Method, device, equipment and storage medium for identifying abnormal address information

Abstract

This invention discloses a method, apparatus, device, and storage medium for identifying abnormal address information, relating to the field of computer technology. The method may include: acquiring address segmentation sequences corresponding to each address information to be identified; for each address segmentation sequence, determining the part-of-speech sequence corresponding to the current address segmentation sequence based on the part-of-speech category corresponding to each address segmentation in the current address segmentation sequence; determining abnormal sequences in each part-of-speech sequence based on a first unsupervised anomaly detection model, and identifying the address information to be identified corresponding to the abnormal sequences as abnormal address information. Using the technical solution of this invention, there is no need to spend a lot of time annotating sample addresses, which can improve recognition efficiency; furthermore, the accuracy of the recognition result no longer depends on the quality of sample annotation, so the stability is high. In addition, it can also meet the recognition needs of continuously changing abnormal addresses, thereby improving the recognition accuracy.

Description

Technical Field

[0001] Embodiments of the present invention relate to the field of computer technology, and in particular, to a method, device, equipment and storage medium for identifying abnormal address information. Background Art

[0002] At present, cyber black industry attacks have seriously affected the normal operation of Internet platforms (such as e-commerce platforms, financial platforms, etc.). For example, in order to illegally obtain coupons issued by e-commerce platforms, or to illegally cash out through interest-free installment benefits provided by e-commerce platforms, black industry attackers fill in abnormal addresses such as false addresses and fuzzy addresses when placing orders on the platform. This behavior affects the normal operation of the logistics and distribution links of e-commerce platforms.

[0003] In response to a series of network black market attacks carried out by black market attackers using abnormal addresses, the existing technology has proposed many solutions for identifying abnormal addresses. The commonly used identification solution is to learn a large number of labeled sample addresses through machine learning algorithms or deep learning algorithms, extract the key features of abnormal addresses, and then identify abnormal addresses based on the summarized key features.

[0004] However, in the process of implementing the present invention, it was found that there are at least the following problems in the prior art: in the identification scheme of the prior art, it takes a lot of time to label the sample addresses, and the accuracy of the identification results depends on the quality of the sample labeling, and the stability is poor; in addition, in actual applications, black market attackers will constantly adjust the rules for filling in abnormal addresses, and the identification scheme of the prior art cannot adapt to the dynamic changes of abnormal addresses, so the recognition accuracy is not high. Summary of the invention

[0005] The embodiments of the present invention provide a method, device, equipment and storage medium for identifying abnormal address information, which can improve the efficiency of identifying abnormal address information and the accuracy of the obtained identification results, and the stability of the identification scheme is relatively high.

[0006] In a first aspect, an embodiment of the present invention provides a method for identifying abnormal address information, the method comprising: obtaining address segmentation sequences corresponding to each address information to be identified; for each address segmentation sequence, based on the part-of-speech category corresponding to each address segmentation in the current address segmentation sequence, determining the segmentation part-of-speech sequence corresponding to the current address segmentation sequence; determining the abnormal sequence in each segmentation part-of-speech sequence based on a first unsupervised anomaly detection model, and determining the address information to be identified corresponding to the abnormal sequence as the abnormal address information.

[0007] In a second aspect, an embodiment of the present invention further provides a device for identifying abnormal address information, which may include an acquisition module and a determination module;

[0008] Specifically, the acquisition module is used to acquire the address segmentation sequence corresponding to each address information to be identified;

[0009] A determination module, for determining, for each address segmentation sequence, a segmentation part-of-speech sequence corresponding to the current address segmentation sequence based on the part-of-speech categories corresponding to each address segmentation in the current address segmentation sequence;

[0010] The determination module is used to determine the abnormal sequence in each word segmentation part-of-speech sequence based on the first unsupervised anomaly detection model, and determine the address information to be identified corresponding to the abnormal sequence as the abnormal address information.

[0011] In a third aspect, an embodiment of the present invention provides a device for identifying abnormal address information, the device for identifying abnormal address information comprising:

[0012] one or more processors;

[0013] A memory for storing one or more programs;

[0014] When the one or more programs are executed by the one or more processors, the one or more processors implement the method for identifying abnormal address information provided by any embodiment of the present invention.

[0015] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a method for identifying abnormal address information as provided in any embodiment of the present invention.

[0016] The embodiments of the above invention have the following advantages or beneficial effects:

[0017] In the technical solution provided by the embodiment of the present invention, based on the fact that the distribution of the part-of-speech categories of each address segmentation in the address segmentation sequence corresponding to the normal address information is greatly different from the distribution of the part-of-speech categories of each address segmentation in the address segmentation sequence corresponding to the abnormal address information, when it is necessary to identify the abnormal address information of each address information to be identified, the address segmentation sequence corresponding to each address information to be identified can be first obtained, and then the part-of-speech sequence corresponding to each address segmentation sequence can be determined based on the part-of-speech category corresponding to each address segmentation in each address segmentation sequence. Afterwards, in order to identify the abnormal sequence in each part-of-speech sequence that has a large difference in distribution from the normal part-of-speech sequence, the embodiment of the present invention, based on the idea of ​​an unsupervised learning algorithm, determines the majority of the part-of-speech sequences in each part-of-speech sequence as a normal sequence, and then identifies the outlier part-of-speech sequence in each part-of-speech sequence that has a large difference in distribution from the majority part-of-speech sequence through the first unsupervised anomaly detection model, and then the identified outlier part-of-speech sequence can be determined as an abnormal sequence, so that the address information to be identified corresponding to the abnormal sequence can be determined as abnormal address information.

[0018] It can be seen that in the technical solution provided by the embodiment of the present invention, the abnormal address information in each address information to be identified can be identified based on the difference in the part-of-speech categories of the address segmentation words in the normal address information and the abnormal address information. Since the recognition process is implemented based on the idea of ​​an unsupervised learning algorithm, the embodiment of the present invention does not need to spend a lot of time to mark the sample addresses, which can improve the recognition efficiency; and since there is no need to mark the sample addresses, the accuracy of the recognition results no longer depends on the quality of the sample marking, so the recognition scheme provided by the embodiment of the present invention has high stability. In addition, the embodiment of the present invention can identify various abnormal address information with different part-of-speech distributions from normal address information, can adapt to the dynamic changes of abnormal addresses, and meet the recognition needs of constantly adjusting and changing abnormal addresses, thereby improving the recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a flowchart of a method for identifying abnormal address information provided by an embodiment of the present invention;

[0020] Figure 2 It is a flowchart of another method for identifying abnormal address information provided by an embodiment of the present invention;

[0021] Figure 3 It is a flowchart of another method for identifying abnormal address information provided by an embodiment of the present invention;

[0022] Figure 4 It is a flowchart of another method for identifying abnormal address information provided by an embodiment of the present invention;

[0023] Figure 5 It is a simplified flowchart of a method for identifying abnormal address information provided by an embodiment of the present invention;

[0024] Figure 6 It is a structural schematic diagram of a device for identifying abnormal address information provided by an embodiment of the present invention;

[0025] Figure 7 It is a structural schematic diagram of an abnormal address information identification device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0026] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention. It should also be noted that, for ease of description, only parts related to the present invention, rather than all structures, are shown in the accompanying drawings.

[0027] The term "and / or" in this article is merely a description of the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone.

[0028] In addition, the terms "including" and "having" and any variations thereof mentioned in the description of the embodiments of the present invention are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device including a series of steps or units is not limited to the listed steps or units, but may optionally include other steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or devices.

[0029] It should be noted that, in the embodiments of the present invention, words such as "exemplary" or "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design scheme described as "exemplary" or "for example" in the embodiments of the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way.

[0030] In the description of the embodiments of the present invention, unless otherwise specified, “plurality” or “each” means two or more than two.

[0031] Before further describing the present invention in detail, the application background of the technical solution provided in the embodiment of the present invention and the identification solution for abnormal addresses in the prior art are described.

[0032] With the rapid development of Internet technology, the number of Internet users continues to increase, and large-scale Internet user groups have generated huge Internet business needs and interactive traffic. In addition to the Internet ecosystem built by major Internet manufacturers through websites, applications (Application, APP), small programs, and online / offline service businesses, there have also emerged a series of black industry chains derived from Internet business, gathering large-scale black industry groups. Black industry attackers in these black industry groups make profits through online black industry attacks such as brushing orders, cashing out, and malicious registration, which seriously affects the normal operation of Internet platforms. Therefore, identifying online black industry attacks on Internet platforms has become the key to ensuring the normal operation of Internet platforms.

[0033] In the prior art, when using Internet services, users often need to fill in address information (for example, users need to fill in the delivery address when placing an order on an e-commerce platform, and need to fill in the registration address when opening an account on a financial platform, etc.), and address information has become a relatively common type of information collected. Therefore, black industry attackers usually use address information to implement network black industry attacks, and fill in false addresses, fuzzy addresses, and abnormal addresses containing special characters when placing orders or registering to circumvent the platform's risk control rules for illegal profit. For example, some e-commerce platforms will issue some large coupons to new users or users who place orders for the first time. In order to circumvent the platform's risk control rules for new users or users who place orders for the first time, black industry attackers will fill in non-existent false addresses when placing orders to illegally obtain large coupons issued by e-commerce platforms. For another example, some e-commerce platforms can provide users with interest-free installment benefits. Black industry attackers can use this benefit to illegally cash out. In order to reduce this cash-out behavior, the platform will set up risk control rules to restrict business orders with concentrated addresses. In order to circumvent the risk control rules set by the platform, black industry attackers will use fuzzy addresses when placing orders (for example, the address information filled in can be: near XX community, call when you arrive) to cash out transactions. For another example, black industry attackers can also use addresses containing special characters to circumvent various risk control rules set by the platform. Therefore, in view of the fact that black industry attackers use address information to implement network black industry attacks, the existing technology can detect network black industry attacks on Internet platforms by identifying abnormal addresses.

[0034] At present, there are two main solutions for identifying abnormal addresses. The first is an identification solution based on statistical analysis, and the second is an identification solution based on machine learning algorithms or deep learning algorithms. Among them, the identification solution based on statistical analysis is: manually statistically analyze a large amount of address information, summarize the key features of the abnormal addresses filled in by black industry attackers (for example, which keywords are included), and then formulate a single rule or multiple rules in sequence according to the summarized features to restrict the address information of registered users or ordering users to achieve the identification of abnormal addresses. This solution has a high recognition accuracy rate for fixed-pattern network black industry attack behaviors with known rules, but manual statistics are too time-consuming, so the second identification solution is currently commonly used. The second identification solution is specifically: learn a large number of labeled sample addresses through machine learning algorithms or deep learning algorithms, extract the key features of abnormal addresses, and then identify abnormal addresses based on the summarized key features. Compared with the first identification scheme, the second identification scheme has more comprehensive coverage and higher applicability. However, the following problems still exist: it takes a lot of time to label the sample addresses, and the accuracy of the identification results depends on the quality of the sample labeling, and the stability is poor; in addition, black market attackers will constantly adjust the rules for filling in abnormal addresses, and the two commonly used identification schemes cannot adapt to the dynamic changes of abnormal addresses, so the identification accuracy is not high.

[0035] In order to solve the problems existing in the above-mentioned prior art, an embodiment of the present invention proposes a method for identifying abnormal address information. Figure 1 A flowchart of a method for identifying abnormal address information provided by an embodiment of the present invention, which can be applied to application scenarios for identifying abnormal addresses such as false addresses, ambiguous addresses, and addresses containing special characters. The method can be executed by an apparatus for identifying abnormal address information provided by an embodiment of the present invention, which can be implemented in software and / or hardware and integrated in an apparatus for identifying abnormal address information.

[0036] like Figure 1 As shown, the method for identifying abnormal address information provided by the embodiment of the present invention specifically includes the following steps:

[0037] S110, obtaining address segmentation sequences corresponding to each piece of address information to be identified.

[0038] In one possible implementation, for an Internet platform that may be attacked by cyber black industry attacks, an embodiment of the present invention may count the address information in the business orders of the platform in each statistical period at intervals of a predetermined statistical period to obtain each address information to be identified. Exemplarily, an embodiment of the present invention may count the address information in the business orders generated by the platform on that day every day with a statistical period of 24 hours. It should be noted that the embodiment of the present invention has the same processing process for each address information to be identified obtained in each statistical period, and the embodiment of the present invention will be described using the processing process for each address information to be identified obtained in any statistical period as an example.

[0039] Among them, the address segmentation sequence can be a sequence containing multiple address segmentations obtained by segmenting the address information to be identified through a segmentation tool. For example, the segmentation tool can be a jieba segmentation tool (a Chinese segmentation tool). Exemplarily, taking the address information to be identified as "XX City XX District XX Street XX Community XX Building" as an example, the address information to be identified is segmented through a segmentation tool, and the address segmentation sequence obtained is: XX City; XX District; XX Street; XX Community; XX Building. Among them, "XX City", "XX District", "XX Street", "XX Community", and "XX Building" are the 5 address segmentations in the address information to be identified.

[0040] Optionally, an embodiment of the present invention can obtain the address word segmentation sequence corresponding to each address information to be identified in the following manner: for each address information to be identified, the current address information to be identified is segmented based on the first named entity recognition model to obtain the current address word segmentation sequence corresponding to the current address information to be identified.

[0041] The first named entity recognition model can be a model for recognizing address entities obtained by pre-training based on a named entity recognition algorithm. The first named entity recognition model can be obtained by training a large amount of address text information. The specific process of obtaining the first named entity recognition model can refer to the relevant description of the prior art, and the embodiments of the present invention will not be repeated here.

[0042] Since some address words in the address information to be identified may be composed of multiple words, the address information to be identified may be segmented using a common word segmentation tool, and the address word segmentation sequence obtained may contain some unreasonable address word segmentations (for example, not a complete address word segmentation). Therefore, the embodiment of the present invention may use a named entity recognition tool (i.e., the first named entity recognition model in the embodiment of the present invention) to process the address information to be identified, determine each address entity in the address information to be identified, and thereby determine each address entity as each address word in the address information to be identified, and obtain the address word segmentation sequence corresponding to the address information to be identified.

[0043] For example, if the address information to be identified is "XX City XX District XX Street Economic and Trade Building", the address information to be identified is segmented using an ordinary segmentation tool, and the address segmentation sequence obtained is: XX City; XX District; XX Street; Economic and Trade; Building. However, Economic and Trade Building is a complete address segment, so this segmentation method is not effective for address information. The address information to be identified is segmented using the first named entity recognition model, and the address segmentation sequence obtained is: XX City; XX District; XX Street; Economic and Trade Building. It can be seen that the address segmentations in the address segmentation sequence obtained by segmenting using the first named entity recognition model are all complete address segments, and the segmentation effect is better.

[0044] In another example, taking the address information to be identified as "the south side of XX community, XX street, XX district, XX city", the first named entity recognition model is used to segment the address information to be identified, and the address segmentation sequence obtained is: XX city; XX district; XX street; XX community; of; south side. In another example, taking the address information to be identified as "XX city, XX district, XX street, no delivery required", the first named entity recognition model is used to segment the address information to be identified, and the address segmentation sequence obtained is: XX city; XX district; XX street; no; need; delivery.

[0045] Optionally, before obtaining the address word segmentation sequences corresponding to each address information to be identified, the method for identifying abnormal address information provided by the embodiment of the present invention may also include: obtaining each address information to be processed; performing data cleaning processing on each address information to be processed based on preset data cleaning rules to obtain each address information to be identified.

[0046] The preset data cleaning rule may be a predetermined rule. Exemplarily, the preset data cleaning rule may be to determine the address information to be processed that has a text length less than a predetermined length, or a null value, or is in a whitelist / blacklist as the information to be screened out. Data cleaning processing is to filter out the information to be screened out in each address information to be processed.

[0047] Since the number of address information to be processed obtained within a statistical cycle is very large, if each address information to be processed is directly subjected to abnormal identification processing, the processing efficiency is extremely low. Therefore, after obtaining each address information to be processed, the embodiment of the present invention can first pre-process each address information to be processed, and filter out some address information to be processed that is obviously normal address information (for example, address information belonging to a preset whitelist) or obviously abnormal address information (for example, address information belonging to a preset blacklist, empty address information, and address information with a text length less than a predetermined length), and then perform abnormal identification processing on each address information to be identified after the screening. In this way, the recognition rate of abnormal address information can be improved.

[0048] It should be noted that in the technical solution provided by the embodiment of the present invention, the collection, collection, update, analysis, processing, use, transmission, storage and other aspects of the user personal information involved are in compliance with the provisions of relevant laws and regulations, are used for legitimate purposes, and do not violate public order and good morals. Take necessary measures for user personal information to prevent illegal access to user personal information data and maintain user personal information security, network security and national security. For example, during the application process of the embodiment of the present invention, it is necessary to collect the address information in the business order. After the address information is collected, the address information can be de-identified / encrypted by technical means.

[0049] S120: For each address segmentation sequence, based on the part-of-speech categories corresponding to each address segmentation in the current address segmentation sequence, determine a part-of-speech sequence corresponding to the current address segmentation sequence.

[0050] In one possible implementation, in the process of determining the address entity in the address information to be identified through the first named entity recognition model, the embodiment of the present invention performs segmentation processing on each address participle in the address information to be identified. In this process, the first named entity recognition model can also identify the part of speech of each address participle and determine the part of speech category corresponding to each address participle.

[0051] The part-of-speech categories may include nouns, verbs, adverbs, adjectives, locatives, quantifiers, auxiliary words, conjunctions, etc. In addition, in actual applications, the address information may also include some special characters, so in the embodiment of the present invention, the part-of-speech categories may include special characters in addition to the categories in the conventional grammatical classification.

[0052] In one possible implementation, an embodiment of the present invention may pre-determine different identifiers for different part-of-speech categories. After obtaining each address segmentation sequence, for each address segmentation in the current address segmentation sequence, the current address segmentation sequence may be represented according to the identifiers corresponding to the part-of-speech categories to which each address segmentation in the current address segmentation sequence belongs, thereby obtaining a segmentation part-of-speech sequence corresponding to the current address segmentation sequence.

[0053] For example, if the identifier of the noun is "1", the identifier of the adverb is "2", the identifier of the verb is "3", and the identifier of the auxiliary word is "4", taking the current address segmentation sequence as "XX city; XX district; XX street; XX community; XX building" as an example, the segmentation part-of-speech sequence corresponding to the current address segmentation sequence is "11111". Taking the current address segmentation sequence as "XX city; XX district; XX street; XX community; of; south side" as an example, the segmentation part-of-speech sequence corresponding to the current address segmentation sequence is "111141". Taking the current address segmentation sequence as "XX city; XX district; XX street; not; needed; distribution" as an example, the segmentation part-of-speech sequence corresponding to the current address segmentation sequence is "111233".

[0054] S130: Determine an abnormal sequence in each word segmentation part-of-speech sequence based on the first unsupervised anomaly detection model, and determine the to-be-identified address information corresponding to the abnormal sequence as abnormal address information.

[0055] Among them, the first unsupervised anomaly detection model can be a statistical model based on an unsupervised anomaly detection algorithm. In the embodiment of the present invention, through the unsupervised learning idea of ​​the statistical model, the majority of the segmentation part-of-speech sequences in each segmentation part-of-speech sequence can be determined as a normal sequence, and then the sequence in each segmentation part-of-speech sequence that has a large difference in distribution from the majority of the segmentation part-of-speech sequences is determined as an outlier segmentation part-of-speech sequence, and these outlier segmentation part-of-speech sequences are also abnormal sequences in the embodiment of the present invention.

[0056] Taking the identifier of noun as "1", the identifier of adverb as "2", the identifier of verb as "3", and the identifier of auxiliary word as "4" as an example, normal address information is generally composed of multiple nouns, and the segmentation part-of-speech sequence corresponding to the normal address information is generally composed of multiple "1", for example, it can be "11111". The segmentation part-of-speech sequence corresponding to the abnormal address information may have permutations and combinations of "1", "2", "3", and "4", for example, it can be "111233", "111141". It can be seen that there is a large difference in the distribution of the segmentation part-of-speech sequence corresponding to the normal address information and the segmentation part-of-speech sequence corresponding to the abnormal address information. Moreover, in actual applications, abnormal address information generally only occupies a small part of each address information to be identified. Therefore, the embodiment of the present invention can identify abnormal address information that accounts for a very small proportion from each address information to be identified based on the idea of ​​abnormality detection. The unsupervised anomaly detection algorithm can well identify outlier sequences in multiple sequences based on the distribution of the sequences. Therefore, the embodiment of the present invention can identify abnormal sequences in each participle part-of-speech sequence based on the first unsupervised anomaly detection model, and then determine the address information to be identified corresponding to the abnormal sequence as abnormal address information.

[0057] In the method for identifying abnormal address information provided by an embodiment of the present invention, based on the characteristic that there is a large difference between the distribution of the part-of-speech categories of each address segmentation in the address segmentation sequence corresponding to the normal address information and the distribution of the part-of-speech categories of each address segmentation in the address segmentation sequence corresponding to the abnormal address information, when it is necessary to identify the abnormal address information of each address information to be identified, the address segmentation sequence corresponding to each address information to be identified can be obtained first, and then the part-of-speech sequence corresponding to each address segmentation sequence can be determined based on the part-of-speech category corresponding to each address segmentation in each address segmentation sequence. Afterwards, in order to identify the abnormal sequence in each part-of-speech sequence that has a large difference in distribution from the normal part-of-speech sequence, the embodiment of the present invention, based on the idea of ​​an unsupervised learning algorithm, determines the majority of the part-of-speech sequences in each part-of-speech sequence as a normal sequence, and then identifies the outlier part-of-speech sequence in each part-of-speech sequence that has a large difference in distribution from the majority part-of-speech sequence through the first unsupervised anomaly detection model, and then the identified outlier part-of-speech sequence can be determined as an abnormal sequence, so that the address information to be identified corresponding to the abnormal sequence can be determined as abnormal address information. It can be seen that the method for identifying abnormal address information provided by the embodiment of the present invention can identify the abnormal address information in each address information to be identified based on the difference in the part-of-speech category of the address segmentation in the normal address information and the abnormal address information. Since the recognition process is implemented based on the idea of ​​an unsupervised learning algorithm, the embodiment of the present invention does not need to spend a lot of time to mark the sample address, which can improve the recognition efficiency; and since there is no need to mark the sample address, the accuracy of the recognition result is no longer dependent on the sample marking quality, so the stability of the method for identifying abnormal address information provided by the embodiment of the present invention is relatively high. In addition, the embodiments of the present invention can identify various abnormal address information with different part-of-speech distributions from normal address information, can adapt to the dynamic changes of abnormal addresses, and meet the recognition requirements of constantly adjusting and changing abnormal addresses, thereby improving the recognition accuracy.

[0058] Reference Figure 2 , is a flow chart of another method for identifying abnormal address information provided by an embodiment of the present invention. The method in this embodiment can be combined with various optional solutions in the method for identifying abnormal address information provided by the previous embodiment, and the method for identifying abnormal address information provided by the previous embodiment is further optimized. Figure 2 As shown, the specific steps include:

[0059] S210: Obtain address word segmentation sequences corresponding to each piece of address information to be identified.

[0060] S220: For each address segmentation sequence, based on the part-of-speech categories corresponding to each address segmentation in the current address segmentation sequence, determine a part-of-speech sequence corresponding to the current address segmentation sequence.

[0061] S230: Based on the first unsupervised anomaly detection model, determine the part-of-speech sequence state probabilities corresponding to each part-of-speech sequence of each word segmentation.

[0062] The part-of-speech sequence state probability corresponding to the current part-of-speech sequence is used to represent the probability of the current part-of-speech sequence appearing in each part-of-speech sequence.

[0063] In a possible implementation, the embodiment of the present invention can determine the distribution probability of the part-of-speech sequence of different distribution conditions (that is, the part-of-speech sequence state probability in the embodiment of the present invention) according to the distribution of each part-of-speech category included in each part-of-speech sequence through the first unsupervised anomaly detection model. Exemplarily, in the aforementioned example, the part-of-speech sequence state probability of the part-of-speech sequence with a distribution condition of "11111" can be 15%, the part-of-speech sequence state probability of the part-of-speech sequence with a distribution condition of "111233" can be 0.001%, and the part-of-speech sequence state probability of the part-of-speech sequence with a distribution condition of "111141" can be 0.002%. It should be noted that the sum of the distribution probabilities of the part-of-speech sequences of various distribution conditions should be 1.

[0064] Optionally, the first unsupervised anomaly detection model in the embodiment of the present invention may be a Hidden Markov Model (HMM), such as Figure 3 As shown, Figure 2 The S230 in can be replaced by S2301 and S2302:

[0065] S2301. Process each part-of-speech sequence based on a first unsupervised anomaly detection model to obtain a first initial probability distribution of each candidate part-of-speech category included in each part-of-speech sequence and a first transition probability distribution between each candidate part-of-speech category.

[0066] The candidate part-of-speech categories may be all the part-of-speech categories that appear in the part-of-speech sequences of the participles, such as nouns, adverbs, prepositions, adjectives, auxiliary words, special characters, and the like.

[0067] Among them, the first initial probability distribution of each candidate part-of-speech category is used to characterize the first initial probability of each candidate part-of-speech category; for each candidate part-of-speech category, the first initial probability of the current candidate part-of-speech category is used to characterize the probability that the current candidate part-of-speech category is the first part-of-speech category in each participle part-of-speech sequence. The first transition probability distribution between each candidate part-of-speech category is used to characterize the first transition probability between each candidate part-of-speech category; for each candidate part-of-speech category, the first transition probability between the first candidate part-of-speech category and the second candidate part-of-speech category is used to characterize the probability that the second candidate part-of-speech category is immediately followed by the first candidate part-of-speech category in each participle part-of-speech sequence. The first candidate part-of-speech category can be the same as or different from the second candidate part-of-speech category.

[0068] For address information, there are also great differences between the grammatical structure of normal address information and the grammatical structure of abnormal address information. For example, the grammatical structure of "no delivery required" followed by two verbs will not appear in normal address information. If the sequence segment of the part-of-speech sequence corresponding to "no delivery required" is "233", then the probability of the sequence segment "233" appearing in each part-of-speech sequence is very low. For another example, the first part-of-speech category of the part-of-speech sequence corresponding to normal address information is generally a noun, and the probability that the first part-of-speech category is other than a noun is very low. Based on this, when performing abnormality detection analysis on each part-of-speech sequence, the embodiment of the present invention can determine the first transition probability distribution between each candidate part-of-speech category according to the probability of each two part-of-speech categories appearing consecutively in each part-of-speech sequence, and determine the first initial probability distribution of each candidate part-of-speech category according to the probability of each candidate part-of-speech category appearing in the first part-of-speech category in each part-of-speech sequence, and then identify the abnormal sequence in each part-of-speech sequence according to the first transition probability distribution and the first initial probability distribution.

[0069] S2302: Based on the first initial probability distribution and the first transition probability distribution, determine the part-of-speech sequence state probabilities corresponding to each participle part-of-speech sequence.

[0070] HMM can be used to describe a Markov process with implicit unknown parameters. The implicit unknown parameters in the process can be determined from the observable sequence. The model is suitable for processing sequence data and supports unsupervised learning and training. Among them, the Markov chain of HMM is a random process of transitioning from one state to another in the state space. In the time series, the probability distribution of the next state can only be determined by the current state and has nothing to do with the previous state. Therefore, in the embodiment of the present invention, HMM can be selected as the first unsupervised anomaly detection model.

[0071] Exemplarily, in an embodiment of the present invention, each participle part-of-speech sequence can be used as an observable sequence of HMM, for example, the observable sequence can be "11111". The hidden state of HMM can include two states, an abnormal state and a normal state. If "1" is used to represent the abnormal state and "0" is used to represent the normal state, then each part-of-speech category in each participle part-of-speech sequence corresponds to an implicit state, and each implicit state of each participle part-of-speech sequence can constitute an implicit state sequence of the participle part-of-speech sequence, for example, it can be "00000", and the length of the implicit state sequence is the same as the length of the corresponding participle part-of-speech sequence.

[0072] For HMM, the parameters that need to be calculated include the initial probability of the observed state (i.e., the first initial probability distribution in the embodiment of the present invention), the transition probability between different observed states (i.e., the first transition probability distribution in the embodiment of the present invention), and the occurrence probability of the observable sequence (i.e., the part-of-speech sequence state probabilities corresponding to the part-of-speech sequences of the word segmentation in the embodiment of the present invention).

[0073] Exemplarily, if each candidate part-of-speech category in each participle part-of-speech sequence includes nouns, adverbs, auxiliary words, and special characters, the proportion of participle part-of-speech sequences starting with nouns is 99%, the proportion of participle part-of-speech sequences starting with adverbs is 0.4%, the proportion of participle part-of-speech sequences starting with auxiliary words is 0.3%, and the proportion of participle part-of-speech sequences starting with special characters is 0.3%, then 99%, 0.4%, 0.3%, and 0.3% can be determined as the first initial probabilities corresponding to nouns, adverbs, auxiliary words, and special characters, respectively. In addition, in the case where each candidate part-of-speech category includes four part-of-speech categories, there are 16 combinations between the candidate part-of-speech categories, and the transition probabilities between these 16 combinations need to be calculated. Taking nouns as an example, the probabilities of nouns transferring to nouns, nouns transferring to adverbs, nouns transferring to auxiliary words, and nouns transferring to special characters need to be calculated respectively, which can be determined according to the order and frequency of occurrence of each part-of-speech category in each participle part-of-speech sequence. For example, in each participle word sequence, the proportions of the previous word being a noun and the next word being a noun, adverb, auxiliary word, and special character are 70%, 10%, 10%, and 10% respectively. Then 70%, 10%, 10%, 10%, and 10% can be determined as the first transition probabilities of noun to noun, noun to adverb, noun to auxiliary word, and noun to special character respectively.

[0074] After determining the first initial probability distribution and the first transition probability distribution, for each segmented part-of-speech sequence, the probability of the segmented part-of-speech sequence appearing in each word part-of-speech sequence can be calculated according to the distribution of each part-of-speech category in the segmented part-of-speech sequence.

[0075] Optionally, determining the part-of-speech sequence state probability corresponding to each participle part-of-speech sequence based on the first initial probability distribution and the first transition probability distribution may include: for each participle part-of-speech sequence, determining the initial probability corresponding to the first part-of-speech category in the current participle part-of-speech sequence based on the first initial probability distribution, and determining the transition probabilities between adjacent part-of-speech categories in the current participle part-of-speech sequence based on the first transition probability distribution; determining the part-of-speech sequence state probability corresponding to the current participle part-of-speech sequence according to the initial probability corresponding to the current participle part-of-speech sequence and the transition probabilities corresponding to the current participle part-of-speech sequence.

[0076] In a possible implementation, the embodiment of the present invention can calculate the state probability of the part-of-speech sequence based on probability multiplication. Exemplarily, taking the current part-of-speech sequence of "111233" as an example, if the first initial probability corresponding to the noun is a, the first transition probability of the noun transferring to the noun is b, the first transition probability of the noun transferring to the adverb is c, the first transition probability of the adverb transferring to the verb is d, and the first transition probability of the adverb transferring to the adverb is e, then the state probability of the part-of-speech sequence corresponding to the current part-of-speech sequence = a*b*b*c*d*e.

[0077] In the embodiment of the present invention, not only the difference in the distribution of each part-of-speech category in the abnormal address information and the normal address information is taken into account, but also the difference in the grammatical structure of each part-of-speech category in the abnormal address information and the normal address information (i.e., the grammatical connection relationship between the part-of-speech categories) is taken into account. Based on this, the embodiment of the present invention combines HMM with named entity recognition technology, and determines the first initial probability distribution of each candidate part-of-speech category and the first transition probability distribution between each candidate part-of-speech category according to the frequency of occurrence of each candidate part-of-speech category in each participle part-of-speech sequence and the front-to-back order between the candidate part-of-speech categories, thereby determining the probability that participle part-of-speech sequences with different distributions may appear in each participle part-of-speech sequence based on the first initial probability distribution and the first transition probability distribution. In this way, the recognition accuracy of abnormal sequences can be improved while reducing the sample annotation amount of address information, thereby improving the recognition accuracy of abnormal address information.

[0078] S240 , determining abnormal sequences from each part-of-speech sequence according to the part-of-speech sequence state probabilities corresponding to each part-of-speech sequence, and determining the to-be-identified address information corresponding to the abnormal sequence as abnormal address information.

[0079] After determining the part-of-speech sequence state probabilities corresponding to each part-of-speech sequence, the embodiment of the present invention can determine abnormal sequences from each part-of-speech sequence based on a predetermined screening rule.

[0080] In one possible implementation, the segmentation part-of-speech sequences with the N distributions whose corresponding part-of-speech sequence state probabilities are the smallest in each segmentation part-of-speech sequence can be determined as abnormal sequences. Wherein, N can be a predetermined integer. In another possible implementation, the segmentation part-of-speech sequence whose corresponding part-of-speech sequence state probability is less than a predetermined part-of-speech sequence state probability can be determined as an abnormal sequence. Exemplarily, in an embodiment of the present invention, sequences with distributions of "111233" and "111141" in each segmentation part-of-speech sequence can be determined as abnormal sequences.

[0081] Optional, such as Figure 4 As shown, Figure 2 The S240 in can be replaced by S2401-S2403:

[0082] S2401. Obtain the segmentation word meaning sequences corresponding to each segmentation part-of-speech sequence.

[0083] The word segmentation meaning sequence corresponding to the word segmentation part sequence may be a sequence obtained by further performing word meaning recognition on each address segmentation in the address segmentation sequence corresponding to the word segmentation part sequence. For example, for address segmentations whose word segmentation part category is noun, a finer granularity division may be performed, such as being divided into names of people, places, animals, institutions, and foods. For another example, for address segmentations whose word segmentation part category is special characters, they may be divided into punctuation characters such as commas, exclamation marks, and colons, operation characters such as multiplication signs, plus signs, and minus signs, and various other types of computer language characters.

[0084] For example, taking the address segmentation sequence of "XX city; XX district; XX street; XX community; XX building" as an example, the segmentation part-of-speech sequence corresponding to the address segmentation sequence is "11111". If "A" is used to represent the place name, the segmentation meaning sequence corresponding to the segmentation part-of-speech sequence can be "AAAAA".

[0085] In a possible implementation, the embodiment of the present invention can perform semantic recognition on each address segmentation in the address segmentation sequence corresponding to each segmentation part-of-speech sequence through a second named entity recognition model to obtain a segmentation meaning sequence corresponding to each segmentation part-of-speech sequence. Among them, the second named entity recognition model can be a model for recognizing the semantics of words / characters obtained by pre-training based on a named entity recognition algorithm. The first named entity recognition model can be obtained by training a large amount of address text information. The specific process of obtaining the second named entity recognition model can refer to the relevant description of the prior art, and the embodiment of the present invention will not be repeated here.

[0086] S2402: Based on the second unsupervised anomaly detection model, determine the word sense sequence state probabilities corresponding to each word segmentation word sense sequence.

[0087] The word sense sequence state probability corresponding to the current word segmentation word sense sequence is used to represent the probability of the current word segmentation word sense sequence appearing in each word segmentation word sense sequence.

[0088] The second unsupervised anomaly detection model can be a statistical model based on an unsupervised anomaly detection algorithm. In an embodiment of the present invention, through the unsupervised learning idea of ​​the statistical model, the majority of the segmentation word sense sequences in each segmentation word sense sequence can be determined as a normal sequence, and then the sequence in each segmentation word sense sequence that has a large difference in distribution from the majority of the segmentation word sense sequences is determined as an outlier segmentation word sense sequence, and then the obtained outlier segmentation word sense sequence can be combined with the outlier segmentation part-of-speech sequence previously obtained by the first unsupervised anomaly detection model to determine the abnormal sequence.

[0089] In a possible implementation, the embodiment of the present invention can determine the distribution probability of the segmentation word sense sequence with different distribution conditions (that is, the word sense sequence state probability in the embodiment of the present invention) through the second unsupervised anomaly detection model according to the distribution of each word sense category included in each segmentation word sense sequence. Exemplarily, in the above example, the word sense sequence state probability of the segmentation word sense sequence with the distribution condition of "AAAAA" can be 16%. It should be noted that the sum of the distribution probabilities of the segmentation word sense sequences with various distribution conditions should be 1.

[0090] S2403. According to the part-of-speech sequence state probabilities corresponding to each part-of-speech sequence and the word sense sequence state probabilities corresponding to each part-of-speech sequence, an abnormal sequence is determined from each part-of-speech sequence, and the to-be-identified address information corresponding to the abnormal sequence is determined as the abnormal address information.

[0091] Optionally, the second unsupervised anomaly detection model in the embodiment of the present invention may be an HMM. Determining the word sense sequence state probabilities corresponding to each word segmentation word sense sequence based on the second unsupervised anomaly detection model may include: processing each word segmentation word sense sequence based on the second unsupervised anomaly detection model to obtain a second initial probability distribution of each candidate word sense category contained in each word segmentation word sense sequence, and a second transition probability distribution between each candidate word sense category; determining the word sense sequence state probabilities corresponding to each word segmentation word sense sequence based on the second initial probability distribution and the second transition probability distribution.

[0092] The candidate word sense categories may be all word sense categories appearing in each word segmentation word sense sequence, such as names of people, places, animals, institutions, foods, commas, exclamation marks, colons, multiplication signs, plus signs, minus signs, etc.

[0093] Among them, the second initial probability distribution of each candidate semantic category is used to characterize the second initial probability of each candidate semantic category; for each candidate semantic category, the second initial probability of the current candidate semantic category is used to characterize the probability that the current candidate semantic category is the first semantic category in each segmentation semantic sequence. The second transition probability distribution between each candidate semantic category is used to characterize the second transition probability between each candidate semantic category; for each candidate semantic category, the second transition probability between the second candidate semantic category and the second candidate semantic category is used to characterize the probability that the second candidate semantic category is followed by the second candidate semantic category in each segmentation semantic sequence. The first candidate semantic category may be the same as or different from the second candidate semantic category.

[0094] Optionally, each segmentation word sense sequence is processed based on a second unsupervised anomaly detection model to obtain a second initial probability distribution of each candidate sense category contained in each segmentation word sense sequence, and a second transition probability distribution between each candidate sense category, which may include: for each segmentation word sense sequence, determining the initial probability corresponding to the first sense category in the current segmentation word sense sequence based on the second initial probability distribution, and determining the transition probabilities between each adjacent sense category in the current segmentation word sense sequence based on the second transition probability distribution; determining the sense sequence state probability corresponding to the current segmentation word sense sequence according to the initial probability corresponding to the current segmentation word sense sequence and the transition probabilities corresponding to the current segmentation word sense sequence.

[0095] The specific process of calculating the second initial probability distribution, the second transition probability distribution through HMM, and calculating the word sense sequence state probability corresponding to each word segmentation word sense sequence according to the second initial probability distribution and the second transition probability distribution can refer to the related description of calculating the first initial probability distribution, the first transition probability distribution, and the part-of-speech sequence state probability in the aforementioned embodiment, which will not be repeated here.

[0096] Optionally, an embodiment of the present invention may determine an abnormal sequence from each participle part-of-speech sequence in the following manner: for each participle part-of-speech sequence, based on the part-of-speech sequence state probability corresponding to the current participle part-of-speech sequence and the word sense sequence state probability corresponding to the current participle word sense sequence corresponding to the current participle part-of-speech sequence, determine the sequence abnormality probability corresponding to the current participle part-of-speech sequence; and determine the part-of-speech sequence whose corresponding sequence abnormality probability meets a preset condition in each participle part-of-speech sequence as an abnormal sequence.

[0097] In a possible implementation, the sequence anomaly probability corresponding to the current part-of-speech sequence can be obtained by performing a mathematical operation on the part-of-speech sequence state probability and the word sense sequence state probability corresponding to the current part-of-speech sequence. For example, the mathematical operation performed on the part-of-speech sequence state probability and the word sense sequence state probability can be an operation such as taking the mean, taking the maximum value, taking the minimum value, taking the weighted average value, etc.

[0098] The preset condition may be a predetermined condition. For example, the preset condition may be that the probability of sequence abnormality is less than a predetermined probability value. Alternatively, each word segmentation part-of-speech sequence may be sorted in order of sequence abnormality probability from small to large, and the sequence of the M distributions ranked at the top is the word segmentation part-of-speech sequence that meets the preset condition. M is a predetermined positive integer.

[0099] In the embodiment of the present invention, not only the difference in the distribution of the part-of-speech categories in the abnormal address information and the normal address information, the difference in the grammatical structure of each part-of-speech category in the abnormal address information and the normal address information are taken into account, but also the difference in the semantics based on the grammatical structure between the abnormal address information and the normal address information is taken into account. It can be seen that the embodiment of the present invention recognizes the abnormal address information from multiple dimensions, combines multiple aspects such as text, grammar, and semantics, broadens the recognition scope, and can ensure full coverage recognition of different types of abnormal address information.

[0100] In the embodiment of the present invention, on the basis of the above-mentioned embodiment, in order to more accurately identify abnormal sequences that are different from the distribution of most of the segmentation part-of-speech sequences from each segmentation part-of-speech sequence, the distribution probability of the segmentation part-of-speech sequences with different distribution conditions (i.e., the part-of-speech sequence state probability) can be determined by the first unsupervised anomaly detection model. Through the part-of-speech sequence state probability, the proportion of segmentation part-of-speech sequences with different distribution conditions in each segmentation part-of-speech sequence can be accurately evaluated, so that the abnormal sequences in each segmentation part-of-speech sequence can be accurately identified, and then the recognition accuracy of abnormal address information can be improved.

[0101] It should be noted that the method for identifying abnormal address information proposed in the embodiment of the present invention and the aforementioned embodiment belong to the same inventive concept. The technical details not fully described in this embodiment can be referred to the aforementioned embodiment, and the beneficial effects of the aforementioned embodiment are also applicable to this embodiment.

[0102] Figure 5 A simplified flow chart of a method for identifying abnormal address information provided by an embodiment of the present invention. Figure 5As shown, the method for identifying abnormal address information provided by the embodiment of the present invention mainly includes four parts. The first part is the data extraction part, which can count the address information in the business orders within the current statistical period to obtain each address information to be processed. The second part is the data cleaning part, which can screen each address information to be processed to obtain each address information to be identified. The third part is the named entity recognition part, which can perform word segmentation processing on each address information to be identified to obtain each address word segmentation sequence; and can perform part-of-speech tagging on each address word segmentation sequence to obtain each word segmentation word sequence, and can also perform word meaning tagging on each address word segmentation sequence to obtain each word segmentation word meaning sequence. The fourth part is the unsupervised anomaly detection part, which can perform word-part anomaly detection processing on each word segmentation word sequence to obtain the word-part sequence state probability corresponding to each word segmentation word sequence; and can perform word meaning anomaly detection processing on each word segmentation word meaning sequence to obtain the word meaning sequence state probability corresponding to each word segmentation word meaning sequence; then, the sequence anomaly probability corresponding to each address information to be processed can be output, so as to identify the abnormal address information according to the sequence anomaly probability.

[0103] The method for identifying abnormal address information provided by the embodiment of the present invention uses a combination of named entity recognition technology and unsupervised anomaly detection technology to identify abnormal address information. Compared with the recognition scheme in the prior art, the amount of calculation is very small. It can not only be applied in an offline environment, but also can be flexibly and effectively deployed and applied in a real-time risk control environment. In addition, the overall process of the method for identifying abnormal address information provided by the embodiment of the present invention is less affected by human factors and has better versatility. In addition, the embodiment of the present invention performs abnormal address recognition from multiple aspects such as text, grammar, and semantics, which can ensure full coverage recognition of different types of abnormal address information and has good stability and robustness.

[0104] The method for identifying abnormal address information provided in the embodiment of the present invention belongs to the same inventive concept as the method for identifying abnormal address information in the aforementioned embodiments. For details not described in detail in the embodiment of the method for identifying abnormal address information, reference can be made to the embodiment of the method for identifying abnormal address information.

[0105] Figure 6 The present invention provides a schematic diagram of the structure of an apparatus for identifying abnormal address information according to an embodiment of the present invention. The apparatus includes: an acquisition module 610 and a determination module 620.

[0106] Exemplarily, the acquisition module 610 may execute S110 in the above method embodiment, and the determination module 620 may execute S120 and S130 in the above method embodiment.

[0107] Specifically, the acquisition module 610 is used to obtain the address segmentation sequences corresponding to each address information to be identified; the determination module 620 is used to determine the segmentation part-of-speech sequence corresponding to the current address segmentation sequence for each address segmentation sequence based on the part-of-speech category corresponding to each address segmentation in the current address segmentation sequence; the determination module 620 is used to determine the abnormal sequence in each segmentation part-of-speech sequence based on the first unsupervised anomaly detection model, and determine the address information to be identified corresponding to the abnormal sequence as the abnormal address information.

[0108] Optionally, in a possible implementation manner, the determination module 620 is specifically configured to:

[0109] Based on the first unsupervised anomaly detection model, the part-of-speech sequence state probabilities corresponding to each participle part-of-speech sequence are determined; wherein the part-of-speech sequence state probability corresponding to the current participle part-of-speech sequence is used to characterize the probability of the current participle part-of-speech sequence appearing in each participle part-of-speech sequence; according to the part-of-speech sequence state probabilities corresponding to each participle part-of-speech sequence, an abnormal sequence is determined from each participle part-of-speech sequence.

[0110] Optionally, in another possible implementation, the first unsupervised anomaly detection model is a hidden Markov model, and the determination module 620 is further configured to:

[0111] Based on the first unsupervised anomaly detection model, each participle part-of-speech sequence is processed to obtain a first initial probability distribution of each candidate part-of-speech category contained in each participle part-of-speech sequence, and a first transition probability distribution between each candidate part-of-speech category; based on the first initial probability distribution and the first transition probability distribution, the part-of-speech sequence state probability corresponding to each participle part-of-speech sequence is determined.

[0112] Optionally, in another possible implementation, the determining module 620 is further configured to:

[0113] For each part-of-speech sequence, the initial probability corresponding to the first part-of-speech category in the current part-of-speech sequence is determined based on the first initial probability distribution, and the transition probabilities between adjacent part-of-speech categories in the current part-of-speech sequence are determined based on the first transition probability distribution; according to the initial probability corresponding to the current part-of-speech sequence and the transition probabilities corresponding to the current part-of-speech sequence, the part-of-speech sequence state probability corresponding to the current part-of-speech sequence is determined.

[0114] Optionally, in another possible implementation, the determining module 620 is further configured to:

[0115] Obtain the segmentation word sense sequences corresponding to each segmentation word sense sequence; determine the word sense sequence state probabilities corresponding to each segmentation word sense sequence based on the second unsupervised anomaly detection model; wherein the word sense sequence state probability corresponding to the current segmentation word sense sequence is used to characterize the probability of the current segmentation word sense sequence appearing in each segmentation word sense sequence; determine the abnormal sequence from each segmentation word sense sequence according to the word sense sequence state probabilities corresponding to each segmentation word sense sequence and the word sense sequence state probabilities corresponding to each segmentation word sense sequence corresponding to each segmentation word sense sequence.

[0116] Optionally, in another possible implementation, the second unsupervised anomaly detection model is a hidden Markov model, and the determination module 620 is further configured to:

[0117] Based on the second unsupervised anomaly detection model, each participle word sense sequence is processed to obtain a second initial probability distribution of each candidate word sense category contained in each participle word sense sequence, and a second transition probability distribution between each candidate word sense category; based on the second initial probability distribution and the second transition probability distribution, the word sense sequence state probability corresponding to each participle word sense sequence is determined.

[0118] Optionally, in another possible implementation, the determining module 620 is further configured to:

[0119] For each part-of-speech sequence, the sequence abnormality probability corresponding to the current part-of-speech sequence is determined based on the part-of-speech sequence state probability corresponding to the current part-of-speech sequence and the word sense sequence state probability corresponding to the current part-of-speech sequence; the part-of-speech sequence whose corresponding sequence abnormality probability meets the preset conditions in each part-of-speech sequence is determined as an abnormal sequence.

[0120] Optionally, in another possible implementation, the acquisition module 610 is specifically configured to:

[0121] For each address information to be identified, the current address information to be identified is segmented based on the first named entity recognition model to obtain a current address segmentation sequence corresponding to the current address information to be identified, and obtain the part of speech category corresponding to each address segmentation in the current address segmentation sequence.

[0122] Optionally, in another possible implementation manner, the apparatus for identifying abnormal address information provided by the embodiment of the present invention may further include a processing module;

[0123] The acquisition module 610 is also used to acquire each address information to be processed before acquiring the address segmentation sequence corresponding to each address information to be identified; the processing module is used to perform data cleaning processing on each address information to be processed based on preset data cleaning rules to obtain each address information to be identified.

[0124] The device for identifying abnormal address information provided in an embodiment of the present invention belongs to the same inventive concept as the method for identifying abnormal address information provided in the aforementioned embodiments. For details not fully described in the embodiment of the device for identifying abnormal address information, reference can be made to the relevant contents of the aforementioned method embodiments, and the corresponding beneficial effects can also be referred to the beneficial effect analysis of the aforementioned method embodiments.

[0125] Figure 7 A schematic diagram of the structure of a device for identifying abnormal address information provided by an embodiment of the present invention. Figure 7 A block diagram of an exemplary abnormal address information identification device 12 suitable for implementing an embodiment of the present invention is shown. Figure 7 The abnormal address information identification device 12 shown is only an example and should not bring any limitation to the function and scope of use of the embodiment of the present invention.

[0126] like Figure 7 As shown, the abnormal address information identification device 12 is in the form of a general-purpose computing device. The components of the abnormal address information identification device 12 may include, but are not limited to: one or more processors or processing units 16, a system memory 28, and a bus 18 connecting different system components (including the system memory 28 and the processing unit 16).

[0127] Bus 18 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processor or a local bus using any of a variety of bus architectures. By way of example, these architectures include, but are not limited to, an Industry Standard Architecture (ISA) bus, a Micro Channel Architecture (MAC) bus, an Enhanced ISA bus, a Video Electronics Standards Association (VESA) local bus, and a Peripheral Component Interconnect (PCI) bus.

[0128] The abnormal address information identification device 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the abnormal address information identification device 12, including volatile and non-volatile media, removable and non-removable media.

[0129] System memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache memory (i.e., Figure 7 The abnormal address information identification device 12 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, the storage system 34 may be used to read and write non-removable, non-volatile magnetic media ( Figure 7 not shown, usually called a "hard drive"). Although Figure 7 Not shown in the figure, a disk drive for reading and writing to a removable non-volatile disk (e.g., a "floppy disk"), and an optical disk drive for reading and writing to a removable non-volatile optical disk (e.g., a CD-ROM, a DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to the bus 18 via one or more data medium interfaces. The system memory 28 may include at least one program product having a set (e.g., at least one) of program modules that are configured to perform the functions of the various embodiments of the present invention.

[0130] A program / utility 40 having a set (at least one) of program modules 42 may be stored, for example, in system memory 28, such program modules 42 including, but not limited to, an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment. Program modules 42 generally perform the functions and / or methods of the embodiments described herein.

[0131] The abnormal address information identification device 12 may also communicate with one or more external devices 14 (e.g., keyboard, pointing device, display 24, etc.), and may also communicate with one or more devices that enable a user to interact with the abnormal address information identification device 12, and / or communicate with any device that enables the abnormal address information identification device 12 to communicate with one or more other computing devices (e.g., network card, modem, etc.). Such communication may be performed through an input / output (I / O) interface 22. Furthermore, the abnormal address information identification device 12 may also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN) and / or public network, such as the Internet) through a network adapter 20. Figure 7 As shown, the network adapter 20 communicates with other modules of the abnormal address information identification device 12 through the bus 18. It should be understood that although Figure 7 Not shown, other hardware and / or software modules may be used in conjunction with the abnormal address information identification device 12, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0132] The processing unit 16 executes various functional applications and data processing by running the program stored in the system memory 28, for example, implementing the steps of the method for identifying abnormal address information provided in an embodiment of the present invention, the method comprising: obtaining address segmentation sequences corresponding to each address information to be identified; for each address segmentation sequence, based on the part-of-speech category corresponding to each address segmentation in the current address segmentation sequence, determining the segmentation part-of-speech sequence corresponding to the current address segmentation sequence; determining the abnormal sequence in each segmentation part-of-speech sequence based on the first unsupervised anomaly detection model, and determining the address information to be identified corresponding to the abnormal sequence as the abnormal address information.

[0133] Of course, those skilled in the art can understand that the processor can also implement the technical solution of the method for identifying abnormal address information provided by any embodiment of the present invention.

[0134] The present embodiment provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the steps of a method for identifying abnormal address information provided in the aforementioned embodiment of the present invention are implemented. The method includes: obtaining address segmentation sequences corresponding to each address information to be identified; for each address segmentation sequence, based on the part-of-speech category corresponding to each address segmentation in the current address segmentation sequence, determining a segmentation part-of-speech sequence corresponding to the current address segmentation sequence; determining an abnormal sequence in each segmentation part-of-speech sequence based on a first unsupervised anomaly detection model, and determining the address information to be identified corresponding to the abnormal sequence as abnormal address information.

[0135] The computer storage medium of the embodiment of the present invention can adopt any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to: an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, a computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.

[0136] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0137] The program code embodied on the computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0138] Computer program code for performing the operation of the present invention may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or an identification device for abnormal address information. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect through the Internet).

[0139] It should be understood by those skilled in the art that the modules or steps of the present invention described above can be implemented by a general-purpose computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, optionally, they can be implemented by a program code executable by a computer device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0140] Note that the above are only preferred embodiments of the present invention and the technical principles used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. A method for identifying abnormal address information, characterized in that: The method comprises: Obtaining the address segmentation sequences corresponding to each address information to be identified; For each of the address segmentation sequences, based on the part-of-speech categories corresponding to each of the address segmentations in the current address segmentation sequence, determine a part-of-speech sequence corresponding to the current address segmentation sequence; Based on the first unsupervised anomaly detection model, an abnormal sequence in each of the segmented part-of-speech sequences is determined, and the to-be-identified address information corresponding to the abnormal sequence is determined as the abnormal address information.

2. The method for identifying abnormal address information according to claim 1, characterized in that: The determining of abnormal sequences in each of the word segmentation part-of-speech sequences based on the first unsupervised anomaly detection model includes: Based on the first unsupervised anomaly detection model, determining the part-of-speech sequence state probabilities corresponding to each of the part-of-speech sequences; wherein the part-of-speech sequence state probability corresponding to the current part-of-speech sequence is used to characterize the probability of the current part-of-speech sequence appearing in each of the part-of-speech sequences; The abnormal sequence is determined from each of the segmentation part-of-speech sequences according to the part-of-speech sequence state probabilities corresponding to each of the segmentation part-of-speech sequences.

3. The method for identifying abnormal address information according to claim 2, characterized in that: The first unsupervised anomaly detection model is a hidden Markov model, and determining the part-of-speech sequence state probabilities corresponding to each of the part-of-speech sequences based on the first unsupervised anomaly detection model includes: Processing each of the segmented part-of-speech sequences based on the first unsupervised anomaly detection model to obtain a first initial probability distribution of each candidate part-of-speech category included in each of the segmented part-of-speech sequences, and a first transition probability distribution between the candidate part-of-speech categories; Based on the first initial probability distribution and the first transition probability distribution, the part-of-speech sequence state probabilities corresponding to each of the part-of-speech sequences are determined.

4. The method for identifying abnormal address information according to claim 3, characterized in that: The determining, based on the first initial probability distribution and the first transition probability distribution, the part-of-speech sequence state probabilities corresponding to each of the participle part-of-speech sequences respectively includes: For each of the segmented part-of-speech sequences, determining an initial probability corresponding to the first part-of-speech category in the current segmented part-of-speech sequence based on the first initial probability distribution, and determining transition probabilities between adjacent part-of-speech categories in the current segmented part-of-speech sequence based on the first transition probability distribution; The part-of-speech sequence state probability corresponding to the current part-of-speech sequence is determined according to the initial probability corresponding to the current part-of-speech sequence and the transition probabilities corresponding to the current part-of-speech sequence.

5. The method for identifying abnormal address information according to claim 2, characterized in that: The step of determining the abnormal sequence from each of the part-of-speech sequences according to the part-of-speech sequence state probabilities respectively corresponding to each of the part-of-speech sequences comprises: Obtaining the segmentation word meaning sequences corresponding to each of the segmentation part-of-speech sequences; Based on the second unsupervised anomaly detection model, determining the word sense sequence state probabilities corresponding to each of the word segmentation word sense sequences; wherein the word sense sequence state probability corresponding to the current word segmentation word sense sequence is used to characterize the probability of the current word segmentation word sense sequence appearing in each of the word segmentation word sense sequences; The abnormal sequence is determined from each of the part-of-speech sequences according to the part-of-speech sequence state probabilities corresponding to each of the part-of-speech sequences and the meaning sequence state probabilities corresponding to each of the part-of-speech sequences corresponding to each of the part-of-speech sequences.

6. The method for identifying abnormal address information according to claim 5, characterized in that: The second unsupervised anomaly detection model is a hidden Markov model, and determining the word sense sequence state probabilities corresponding to each of the word segmentation word sense sequences based on the second unsupervised anomaly detection model includes: Processing each of the segmentation word sense sequences based on the second unsupervised anomaly detection model to obtain a second initial probability distribution of each candidate word sense category included in each of the segmentation word sense sequences and a second transition probability distribution between the candidate word sense categories; Based on the second initial probability distribution and the second transition probability distribution, the word sense sequence state probabilities corresponding to each of the word segmentation word sense sequences are determined.

7. The method for identifying abnormal address information according to claim 5, characterized in that: The determining the abnormal sequence from each of the part-of-speech sequences according to the part-of-speech sequence state probabilities corresponding to each of the part-of-speech sequences and the sense sequence state probabilities corresponding to each of the part-of-speech sequences corresponding to each of the part-of-speech sequences comprises: For each of the part-of-speech sequences, determining a sequence abnormality probability corresponding to the current part-of-speech sequence based on a part-of-speech sequence state probability corresponding to the current part-of-speech sequence and a sense sequence state probability corresponding to the current part-of-speech sequence corresponding to the current part-of-speech sequence; Among the segmentation part-of-speech sequences, the segmentation part-of-speech sequences whose corresponding sequence abnormality probabilities meet preset conditions are determined as the abnormal sequences.

8. The method for identifying abnormal address information according to claim 1, characterized in that: The step of obtaining the address segmentation sequences corresponding to the address information to be identified includes: For each address information to be identified, segmentation processing is performed on the current address information to be identified based on the first named entity recognition model to obtain a current address segmentation sequence corresponding to the current address information to be identified, and the part of speech category corresponding to each address segmentation in the current address segmentation sequence is obtained.

9. The method for identifying abnormal address information according to any one of claims 1 to 8, characterized in that: Before acquiring the address segmentation sequences corresponding to the address information to be identified, the method further includes: Obtain information of each address to be processed; The address information to be processed is cleaned based on preset data cleaning rules to obtain the address information to be identified.

10. A device for identifying abnormal address information, characterized in that: include: An acquisition module is used to acquire the address segmentation sequence corresponding to each address information to be identified; A determination module, configured to determine, for each of the address segmentation sequences, a segmentation part-of-speech sequence corresponding to the current address segmentation sequence based on the part-of-speech categories corresponding to each of the address segmentations in the current address segmentation sequence; The determination module is used to determine the abnormal sequence in each of the segmented part-of-speech sequences based on the first unsupervised anomaly detection model, and determine the to-be-identified address information corresponding to the abnormal sequence as the abnormal address information.

11. A device for identifying abnormal address information, characterized in that: The abnormal address information identification device includes: one or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method for identifying abnormal address information as described in any one of claims 1-9.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method for identifying abnormal address information as described in any one of claims 1 to 9 is implemented.

Images