CN114049215A - Abnormal transaction identification method, device and application - Google Patents
Abnormal transaction identification method, device and application Download PDFInfo
- Publication number
- CN114049215A CN114049215A CN202210008148.4A CN202210008148A CN114049215A CN 114049215 A CN114049215 A CN 114049215A CN 202210008148 A CN202210008148 A CN 202210008148A CN 114049215 A CN114049215 A CN 114049215A
- Authority
- CN
- China
- Prior art keywords
- wind control
- control rule
- target
- transaction
- threshold
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
- G06Q40/04—Trading; Exchange, e.g. stocks, commodities, derivatives or currency exchange
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/35—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
Abstract
The application provides an abnormal transaction identification method, an abnormal transaction identification device and application, wherein the method comprises the steps of extracting key words from different types of wind control rules by adopting different classifiers, automatically acquiring input data under fields based on the key words and processing the input data, and accordingly obtaining classification results. According to the invention, abnormal transaction identification can be automatically carried out based on various wind control rules, and the existing manual discrimination mode is replaced, so that the accuracy and efficiency are higher.
Description
Technical Field
The application relates to the field of neural network risk assessment, in particular to an abnormal transaction identification method, an abnormal transaction identification device and application.
Background
Along with the rapid development of online financial business, the related transaction amount is rapidly increased, and meanwhile, abnormal transactions are more and more. The abnormal transaction not only causes great loss to the user, but also seriously destroys the financial order, hurts the credit of the financial institution and causes extremely bad influence on the society.
The conventional transaction wind control system for identifying abnormal transactions comprises three processes of input, processing and output: inputting, namely cleaning and standardizing input data; processing, namely calculating input data based on various wind control rules to obtain a calculation result; and outputting and presenting the calculation result to a system user. The accuracy of the calculation result depends on the selection of the wind control rule to a greater extent, so that the wind control rules are not only increasing in number, but also increasing in variety, for example, the wind control rules include forbidden rules or numerator/denominator-based numerical class rules. The different types of the wind control rules lead to the fact that abnormal transactions cannot be automatically identified at present, but the efficiency of manually identifying the abnormal transactions is extremely low due to the large number of the wind control rules.
Based on this, a solution is urgently sought at present for the problems of low accuracy and low efficiency caused by the fact that the conventional transaction wind control system cannot automatically identify abnormal transactions.
Disclosure of Invention
The embodiment of the application provides an abnormal transaction identification method, an abnormal transaction identification device and application, and can solve the problem that target transactions cannot be automatically identified and classified according to existing wind control rules. According to the invention, abnormal transaction identification can be automatically carried out based on various wind control rules, and the existing manual discrimination mode is replaced, so that the accuracy and efficiency are higher.
In a first aspect, an embodiment of the present application provides an abnormal transaction identification method, where the method includes: acquiring at least one wind control rule corresponding to a target transaction to be detected, wherein each wind control rule comprises a no-throw type wind control rule or a threshold value type wind control rule; after word segmentation is carried out on the no-entry type wind control rule, inputting the no-entry type wind control rule into a trained first classifier for recognition, and obtaining a field name output by the trained first classifier and no-entry words; acquiring first target data according to the field name, and matching the first target data with the forbidden word to obtain a first classification result of the target transaction; after segmenting words of the threshold value type wind control rule, inputting the words into a trained second classifier for recognition to obtain an original field name, a field processing word and a threshold value output by the trained second classifier; processing the original data under the original field name according to the field processing words to obtain second target data, and comparing the second target data with a threshold value to obtain a second classification result of the target transaction; a classification result for the target transaction is determined based on the first classification result and/or the second classification result.
In some embodiments, the second classification result further includes a third classification result, wherein obtaining the third classification result includes: and when the second classifier does not recognize the field processing words, directly taking the original data under the original field name as second target data to be compared with a threshold value, and obtaining a third classification result of the target transaction.
In some embodiments, the "acquiring at least one wind control rule corresponding to the target transaction to be detected" includes: and obtaining the transaction category to which the target transaction belongs, and determining at least one wind control rule corresponding to the transaction category from a preset risk strategy according to the preset corresponding relation between the transaction category and the wind control rule.
In some of these embodiments, "matching the first target data with the contra-cast" includes: acquiring the field name number of the field names identified from the no-entry type wind control rule and the no-entry word number of the no-entry words; if the number of the field names is equal to 1 and the number of the forbidden words is greater than 1, matching the first target data under the field names with each forbidden word; and if the number of the field names and the number of the forbidden words are both larger than 1, associating the field names with the forbidden words according to the field names and the occurrence positions of the forbidden words in the forbidden type wind control rule, and matching the first target data under each field name with the associated forbidden words.
In some of these embodiments, "comparing the second target data to the threshold" includes: acquiring the number of action words of field processing words identified from the threshold type wind control rule and the threshold number of the threshold, and when the number of the action words is not 0: acquiring the number of the target field names of the corresponding target field names of the second target data, and matching the second target data under the target field names with the threshold when the number of the target field names is equal to the number of the threshold and is equal to 1; and when the number of the target field names is equal to the threshold number and is more than 1, associating the target field names with the threshold according to the original field names and the appearance positions of the threshold in the threshold type wind control rule, and comparing the second target data under each target field name with the associated threshold.
In some of these embodiments, where the number of action words is 0: acquiring the number of original fields of the original fields, and matching original data under the original fields with a threshold value when the number of the original fields is equal to the threshold value and is equal to 1; and when the number of the original fields is equal to the number of the threshold values and is more than 1, associating the original field names with the threshold values according to the original field names and the appearance positions of the threshold values in the threshold value type wind control rule, and comparing the original data under each original field name with the associated threshold values.
In some of these embodiments, the training method of the "trained first classifier" includes: the method comprises the steps of obtaining a plurality of no-throw type wind control rule samples, carrying out pre-labeling processing on each no-throw type wind control rule sample to obtain a first classification label, and constructing a first training data set by the plurality of no-throw type wind control rule samples and the corresponding first classification label; training the first classification model according to the first training data set to obtain a trained first classifier; the first classification model comprises one of a Gaussian mixture model classifier or a K nearest neighbor classifier, each forbidden-to-throw type wind control rule sample is used as the input of the first classification model, the first classification label is used as the output of the first classification model, and the first classification model is trained.
In some of these embodiments, the training method of the "trained second classifier" includes: obtaining a plurality of threshold type wind control rule samples, performing pre-labeling processing on each threshold type wind control rule sample to obtain a second classification label, and constructing a second training data set by using the plurality of threshold type wind control rule samples, namely the corresponding second classification labels; training the second classification model according to a second training data set to obtain a trained second classifier; and the second classification model comprises a support vector machine classifier, each threshold type wind control rule sample is used as the input of the second classification model, the second classification label is used as the output of the second classification model, and the second classification model is trained.
In a second aspect, an embodiment of the present application provides an abnormal transaction identification apparatus, including: the system comprises an acquisition module, a processing module and a processing module, wherein the acquisition module is used for acquiring at least one wind control rule corresponding to a target transaction to be detected, and each wind control rule comprises a no-throw type wind control rule or a threshold value type wind control rule; the first word segmentation module is used for segmenting the no-switching type wind control rule and inputting the segmented word into a trained first classifier for recognition to obtain a field name output by the trained first classifier and a no-switching word; the first calculation module is used for acquiring first target data according to the field name, and matching the first target data with the forbidden word to obtain a first classification result of the target transaction; the second word segmentation module is used for segmenting the threshold value type wind control rule and inputting the segmented word into a trained second classifier for recognition to obtain an original field name, a field processing word and a threshold value output by the trained second classifier; the data processing module is used for processing the original data under the original field name according to the field processing words to obtain second target data, and the second calculating module is used for comparing the second target data with a threshold value to obtain a second classification result of the target transaction; a classification result for the target transaction is determined based on the first classification result and/or the second classification result.
In a third aspect, the present application provides a readable storage medium, wherein the readable storage medium stores therein a computer program, the computer program includes program code for controlling a process to execute a process, the process includes the abnormal transaction identification method according to the first aspect.
The main contributions and innovation points of the invention are as follows:
according to the method and the device, the types of the wind control rules are distinguished firstly, the wind control rules of different types are input into different classifiers to be identified to obtain the keywords, the meanings represented by the wind control rules are represented according to the identified keywords, and the classification results are finally obtained by adopting corresponding processing modes according to the actual meanings, so that related personnel can perform subsequent processing on the transaction targets based on the classification results.
The details of one or more embodiments of the application are set forth in the accompanying drawings and the description below to provide a more thorough understanding of the application.
Drawings
The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this application, illustrate embodiment(s) of the application and together with the description serve to explain the application and not to limit the application. In the drawings:
fig. 1 is a flowchart of an abnormal transaction identification method according to an embodiment of the present application.
Fig. 2 is a schematic diagram of a process of classifying a sample object according to a first embodiment of the present application.
Fig. 3 is a block diagram of a structure of an abnormal transaction identification apparatus according to a second embodiment of the present application.
Fig. 4 is a schematic hardware structure diagram of an electronic device according to a third embodiment of the present application.
Detailed Description
Reference will now be made in detail to the exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, like numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with one or more embodiments of the present specification. Rather, they are merely examples of apparatus and methods consistent with certain aspects of one or more embodiments of the specification, as detailed in the claims which follow.
It should be noted that: in other embodiments, the steps of the corresponding methods are not necessarily performed in the order shown and described herein. In some other embodiments, the method may include more or fewer steps than those described herein. Moreover, a single step described in this specification may be broken down into multiple steps for description in other embodiments; multiple steps described in this specification may be combined into a single step in other embodiments.
The scheme can be used for automatically identifying abnormal transactions through the existing wind control rules. Since the wind control rules generally come from the accumulation of the enterprises in the financial business, the structure of the wind control rules is not fixed, that is, each wind control rule has a quite flexible grammatical structure, and there may be multiple layers of judgment and processing logic for data in each wind control rule.
For example, a certain wind control rule is: only 3A rated bond issued by a civil enterprise can be paid. In this wind-control rule, if the investment bond is to satisfy both the civil enterprise issue and the 3A rating, then the expression of the transaction that can pass through the wind-control rule is [3A rating ] & [ civil enterprise issue ]. And another similar wind control rule is: no bond issued by a 3A rated private enterprise could be thrown. In this wind-control rule, the investment bond cannot simultaneously meet the no-vote 3A rating and the no-vote civil enterprise issues, and the expression of the transaction passing through the wind-control rule is! { [3A rating ] & [ civil enterprise issue ] }. Therefore, if the field name provided by the wind control rule alone cannot represent the meaning of the original wind control rule, the field name is far from sufficient for identifying abnormal transactions, and based on the situation, the existing wind control system judges whether the target transaction is risky by adopting a mode of matching with manual understanding of the wind control rule.
In other words, the existing wind control rules lack standardized expression, so that the reliability of the result identified by the conventional abnormal identification model is low, and most of the risk transactions are identified in a mode of being matched with manual understanding of the wind control rules, so that the efficiency is low. Based on the method and the device, the keywords are extracted from the wind control rules, the meanings of the keywords in the wind control rules are determined according to whether the wind control rules belong to the no-throw type wind control rules or the threshold value type wind control rules, and the classification results are finally obtained by adopting a corresponding processing mode according to the essential meanings, so that related personnel can perform subsequent processing on the transaction targets based on the classification results.
Before explaining a specific method of the present solution, concepts to which the present solution may relate are explained first:
target transaction: the bond transaction behavior record data can be obtained from a transaction database.
And (3) forbidden throwing type wind control rule: and a wind control rule for limiting the investment type or the investment forbidding type, such as only one bond can be paid or not.
Threshold type wind control rule: and screening the wind control rule of the transaction behavior according with the threshold value through the threshold value, wherein the investment quantity of a certain bond does not exceed the preset threshold value or is in a preset interval.
The field name: the identifier is used for indicating the stored content in each column of the database, for example, each column of the table is used for storing an account name, the stored data sequentially comprises an account a, an account B, an account C and an account d.
Forbidding word throwing: words representing investment prohibited behaviors, if bonds issued by the public enterprise cannot be thrown, the words issued by the public enterprise are forbidden; if the bond issued by the civil enterprise can only be thrown, the words except the national enterprise are forbidden to be thrown.
A field processing word: words that characterize how the raw data is processed, such as: and adding the M account and the N account to each other to not exceed a threshold S, and then adding the M account and the N account to each other to form a field processing word.
Fig. 1 is a flowchart of main steps of an abnormal transaction identification method according to a first embodiment of the present application.
To achieve this object, as shown in fig. 1, the abnormal transaction identifying method includes steps S101 to S106 as follows.
Step S101, at least one wind control rule corresponding to a target transaction to be detected is obtained, wherein each wind control rule comprises a no-throw type wind control rule or a threshold value type wind control rule.
In the step, all the wind control rules for identifying the target transaction are obtained firstly, wherein the number of the wind control rules can be multiple, and the wind control rules can be classified into forbidden operation type wind control rules or threshold value type wind control rules according to different types of the wind control rules. The classification aims to respectively process different types of wind control rules in subsequent processing steps, so that the reliability of a calculation result is improved.
It should be noted that, in actual operation, the original wind control rule may also include a plurality of different types of rules, for example, one original rule includes both the no-throw type wind control rule and the threshold type wind control rule. For the situation, the method is used for firstly carrying out clause division on the original wind control rule according to the type, and dividing the original wind control rule into an atomic forbidden-throw type wind control rule or a threshold value type wind control rule.
In this step, the prohibited-investment type wind control rule may be a "must" type rule that only a certain bond can be invested, or a "must not" type rule that a certain bond cannot be invested. According to the scheme, the method and the device are divided into the forbidden investment type wind control rules, aiming at the forbidden investment type wind control rules of the 'necessary' type rules, when no negative word exists before the forbidden investment word is collected, the rule is the 'necessary' type rule, when the transaction data are matched with the forbidden investment words, the matched transaction data are characterized as no-risk transactions, and the unmatched transaction data are characterized as risk transactions. And the 'must not' type rule is just opposite, when a negative word exists before the word throwing is collected, the rule is the 'must not' type rule, when the transaction data is matched with the word throwing prohibition, the matched transaction data is characterized as a risk transaction, and the unmatched transaction data is characterized as a no risk transaction.
In this step, the threshold value type pneumatic control rule may be a "direct comparison" type rule in which the original transaction data is compared with the threshold value, or may be an "indirect comparison" type rule in which the calculation results of a plurality of direct transaction data are compared with the threshold value. In the scheme, aiming at a direct comparison rule, the system automatically compares original data with a threshold value, transaction data exceeding the threshold value are characterized as risk transactions, and transaction data within the threshold value are characterized as risk-free transactions. And aiming at the principle of indirect comparison, the system can automatically calculate the original data and compare the calculated target data with a threshold value, wherein the target data exceeding the threshold value are characterized as risk-free transactions, and the target data within the threshold value are characterized as risk-free transactions.
In step S101, the wind control rules are classified to process different types of wind control rules, and the meanings of the keywords identified by the model in the corresponding types of wind control rules are determined by classification, so that corresponding processing methods are adopted according to the actual meanings, and a classification result is obtained finally.
In one possible embodiment, "acquiring at least one wind control rule corresponding to a target transaction to be detected" includes: and obtaining the transaction category to which the target transaction belongs, and determining at least one wind control rule corresponding to the transaction category from a preset risk strategy according to the preset corresponding relation between the transaction category and the wind control rule.
In this embodiment, a preset risk policy is used for storing all the wind control rules, each wind control rule is stored in only one field, and the corresponding wind control rules are determined for the target transaction by recording the corresponding relationship between the transaction type and the wind control rule.
Specifically, in the current target transaction, m transaction records correspond to n wind control rules, and if the wind control rule of each transaction record is directly recorded, m × n rules are generated, which directly results in large quantity of the wind control rules and high maintenance cost.
Exemplarily, if the target transaction a and the target transaction B need to be classified by the risk rule α, directly obtaining the risk rule α according to the corresponding relationship between the target transaction a and α to perform risk transaction detection; and obtaining a risk rule alpha according to the corresponding relation between the target transaction B and the target transaction alpha to detect the risk transaction. By the method, the problem that the risk rules for different transaction types are the same but are stored in different fields at present can be greatly reduced, and the maintenance efficiency of the risk rules is improved.
In this embodiment, the transaction category includes an account group or a transaction area or a bond type or other category customized by a manager, and the like. The set of corresponding risk rules may differ depending on the transaction category. In this embodiment, the set of risk rules is not divided by transaction category any more, but all the wind control rules corresponding to the transaction category are queried from the preset risk policy directly according to the corresponding relationship between the transaction category and the risk rules.
And S102, segmenting words of the no-entry type wind control rule, and inputting the words into the trained first classifier for recognition to obtain field names and no-entry words output by the trained first classifier.
In this step, the wind control rule is segmented to obtain a process text composed of word groups, which is denoted as T = (W1, W2, W3.,. Wn), where W1, W2, W3.,. Wn refers to individual word groups, the process text T is input into a trained first classifier for recognition, and a field name and a forbidden word are output.
The word segmentation can adopt a common algorithm based on character string matching to compare the Chinese character string to be segmented with the vocabulary entries in the dictionary one by one through a certain strategy, and the word segmentation mode is carried out if the comparison is successful; or a common statistical-based method is adopted, the combination frequency of adjacent co-occurring characters in the material is counted, the co-occurrence information of the characters is calculated, and a word of which the co-occurrence information is higher than a threshold value is taken as a word; or "hidden markov model based segmentation algorithm", etc., and will not be redundantly described here. The following description focuses on specific implementation steps capable of improving the classification accuracy of field names and forbidden words in the scheme.
The keywords mainly extracted aiming at the forbidden-throw type wind control rule are field names and forbidden-throw words, and are different from the situation that 'nouns' and 'verbs' are distinguished under a conventional scene, and the field names and the forbidden-throw words are noun phrases, so that a large number of samples are needed for training a model, and otherwise, the accuracy of distinguishing the field names and the forbidden-throw words by the model is difficult to meet the expectation. In the scheme, the 'inter-field distance' is introduced to further distinguish the two.
Taking "a bond issued by a civil enterprise that cannot be rated by 3A" as an example, in the wind control rule, the positive word/negative word is "impossible", the recognized nouns are "3A rating", "civil enterprise issue", and "bond", respectively, 3A rating, civil enterprise issue, and bond are sorted from near to far in order of the distance between each noun and "impossible", the noun that is farthest from "impossible" is a field name, and the noun that is closest to "impossible" is "prohibited word". Therefore, after the classification result of the first classifier is obtained, the classification result which is not expected to be met can be eliminated according to whether the classification result meets the distance relation of the 'distance between fields', or the classification result is manually classified, so that the influence on the reliability of the result caused by the classification error of the 'field name' and the 'forbidden word' is avoided.
In one possible embodiment, the training method of the "trained first classifier" includes: the method comprises the steps of obtaining a plurality of no-throw type wind control rule samples, carrying out pre-labeling processing on each no-throw type wind control rule sample to obtain a first classification label, and constructing a first training data set by the plurality of no-throw type wind control rule samples and the corresponding first classification label; training the first classification model according to the first training data set to obtain a trained first classifier; the first classification model comprises one of a Gaussian mixture model classifier or a K nearest neighbor classifier, each forbidden-to-throw type wind control rule sample is used as the input of the first classification model, the first classification label is used as the output of the first classification model, and the first classification model is trained.
The training mode of the model and the obtained trained first classifier are different from the prior art in that the field name can be recognized, and the word throwing prohibition can be recognized, so that the beneficial effect of processing logic without manually determining the field name is brought, and the subsequent abnormal transaction recognition efficiency can be improved by automatically recognizing the keywords and automatically matching according to the meanings of the keywords in the wind control rule.
Step S103, acquiring first target data according to the field name, and matching the first target data with the forbidden word to obtain a first classification result of the target transaction.
The step can automatically identify batch transaction data, specifically, if the field name includes a plurality of first target data, each first target data is matched with the forbidden word to obtain a matching result, and a first classification result is determined based on the matching result.
Referring to the description of step S101, according to whether the positive word or the negative word is identified in the wind control rule, the matching logic is opposite, that is: if a positive word is identified, the transaction data on the match is characterized as a no risk transaction and the transaction data on the mismatch is characterized as a risk transaction, or vice versa. Therefore, after the matching result is obtained, the first classification result is further obtained through the recognition of the "positive word" or the "negative word". Wherein the identification of "positive words" or "negative words" may also employ a first classifier: the model is trained by manually marking 'positive words' or 'negative words', so that the trained model can automatically identify and classify corresponding words from sentences, and introduction is not required.
For the above step S103, the scheme may match the first target data of the no-entry word with the no-entry word according to the identified field name, and confirm the matching result, and further obtain the first classification result based on the matching result and the identified positive word/negative word. In one possible embodiment, the field names and the banned words identified may be matched one by one. As follows:
"matching the first target data with the prohibited word" includes: acquiring the field name number of the field names identified from the no-entry type wind control rule and the no-entry word number of the no-entry words; if the number of the field names is equal to 1 and the number of the forbidden words is greater than 1, matching the first target data under the field names with each forbidden word; and if the number of the field names and the number of the forbidden words are both larger than 1, associating the field names with the forbidden words according to the field names and the occurrence positions of the forbidden words in the forbidden type wind control rule, and matching the first target data under each field name with the associated forbidden words.
Specifically, in the present embodiment, the meaning of each keyword expressed in the wind control rule is determined according to the number of keywords and the appearance position of each keyword in the wind control rule.
Illustratively, when the number of field names is 1, it is stated that the wind control rule performs exception calculation only for one field name. Therefore, no matter how many forbidden words are contained, the forbidden words are the definition of the field names. Therefore, each first target data is directly traversed and is matched with each forbidden word in sequence.
And when the number of the field names is more than 1, for example, the result of the field names obtained by the first classifier is { bond type, trade area } voting forbidding { pearl triangle area, civil enterprise issue, 3A rating }. The two identified results are both an unordered array, and if the association mapping is directly established, the two identified results comprise 6 association mappings, which are respectively as follows: bond type → pearl triangle area, trade area → civil enterprise issue, 3A rating; type of bond → issuance of civil enterprise, trade area → pearl triangle area, 3A rating; type of bond → 3A rating, trade area → pearl triangle area, civil enterprise issue; bond type → pearl triangle area, civil enterprise issue, trade area → 3A rating; type of bond → area of triangle of pearl, 3A rating, trade area → civil enterprise issue; type of bond → issue of civil enterprise, 3A rating, trade area → pearl triangle area. But actually only the associated mapping (c) is consistent with what the wind control rule actually will mean. Therefore, besides the existing semantic matching method, the scheme also provides that the specific semantics are determined according to the appearance positions of the keywords in the wind control rule.
For example, the appearance position may refer to the appearance position of the short sentence in the wind control rule in which the keyword is located in the wind control rule, for example, if the first short sentence in the wind control rule includes the type of bond, issued by a civil enterprise, and a 3A rating, an association relationship between the three keywords is established; and if the second short sentence comprises a trading area and a triangle area, establishing the association relationship of the two keywords.
In addition, if only one keyword exists in a sentence, that is, each keyword appears in different short sentences, the association relationship can be established by the distance between the keywords. Such as: in the sentences of the wind control rule, the distance between the bond type and the issuance of the civil enterprise is far shorter than the distance between the bond type and the triangular bead area, and then the bond type and the issuance of the civil enterprise are associated. By the embodiment, the association mapping representing the correct semantics of the wind control rule can be obtained, and the reliability of the calculation result is improved.
And step S104, segmenting the threshold value type wind control rule, inputting the segmented threshold value type wind control rule into the trained second classifier, and identifying to obtain the original field name, the field processing word and the threshold value output by the trained second classifier.
In this step, the wind control rule is segmented to obtain a process text composed of word groups, which is denoted as T '= (W1, W2, W3.., Wn), where W1, W2, W3.., Wn refers to individual word groups, the process text T' is input into a trained second classifier for recognition, and an original field name, a field processing word, and a threshold are output.
It should be noted that, in this step, the original field name is to be distinguished from the subsequent target field name, and what the second classifier outputs when actually recognizing is the field name.
In pre-labeling, the original field name is represented by a noun and the field handling word is a verb, so the classification of the two is essentially a classification of nouns and verbs in the wind control rules. The threshold value is generally before or after the judgment words "greater than", "less than" and "not greater than", so that the threshold value can be directly identified according to the judgment words.
In one possible embodiment, the training method of the "trained second classifier" includes: obtaining a plurality of threshold type wind control rule samples, performing pre-labeling processing on each threshold type wind control rule sample to obtain a second classification label, and constructing a second training data set by using the plurality of threshold type wind control rule samples, namely the corresponding second classification labels; training the second classification model according to a second training data set to obtain a trained second classifier; and the second classification model comprises a support vector machine classifier, each threshold type wind control rule sample is used as the input of the second classification model, the second classification label is used as the output of the second classification model, and the second classification model is trained.
The technical points of the trained second classifier comprise: the original field name and the field processing word can be recognized, and the target field can be obtained by automatically calculating according to the two types of relevant words. Namely, the target field is acquired without manual operation, and the judgment efficiency is improved.
And S105, processing the original data under the original field name according to the field processing words to obtain second target data, and comparing the second target data with a threshold value to obtain a second classification result of the target transaction.
In step S104, it can be known that the output result of the second classifier includes at least two types, one of which includes the original field name, the field processing word and the threshold value, and the other one includes the original field name and the threshold value, and the field processing word is not recognized.
Therefore, in this step, for the first case, when a field processing word is identified, it is indicated that data under the original field name needs to be processed, and the processed data under the target field name is compared with the threshold, for example, the wind control rule 1 is: the ratio of the bond investment amount to the bond issue amount is less than 30%. In the wind control rule 1, if the field processing word is identified as the "ratio", the "bond investment amount" and the "bond issue amount" are both the original field names, and the original data under the two are calculated to obtain the ratio as the second target data. The wind control rule 2 is: the bond investment amount is not higher than 30 Wyuan. In the wind control rule 2, if no field processing word is recognized, the original data of the bond investment amount is used as the second target data.
Based on this, the steps that this scheme adopted include: and when the second classifier does not recognize the field processing action word and the field processing word, directly taking the original data under the original field name as second target data to be compared with a threshold value, and obtaining a third classification result of the target transaction.
Further, an embodiment of the present disclosure further provides a method for establishing association mapping one by one for a plurality of identified field processing words and a plurality of thresholds: the "comparing the second target data with the threshold" includes: acquiring the number of action words of field processing action word field processing words identified from the threshold type wind control rule and the threshold number of thresholds, and when the number of action words is not 0: acquiring the number of the target field names of the corresponding target field names of the second target data, and matching the second target data under the target field names with the threshold when the number of the target field names is equal to the number of the threshold and is equal to 1; and when the number of the target field names is equal to the threshold number and is more than 1, associating the target field names with the threshold according to the appearance sequence of the original field names and the threshold in the threshold type wind control rule, and comparing the second target data under each target field name with the associated threshold.
In addition, when the number of action words is 0: acquiring the number of original fields of the original fields, and matching original data under the original fields with a threshold value when the number of the original fields is equal to the threshold value and is equal to 1; and when the number of the original fields is equal to the number of the threshold and is more than 1, associating the original field names with the threshold according to the appearance sequence of the original field names and the threshold in the threshold type wind control rule, and comparing the original data under each original field name with the associated threshold.
For the explanation of this embodiment, reference may be made to the description of step S103, and similar differences will be repeated. The present embodiment is distinguished in that: the fields associated with the threshold differ as to whether or not a field handling word is present: when a field handling word is present, the target name field is associated with the threshold, and when a field handling word is absent, the original name field is associated with the threshold.
And step S106, determining a classification result of the target transaction based on the first classification result and/or the second classification result.
In the step, the first classification result and/or the second classification result obtained by different processing modes aiming at different types of wind control rules are combined to obtain a final classification result. Such as: the target transaction corresponds to 3 wind control rules, the identified results are risky, risk-free and risk respectively, and then the final classification result of the target transaction may be risky based on the identified classification result.
It should be noted that, there may be a plurality of specific operation rules for obtaining the final classification result according to the first classification result and/or the second classification result, for example: when any one of the first classification result or the second classification result is at risk, the final classification result is at risk; or the following steps: when any one of the first classification result or the second classification result is risk-free, the final classification result is risk-free; or the following steps: and distributing weights for the first classification result and the second classification result, calculating a final risk value, wherein if the risk value accords with a preset risk threshold value, the classification result is risk-free, and if the risk value does not accord with the preset risk threshold value, the classification result is risk-free.
In summary, referring to fig. 2, for the above steps S101 to S106, the present disclosure proposes a method for automatically identifying a risk of a target transaction to be detected, in the method, wind control rule types are first distinguished, and different types of wind control rules are input into different classifiers to identify a keyword, unlike the prior art, the keyword identified in the present disclosure is a word capable of expressing a specific meaning of the wind control rule, and if "field name" and "prohibited word" are identified in the prohibited wind control rule, indicating that data expressed by the wind control rule under the field name conforms to the prohibited word, the transaction is prompted with a risk; in the threshold value type wind control rule, the original field name, the field processing word and the threshold value are identified, and the meaning indicated by the wind control rule is that the processed data under the original field name exceeds the threshold value, and then the transaction is prompted to be risky. Based on the method and the system, the keywords identified by the classifier can be guaranteed not to have deviation with the meanings represented by the wind control rules, and therefore the reliability of abnormal transactions automatically identified by the model is improved.
Fig. 3 is a block diagram showing a configuration of a device for determining a cause of error problem type according to a second embodiment of the present application. As shown in fig. 2, the present application also proposes an abnormal transaction identification apparatus, including:
the obtaining module 301 is configured to obtain at least one wind control rule corresponding to a target transaction to be detected, where each wind control rule includes a no-entry type wind control rule or a threshold type wind control rule.
The first word segmentation module 302 is configured to input the word of the no-entry type wind control rule into the trained first classifier for recognition, so as to obtain a field name output by the trained first classifier and a no-entry word.
The first calculating module 303 is configured to obtain first target data according to the field name, match the first target data with the prohibited word, and obtain a first classification result of the target transaction.
And the second word segmentation module 304 is configured to segment words of the threshold type wind control rule and then input the segmented words into the trained second classifier for recognition, so as to obtain an original field name, a field processing word and a threshold output by the trained second classifier.
And the data processing module 305 is configured to process the original data under the original field name according to the field processing word, so as to obtain second target data.
A second calculating module 306, configured to compare the second target data with a threshold value to obtain a second classification result of the target transaction; a classification result for the target transaction is determined based on the first classification result and/or the second classification result.
The first segmentation module 302 is provided with a segmentation model and a first classifier, and the second segmentation module 304 is provided with a segmentation model and a second classifier, which are described in the first embodiment. Further, since the apparatus is operated by the method described above, the repetitive description is not intended.
EXAMPLE III
The present embodiment also provides an electronic device, referring to fig. 4, comprising a memory 404 and a processor 402, wherein the memory 404 stores a computer program, and the processor 402 is configured to execute the computer program to perform the steps of any of the above method embodiments.
Specifically, the processor 402 may include a Central Processing Unit (CPU), or A Specific Integrated Circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.
The processor 402 may implement any of the above-described embodiments of the anomalous transaction identification methods by reading and executing computer program instructions stored in the memory 404.
Optionally, the electronic apparatus may further include a transmission device 406 and an input/output device 408, where the transmission device 406 is connected to the processor 402, and the input/output device 408 is connected to the processor 402.
The transmitting device 406 may be used to receive or transmit data via a network. Specific examples of the network described above may include wired or wireless networks provided by communication providers of the electronic devices. In one example, the transmission device includes a Network adapter (NIC) that can be connected to other Network devices through a base station to communicate with the internet. In one example, the transmitting device 406 may be a Radio Frequency (RF) module, which is used to communicate with the internet in a wireless manner.
The input and output devices 408 are used to input or output information. In this embodiment, the input information may be a target transaction to be detected, and the output information may be a trained first classifier, a trained second classifier, a classification result, and the like.
Optionally, in this embodiment, the processor 402 may be configured to execute the following steps by a computer program:
s101, at least one wind control rule corresponding to a target transaction to be detected is obtained, wherein each wind control rule comprises a no-throw type wind control rule or a threshold value type wind control rule.
S102, after word segmentation is carried out on the no-entry type wind control rule, the word is input into the trained first classifier to be recognized, and the field name and the no-entry word output by the trained first classifier are obtained.
S103, acquiring first target data according to the field name, and matching the first target data with the forbidden word to obtain a first classification result of the target transaction.
And S104, segmenting the threshold value type wind control rule, inputting the segmented threshold value type wind control rule into the trained second classifier, and identifying to obtain the original field name, the field processing action word and the field processing word output by the trained second classifier and the threshold value.
And S105, processing the original data under the original field name according to the field processing action word and the field processing word to obtain second target data, and comparing the second target data with a threshold value to obtain a second classification result of the target transaction.
And S106, determining a classification result of the target transaction based on the first classification result and/or the second classification result.
It should be noted that, for specific examples in this embodiment, reference may be made to examples described in the foregoing embodiments and optional implementations, and details of this embodiment are not described herein again.
In general, the various embodiments may be implemented in hardware or special purpose circuits, software, logic or any combination thereof. Some aspects of the invention may be implemented in hardware, while other aspects may be implemented in firmware or software which may be executed by a controller, microprocessor or other computing device, although the invention is not limited thereto. While various aspects of the invention may be illustrated and described as block diagrams, flow charts, or using some other pictorial representation, it is well understood that these blocks, apparatus, systems, techniques or methods described herein may be implemented in, as non-limiting examples, hardware, software, firmware, special purpose circuits or logic, general purpose hardware or controller or other computing devices, or some combination thereof.
Embodiments of the invention may be implemented by computer software executable by a data processor of the mobile device, such as in a processor entity, or by hardware, or by a combination of software and hardware. Computer software or programs (also referred to as program products) including software routines, applets and/or macros can be stored in any device-readable data storage medium and they include program instructions for performing particular tasks. The computer program product may comprise one or more computer-executable components configured to perform embodiments when the program is run. The one or more computer-executable components may be at least one software code or a portion thereof. Further in this regard it should be noted that any block of the logic flow as in the figures may represent a program step, or an interconnected logic circuit, block and function, or a combination of a program step and a logic circuit, block and function. The software may be stored on physical media such as memory chips or memory blocks implemented within the processor, magnetic media such as hard or floppy disks, and optical media such as, for example, DVDs and data variants thereof, CDs. The physical medium is a non-transitory medium.
It should be understood by those skilled in the art that various features of the above embodiments can be combined arbitrarily, and for the sake of brevity, all possible combinations of the features in the above embodiments are not described, but should be considered as within the scope of the present disclosure as long as there is no contradiction between the combinations of the features.
The above examples are merely illustrative of several embodiments of the present application, and the description is more specific and detailed, but not to be construed as limiting the scope of the present application. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the concept of the present application, which falls within the scope of protection of the present application. Therefore, the protection scope of the present application shall be subject to the appended claims.
Claims (10)
1. An abnormal transaction identification method is characterized by comprising the following steps:
acquiring at least one wind control rule corresponding to a target transaction to be detected, wherein each wind control rule comprises a no-throw type wind control rule or a threshold value type wind control rule;
after word segmentation is carried out on the no-entry type wind control rule, inputting the no-entry type wind control rule into a trained first classifier for recognition, and obtaining a field name output by the trained first classifier and no-entry words;
acquiring first target data according to the field name, and matching the first target data with the forbidden word to obtain a first classification result of the target transaction;
after segmenting words of the threshold value type wind control rule, inputting the words into a trained second classifier for recognition to obtain an original field name, a field processing word and a threshold value output by the trained second classifier;
processing the original data under the original field name according to the field processing words to obtain second target data, and comparing the second target data with a threshold value to obtain a second classification result of the target transaction;
a classification result for the target transaction is determined based on the first classification result and/or the second classification result.
2. The abnormal transaction identification method of claim 1, wherein the second classification result further comprises a third classification result, wherein obtaining the third classification result comprises:
and when the second classifier does not recognize the field processing words, directly taking the original data under the original field name as second target data to be compared with a threshold value, and obtaining a third classification result of the target transaction.
3. The abnormal transaction identification method according to claim 1, wherein the step of obtaining at least one wind control rule corresponding to the target transaction to be detected comprises the steps of:
and obtaining the transaction category to which the target transaction belongs, and determining at least one wind control rule corresponding to the transaction category from a preset risk strategy according to the preset corresponding relation between the transaction category and the wind control rule.
4. The abnormal transaction identifying method according to claim 1, wherein "matching the first target data with the banned word" includes:
acquiring the field name number of the field names identified from the no-entry type wind control rule and the no-entry word number of the no-entry words;
if the number of the field names is equal to 1 and the number of the forbidden words is greater than 1, matching the first target data under the field names with each forbidden word;
and if the number of the field names and the number of the forbidden words are both larger than 1, associating the field names with the forbidden words according to the field names and the occurrence positions of the forbidden words in the forbidden type wind control rule, and matching the first target data under each field name with the associated forbidden words.
5. The abnormal transaction identification method according to claim 1, wherein the comparing the second target data with the threshold value comprises:
acquiring the number of action words of field processing words identified from the threshold type wind control rule and the threshold number of the threshold, and when the number of the action words is not 0:
acquiring the number of the target field names of the corresponding target field names of the second target data, and matching the second target data under the target field names with the threshold when the number of the target field names is equal to the number of the threshold and is equal to 1;
and when the number of the target field names is equal to the threshold number and is more than 1, associating the target field names with the threshold according to the original field names and the appearance positions of the threshold in the threshold type wind control rule, and comparing the second target data under each target field name with the associated threshold.
6. The abnormal transaction recognition method according to claim 5, wherein in the case where the number of action words is 0:
acquiring the number of original fields of the original fields, and matching original data under the original fields with a threshold value when the number of the original fields is equal to the threshold value and is equal to 1;
and when the number of the original fields is equal to the number of the threshold values and is more than 1, associating the original field names with the threshold values according to the original field names and the appearance positions of the threshold values in the threshold value type wind control rule, and comparing the original data under each original field name with the associated threshold values.
7. The abnormal transaction recognition method of claim 1, wherein the training method of the "trained first classifier" comprises:
the method comprises the steps of obtaining a plurality of no-throw type wind control rule samples, carrying out pre-labeling processing on each no-throw type wind control rule sample to obtain a first classification label, and constructing a first training data set by the plurality of no-throw type wind control rule samples and the corresponding first classification label;
training the first classification model according to the first training data set to obtain a trained first classifier; the first classification model comprises one of a Gaussian mixture model classifier or a K nearest neighbor classifier, each forbidden-to-throw type wind control rule sample is used as the input of the first classification model, the first classification label is used as the output of the first classification model, and the first classification model is trained.
8. The abnormal transaction recognition method of claim 1, wherein the training method of the "trained second classifier" comprises:
obtaining a plurality of threshold type wind control rule samples, performing pre-labeling processing on each threshold type wind control rule sample to obtain a second classification label, and constructing a second training data set by using the plurality of threshold type wind control rule samples, namely the corresponding second classification labels;
training the second classification model according to a second training data set to obtain a trained second classifier; and the second classification model comprises a support vector machine classifier, each threshold type wind control rule sample is used as the input of the second classification model, the second classification label is used as the output of the second classification model, and the second classification model is trained.
9. An abnormal transaction identifying apparatus, comprising:
the system comprises an acquisition module, a processing module and a processing module, wherein the acquisition module is used for acquiring at least one wind control rule corresponding to a target transaction to be detected, and each wind control rule comprises a no-throw type wind control rule or a threshold value type wind control rule;
the first word segmentation module is used for segmenting the no-switching type wind control rule and inputting the word into the trained first classifier for recognition to obtain a field name output by the trained first classifier and a no-switching word;
the first calculation module is used for acquiring first target data according to the field name, and matching the first target data with the forbidden word to obtain a first classification result of the target transaction;
the second word segmentation module is used for segmenting the threshold value type wind control rule and inputting the segmented word into a trained second classifier for recognition to obtain an original field name, a field processing word and a threshold value output by the trained second classifier;
a data processing module for processing the original data under the original field name according to the field processing words to obtain second target data,
the second calculation module is used for comparing the second target data with the threshold value to obtain a second classification result of the target transaction;
a classification result for the target transaction is determined based on the first classification result and/or the second classification result.
10. A readable storage medium having stored therein a computer program comprising program code for controlling a process to execute a process, the process comprising the abnormal transaction identification method according to any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210008148.4A CN114049215A (en) | 2022-01-06 | 2022-01-06 | Abnormal transaction identification method, device and application |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210008148.4A CN114049215A (en) | 2022-01-06 | 2022-01-06 | Abnormal transaction identification method, device and application |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114049215A true CN114049215A (en) | 2022-02-15 |
Family
ID=80213435
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210008148.4A Pending CN114049215A (en) | 2022-01-06 | 2022-01-06 | Abnormal transaction identification method, device and application |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114049215A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116151841A (en) * | 2022-12-28 | 2023-05-23 | 连连银通电子支付有限公司 | Keyword recognition-based control method and device, electronic equipment and storage medium |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107316134A (en) * | 2017-06-16 | 2017-11-03 | 深圳乐信软件技术有限公司 | A kind of risk control method, device, server and storage medium |
| CN108197777A (en) * | 2017-12-14 | 2018-06-22 | 阿里巴巴集团控股有限公司 | A kind of method, apparatus and equipment for adjusting air control rule |
| CN108364132A (en) * | 2018-02-11 | 2018-08-03 | 深圳市快付通金融网络科技服务有限公司 | A kind of air control method, apparatus, computer installation and computer readable storage medium |
| CN111915312A (en) * | 2020-08-06 | 2020-11-10 | 支付宝(杭州)信息技术有限公司 | Risk identification method and device and electronic equipment |
| CN113011889A (en) * | 2021-03-10 | 2021-06-22 | 腾讯科技（深圳）有限公司 | Account abnormity identification method, system, device, equipment and medium |
| CN113052266A (en) * | 2021-04-27 | 2021-06-29 | 中国工商银行股份有限公司 | Transaction mode type identification method and device |
| WO2021169208A1 (en) * | 2020-02-25 | 2021-09-02 | 平安科技（深圳）有限公司 | Text review method and apparatus, and computer device, and readable storage medium |
| CN113704805A (en) * | 2021-10-27 | 2021-11-26 | 华控清交信息科技(北京)有限公司 | Wind control rule matching method and device and electronic equipment |
-
2022
- 2022-01-06 CN CN202210008148.4A patent/CN114049215A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107316134A (en) * | 2017-06-16 | 2017-11-03 | 深圳乐信软件技术有限公司 | A kind of risk control method, device, server and storage medium |
| CN108197777A (en) * | 2017-12-14 | 2018-06-22 | 阿里巴巴集团控股有限公司 | A kind of method, apparatus and equipment for adjusting air control rule |
| CN108364132A (en) * | 2018-02-11 | 2018-08-03 | 深圳市快付通金融网络科技服务有限公司 | A kind of air control method, apparatus, computer installation and computer readable storage medium |
| WO2021169208A1 (en) * | 2020-02-25 | 2021-09-02 | 平安科技（深圳）有限公司 | Text review method and apparatus, and computer device, and readable storage medium |
| CN111915312A (en) * | 2020-08-06 | 2020-11-10 | 支付宝(杭州)信息技术有限公司 | Risk identification method and device and electronic equipment |
| CN113011889A (en) * | 2021-03-10 | 2021-06-22 | 腾讯科技（深圳）有限公司 | Account abnormity identification method, system, device, equipment and medium |
| CN113052266A (en) * | 2021-04-27 | 2021-06-29 | 中国工商银行股份有限公司 | Transaction mode type identification method and device |
| CN113704805A (en) * | 2021-10-27 | 2021-11-26 | 华控清交信息科技(北京)有限公司 | Wind control rule matching method and device and electronic equipment |
Non-Patent Citations (1)
| Title |
|---|
| 马刚: "《基于语义的Web数据挖掘》", 31 January 2014, 大连：东北财经大学出版社 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116151841A (en) * | 2022-12-28 | 2023-05-23 | 连连银通电子支付有限公司 | Keyword recognition-based control method and device, electronic equipment and storage medium |
| CN116151841B (en) * | 2022-12-28 | 2023-09-19 | 连连银通电子支付有限公司 | Keyword recognition-based control method and device, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109460455B (en) | Text detection method and device | |
| CN109543032A (en) | File classification method, device, computer equipment and storage medium | |
| TW201737118A (en) | Method and device for webpage text classification, method and device for webpage text recognition | |
| CN109271489B (en) | Text detection method and device | |
| CN101751550B (en) | Fast fingerprint searching method and fast fingerprint searching system thereof | |
| CN110929525B (en) | Network loan risk behavior analysis and detection method, device, equipment and storage medium | |
| CN115062148B (en) | Risk control method based on database | |
| CN106203808A (en) | Enterprise Credit Risk Evaluation method and apparatus | |
| CN116361815B (en) | Code sensitive information and hard coding detection method and device based on machine learning | |
| CN109933648B (en) | Real user comment distinguishing method and device | |
| CN114049215A (en) | Abnormal transaction identification method, device and application | |
| CN110610007A (en) | Maintenance vehicle condition intelligent identification method and device based on NLP | |
| Aljabri et al. | Fake news detection using machine learning models | |
| CN116701641B (en) | Hierarchical classification method and device for unstructured data | |
| CN114117038A (en) | Document classification method, device and system and electronic equipment | |
| CN111241269B (en) | Short message text classification method and device, electronic equipment and storage medium | |
| CN116401343A (en) | Data compliance analysis method | |
| CN115358340A (en) | Credit credit collection short message distinguishing method, system, equipment and storage medium | |
| CN114064893A (en) | Abnormal data auditing method, device, equipment and storage medium | |
| CN112308251A (en) | Work order assignment method and system based on machine learning | |
| CN113378244B (en) | Intelligent electronic signature calling system and method based on data analysis | |
| JPH0484366A (en) | Document kind discriminating device | |
| CN111488738A (en) | Illegal information identification method and device | |
| CN112686055B (en) | Semantic recognition method and device, electronic equipment and storage medium | |
| CN109344254A (en) | A kind of address information classification method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20220215 |