A method and device for address identification, and storage medium

Abstract

The embodiments of the present application provide an address recognition method and device, and a storage medium, including: extracting at least one group of word vectors and at least one word statistical quantity from the address information to be recognized according to a preset classification dimension; using a preset neural network model to predict the association probability between a word vector and the next adjacent word vector in each group of word vectors in turn, and obtain a group of association probabilities corresponding to each group of word vectors; the preset neural network model is a model obtained through unsupervised training; at least one group of association probabilities and at least one word statistical quantity corresponding to at least one group of word vectors are input into a preset classification model to obtain a classification result corresponding to the address information to be recognized, and the classification result is used to identify the authenticity of the address information to be recognized; the preset classification model is a model obtained through supervised training.

Description

Technical Field

[0001] The present application relates to the field of electronic applications, and in particular to an address recognition method and device, and a storage medium. Background Art

[0002] Cybercrime has posed a huge threat to e-commerce platforms. Attackers from the cybercrime mainly register accounts through major e-commerce platforms and use these accounts to cheat merchants by placing fake orders and reputation, which impacts the e-commerce evaluation system and damages the interests of e-commerce platforms, normal merchants and buyers. In particular, in the logistics and distribution links, a large number of cybercrime attacks affect the normal ecology of e-commerce, logistics distribution and local life industries by placing fake orders or malicious orders. In order to discover the address cheating risks, it is urgent to identify the false addresses in transaction orders.

[0003] However, the current rule-based risk scoring technology requires that the rules be determined based on the business experience of experts, resulting in slow updates, small coverage, and a high possibility of missed detections. Therefore, a risk scoring method based on supervised classification technology was proposed. However, false addresses are constantly changing, and the number of true address samples is far greater than the number of false address samples, resulting in the inability to accurately identify false addresses for complex false address detection. Summary of the invention

[0004] The embodiments of the present application provide an address recognition method and device, and a storage medium, which can improve the accuracy of identifying false addresses.

[0005] The technical solution of this application is implemented as follows:

[0006] In a first aspect, an embodiment of the present application provides an address identification method, the method comprising:

[0007] Extracting at least one set of word vectors and at least one word statistical quantity from the address information to be identified according to the preset classification dimension;

[0008] Using a preset neural network model, sequentially predicting the association probability between a word vector and the next adjacent word vector in each group of word vectors, to obtain a set of association probabilities corresponding to each group of word vectors; the preset neural network model is a model obtained through unsupervised training;

[0009] Input at least one group of association probabilities corresponding to the at least one group of word vectors and the at least one word statistical quantity into a preset classification model to obtain a classification result corresponding to the address information to be identified, and the classification result is used to identify the authenticity of the address information to be identified; the preset classification model is a model obtained through supervised training.

[0010] In a second aspect, an embodiment of the present application provides an address recognition device, the device comprising:

[0011] An extraction unit, configured to extract at least one set of word vectors and at least one word statistical quantity from the address information to be identified according to a preset classification dimension;

[0012] A prediction unit, used to predict the association probability between a word vector and the next adjacent word vector in each group of word vectors in turn by using a preset neural network model, to obtain a set of association probabilities corresponding to each group of word vectors; the preset neural network model is a model obtained by unsupervised training;

[0013] A classification unit is used to input at least one group of association probabilities corresponding to the at least one group of word vectors and the at least one word statistical quantity into a preset classification model to obtain a classification result corresponding to the address information to be identified, and the classification result is used to identify the authenticity of the address information to be identified; the preset classification model is a model obtained through supervised training.

[0014] In a third aspect, an embodiment of the present application proposes an address identification device, which includes: a processor, a memory, and a communication bus; when the processor executes a running program stored in the memory, the method as described above is implemented.

[0015] In a fourth aspect, an embodiment of the present application provides a storage medium on which a computer program is stored, and when the computer program is executed by a processor, the method as described above is implemented.

[0016] The embodiment of the present application provides an address recognition method and device, and a storage medium, the method comprising: extracting at least one group of word vectors and at least one word statistic from the address information to be recognized according to a preset classification dimension; using a preset neural network model, sequentially predicting the association probability between a word vector and the next adjacent word vector in each group of word vectors, and obtaining a group of association probabilities corresponding to each group of word vectors; the preset neural network model is a model obtained through unsupervised training; inputting at least one group of association probabilities and at least one word statistic corresponding to at least one group of word vectors into the preset classification model, and obtaining a classification result corresponding to the address information to be recognized, and the classification result is used to identify the authenticity of the address information to be recognized; the preset classification model is a model obtained through supervised training. The above implementation scheme adopts a combination of a preset neural network model obtained through unsupervised training and a preset classification model obtained through supervised training. According to the preset classification dimension, the address information to be recognized is converted into word segmentation information of a text sequence pattern, and the preset neural network model of unsupervised training is used to calculate the association probability of the feature information in the address text, and then the preset classification model is used to perform category recognition, so that the key features that can distinguish true and false addresses can be extracted, thereby improving the accuracy of false address recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A flowchart of an address recognition method provided in an embodiment of the present application;

[0018] Figure 2 A flowchart of an exemplary address recognition process provided in an embodiment of the present application;

[0019] Figure 3 A schematic diagram of the structure of an address recognition device provided in an embodiment of the present application Figure 1 ;

[0020] Figure 4 A schematic diagram of the structure of an address recognition device provided in an embodiment of the present application Figure 2 . DETAILED DESCRIPTION

[0021] It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0022] The present application embodiment provides an address recognition method, such as Figure 1 As shown, the method may include:

[0023] S101. Extract at least one set of word vectors and at least one word statistical quantity from the address information to be identified according to a preset classification dimension.

[0024] An address recognition method proposed in an embodiment of the present application is suitable for scenarios in which false addresses are identified in transaction orders on e-commerce platforms, logistics distribution, and local life platforms.

[0025] In the embodiment of the present application, the device for performing address recognition can be any device with data processing and storage functions, such as: a tablet computer, a mobile phone, a personal computer (PC), a laptop computer, a wearable device, etc.

[0026] In the embodiment of the present application, the corresponding address information to be identified is obtained from the transaction order, and firstly, the address information to be identified is subjected to data preprocessing such as word segmentation and word frequency statistics. Specifically, the address information to be identified is segmented to obtain the word segmentation result; the word frequency statistics are performed on the word segmentation result to obtain the first word segmentation and the second word segmentation; wherein the word frequency of the first word segmentation is higher than the preset word frequency threshold, and the word frequency of the second word segmentation is lower than the preset word frequency threshold. After that, at least one set of word vectors and at least one word statistical quantity are extracted from the preprocessed address information to be identified according to the preset classification dimension.

[0027] It should be noted that the embodiment of the present application can use the jieba segmentation method and other segmentation methods to segment the address information to be identified. The specific segmentation method can be selected according to the actual situation, and the embodiment of the present application does not make any specific limitation.

[0028] Specifically, the address recognition device divides the first word segment into at least one group of words according to a preset classification dimension, and extracts word vectors for at least one group of words to obtain at least one group of word vectors; the address recognition device counts the number of second word segmentations, and the number of second word segmentations is a statistical number in at least one word statistical number.

[0029] It should be noted that the embodiment of the present application can use sougou pre-trained word vectors to extract word vectors for the first word segment. The specific word vector extraction method can be selected according to actual conditions, and the embodiment of the present application does not make specific limitations.

[0030] It can be understood that by extracting the word vector for the first word segment whose word frequency is higher than the preset word frequency threshold and then inputting it into the preset neural network model for probability calculation, the vocabulary dimension of the preset neural network model can be reduced, thereby greatly reducing the amount of data processed by the preset neural network model and improving the processing speed of the preset neural network model.

[0031] In an embodiment of the present application, the preset classification dimensions include seven classification dimensions: target address, stop words, English characters, Chinese numerals, Arabic numerals, the number of rare characters, and the number of characters between stop words; wherein the target address is address information with a word frequency higher than a preset word frequency threshold and excluding stop words, English characters, Chinese numerals, and Arabic numerals; rare characters are address information with a word frequency lower than the preset word frequency threshold; and stop words are words that represent administrative divisions.

[0032] Exemplarily, stop words are words of administrative divisions such as provinces, cities, districts, counties, and streets; English characters are characters such as a, b, and c; Chinese characters are numbers such as one, two, and three; and Arabic numerals are numbers such as 1, 2, and 3.

[0033] It should be noted that the classification dimensions of target address, stop words, English characters, Chinese characters and Arabic data need to be input into the preset neural network model for probability calculation. Therefore, the first word segmentation can be divided into five groups of words under the above five classification dimensions.

[0034] Furthermore, in an embodiment of the present application, the number of characters between each two adjacent word vectors in a group of word vectors corresponding to stop words in the address information to be identified is searched to obtain a group of character numbers; a preset number of characters is filtered from the group of character numbers, and the preset number of characters is processed to obtain a statistical value; a statistical value is a statistical number in at least one word statistical number.

[0035] It can be seen that for the seven classification dimensions, at least one group of word vectors may include five groups of word vectors corresponding to the five classification dimensions of target address, stop words, English characters, Chinese numerals and Arabic data, and at least one word statistical quantity may include two word statistical quantities corresponding to the two classification dimensions of the number of second participles and a statistical value.

[0036] In an embodiment of the present application, the preset number of characters can be three characters, and the process of screening the preset number of characters from a group of characters may include: the address recognition device sorts a group of characters in order from large to small, and searches for the top 3 characters from the sorted group of characters, and then weights the top 3 characters according to a certain weight value to obtain the final value.

[0037] Exemplarily, the weight values ​​corresponding to the first 3 numbers of characters may be 0.2, 0.7, and 0.1.

[0038] S102. Using a preset neural network model, predict the association probability between a word vector and the next adjacent word vector in each group of word vectors in turn, and obtain a set of association probabilities corresponding to each group of word vectors; the preset neural network model is a model obtained through unsupervised training.

[0039] In an embodiment of the present application, the address recognition device inputs at least one group of word vectors into a preset neural network model, and for each group of word vectors, uses the preset neural network model to predict the association relationship between every two adjacent word vectors to obtain a set of association probabilities corresponding to each group of word vectors.

[0040] It should be noted that the preset neural network model corresponds to the preset length. Before the address recognition device inputs at least one set of word vectors into the preset neural network model, the length of each set of word vectors is adjusted to the preset length.

[0041] Exemplarily, the preset length is 20, and a group of word vectors with a length greater than 20 are truncated to 20; for a group of word vectors with a length less than 20, the length is padded to 20 by adding a placeholder at the end.

[0042] It should be noted that the above truncation operation can be to discard the part exceeding 20, or to transfer the part exceeding 20 to another set of word vectors. The specific truncation operation can be selected according to the actual situation, and the embodiment of the present application does not make specific limitations.

[0043] In an embodiment of the present application, the preset neural network model is a char-RNN network, wherein the char-RNN network includes an embedding layer, an RNN neural network layer and a linear mapping layer.

[0044] In an embodiment of the present application, at least one group of word vectors is input into the embedding layer, and the input form can be a vocabulary number, such as [34, 56, 199, 500, 500] can represent [Beijing, Haidian, Garden, Placeholder, Placeholder, Placeholder]. In the embedding layer, the [b, s] matrix is ​​mapped to the [b, s, l] matrix, where b represents the batch size, s represents the sentence length, and l represents the word vector length.

[0045] In the embodiment of the present application, the RNN neural network layer is composed of two unidirectional LSTM layers, wherein the number of neurons in each LSTM layer is 128.

[0046] In the embodiment of the present application, the input dimension of the linear mapping layer is 128. In the linear mapping layer, 128 dimensions are mapped to the vocabulary dimension, and then the softmax function is used to determine which word vector the word vector is mapped to.

[0047] In the embodiment of the present application, the preset neural network model is obtained through unsupervised training. Specifically, the initial neural network model is unsupervisedly trained using real sample address information to obtain the preset neural network model.

[0048] It should be noted that the real sample address information is divided into a training set and a validation set. The training set is used to train the initial neural network model, and the validation set is used to verify the performance of the trained neural network model. Since the preset neural network model is obtained through unsupervised training, in the preset neural network model, the indicator used to measure whether the performance of the preset neural network model is improved is the loss value on the validation set, where the loss function measures the loss value of the probability from the previous word to the next word, and a reduction in the loss value of the validation set represents an improvement in the performance of the preset neural network model.

[0049] S103. Input at least one group of association probabilities and at least one word statistical quantity corresponding to at least one group of word vectors into a preset classification model to obtain a classification result corresponding to the address information to be identified, and the classification result is used to identify the authenticity of the address information to be identified; the preset classification model is a model obtained through supervised training.

[0050] In the embodiment of the present application, the preset classification model is a random forest model.

[0051] In an embodiment of the present application, the classification results of the preset classification model for at least one set of association probabilities and at least one word statistical number include false addresses and real addresses, and can also be scoring results for at least one set of association probabilities and at least one word statistical number. The address recognition device determines the risk level of the address information to be identified based on the scoring results, and then identifies whether the address information to be identified is a real address or a false address.

[0052] In an embodiment of the present application, the preset classification model is trained using a supervised training method. Specifically, the true sample address information and the false sample address information are first processed based on the preset neural network model to obtain the sample output result; then the sample output result is divided into a training set and a test set, and the initial classification model is supervised trained based on the sample output result to obtain the preset classification model.

[0053] In the embodiment of the present application, the training set is input into the initial classification model for training, and then the trained classification model is classified and tested using the test set to finally obtain a preset classification model.

[0054] For example, Figure 2 As shown in the figure, the original address is preprocessed, and then the original address is divided into seven dimensions of core address content, stop words such as provinces, cities and districts, English characters, Chinese characters, Arabic numerals, the number of rare words and the length of characters between stop words. The core address content, stop words such as provinces, cities and districts, English characters, Chinese characters, Arabic numerals and the five dimensions of word segmentation are input into char-RNN, and probability vector 1, probability vector 2, probability vector 3, probability vector 4 and probability vector 5 are output; probability vector 1, probability vector 2, probability vector 3, probability vector 4, probability vector 5, the number of rare words and the length of characters between stop words are input into the Radom Forest model to obtain the classification result of the original address. That is, whether the original address is a real address or a false address.

[0055] It can be understood that a combination of the preset neural network model obtained by unsupervised training and the preset classification model obtained by supervised training is adopted. According to the preset classification dimension, the address information to be identified is converted into the word segmentation information of the text sequence pattern. The preset neural network model of unsupervised training is used to calculate the association probability of the feature information in the address text. After that, the preset classification model is used for category recognition, and the key features that can distinguish between true and false addresses can be extracted, thereby improving the accuracy of false address recognition.

[0056] The present application embodiment provides an address recognition device 1. Figure 3 As shown, the address recognition device 1 includes:

[0057] An extraction unit 10, configured to extract at least one set of word vectors and at least one word statistical quantity from the address information to be identified according to a preset classification dimension;

[0058] The prediction unit 11 is used to predict the association probability between a word vector and the next adjacent word vector in each group of word vectors in turn by using a preset neural network model, and obtain a group of association probabilities corresponding to each group of word vectors; the preset neural network model is a model obtained by unsupervised training;

[0059] The classification unit 12 is used to input at least one group of association probabilities corresponding to the at least one group of word vectors and the at least one word statistical quantity into a preset classification model to obtain a classification result corresponding to the address information to be identified, and the classification result is used to identify the authenticity of the address information to be identified; the preset classification model is a model obtained through supervised training.

[0060] Optionally, the device further comprises: a training unit;

[0061] The training unit is used to perform unsupervised training on the initial neural network model using the true sample address information to obtain the preset neural network model; based on the preset neural network model, the true sample address information and the false sample address information are processed to obtain the sample output result; based on the sample output result, the initial classification model is supervised trained to obtain the preset classification model.

[0062] Optionally, the device further comprises: a word segmentation unit, a statistical unit and a division unit;

[0063] The word segmentation unit is used to segment the address information to be identified to obtain a word segmentation result;

[0064] The statistical unit is used to perform word frequency statistics on the word segmentation results to obtain a first word segmentation and a second word segmentation; the word frequency of the first word segmentation is higher than a preset word frequency threshold, and the word frequency of the second word segmentation is lower than the preset word frequency threshold;

[0065] The division unit is used to divide the first word segmentation into at least one group of words according to the preset classification dimension, and extract word vectors for the at least one group of words to obtain the at least one group of word vectors;

[0066] The statistical unit is further used to count the number of the second participles, where the number of the second participles is a statistical number in the at least one word statistical number.

[0067] Optionally, the preset classification dimensions include the target address, stop words, English characters, Chinese numerals, Arabic numerals, the number of rare characters and the number of characters between stop words; wherein the target address is address information with a word frequency higher than a preset word frequency threshold and excluding stop words, English characters, Chinese numerals and Arabic numerals; the rare characters are address information with a word frequency lower than a preset word frequency threshold; and the stop words are words representing administrative divisions.

[0068] Optionally, the device further comprises: a searching unit and a screening unit;

[0069] The search unit is used to search for the number of characters between each two adjacent word vectors in a group of word vectors corresponding to the stop words from the address information to be identified, so as to obtain a group of character numbers;

[0070] The screening unit is used to screen a preset number of characters from a group of character numbers, and process the preset number of characters to obtain a statistical value; the statistical value is a statistical value of the at least one word statistical number.

[0071] Optionally, the preset neural network model is a character-level recurrent neural network Char-RNN model.

[0072] Optionally, the preset classification model is a random forest model.

[0073] The embodiment of the present application provides an address recognition device, which extracts at least one group of word vectors and at least one word statistic from the address information to be recognized according to a preset classification dimension; uses a preset neural network model to predict the association probability between a word vector and the next adjacent word vector in each group of word vectors in turn, and obtains a group of association probabilities corresponding to each group of word vectors; the preset neural network model is a model obtained through unsupervised training; at least one group of association probabilities and at least one word statistic corresponding to at least one group of word vectors are input into the preset classification model to obtain a classification result corresponding to the address information to be recognized, and the classification result is used to identify the authenticity of the address information to be recognized; the preset classification model is a model obtained through supervised training. It can be seen that the address recognition device proposed in this embodiment adopts a combination of a preset neural network model obtained by unsupervised training and a preset classification model obtained by supervised training. According to the preset classification dimension, the address information to be recognized is converted into word segmentation information of a text sequence pattern, and the preset neural network model of unsupervised training is used to calculate the association probability of the feature information in the address text, and then the preset classification model is used to perform category recognition, so as to extract the key features that can distinguish true and false addresses, thereby improving the accuracy of false address recognition.

[0074] Figure 4 A schematic diagram of the structure of an address recognition device 1 provided in an embodiment of the present application Figure 2 In practical applications, based on the same disclosure concept of the above embodiments, Figure 4 As shown, the address recognition device 1 of this embodiment includes: a processor 13 , a memory 14 and a communication bus 15 .

[0075] In the process of a specific embodiment, the above-mentioned extraction unit 10, prediction unit 11, classification unit 12, training unit, word segmentation unit, statistical unit, division unit, search unit and screening unit can be implemented by a processor 13 located on the terminal 1, and the above-mentioned processor 13 can be at least one of a specific application integrated circuit (ASIC, Application Specific Integrated Circuit), a digital signal processor (DSP, Digital Signal Processor), a digital signal processing image processing device (DSPD, Digital Signal Processing Device), a programmable logic image processing device (PLD, Programmable Logic Device), a field programmable gate array (FPGA, Field Programmable Gate Array), a CPU, a controller, a microcontroller, and a microprocessor. It can be understood that for different devices, the electronic device used to implement the above-mentioned processor function can also be other, and this embodiment does not specifically limit it.

[0076] In the embodiment of the present application, the communication bus 15 is used to realize the connection communication between the processor 13 and the memory 14; when the processor 13 executes the running program stored in the memory 14, the following address recognition method is implemented:

[0077] Extracting at least one set of word vectors and at least one word statistical quantity from the address information to be identified according to the preset classification dimension;

[0078] Using a preset neural network model, sequentially predicting the association probability between a word vector and the next adjacent word vector in each group of word vectors, to obtain a set of association probabilities corresponding to each group of word vectors; the preset neural network model is a model obtained through unsupervised training;

[0079] Input at least one group of association probabilities corresponding to the at least one group of word vectors and the at least one word statistical quantity into a preset classification model to obtain a classification result corresponding to the address information to be identified, and the classification result is used to identify the authenticity of the address information to be identified; the preset classification model is a model obtained through supervised training.

[0080] Furthermore, the above-mentioned processor 13 is also used to use the true sample address information to perform unsupervised training on the initial neural network model to obtain the preset neural network model; based on the preset neural network model, the true sample address information and the false sample address information are processed to obtain the sample output result; based on the sample output result, the initial classification model is supervised trained to obtain the preset classification model.

[0081] Furthermore, the processor 13 is further configured to segment the address information to be identified to obtain a segmentation result; perform word frequency statistics on the segmentation result to obtain a first segmentation and a second segmentation; the word frequency of the first segmentation is higher than a preset word frequency threshold, and the word frequency of the second segmentation is lower than the preset word frequency threshold; divide the first segmentation into at least one group of segments according to the preset classification dimension, and perform word vector extraction on the at least one group of segments to obtain the at least one group of word vectors; and count the number of the second segmentations, where the number of the second segmentations is one of the statistical numbers of the at least one word.

[0082] Furthermore, the preset classification dimensions include the target address, stop words, English characters, Chinese numerals, Arabic numerals, the number of rare characters and the number of characters between stop words; wherein the target address is address information with a word frequency higher than a preset word frequency threshold and excluding stop words, English characters, Chinese numerals and Arabic numerals; the rare characters are address information with a word frequency lower than a preset word frequency threshold; and the stop words are words representing administrative divisions.

[0083] Furthermore, the processor 13 is also used to search for the number of characters between each two adjacent word vectors in a group of word vectors corresponding to the stop words in the address information to be identified, and obtain a group of character numbers; filter a preset number of characters from the group of character numbers, and process the preset number of characters to obtain a statistical value; the statistical value is a statistical number in the at least one word statistical number.

[0084] Furthermore, the preset neural network model is a character-level recurrent neural network Char-RNN model.

[0085] Furthermore, the preset classification model is a random forest model.

[0086] An embodiment of the present application provides a storage medium having a computer program stored thereon. The computer-readable storage medium stores one or more programs. The one or more programs can be executed by one or more processors and applied to an address recognition device. The computer program implements the address recognition method as described above.

[0087] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article 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, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0088] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present disclosure, or the part that contributes to the relevant technology, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a disk, or an optical disk), and includes a number of instructions for enabling an image display device (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in each embodiment of the present disclosure.

[0089] 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.

Claims

1. An address recognition method, characterized in that: The method comprises: Extracting at least one set of word vectors and at least one word statistical quantity from the address information to be identified according to the preset classification dimension; Using a preset neural network model, sequentially predicting the association probability between a word vector and the next adjacent word vector in each group of word vectors, to obtain a set of association probabilities corresponding to each group of word vectors; the preset neural network model is a model obtained through unsupervised training; Inputting at least one group of association probabilities corresponding to the at least one group of word vectors and the at least one word statistical quantity into a preset classification model to obtain a classification result corresponding to the address information to be identified, wherein the classification result is used to identify the authenticity of the address information to be identified; the preset classification model is a model obtained through supervised training; Before extracting at least one set of word vectors and at least one word statistical quantity from the address information to be identified according to the preset classification dimension, the method further includes: Performing word segmentation on the address information to be identified to obtain a word segmentation result; Performing word frequency statistics on the word segmentation results to obtain a first word segmentation and a second word segmentation; the word frequency of the first word segmentation is higher than a preset word frequency threshold, and the word frequency of the second word segmentation is lower than the preset word frequency threshold; Accordingly, extracting at least one set of word vectors and at least one word statistical quantity from the address information to be identified according to the preset classification dimension includes: According to the preset classification dimension, the first word segmentation is divided into at least one group of words, and word vectors are extracted for the at least one group of words to obtain the at least one group of word vectors; The number of the second participles is counted, where the number of the second participles is a statistical number in the at least one word statistical number.

2. The method according to claim 1, characterized in that Before extracting at least one set of word vectors and at least one word statistical quantity from the address information to be identified according to the preset classification dimension, the method further includes: Using the real sample address information to perform unsupervised training on the initial neural network model to obtain the preset neural network model; Based on the preset neural network model, the true sample address information and the false sample address information are processed to obtain a sample output result; The initial classification model is supervisedly trained based on the sample output results to obtain the preset classification model.

3. The method according to claim 1, characterized in that The preset classification dimensions include the target address, stop words, English characters, Chinese numerals, Arabic numerals, the number of rare characters and the number of characters between stop words; wherein the target address is address information with a word frequency higher than a preset word frequency threshold and excluding stop words, English characters, Chinese numerals and Arabic numerals; the rare characters are address information with a word frequency lower than a preset word frequency threshold; and the stop words are words representing administrative divisions.

4. The method according to claim 3, characterized in that The method further comprises: From the address information to be identified, searching for the number of characters between every two adjacent word vectors in a group of word vectors corresponding to the stop words to obtain a group of character numbers; A preset number of characters is selected from a group of character numbers, and the preset number of characters is processed to obtain a statistical value; the statistical value is a statistical value in the at least one word statistical value.

5. The method according to claim 1, characterized in that The preset neural network model is a character-level recurrent neural network Char-RNN model.

6. The method according to claim 1, characterized in that The preset classification model is a random forest model.

7. An address recognition device, characterized in that: The device comprises: An extraction unit, configured to extract at least one set of word vectors and at least one word statistical quantity from the address information to be identified according to a preset classification dimension; A prediction unit, used to predict the association probability between a word vector and the next adjacent word vector in each group of word vectors in turn by using a preset neural network model, to obtain a set of association probabilities corresponding to each group of word vectors; the preset neural network model is a model obtained by unsupervised training; a classification unit, configured to input at least one group of association probabilities corresponding to the at least one group of word vectors and the at least one word statistical quantity into a preset classification model to obtain a classification result corresponding to the address information to be identified, wherein the classification result is used to identify the authenticity of the address information to be identified; the preset classification model is a model obtained through supervised training; A word segmentation unit, used for segmenting the address information to be identified to obtain a word segmentation result; A statistical unit, configured to perform word frequency statistics on the word segmentation result to obtain a first word segmentation and a second word segmentation; the word frequency of the first word segmentation is higher than a preset word frequency threshold, and the word frequency of the second word segmentation is lower than the preset word frequency threshold; A division unit, configured to divide the first word segmentation into at least one group of words according to the preset classification dimension, and extract word vectors for the at least one group of words to obtain the at least one group of word vectors; The statistical unit is further used to count the number of the second participles, where the number of the second participles is a statistical number in the at least one word statistical number.

8. An address recognition device, characterized in that: The device comprises: a processor, a memory and a communication bus; when the processor executes the running program stored in the memory, the method according to any one of claims 1 to 7 is implemented.

9. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images