain the required training data, improves the address data acquisition efficiency, and better ensures the cascading relationship and correctness of the address by using the address templates in the template library, thereby improving the accuracy of model training and facilitating maintenance and generation.

[0064] In one embodiment, in step S40, i.e., by calling the address data and the calling address template, a plurality of training text data is generated, including: sequentially filling each calling address data into each calling address template until a plurality of training text data meeting the training quantity requirement is obtained.

[0065] In this embodiment, the process of generating a plurality of training text data from the calling address data and the calling address template is further optimized, i.e., sequentially filling each calling address data into each calling address template until a plurality of training text data meeting the training quantity requirement is obtained. For example, assuming there are N calling address data (N1, N2,..., NN) and K calling address templates (K1, K2,..., KK), in this embodiment, the calling address data N1 will be filled into the K calling address templates to obtain K training text data, then the calling address data N2 will be filled into the K calling address templates to obtain K training text data, and the calling address data N3 will be filled into the K calling address templates to obtain K training text data, and so on, and then the calling address data NN will be filled into the K calling address templates to obtain K training text data. However, in the above filling process, the number of training text data obtained will be calculated continuously. For example, if the training text data quantity requirement of the target neural network model is 3*K, then from the calling address data N1, sequentially filling will reach the calling address data N3 filled into the K calling address templates to obtain 3*K training text data, which meets the training requirement, and the address data filling action will be stopped.

[0066] It should be noted that the above training requirement of 3*K training text data in this embodiment is only an example, and when it is other training text data, such as 4*K training text data, the generation of training address data will be stopped when the calling address data N4 is filled into the K calling address templates, which will not be illustrated here and is not limited.

[0067] As can be seen, since different address data and different expressions (i.e., the final training text data) need to be generated, the embodiment can generate the required data according to the existing address database and address template. For example, one of the address templates is "I am at {address}", then {address} can be replaced by different addresses in the address database. For the template application method, one can ensure the correctness of address filling, and also ensure that the filled address has a cascading relationship. Maintaining different templates can obtain different sentence patterns, and replacing different addresses can obtain almost infinite data as training text data. In addition, in this embodiment, the training text data is generated according to the training requirement, which can effectively reduce the redundant address data filling process and improve the processing efficiency.

[0068] In combination with the above embodiments, in an embodiment, as shown in Figure 3 the step S20, that is, calling multiple address templates from the pre-constructed address template library to obtain the calling address templates, includes the following steps:

[0069] S21: determining the template calling quantity according to the training quantity requirement and the weight of each calling address data.

[0070] S22: randomly calling multiple address templates corresponding to the template calling quantity from the pre-constructed address template library to obtain the calling address templates.

[0071] In this embodiment, the process of calling address templates from the pre-constructed address template library is further optimized. Specifically, when calling the address templates, the template calling quantity is first determined according to the training quantity requirement and the weight of each calling address data. As known from the foregoing two embodiments, the weight of the calling address data affects the repetition times of the calling address data, that is, the weight affects the generation quantity of the final training text data. Therefore, in order to more accurately call the appropriate quantity of address templates from the address template library, the template calling quantity is first determined according to the training quantity requirement and the weight of each calling address data, and then multiple address templates corresponding to the template calling quantity are randomly called from the pre-constructed address template library to obtain the calling address templates. In this way, the redundant calling process of the address templates can be reduced, and the randomness and diversity of the training text data are increased by randomly calling the address templates.

[0072] In an embodiment, as shown in Figure 4 the address database is constructed by the following method:

[0073] S101: obtaining original address data, wherein the original address data is the real address data searched out.

[0074] S102: taking the original address data as target address data.

[0075] S103: giving the target address data corresponding weight and address storage time stamp according to the compliance requirement degree of the target address data.

[0076] S104: storing multiple target address data after being given weight and address storage time stamp into the address database.

[0077] The embodiment is an embodiment of constructing an address database. First, original address data is acquired. It should be noted that the original address data is real address data searched out. The so-called real address data refers to address data corresponding to the original address data that actually exists, rather than address data containing false addresses randomly generated through text enhancement. In a specific implementation, a large amount of address data can be randomly acquired through a crawler and the like, and the real address data is determined as the original address data by analyzing the authenticity of the address data. In the authenticity analysis process, the authenticity of the acquired address data can be judged by connecting an official system. For example, the address data of Guangzhou City, Hunan Province is obviously not real address data, because there is no such address. The specific authenticity analysis method is not limited here.

[0078] After obtaining the real original address data, the original address data is target address data, and a corresponding weight and address storage timestamp are assigned to the target address data according to the degree of compliance of the target address data. The target address data to which the weight and the address storage timestamp are assigned is stored in the address database. It should be noted that the degree of compliance of the target address data is determined according to whether the target address data complies with the requirements, including whether the address information is detailed and / or whether the address information has errors, and / or whether the address information is obviously inconsistent with the real address, and the like. The more detailed the address information, the fewer the errors, and the more real the address information, the higher the corresponding weight. For example, the weight of some target address data can be set to 0, so that the target address data is not effective by default and cannot be called, but is still saved in the address database. The weight of some target address data can be set to be greater than 0. In addition, an address storage timestamp is added to each target address data.

[0079] In the embodiment, real address data is acquired to facilitate improving the authenticity of subsequent constructed training text data, thereby ensuring the correct cascading relationship and correctness of the address information in the training text data. Moreover, assigning a weight can ensure that the required training text data is called and generated according to the weight. Different measures can be taken for part of the address, for example, some addresses that do not meet the requirements are set to 0, so that the address is not effective by default. If the weight is greater than 1, the address is repeated n times when generating the address subsequently, which represents that the address is very important. Through the storage timestamp, it can be distinguished when the address is stored in the database, which can facilitate the maintenance of the address.

[0080] It should be noted that the above method of constructing the address database is only one example, and in other embodiments, the storage timestamp can not be added, and in another embodiment, the original address data can also be a constructed virtual address. However, for real addresses, it is more beneficial to generate more realistic training text data and improve the authenticity and accuracy of the training of the target training model. The specific implementation is not limited.

[0081] In an embodiment, in S102, the original address data is used as the target address data, including: dividing the address information of the original address data according to a preset division rule to obtain target address data including multi-level address information, wherein each target address data includes address information of the same level.

[0082] In this embodiment, on the basis of the above address database construction, further optimization processing is performed. First, for the storage of the target address data, it needs to be confirmed to be stored in a unified style. For example, since the address is mainly divided into province, city, district / county, and detailed address, the detailed address includes town / street, village, road number, building name, building number, unit number, and room number. In this embodiment, the address information of the original address data can be divided according to the above division rule to extract address information at different positions and perform a division. For example, it can be divided into the following 7 levels, as shown in the following table:

[0083]

[0084]

[0085] It should be noted that the above division rule is only an example, and other division methods can also be used, for example, more levels can be divided, and the above division rule can be changed according to the address division characteristics of different countries or regions. The specific implementation is not limited. In this embodiment, the original address data is divided to standardize addresses of different lengths for storage, and it is also beneficial to the generation of subsequent training text data.

[0086] In an embodiment, as shown in Figure 5 , the address template library is constructed by the following method:

[0087] S201: Obtain service data provided by a service application, parse address templates from the service data as original address templates; and / or generate address templates by an address template generation model as original address templates.

[0088] S202: Perform semantic analysis on a plurality of original address templates to obtain the semantic fluency of each original address template.

[0089] S203: From the plurality of original address templates, a target original address template with a semantic fluency greater than a preset fluency is screened out.

[0090] S204: The target original address template that passes the artificial inspection is stored in the address template library.

[0091] This embodiment is an embodiment of constructing an address template library. First, various address templates need to be obtained. This embodiment provides at least two ways to obtain address templates. The business data provided by the business application is obtained, and the address template is parsed from the business data as an original address template; and / or, the address template is generated by the address template generation model as an original address template. The text original template can be extracted from the business data, and then the address type of the address entity in the text original template is obtained; according to the address type of the address entity in the text original template, the generalization word corresponding to the address type is selected to replace the address entity in the text original template, and the original address template is obtained. For example, the address information at the corresponding position is replaced by the generalization words such as "{province}", "{city}", "{county}", or the entire address information is directly replaced by "{address}", which is not limited. The trained address template generation model can also be used to generate the address template as the original address template, or the two methods can be combined to obtain a plurality of original address templates.

[0092] Subsequently, the plurality of original address templates are subjected to semantic analysis to obtain the semantic fluency of each original address template, and from the plurality of original address templates, a target original address template with a semantic fluency greater than a preset fluency is screened out. The target original address template that passes the artificial inspection is stored in the address template library. In this embodiment, the semantic fluency of each address template is judged, so that the original address template with insufficient fluency is eliminated and subjected to artificial inspection. This is beneficial to improve the fluency of the subsequently generated training text data, and is beneficial to improve the model training precision and convergence efficiency when used as training data.

[0093] It should be noted that in other embodiments, no fluency analysis can be performed, which is not limited in the present application.

[0094] In an embodiment, the above address database or address template library can be a blockchain database, and the generated training text data can also be stored in the blockchain database, which is not limited.

[0095] Blockchain is a new application mode of distributed data storage, peer-to-peer transmission, consensus mechanism, encryption algorithm and other computer technologies. Blockchain, in essence, is a decentralized database, which is a chain of data blocks associated using cryptographic methods, each of which contains information about a batch of network transactions, used to verify the validity of the information (anti-fake) and generate the next block. The blockchain can include a blockchain underlying platform, a platform product service layer, and an application service layer.

[0096] The blockchain underlying platform can include user management, basic services, smart contracts, and operation monitoring processing modules. Among them, the user management module is responsible for the identity information management of all blockchain participants, including maintaining the generation of public and private keys (account management), key management, and the maintenance of the correspondence between the real identity of the user and the blockchain address (permission management), etc., and under authorization, supervises and audits the transaction situation of certain real identities, provides rule configuration for risk control (risk audit); the basic service module is deployed on all blockchain node devices to verify the validity of business requests, and after consensus, the valid requests are recorded on the storage. For a new business request, the basic service first parses and authenticates the interface (interface adaptation), then encrypts the business information through the consensus algorithm (consensus management), and after encryption, the complete and consistent transmission is transmitted to the shared ledger (network communication), and is recorded and stored; the smart contract module is responsible for the registration and issuance of contracts, as well as the triggering and execution of contracts. Developers can define contract logic through a certain programming language, publish it to the blockchain (contract registration), call keys or other event triggers for execution according to the logic of the contract terms, complete the contract logic, and also provide contract upgrade and cancellation functions; the operation monitoring module is mainly responsible for the deployment, configuration modification, contract setting, cloud adaptation in the product release process, and the real-time state visualization output in the product running, such as alarm, monitoring network situation, monitoring node device health status, etc.

[0097] It should be understood that the size of the serial number of each step in the above embodiment does not mean the order of execution, and the execution order of each process should be determined according to its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0098] In an embodiment, as Figure 6As shown, the present application also provides an address data processing system, which comprises an address data processing device, an address database and an address template library, wherein the address data processing device is configured to: when a target neural network model needs to be trained, sequentially call multiple target address data from the address database according to the weight size to obtain called address data, wherein each target address data is assigned a corresponding weight according to the degree of compliance; call multiple address templates from the address template library to obtain called address templates; generate multiple training text data through the called address data and the called address templates, wherein the greater the weight of the called address data, the more times the called address data is filled into each called address template to generate training text data; and deliver the training text data to the target neural network model.

[0099] It should be noted that the specific limitations of the modules in the address data processing system can be referred to the corresponding limitations in the address data processing method described above, and will not be repeated here for brevity.

[0100] In this embodiment, an address data processing system is mainly proposed, which includes address processing, template extraction, and a scheme for combining templates and addresses to generate training text data. The address database and the address template library can be updated and maintained separately, and they are also related to the generation of training text data, which can better coordinate the entire data generation system.

[0101] In an embodiment, an address data processing device 10 is provided, which corresponds to the address data processing method in the above embodiments. As shown, the address data processing device 10 comprises an address data calling module 101, an address template calling module 102, a generating module 103 and a sending module 104. The functions of each module are described in detail as follows: Figure 7 The address data calling module 101 is configured to, when a target neural network model needs to be trained, sequentially call multiple target address data from a pre-constructed address database according to the weight size to obtain called address data, wherein each target address data is assigned a corresponding weight according to the degree of compliance;

[0102] The address template calling module 102 is configured to call multiple address templates from a pre-constructed address template library to obtain called address templates;

[0103] The generating module 103 is configured to generate multiple training text data through the called address data and the called address templates, wherein the greater the weight of the called address data, the more times the called address data is filled into each called address template to generate training text data;

[0104]

[0105] ​The sending module 104 is configured to send the training text data to the target neural network model.

[0106] In an embodiment, the generating module 103 is specifically configured to:

[0107] The each calling address data is sequentially filled into the each calling address template until the multiple training text data meeting the training quantity requirement are obtained.

[0108] In an embodiment, the address template calling module 102 is specifically configured to:

[0109] According to the training quantity requirement and the weight of each calling address data, the number of template calls is determined;

[0110] A plurality of address templates corresponding to the number of template calls are randomly called from the pre-constructed address template library to obtain the calling address template.

[0111] In an embodiment, the address data processing apparatus 10 further comprises an address database construction module, which is configured to:

[0112] Obtain the original address data, wherein the original address data is the real address data searched out;

[0113] The original address data is used as the target address data;

[0114] According to the degree of meeting the requirement of the target address data, the corresponding weight and address storage time stamp are given to the target address data;

[0115] The multiple target address data to which the weight and the address storage time stamp are given are stored in the address database.

[0116] In an embodiment, the address database construction module is further configured to:

[0117] According to the preset segmentation rule, the address information of the original address data is segmented to obtain the target address data comprising multiple levels of address information, wherein each target address data comprises address information of the same number of levels.

[0118] In an embodiment, the address data processing apparatus 10 further comprises an address template library construction module, which is configured to:

[0119] Obtain the business data provided by the business application, parse the address template from the business data as the original address template; and / or generate the address template as the original address template through the address template generation model;

[0120] The multiple original address templates are subjected to semantic analysis to obtain the semantic fluency of each original address template;

[0121] From the plurality of original address templates, a target original address template with a semantic fluency greater than a preset fluency is screened out;

[0122] The target original address template that passes the manual inspection is stored in the address template library.

[0123] In summary, the embodiment of the present application provides an address data processing apparatus. Since the greater the weight of the called address data is, the more times the called address data is filled into each called address template to generate training text data, and each target address data is given a corresponding weight according to a required degree, therefore, according to the weight, appropriate target address data can be quickly retrieved and effective training text data can be quickly generated when training a model, which is conducive to quickly obtaining required training data, improves the address data acquisition efficiency, and compared with a method of generating training text data by only using a model to perform address disturbance and the like, this method of applying address templates in the template library can also better ensure the cascading relationship and correctness of the address, improve the accuracy of model training, and is convenient for maintenance and generation.

[0124] The specific limitations of the address data processing apparatus can be referred to the limitations of the address data processing method in the foregoing, and will not be described herein. Each module in the address data processing apparatus can be realized by software, hardware, and a combination thereof, in whole or in part. Each module can be embedded in or independent of a processor in a computer device in a hardware form, or can be stored in a memory in a computer device in a software form, so as to be called and executed by a processor to perform the operations corresponding to each module.

[0125] In one embodiment, a computer device is provided, which can be a server, and an internal structure diagram thereof can be as shown in Figure 8 The computer device includes a processor, a memory, a network interface, and a database connected through a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is configured to communicate with an external terminal through a network connection. The computer program is executed by the processor to implement an address data processing method.

[0126] In one embodiment, a computer device is provided, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor implements the following steps when executing the computer program:

[0127] When the target neural network model needs to be trained, multiple target address data are called from the pre-constructed address database in order according to the weight size, and calling address data are obtained, wherein each target address data is given a corresponding weight according to the degree of compliance;

[0128] Multiple address templates are called from the pre-constructed address template library, and calling address templates are obtained;

[0129] Through the calling address data and the calling address templates, multiple training text data are generated, wherein the greater the weight of the calling address data, the more times the calling address data are filled into each calling address template to generate the training text data;

[0130] The training text data are delivered to the target neural network model.

[0131] In an embodiment, a computer readable storage medium is provided, and the computer readable storage medium stores a computer program:

[0132] When the target neural network model needs to be trained, multiple target address data are called from the pre-constructed address database in order according to the weight size, and calling address data are obtained, wherein each target address data is given a corresponding weight according to the degree of compliance;

[0133] Multiple address templates are called from the pre-constructed address template library, and calling address templates are obtained;

[0134] Through the calling address data and the calling address templates, multiple training text data are generated, wherein the greater the weight of the calling address data, the more times the calling address data are filled into each calling address template to generate the training text data;

[0135] The training text data are delivered to the target neural network model.

[0136] The application is operable in a multitude of various computer system environments or configurations. For example, a personal computer, a server computer, a handheld or portable device, a tablet, a multiprocessor system, a microprocessor-based system, a set top box, programmable consumer electronics, network PC, minicomputer, mainframe computer, distributed computing environments that include any of the above systems or devices, and the like. The application can be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, and the like, that perform particular tasks or implement particular abstract data types. The application can also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules can be located in both local and remote computer storage media including memory storage devices

[0137] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiments. Any reference to memory, storage, database, or other medium used in the embodiments provided by the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0138] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above-mentioned functional units and modules is exemplified, and in actual application, the above-mentioned functions can be completed by different functional units and modules according to needs, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above.

[0139] The above-described embodiments are only used to illustrate the technical solutions of the present application, but not limit them; although the present application is described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. An address data processing method, characterized by, The method comprises: When a target neural network model needs to be trained, a plurality of target address data are called from a pre-constructed address database in order according to weight size, to obtain calling address data, wherein each target address data is given a corresponding weight according to a required degree of compliance; A plurality of address templates are called from a pre-constructed address template library, to obtain calling address templates; A plurality of training text data are generated through the calling address data and the calling address templates, wherein the greater the weight of the calling address data, the more times the calling address data are filled into each calling address template to generate training text data; The training text data are delivered to the target neural network model.

2. The address data processing method of claim 1, wherein, The generation of the plurality of training text data through the calling address data and the calling address templates comprises: Each calling address data is filled into each calling address template in order, until a plurality of training text data meeting a training quantity requirement are obtained.

3. The address data processing method of claim 2, wherein, The calling of the plurality of address templates from the pre-constructed address template library to obtain the calling address templates comprises: A template calling quantity is determined according to the training quantity requirement and the weight of each calling address data; A plurality of address templates corresponding to the template calling quantity are randomly called from the pre-constructed address template library, to obtain the calling address templates.

4. The address data processing method according to any one of claims 1 to 3, characterized in that, The address database is constructed in the following manner: Original address data are obtained, wherein the original address data are real address data searched out; The original address data are taken as the target address data; The target address data are given corresponding weights and address storage time stamps according to a required degree of compliance of the target address data; The plurality of target address data given the weights and the address storage time stamps are stored in the address database.

5. The address data processing method of claim 4, wherein, The original address data are taken as the target address data, which comprises: The address information of the original address data is segmented according to a preset segmentation rule, to obtain target address data comprising multi-level address information, wherein each target address data comprises address information of the same number of levels.

6. The address data processing method according to any one of claims 1 to 3, wherein The address template library is constructed in the following manner: Business data provided by a business application are obtained, and address templates are parsed from the business data as original address templates; and / or, address templates are generated by an address template generation model as original address templates; Each original address template is subjected to semantic analysis to obtain a semantic fluency degree of each original address template; From the plurality of original address templates, target original address templates having a semantic fluency degree greater than a preset fluency degree are screened out; The target original address templates that pass artificial inspection are stored in the address template library.

7. An address data processing system, characterized by The system comprises an address data processing device, an address database and an address template library, wherein the address data processing device is configured to: When a target neural network model needs to be trained, a plurality of target address data are called from the address database in order according to weight size, to obtain calling address data, wherein each target address data is given a corresponding weight according to a required degree of compliance; Call a plurality of address templates from the address template library to obtain called address templates; Generate a plurality of training text data through the called address data and the called address templates, wherein the greater the weight of the called address data is, the more times the called address data is filled into each called address template to generate training text data; Send the training text data to a target neural network model.

8. An address data processing apparatus, characterized by comprising: Comprise: An address data calling module, configured to call a plurality of target address data from a pre-constructed address database according to the weight size in sequence when a target neural network model needs to be trained to obtain called address data, wherein each target address data is given a corresponding weight according to the degree of compliance; An address template calling module, configured to call a plurality of address templates from a pre-constructed address template library to obtain called address templates; A generation module, configured to generate a plurality of training text data through the called address data and the called address templates, wherein the greater the weight of the called address data is, the more times the called address data is filled into each called address template to generate training text data; A sending module, configured to send the training text data to a target neural network model.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the computer program to realize the steps of the address data processing method in any one of claims 1 to 6.

10. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 9. The computer program is executed by the processor to realize the steps of the address data processing method in any one of claims 1 to 6.

Images