Account identification method and apparatus, electronic device, and storage medium

Abstract

The application provides a method and device for account identification, electronic equipment and storage medium. The method comprises: obtaining a first text corresponding to a to-be-identified account, wherein the first text is obtained by splicing at least one text information associated with the to-be-identified account; obtaining a keyword sequence associated with a to-be-matched risk label, wherein the keyword sequence includes a plurality of keywords having semantic correlation with the to-be-matched risk label; performing splicing processing on the first text and the keyword sequence to obtain a target text; and predicting a matching result of the to-be-identified account and the to-be-matched risk label according to the target text. The method can add prior knowledge about risk conditions to the process of risk prediction by adding the keyword sequence when predicting, thereby improving the accuracy and coverage of risk label prediction, and further reducing the false interception rate of attack accounts and the error rate of account handling operations.

Description

Technical Field

[0001] This application relates to the field of computer technology, and more particularly to a method, apparatus, electronic device, and storage medium for account identification. Background Technology

[0002] With the development of mobile payment and internet finance, people can manage their personal funds anytime, anywhere, making personal fund management and investment increasingly convenient and diversified. However, illegal activities such as gambling have also emerged, making it urgent to identify risky accounts involved in such activities a pressing issue.

[0003] Currently, the identification of risky accounts typically employs text classification algorithms to categorize user account information into tags, thereby identifying the user account's tags, and then determining whether the account belongs to a risky account based on the tags.

[0004] However, text classification algorithms often encounter problems such as incorrect label classification or missed label recognition when classifying labels, resulting in a high rate of false blocking and error rate in account handling operations. This affects normal user operations, and correcting errors requires additional cost and wastes resources. Summary of the Invention

[0005] Based on the above-mentioned technical problems, this application provides an account identification method to improve the accuracy and coverage of risk label prediction, thereby reducing the false blocking rate of malicious accounts and the error rate of account handling operations.

[0006] Other features and advantages of this application will become apparent from the following detailed description, or may be learned in part from practice of this application.

[0007] According to one aspect of the embodiments of this application, a method for account identification is provided, comprising:

[0008] Obtain the first text corresponding to the account to be identified, wherein the first text is obtained by concatenating at least one text information associated with the account to be identified;

[0009] Obtain a keyword sequence associated with the risk label to be matched, wherein the keyword sequence includes multiple keywords that are semantically related to the risk label to be matched;

[0010] The first text and the keyword sequence are concatenated to obtain the target text;

[0011] Predict the matching result between the account to be identified and the risk label to be matched based on the target text.

[0012] According to one aspect of the embodiments of this application, an account identification device is provided, comprising:

[0013] The text acquisition module is used to acquire the first text corresponding to the account to be identified, wherein the first text is obtained by concatenating at least one text information associated with the account to be identified;

[0014] A sequence acquisition module is used to acquire a keyword sequence associated with the risk label to be matched, wherein the keyword sequence includes multiple keywords that are semantically related to the risk label to be matched;

[0015] The splicing module is used to splice the first text and the keyword sequence to obtain the target text;

[0016] The prediction module is used to predict the matching result between the account to be identified and the risk label to be matched based on the target text.

[0017] In some embodiments of this application, based on the above technical solutions, the account identification device further includes:

[0018] The second text acquisition module is used to acquire multiple second texts from multiple accounts, wherein the second text of each account is obtained by concatenating at least one text information associated with each account;

[0019] A related word set acquisition module is used to acquire a related word set, wherein the related word set includes at least one related word, and the related word is determined based on the frequency of occurrence of words in the plurality of second texts;

[0020] The feature volume module is used to extract features from the multiple second texts based on the set of related words to obtain keyword feature vectors;

[0021] The classification module is used to classify the keyword feature vector using a tag classification model to obtain at least one keyword for the risk tag to be matched.

[0022] A combination module is used to combine the keywords into the keyword sequence.

[0023] In some embodiments of this application, based on the above technical solutions, the feature volume module includes:

[0024] The feature vector determination unit is used to determine an intermediate feature vector based on the frequency of occurrence of related words in the set of related words in the plurality of second texts;

[0025] The singular value decomposition unit is used to perform singular value decomposition on the intermediate feature vector to obtain the keyword feature vector.

[0026] In some embodiments of this application, based on the above technical solutions, the feature volume module includes:

[0027] The frequency determination unit is used to determine the frequency of occurrence of all related words in the related word set in each of the plurality of second texts in the second text;

[0028] The text weight determination unit is used to determine the text weight of all related words in the second text based on the frequency of occurrence of all related words in the second text, the text length of the second text, and the average text length of the plurality of second texts.

[0029] The keyword vector determination unit is used to determine the keyword feature vector based on the text weights of all related words in the plurality of second texts.

[0030] In some embodiments of this application, based on the above technical solutions, the related word set acquisition module includes:

[0031] The word segmentation unit is used to segment multiple second texts from multiple accounts to obtain a word set, wherein the word set includes at least one word;

[0032] The set determination unit is used to determine the related word set based on the frequency of occurrence of each word in the word set in the plurality of second texts.

[0033] In some embodiments of this application, based on the above technical solutions, the prediction module includes:

[0034] The text conversion unit is used to convert the target text through the attention layer in the risk identification model to obtain a text vector.

[0035] The vector transformation unit is used to perform vector transformation on the text vector through the encoder in the risk identification model to obtain a text-enhanced vector representation.

[0036] The logistic regression unit is used to perform logistic regression calculations based on the text enhancement vector to obtain the matching result.

[0037] In some embodiments of this application, based on the above technical solutions, the account identification device further includes:

[0038] The training data acquisition module is used to acquire training data and corresponding matching results. The training data includes keyword sequence samples and second text samples from multiple accounts.

[0039] The training prediction module is used to feed the keyword sequence sample and the second text input sample into the model to be trained for prediction, and obtain the training prediction result.

[0040] The model training module is used to train the model to be trained based on the training prediction results and the matching results to obtain the risk identification model.

[0041] In some embodiments of this application, based on the above technical solutions, the account identification device further includes:

[0042] The first matching result acquisition module is used to acquire multiple matching results with multiple risk labels to be matched, wherein the matching result is a risk identifier indicating whether the account to be identified has a risk;

[0043] The transaction prohibition module is used to prohibit transaction operations related to the account to be identified if any matching result indicates that the account to be identified is at risk.

[0044] In some embodiments of this application, based on the above technical solutions, the account identification device further includes:

[0045] The risk level determination module is used to determine the risk level of the account to be identified based on the number of matching results indicating that the account to be identified is at risk if multiple matching results indicate that the account to be identified is at risk.

[0046] In some embodiments of this application, based on the above technical solutions, the account identification device further includes:

[0047] The second matching result acquisition module is used to acquire multiple matching results with multiple risk labels to be matched, wherein the matching result is a risk score indicating that the account to be identified has a risk;

[0048] The risk identification module is used to determine that the account to be identified is at risk if the risk score of any matching result is greater than the score threshold.

[0049] The risk alarm module is used to calculate the sum of the risk scores of the multiple matching results if the risk scores of each matching result are less than the scoring threshold, and to issue an alarm to the account to be identified if the sum of the risk scores is greater than the risk threshold.

[0050] In some embodiments of this application, based on the above technical solutions, the account identification device further includes:

[0051] The first text expression module is used to input the first text into the first text model to obtain the first text expression;

[0052] The second text expression module is used to input the keyword sequence into the second text model to obtain the second text expression, wherein the first text model and the second text model share model parameters;

[0053] A similarity calculation module is used to determine the similarity between the first text and the keyword sequence based on the first text expression and the second text expression;

[0054] The matching result determination module is used to determine the matching result between the account to be identified and the risk label to be matched based on the similarity.

[0055] In some embodiments of this application, based on the above technical solutions, the prediction module includes:

[0056] A word vector unit is used to map the target text to a word vector space to generate a feature image;

[0057] A convolutional unit is used to convolve the feature image through a convolutional layer to obtain multiple feature maps;

[0058] A pooling unit is used to perform pooling operations on the multiple feature maps through a pooling layer to obtain the maximum feature value of each feature map.

[0059] A fully connected unit is used to input multiple maximum feature values ​​into a fully connected layer to calculate the matching result.

[0060] According to one aspect of the embodiments of this application, an account identification device is provided. The account identification prompting device includes: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to perform the account identification method as described above by executing the executable instructions.

[0061] According to one aspect of the embodiments of this application, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the account identification method as described in the above technical solutions.

[0062] According to one aspect of the embodiments of this application, a computer program product or computer program is provided, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the account identification method provided in the various optional implementations described above.

[0063] In the embodiments of this application, the account-related text of the account to be identified and the keyword sequence corresponding to the risk label to be matched are first obtained. Then, the account-related text and the keyword sequence are concatenated and input into the risk identification model to predict the matching result of the account for the risk label. Since the keyword sequence is determined based on the account-related text of multiple accounts, adding the keyword sequence during prediction can add prior knowledge about the risk status to the risk prediction process, thereby improving the accuracy and coverage of risk label prediction, and thus reducing the false blocking rate of malicious accounts and the error rate of account handling operations.

[0064] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0065] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort. In the drawings:

[0066] Figure 1 The schematic diagram illustrates an exemplary system architecture of the technical solution of this application in an application scenario;

[0067] Figure 2 A flowchart illustrating an account identification method according to an embodiment of this application is shown;

[0068] Figure 3 A flowchart illustrating the keyword sequence acquisition method in an embodiment of this application is shown;

[0069] Figure 4 This is a schematic diagram of the overall keyword sequence extraction process in the embodiments of this application;

[0070] Figure 5 This is a schematic diagram of the risk identification model in the embodiments of this application;

[0071] Figure 6 This is a schematic diagram of the overall account identification process in the embodiments of this application;

[0072] Figure 7 A schematic block diagram illustrating the composition of the account identification device in an embodiment of this application is shown.

[0073] Figure 8 A schematic diagram of the structure of a computer system suitable for implementing the electronic device of the present application is shown. Detailed Implementation

[0074] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make this application more comprehensive and complete, and to fully convey the concept of the exemplary embodiments to those skilled in the art.

[0075] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a thorough understanding of embodiments of this application. However, those skilled in the art will recognize that the technical solutions of this application can be practiced without one or more of the specific details, or other methods, components, apparatuses, steps, etc., can be employed. In other instances, well-known methods, apparatuses, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of this application.

[0076] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0077] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0078] It should be understood that this application can be applied to online payment or mobile payment scenarios. Users accumulate a large amount of text information during their daily use of online payment applications, such as transaction records, personal signatures, group names, historical nicknames, and complaint texts. The solution in this application analyzes this text information to determine whether a user has engaged in illegal activities during online transactions, thereby identifying whether the user is a malicious user and taking appropriate action against them, such as banning accounts, prohibiting transactions, or freezing funds, thus ensuring the security of the online transaction environment.

[0079] Specifically, in mobile payment scenarios, servers can obtain information such as a user's past transactions, group names, historical nicknames, and complaint texts. For example, a user's account might frequently involve large inflows of funds, participate in multiple groups with names including words like "cashback" or "rebate," and have been involved in financial disputes with other users, resulting in complaints. The solution proposed in this application can predict this user's behavior based on prior knowledge gathered from similar information collected from other users, and identify the user's account as a malicious account, thereby taking action against the account's illegal activities.

[0080] In this application, the data user information acquisition technology involved, when applied to specific products or technologies in various embodiments of this application, ensures that the relevant data collection, use and processing processes comply with national laws and regulations. Before collecting user information, the information processing rules have been communicated and the individual consent of the target object has been obtained. User information is processed in strict accordance with legal and regulatory requirements and personal information processing rules, and technical measures are taken to ensure the security of relevant data.

[0081] The solution presented in this application can be applied to personal computers, servers, or server systems composed of multiple servers, and can be implemented using artificial intelligence. With the research and advancement of artificial intelligence technology, it is being researched and applied in various fields, such as smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, autonomous driving, drones, robots, smart healthcare, and smart customer service. It is believed that with technological development, artificial intelligence will be applied in even more fields and play an increasingly important role.

[0082] Artificial intelligence (AI) is the theory, methods, technology, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce a new kind of intelligent machine that can react in a way similar to human intelligence. AI studies the design principles and implementation methods of various intelligent machines, enabling them to possess the functions of perception, reasoning, and decision-making.

[0083] Artificial intelligence (AI) is a comprehensive discipline encompassing a wide range of fields, including both hardware and software technologies. Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies primarily include computer vision, speech processing, natural language processing, and machine learning / deep learning.

[0084] The solutions provided in this application involve artificial intelligence technologies such as natural language processing and machine learning.

[0085] Natural Language Processing (NLP) is an important field within computer science and artificial intelligence. It studies the theories and methods for enabling effective communication between humans and computers using natural language. NLP is a science that integrates linguistics, computer science, and mathematics. Therefore, research in this field involves natural language—the language people use in daily life—and thus it has a close relationship with linguistic research. NLP techniques typically include text processing, semantic understanding, machine translation, question answering, and knowledge graphs.

[0086] Machine learning (ML) is a multidisciplinary field involving probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory. It specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to endow computers with intelligence; its applications span all areas of artificial intelligence. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and instructional learning.

[0087] For ease of understanding, this application's solution will be used as an example in a mobile payment application scenario. Please refer to [link / reference]. Figure 1 , Figure 1 The schematic diagram illustrates an exemplary system architecture of the technical solution of this application in an application scenario.

[0088] Depend on Figure 1As can be seen, this scenario includes servers and terminal devices such as mobile phones and laptops. The mobile payment application's client runs on these terminal devices and communicates with the server. The server collects relevant text information about users based on the transaction behavior of the terminal devices, including complaint texts, transaction records, group names, historical nicknames, and personal signatures. Then, the server can predict the risk label matched to a user's account based on instructions from the mobile payment application administrator, scheduled tasks assigned by the administrator, or responses to user requests and complaints. Risk labels are predefined by the administrator or application developers based on experience and correspond to violations, such as illegal telecommunications activities, illegal pornography activities, illegal rebate activities, or gambling. After predicting the risk label of a user's account, the server can take relevant actions against the corresponding user account, such as freezing or banning it.

[0089] Figure 1 The server in this context can take various forms, such as a single server, a server cluster, or a cloud server. The server supports mobile payment services for clients, including various types of services such as fund transfers, logins, communication, and reporting / complaints. In this application, the server also identifies user account usage behavior based on user text information to determine the risk label matched to the user account, and takes relevant actions to manage the user account based on the risk label, thereby maintaining a healthy mobile payment transaction environment.

[0090] The communication between the client and the server can be conducted using any wired or wireless communication method, and this application does not impose any restrictions on this.

[0091] Understandable, Figure 1 The scenarios shown are merely examples of scenarios in which the solution of this application is applied. In actual application scenarios, other suitable network structures may be adopted, such as adding proxy servers and multi-level networks. This application does not impose any restrictions on this.

[0092] The solution proposed in this application can also be combined with a blockchain system. Specifically, the server can be a node in the blockchain system. Each time a user makes a transaction, the server adds the user's transaction history and related information to the blockchain system for storage. The solution proposed in this application can be executed by any node in the blockchain that has undergone data synchronization, thereby enabling a network of multiple servers to execute the solution provided in this application.

[0093] Blockchain is a novel application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and cryptographic algorithms. Essentially, a blockchain is a decentralized database, a chain of data blocks linked together using cryptographic methods. Each data block contains information about a batch of network transactions, used to verify the validity of the information (anti-counterfeiting) and generate the next block. A blockchain can include an underlying platform, a platform product and service layer, and an application service layer.

[0094] The underlying blockchain platform can include processing modules such as user management, basic services, smart contracts, and operations management. The user management module is responsible for managing the identification information of all blockchain participants, including maintaining public and private key generation (account management), key management, and maintaining the correspondence between target objects and blockchain addresses (access management). Furthermore, under authorization, it monitors and audits the transactions of certain target objects and provides risk control rule configuration (risk control audit). The basic services module is deployed on all blockchain node devices to verify the validity of business requests. After consensus is reached on valid requests, they are recorded in storage. For a new business request, the basic services first perform interface adaptation parsing and authentication (interface adaptation), and then encrypt the business information using a consensus algorithm (consensus management). After encryption, the data is transmitted completely and consistently to the shared ledger (network communication) and recorded and stored. The smart contract module is responsible for contract registration, issuance, triggering, and execution. Developers can define contract logic using a programming language and publish it to the blockchain (contract registration). According to the contract terms, the key or other events are invoked to trigger execution and complete the contract logic. It also provides functions for contract upgrades and cancellations. The operations management module is mainly responsible for deployment, configuration modification, contract settings, cloud adaptation, and real-time status visualization during product launch, such as alarms, network status management, and node device health status management.

[0095] The platform's product service layer provides the basic capabilities and implementation frameworks for typical applications. Developers can leverage these basic capabilities, along with the specific characteristics of their business needs, to implement blockchain-based business logic. The application service layer provides blockchain-based application services to business stakeholders.

[0096] The technical solution provided in this application will be described in detail below with reference to specific embodiments.

[0097] Please see Figure 2 , Figure 2 A flowchart illustrating an account identification method according to an embodiment of this application is shown. This method can be applied to the server described above and executed by a background program on the server. The method of this embodiment may include the following steps S201 to S204:

[0098] Step S201: Obtain the first text corresponding to the account to be identified, wherein the first text is obtained by concatenating at least one text information associated with the account to be identified.

[0099] In this embodiment, the account identification device obtains a first text from the historical information of the account to be identified. The first text is obtained by concatenating at least one text information associated with the account to be identified. Specifically, the at least one text information includes one or more of the following: complaint text, transaction text, group name, historical nickname, and personal signature. The account identification device concatenates the above at least one text information to form the first text. The specific concatenation method can be: complaint text | transaction text | group name | historical nickname | personal signature. It is understood that the account to be identified does not necessarily include all of the above types of information, and information not included in the account to be identified can be omitted. The action of the account identification device obtaining at least one text information to concatenate the first text can be performed during the execution of this step, or it can be pre-executed by the account identification device and stored in the server, and the first text corresponding to the account to be identified can be read from the server during the execution of this step.

[0100] Step S202: Obtain the keyword sequence associated with the risk label to be matched, wherein the keyword sequence includes multiple keywords that are semantically related to the risk label to be matched.

[0101] In this embodiment, the account identification device obtains a keyword sequence associated with the risk label to be matched. The keyword sequence includes multiple keywords that are semantically relevant to the risk label to be matched. Semantic relevance refers to the semantic connection between the keywords and the risk label to be matched. For example, assuming the risk label to be matched is "illegal rebate activities," the keywords might be related terms such as "rebate," "cashback," "commission," "money back," and "rebate." The risk label to be matched is selected from predefined risk labels, each corresponding to a violation or illegal act. The risk label to be matched can be determined based on the conditions that trigger the execution of this method. For example, if a user files a complaint against an account to be identified, the user can select the label they believe in from the complaint request, and the account identification device will designate the risk label in the complaint request as the risk label to be matched. The risk label to be matched can also be determined by iterating through all risk labels. For example, if there are 10 risk labels, the risk label to be matched can be determined sequentially from these 10 risk labels.

[0102] Keyword sequences can be pre-generated by the account identification device and stored on the server, and retrieved from the server using the risk label as an identifier when needed, or they can be generated directly when needed.

[0103] Step S203: The first text and the keyword sequence are concatenated to obtain the target text.

[0104] In this embodiment, the account recognition device concatenates the first text and the keyword sequence together to obtain the target text. Specifically, the concatenation order is usually in the order of keyword sequence + first text. During the concatenation operation, a delimiter can be added to separate the first text and the keyword sequence.

[0105] Step S204: Predict the matching result between the account to be identified and the risk label to be matched based on the target text.

[0106] In this embodiment, the account recognition device predicts the matching result between the account to be identified and the tag to be matched based on the target text. Depending on the testing method, the matching result can directly indicate whether the account to be identified matches the tag, or it can indicate the probability of the account to be identified matching the tag. The prediction can be performed using sentence matching. For example, the account recognition device can identify and count the keywords appearing in the first text based on the keyword sequence in the target text. Then, it can perform statistical analysis based on the number, position, frequency, proportion in the first text, and relevance to other text in the first text of the identified keywords. It can then judge according to preset standards for each attribute to obtain a comprehensive judgment result, or score the matching situation as the matching result. The prediction method can also use machine learning models or vector machines to perform sentence analysis on the target text to obtain the degree of relevance between the first text and the keywords in the keyword sequence, and then determine the matching result according to the degree of relevance.

[0107] After obtaining the matching results, the server can take further actions on the user account based on the matching results. For example, it can suspend a transaction suspected of illegal activity, freeze funds, and issue an alert on the risk status of the account.

[0108] In the embodiments of this application, the account-related text of the account to be identified and the keyword sequence corresponding to the risk label to be matched are first obtained. Then, the account-related text and the keyword sequence are concatenated and input into the risk identification model to predict the matching result of the account for the risk label. Since the keyword sequence is determined based on the account-related text of multiple accounts, adding the keyword sequence during prediction can add prior knowledge about the risk status to the risk prediction process, thereby improving the accuracy and coverage of risk label prediction, and thus reducing the false blocking rate of malicious accounts and the error rate of account handling operations.

[0109] In one embodiment of this application, such as Figure 3As shown, in order to obtain the keyword sequence, this method also includes the following steps S301 to S305, which are described in detail below:

[0110] Step S301: Obtain multiple second texts for multiple accounts, wherein the second text for each account is obtained by concatenating at least one text information associated with each account.

[0111] In this embodiment, the account identification device acquires multiple second texts for multiple accounts, wherein the second text for each account is obtained by concatenating at least one text information associated with each account. Specifically, the content of the second text is the same as that of the first text, except that the content of the second text can be generated based on historical information from other accounts. The second text can be information pre-generated and stored in a server, or it can be generated by reading at least one corresponding text information and concatenating them when acquiring the second text.

[0112] Multiple accounts typically refer to all users in the scenario in which this method is applied, including a large number of ordinary users and a relatively smaller number of malicious users. For example, in a mobile payment application scenario, multiple accounts could include all registered accounts of the mobile payment application. In one embodiment, the scope of multiple accounts can be defined as highly active users, that is, by statistically analyzing information such as the frequency of account use or transaction behavior, accounts with high usage frequency or numerous transactions are selected as the scope of accounts for generating keyword sequences.

[0113] Step S302: Obtain a set of related words, wherein the set of related words includes at least one related word, and the related words are determined based on the frequency of occurrence of words in multiple second texts.

[0114] Specifically, the relevant word set is generated based on the secondary text of a large number of user accounts. Typically, the words included in the relevant word set are generated based on the secondary text of all accounts. A word is identified as a relevant word if it appears more than a certain number of times in all the secondary texts. For example, if a word appears more than 5 times in all the secondary texts, it can be identified as a relevant word. In one embodiment, the length and quantity of the secondary texts in which the word appears can also be considered. For example, in addition to the 5 occurrences, the length of the secondary text in which the word appears is also considered. For instance, if word A appears 10 times and the total length of the secondary text containing word A is 20,000 characters, while word B appears 5 times and the total length of the secondary text containing word B is 2,000 characters, then the ratio of the frequency of word B to the text length is significantly higher than that of word A. If the set ratio threshold is lower than that of word B but higher than that of word A, word B can be selected as a relevant word while word A will not.

[0115] Step S303: Extract features from multiple second texts based on the relevant word set to obtain keyword feature vectors.

[0116] Feature extraction involves transforming multiple second-text documents into a multi-dimensional feature vector based on the occurrence of words from a related word set within those documents. Each element in the feature vector corresponds to a related word in a single second-text document. Depending on the method used to calculate the feature values, the values ​​of the elements can represent the importance of the word, its relevance, or other meanings derived from the calculation method. For example, if N second-text documents are obtained, and the related word set contains V words, then the keyword feature vector could be an N... A vector of V.

[0117] Step S304: For the risk label to be matched, classify the keyword feature vector using a label classification model to obtain at least one keyword.

[0118] Specifically, a tag classification model is a machine learning model used to determine whether related words are relevant to risk tags to be matched. It can be implemented using a decision tree model, determining the relevance between related words and corresponding risk tags based on keyword feature vectors. The tag classification model can be a binary classification model, where each different risk tag has its own corresponding tag classification model, and the corresponding model is used to retrieve keywords. Alternatively, it can be a multi-class classification model, processing keyword feature vectors to output the probability that related words are keywords for each risk tag. In this way, keywords corresponding to multiple risk tags can be retrieved simultaneously based on a pre-set threshold.

[0119] After obtaining the keywords corresponding to the risk tags to be matched, the keywords can be sorted according to their importance, and a predetermined number of keywords can be selected to filter out the most important ones. Importance can be ranked by calculating scores or probabilities using a tag classification model, by the number of times or frequency of keyword occurrences, or by semantic relevance to the risk tags, etc., without any restrictions here.

[0120] Step S305: Combine the keywords into a keyword sequence.

[0121] The account identification device arranges and concatenates the acquired keywords to form a keyword sequence. Specifically, the position of keywords in the keyword sequence can be arranged according to the importance of each keyword. Separators can be added between each keyword to facilitate subsequent processing and avoid the need for word segmentation due to differences in keyword length. For example, if the keywords corresponding to the tag of illegal rebate activities include rebate, cashback, commission rebate, money back, rebate, cash back, profit, subscription, advance payment, invite you to join the group, short cycle, five times, greed for small gains, then the corresponding keyword sequence could be "rebate #cashback #commission rebate #money back #rebate #cash back #profit #subscription #advance payment #invite you to join the group #short cycle #five times #greed for small gains".

[0122] It is understandable that the above process is for obtaining a keyword sequence for a single tag to be matched. If there are multiple risky tags to be matched, the above process is repeated for each risky tag to be matched. That is, if there are M tags, the above process can be repeated M times to obtain M keyword sequences.

[0123] In the embodiments of this application, a keyword sequence for matching risk labels is generated from the second text of multiple accounts based on a set of related words. The keyword sequence is generated based on the relationship between words and risk labels, and also takes into account the importance of words, so that the keyword sequence can fully reflect the specific situation corresponding to the risk label, thereby forming prior knowledge for identifying risk status, which is conducive to improving the accuracy of account risk identification.

[0124] In one embodiment of this application, in order to calculate the keyword feature vector, the above step S303, which extracts features from multiple second texts based on the relevant word set, further includes the following steps, which are described in detail below:

[0125] The intermediate feature vector is determined based on the frequency of occurrence of related words in multiple second texts from the related word set.

[0126] Singular value decomposition is performed on the intermediate feature vector to obtain the keyword feature vector.

[0127] Specifically, determining the intermediate feature vector based on the frequency of related words across multiple secondary texts can be achieved using the term frequency-inverse document frequency (TDF) method. Term frequency (TF) refers to the frequency of a related word across multiple secondary texts, which can be calculated by normalizing the word count. Inverse document frequency (IVF) is a measure of the general importance of related words, calculated by dividing the total number of secondary texts by the number of texts containing the related word. The product of TF and IVF is then calculated to obtain the feature value of the related word. This calculation is performed for each related word to obtain the intermediate feature vector. Using the TF-IV method, a higher feature value can be obtained when a word has a high frequency in a few secondary texts but a low overall frequency across multiple secondary texts. This allows filtering out words common among normal users while retaining words with higher frequencies among a relatively small number of aggressive users.

[0128] Since the number of related words and the amount of second text are usually large, the dimensionality of the intermediate feature vector will be relatively high. Therefore, singular value decomposition (SVD) is performed on the intermediate feature vector to obtain the keyword feature sequence. Singular value decomposition is a matrix factorization method that decomposes the feature vector into the product of three matrices, thereby mapping the high-dimensional keyword feature sequence to a low-dimensional space.

[0129] The following is a general explanation of the keyword sequence extraction process. For a clearer understanding, please refer to [link to relevant documentation]. Figure 4 , Figure 4 This is a schematic diagram illustrating the overall keyword sequence extraction process in an embodiment of this application. For example... Figure 4 As shown, firstly, after obtaining the second text of multiple user accounts, the obtained second text is input into the feature extraction module to calculate the feature vector. Then, the feature vector is fed into the singular value decomposition module for dimensionality reduction. Next, the dimensionality reduction result is input into the label classification model to obtain the words associated with the risk label to be matched. Subsequently, based on the feature importance of the words output by the label classification model, the most important keywords among the words associated with the risk label to be matched can be determined, and thus concatenated into a keyword sequence.

[0130] In the embodiments of this application, features are extracted based on the frequency of occurrence of related words in multiple second texts, and then singular value decomposition is performed on the extracted features to reduce dimensionality, which can reduce the computational load of the scheme and thus improve the overall prediction efficiency.

[0131] In one embodiment of this application, in order to calculate the keyword feature vector, the above step S303, which extracts features from multiple second texts based on the relevant word set, further includes the following steps, which are described in detail below:

[0132] For each of the multiple second texts, determine the frequency of occurrence of all related words in the related word set within the second text;

[0133] The text weights of all relevant words in the second text are determined based on the frequency of occurrence of all relevant words in the second text, the text length of the second text, and the average text length of multiple second texts.

[0134] The keyword feature vector is determined based on the text weights of all relevant words in multiple second texts.

[0135] For all relevant words, their frequency of occurrence in each secondary text can be counted. Then, based on the frequency of occurrence of all relevant words in the secondary text, the text length of the secondary text, and the average text length of multiple secondary texts, the text weight of all relevant words in the secondary text is determined. Specifically, for a specific relevant word j, its frequency of occurrence in a specific secondary text i is counted. Based on the frequency of occurrence of specific relevant word j in secondary text i, the text length of secondary text i, and the average text length of all secondary texts, the text weight of relevant word j in secondary text i is determined. Calculating the text weight of all relevant words for secondary text i yields the feature vector of the secondary text, and calculating the text weight of all relevant words for all secondary texts yields the feature matrix, which is the keyword feature vector.

[0136] Specifically, the text weight of a related word j in a second text i can be calculated using the following formula:

[0137]

[0138] Among them, w ij The related word j is in the text of the second text i, f ij is the frequency of related word j in second text i. N is the total number of occurrences in all second texts. j This is the number of second texts in which the related word j appears at least once. N / n j It refers to the frequency of reverse document lookup. dl It is the length of the second text. avg_dl It is the average document length of all second-order texts.

[0139] For all related words, the text weights are calculated using the formula above, thus obtaining the keyword feature vectors. For example, if there are N secondary texts and V related words, then N... The keyword feature vector is obtained through V calculations.

[0140] In the embodiments of this application, the length of the second text is taken into account when calculating the keyword feature vector, thereby avoiding word frequency differences caused by the difference in length of the second text, thus improving the expressive effect of feature extraction.

[0141] In one embodiment of this application, in order to obtain the set of related words, step S302 above, obtaining the set of related words, further includes the following steps, which are described in detail below:

[0142] Segment multiple second texts from multiple accounts to obtain a word set, wherein the word set includes at least one word;

[0143] The relevant word set is determined based on the frequency of occurrence of each word in the word set in multiple second texts.

[0144] Here, "multiple accounts" typically refers to all accounts in the application scenario. The scope of the multiple second texts used to determine the set of related words can be larger than the scope of the multiple second texts used to determine the keyword sequence in the above embodiments. That is, the scope of the multiple second texts used to determine the set of related words can be all user accounts, while the scope of the multiple second texts used to determine the keyword sequence can be a subset of accounts selected from all users, such as active accounts from the last 5 years.

[0145] Secondary text can be segmented using methods such as Jieba. After segmenting all secondary texts and removing duplicate words, a word set is obtained. Then, filtering is performed based on the frequency of each word across multiple secondary texts to obtain a related word set. For example, words that appear more than 5 times in multiple secondary texts can be selected to form a related word set.

[0146] In this embodiment, a method for obtaining a set of related words is provided. By segmenting multiple second texts and then performing word frequency filtering, low-frequency words with weak distinguishing ability can be filtered out, thereby improving the feasibility of the solution.

[0147] In one embodiment of this application, in order to predict the matching result, step S204 above, which predicts the matching result between the account to be identified and the risk label to be matched based on the target text, includes the following steps, which are described in detail below:

[0148] The target text is transformed into a text vector by using the attention layer in the risk identification model;

[0149] The text vector is transformed by the encoder in the risk identification model to obtain the text-enhanced vector representation;

[0150] Logistic regression is performed based on the text augmentation vectors to obtain the matching results.

[0151] Specifically, after the target text is input into the risk identification model, it is first converted into a feature vector. This feature vector then passes through an attention layer to consider different fusion methods of the text, thereby semantically enhancing it and obtaining a text vector. Next, the text vector is input into multiple encoders in the model for vector transformation calculations, specifically including residual connections, normalization, and two linear transformations, thus further semantically enhancing the text vector. Risk identification models typically include multiple encoders. These encoders execute sequentially, with the output of the previous encoder serving as the input to the next. The output of the last encoder is the text-enhanced vector representation.

[0152] Finally, by performing logistic regression on the text augmentation vector representation using the sigmoid function, the final matching result can be obtained.

[0153] Risk identification models can be implemented using machine learning models such as BERT, GPT, and ELMo. For ease of explanation, we will use the BERT model as an example; please refer to [link to relevant documentation]. Figure 5 , Figure 5 This is a schematic diagram of the risk identification model in an embodiment of this application. Figure 5 As shown, when the target text is concatenated, the identifier "[CLS]" is added at the beginning, and the identifier "[SEP]" is added at the end of the keyword sequence and the first text. When the target text is input into the model for calculation, the output corresponding to the identifier "[CLS]" is 0 or 1, which is the matching result.

[0154] It is understood that in this embodiment, the sigmoid function is used as an example, so the model outputs a binary classification result of 0 or 1. In other embodiments, other functions such as the softmax function can also be used to input results such as matching probabilities.

[0155] In the embodiments of this application, a method for prediction based on target text is provided. The target text is predicted by a risk identification model to obtain matching results, thereby improving the feasibility of the solution.

[0156] In one embodiment of this application, in order to obtain a risk identification model, the application further includes the following steps, which are described in detail below:

[0157] Obtain training data and corresponding matching results. The training data includes keyword sequence samples and second text samples from multiple accounts.

[0158] The keyword sequence samples and the second text input samples are fed into the model to be trained for prediction, and the training prediction results are obtained.

[0159] Based on the training prediction results and matching results, the model to be trained is trained to obtain the risk identification model.

[0160] The keyword sequence samples in the training data are generated based on the historical text of the accounts, and the second text samples from multiple accounts can be extracted from the identified malicious and normal accounts. The concatenation method between keywords in the keyword sequence samples and the concatenation order between the keyword sequence samples and the second text samples in the training data are the same as those described in the above embodiments.

[0161] The training data is input into the model to be trained for prediction, thus obtaining the training prediction result. The model to be trained can be constructed by adding a small number of neural network layers to the output of a pre-trained BERT model. Based on the training prediction result and the actual matching result corresponding to the input training data, the parameters of the small number of neural network layers added can be adjusted according to the loss function to obtain the risk identification model.

[0162] In the embodiments of this application, a training method for the risk identification model is provided, specifically introducing a method of training the risk identification model using keyword sequence samples, second text samples, and corresponding matching results, thereby improving the feasibility of the solution.

[0163] Based on the risk identification model, the overall account identification process of this application can be found in [reference needed]. Figure 6 , Figure 6 This is a schematic diagram of the overall account identification process in this application embodiment. The user's text includes the first text of the account to be identified and the second text of all accounts. The second text is input into a keyword sequence model to obtain keyword sequences corresponding to each risk label. The obtained keyword sequences and the first text of the account to be identified are then input into a risk identification model to obtain the final identification result. It is understood that the determination of the keyword sequence is independent of the first text identification process; that is, the keyword sequence determination process, including the determination of the related word set, can be executed independently, and the results are stored on the server for subsequent account identification.

[0164] In one embodiment of this application, after obtaining the matching result, the method of this application may further include the following steps, which are described in detail below:

[0165] Obtain multiple matching results with multiple risk labels to be matched, where the matching result is a risk label indicating whether the account to be identified has a risk;

[0166] If any matching result indicates that the account to be identified is at risk, then transaction operations related to the account to be identified are prohibited.

[0167] Specifically, for the account to be identified, the account identification method described in the above embodiments can be executed for multiple risk labels to be matched, thereby obtaining the matching results between the account to be identified and multiple risk labels to be matched. The matching result can be a binary classification result of 0 or 1, where 1 indicates that the account to be identified matches the risk label to be matched, that is, the account to be identified has the violation represented by the risk label to be matched, and 0 indicates that there is no match.

[0168] If any matching result indicates that the account to be identified is at risk, that is, any matching result is 1, it can be determined that the account to be identified has engaged in illegal activities, thereby prohibiting transaction operations related to the account to be identified, such as deposit and withdrawal activities from the account to be identified.

[0169] In the embodiments of this application, a method for prohibiting the transaction operations of an account is provided. The method determines the risk status of the account to be identified based on multiple identification results, and if any matching result indicates that the account to be identified is at risk, the account is considered to be at risk. This ensures the accurate identification of malicious accounts and improves the coverage of malicious account identification.

[0170] In one embodiment of this application, in order to further determine the risk level of an account, after obtaining multiple matching results with multiple risk tags to be matched, the method of this application may further include the following steps, which are described in detail below:

[0171] If multiple matching results indicate that the account to be identified is at risk, the risk level of the account to be identified is determined based on the number of matching results indicating that the account to be identified is at risk.

[0172] Specifically, if multiple matching results indicate that the account to be identified poses a risk, the number of matching results can be counted, that is, the number of risk tags that match the account to be identified. The more matching risk tags there are, the more violations the user account has been involved in, which means its risk level is higher. Correspondingly, the higher the risk level of a user account, the more severe the corresponding actions can be taken against the user account, such as from prohibiting trading to permanently banning the account.

[0173] In the embodiments of this application, the risk level of a user account is determined by the number of risk tags to be matched with the account, so that different actions can be taken on the account according to the risk level, thereby improving the flexibility of the solution.

[0174] In one embodiment of this application, in order to identify risks associated with user accounts, the method of this application may further include the following steps, which are described in detail below:

[0175] Obtain multiple matching results with multiple risk labels to be matched, where the matching result is a risk score indicating that the account to be identified has a risk.

[0176] If the risk score of any matching result is greater than the scoring threshold, the account to be identified is determined to be at risk.

[0177] If the risk scores of each matching result are all less than the scoring threshold, the sum of the risk scores of multiple matching results is calculated. If the sum of the risk scores is greater than the risk threshold, an alert is issued to the account to be identified.

[0178] Specifically, in this embodiment, the risk score is a decimal ranging from 0 to 1, and its value represents the probability that the account to be identified matches the risk label to be matched. Multiple matching results are calculated independently. A scoring threshold can be set for the risk score; when the risk score is greater than the threshold, the account to be identified is determined to be risky. It is understood that different scoring thresholds can be set for different risk labels.

[0179] In this embodiment, if the risk score of any matching result is greater than the corresponding scoring threshold, the account to be identified is determined to be at risk, and further corresponding processing can be carried out on the account. Furthermore, if the risk scores of all matching results are less than the scoring threshold, the account to be identified is determined to be risk-free. In this case, the sum of the risk scores of multiple matching results can be calculated. The sum of the risk scores is compared with the risk threshold. If the sum of the risk scores is greater than the risk threshold, it indicates that although the account has not reached the level of being at risk, it still has the possibility of involving illegal operations. Therefore, a corresponding alert can be sent to the user account to regulate the user's behavior. For example, for an account to be identified, its risk scores matching three risk tags are 0.4, 0.3, and 0.4 respectively. Assuming the scoring threshold is 0.5 for each, the account to be identified does not match any of these three risk tags. However, if the risk threshold is 1, since the sum of the three risk scores, 1.1, is greater than 1, an alert message can be sent to the account to be identified to remind the user that they may be threatened by related risks.

[0180] In the embodiments of this application, a method for risk identification of accounts is provided. The method measures the matching probability between the account to be identified and the risk label by using a risk score, and further alerts the account based on the sum of the risk scores, thereby improving the coverage of risk identification, providing early warnings for potential non-compliant accounts, reducing the possibility of users' accounts being penalized due to non-subjective non-compliant operations, and thus improving the user experience.

[0181] In one embodiment of this application, another method can also be used for prediction. Specifically, after obtaining the keyword sequence associated with the risk label to be matched in step S202 above, the method of this application may further include the following steps, which are described in detail below:

[0182] The first text is input into the first text model to obtain the first text representation;

[0183] The keyword sequence is input into the second text model to obtain the second text representation, wherein the first text model and the second text model share the model parameters;

[0184] Based on the first text expression and the second text expression, determine the similarity between the first text and the keyword sequence;

[0185] Based on similarity, determine the matching result between the account to be identified and the risk label to be matched.

[0186] In this embodiment, a Siamese model is used to predict the similarity between the first text and the keyword sequence. Specifically, the Siamese model includes a first text model and a second text model, which have identical structures and share parameters during training. The first and second text models can be implemented using BERT models, and the text information received by both models is converted into corresponding text embeddings. Specifically, the first text is input into the first text model to obtain the first text representation, and the keyword sequence is input into the second text model to obtain the second text representation. Both the first and second text representations are in the form of feature vectors, and their calculation processes can be performed in parallel. Then, the first and second text representations are input into the similarity module of the Siamese model for similarity calculation, and the calculation result is then input into the softmax function for logistic regression calculation to obtain the final similarity.

[0187] For similarity, a similarity threshold can be preset, and the calculated similarity can be compared with the similarity threshold to determine the matching result. Alternatively, after calculating the similarity of a large number of accounts, obvious outliers can be selected as the matching result. For example, if the similarity of a large number of accounts is 0.02, while the similarity of a small number of users is 0.2, then it can be determined that the accounts of this small number of users match the risk label to be matched.

[0188] In the embodiments of this application, a method is provided to predict the matching result between an account to be identified and a risk label to be matched. The method uses a Siamese model to determine the similarity between the first text and the keyword sequence, thereby determining the matching result between the account to be identified and the risk label to be matched. Since the prior knowledge of the keyword sequence is combined, the identification of the risk status of the account can be more accurate compared to relying on the semantics of the first text itself for analysis.

[0189] In one embodiment of this application, a matching result can also be calculated. Step S204, which predicts the matching result between the account to be identified and the risk label to be matched based on the target text, may include the following steps, detailed below:

[0190] Map the target text to the word vector space to generate a feature image;

[0191] Multiple feature maps are obtained by convolving the feature image through a convolutional layer;

[0192] The pooling layer performs pooling operations on multiple feature maps to obtain the maximum eigenvalue of each feature map.

[0193] Multiple maximum feature values ​​are input into a fully connected layer to calculate the matching results.

[0194] In this embodiment, TextCNN is used to calculate the matching result between the account to be identified and the risk label to be matched. Specifically, the target text input to the model is first fed into the input layer to be mapped to the word vector space, generating a feature image. The feature image is usually in the form of a two-dimensional matrix, and the transformation can be achieved using word embedding methods such as word2vec and glove.

[0195] Following the feature images, convolutional layers are used to convolve them, resulting in multiple feature maps. In this step, the width of the convolutional kernel is the same as the dimension of the word vectors, and the kernel only moves in the height direction. Therefore, each time the kernel slides over a complete word, it ensures that words are treated as the smallest unit of text. Each convolutional operation is equivalent to extracting a feature vector. By defining different windows, different feature vectors can be extracted, forming the output of the convolutional layer and thus producing multiple feature maps.

[0196] After obtaining multiple feature maps, they are input into a pooling layer for pooling. In this step, max pooling is used, which involves selecting the largest feature from each feature map generated by the sliding window and then concatenating these features to form a vector representation. In one embodiment, other pooling methods can also be used, such as selecting the K largest features from each feature map and averaging each dimension of the feature vector. The overall effect is to obtain a fixed-length vector representation from sentences of different lengths through pooling.

[0197] The pooled result is fed into the fully connected layer, and the probability of the account to be identified matching the risk label to be matched can be output using the Softmax activation function, or the binary classification result can be output to directly represent the matching result.

[0198] In the embodiments of this application, the target text is processed by modules such as convolutional layers, pooling layers, and fully connected layers to calculate the matching results. Compared with the large amount of computation of pre-trained models, the method of this embodiment can significantly reduce the number of model parameters and correspondingly reduce the requirements for computing devices, thereby improving the speed of model training and computation, and ultimately improving the running speed of the solution and reducing the implementation cost of the solution.

[0199] It should be noted that although the steps of the method in this application are described in a specific order in the accompanying drawings, this does not require or imply that the steps must be performed in that specific order, or that all the steps shown must be performed to achieve the desired result. Additional or alternative steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0200] The following describes the implementation of the apparatus of this application, which can be used to perform the account identification method in the above embodiments of this application. Figure 7 A schematic block diagram illustrating the composition of an account identification device in an embodiment of this application is shown. For example... Figure 7 As shown, the account identification device 400 mainly includes:

[0201] The text acquisition module 410 is used to acquire the first text corresponding to the account to be identified, wherein the first text is obtained by concatenating at least one text information associated with the account to be identified;

[0202] The sequence acquisition module 420 is used to acquire a keyword sequence associated with the risk label to be matched, wherein the keyword sequence includes multiple keywords that are semantically related to the risk label to be matched;

[0203] The splicing module 430 is used to splice the first text and the keyword sequence to obtain the target text;

[0204] The prediction module 440 is used to predict the matching results between the account to be identified and the risk label to be matched based on the target text.

[0205] In some embodiments of this application, based on the above technical solutions, the account identification device 400 further includes:

[0206] The second text acquisition module is used to acquire multiple second texts from multiple accounts, wherein the second text of each account is obtained by concatenating at least one text information associated with each account;

[0207] The related word set acquisition module is used to acquire a related word set, wherein the related word set includes at least one related word, and the related word is determined based on the frequency of occurrence of words in multiple second texts;

[0208] The feature volume module is used to extract features from multiple second texts based on the relevant word set to obtain keyword feature vectors.

[0209] The classification module is used to classify the keyword feature vectors based on the risk labels to be matched using a label classification model, thereby obtaining at least one keyword.

[0210] The combination module is used to combine keywords into keyword sequences.

[0211] In some embodiments of this application, based on the above technical solutions, the feature volume module includes:

[0212] The feature vector determination unit is used to determine the intermediate feature vector based on the frequency of occurrence of related words in multiple second texts in the related word set;

[0213] The singular value decomposition unit is used to perform singular value decomposition on the intermediate feature vector to obtain the keyword feature vector.

[0214] In some embodiments of this application, based on the above technical solutions, the feature volume module includes:

[0215] The frequency determination unit is used to determine the frequency of occurrence of all related words in the related word set in each of the multiple second texts;

[0216] The text weight determination unit is used to determine the text weight of all related words in the second text based on the frequency of occurrence of all related words in the second text, the text length of the second text, and the average text length of multiple second texts.

[0217] The keyword vector determination unit is used to determine the keyword feature vector based on the text weights of all relevant words in multiple second texts.

[0218] In some embodiments of this application, based on the above technical solutions, the related word set acquisition module includes:

[0219] The word segmentation unit is used to segment multiple second texts from multiple accounts to obtain a word set, which includes at least one word.

[0220] The set determination unit is used to determine the related word set based on the frequency of occurrence of each word in the word set in multiple second texts.

[0221] In some embodiments of this application, based on the above technical solutions, the prediction module 440 includes:

[0222] The text conversion unit is used to convert the target text into a text vector by passing the attention layer in the risk identification model.

[0223] The vector transformation unit is used to perform vector transformation on the text vector through the encoder in the risk identification model to obtain the text-enhanced vector representation;

[0224] The logistic regression unit is used to perform logistic regression calculations based on the text augmentation vectors to obtain matching results.

[0225] In some embodiments of this application, based on the above technical solutions, the account identification device 400 further includes:

[0226] The training data acquisition module is used to acquire training data and corresponding matching results. The training data includes keyword sequence samples and second text samples from multiple accounts.

[0227] The training prediction module is used to feed the keyword sequence samples and the second text input samples into the model to be trained for prediction, and obtain the training prediction results.

[0228] The model training module is used to train the model to be trained based on the training prediction results and matching results, so as to obtain the risk identification model.

[0229] In some embodiments of this application, based on the above technical solutions, the account identification device 400 further includes:

[0230] The first matching result acquisition module is used to acquire multiple matching results with multiple risk labels to be matched, wherein the matching result is a risk label indicating whether the account to be identified has a risk;

[0231] The transaction prohibition module is used to prohibit transaction operations related to the account to be identified if any matching result indicates that the account to be identified is at risk.

[0232] In some embodiments of this application, based on the above technical solutions, the account identification device 400 further includes:

[0233] The risk level determination module is used to determine the risk level of an account to be identified based on the number of matching results indicating that the account to be identified is at risk if multiple matching results indicate that the account to be identified is at risk.

[0234] In some embodiments of this application, based on the above technical solutions, the account identification device 400 further includes:

[0235] The second matching result acquisition module is used to acquire multiple matching results with multiple risk labels to be matched, wherein the matching result is a risk score indicating that the account to be identified has risks;

[0236] The risk identification module is used to determine that the account to be identified is at risk if the risk score of any matching result is greater than the score threshold.

[0237] The risk alert module is used to calculate the sum of risk scores for multiple matching results if the risk scores of each matching result are less than the scoring threshold. If the sum of risk scores is greater than the risk threshold, an alert is issued to the account to be identified.

[0238] In some embodiments of this application, based on the above technical solutions, the account identification device 400 further includes:

[0239] The first text expression module is used to input the first text into the first text model to obtain the first text expression;

[0240] The second text expression module is used to input the keyword sequence into the second text model to obtain the second text expression, wherein the first text model and the second text model share model parameters;

[0241] The similarity calculation module is used to determine the similarity between the first text and the keyword sequence based on the first text expression and the second text expression;

[0242] The matching result determination module is used to determine the matching result between the account to be identified and the risk label to be matched based on similarity.

[0243] In some embodiments of this application, based on the above technical solutions, the prediction module 440 includes:

[0244] Word vector units are used to map target text to a word vector space to generate feature images;

[0245] A convolutional unit is used to convolve a feature image through a convolutional layer to obtain multiple feature maps;

[0246] The pooling unit is used to perform pooling operations on multiple feature maps through the pooling layer to obtain the maximum eigenvalue of each feature map.

[0247] Fully connected units are used to input multiple maximum feature values ​​into the fully connected layer to calculate the matching results.

[0248] It should be noted that the apparatus provided in the above embodiments and the method provided in the above embodiments belong to the same concept, and the specific way in which each module performs the operation has been described in detail in the method embodiments, and will not be repeated here.

[0249] Figure 8 A schematic diagram of the structure of a computer system suitable for implementing the electronic device of the present application is shown.

[0250] It should be noted that, Figure 8 The computer system 500 of the electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0251] like Figure 8 As shown, the computer system 500 includes a Central Processing Unit (CPU) 501, which can perform various appropriate actions and processes based on programs stored in Read-Only Memory (ROM) 502 or programs loaded from storage section 508 into Random Access Memory (RAM) 503. The RAM 503 also stores various programs and data required for system operation. The CPU 501, ROM 502, and RAM 503 are interconnected via a bus 504. An Input / Output (I / O) interface 505 is also connected to the bus 504.

[0252] The following components are connected to I / O interface 505: an input section 506 including a keyboard, mouse, etc.; an output section 507 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 508 including a hard disk, etc.; and a communication section 509 including a network interface card such as a LAN (Local Area Network) card, modem, etc. The communication section 509 performs communication processing via a network such as the Internet. A drive 510 is also connected to I / O interface 505 as needed. Removable media 511, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 510 as needed so that computer programs read from them can be installed into storage section 508 as needed.

[0253] Specifically, according to embodiments of this application, the processes described in the various method flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 509, and / or installed from removable medium 511. When the computer program is executed by central processing unit (CPU) 501, it performs various functions defined in the system of this application.

[0254] It should be noted that the computer-readable medium shown in the embodiments of this application can be a computer-readable signal medium, a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such transmitted data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wired, etc., or any suitable combination thereof.

[0255] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0256] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to the embodiments of this application, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0257] Through the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, touch terminal, or network device, etc.) to execute the methods according to the embodiments of this application.

[0258] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein.

[0259] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A method for account identification, characterized in that, include: Obtain the first text corresponding to the account to be identified, wherein the first text is obtained by concatenating at least one text information associated with the account to be identified; Obtain multiple second texts from multiple accounts, wherein the second text of each account is obtained by concatenating at least one text information associated with each account; Obtain a set of related words, wherein the set of related words includes at least one related word, and the related word is determined based on the frequency of occurrence of words in the plurality of second texts; Based on the set of related words, feature extraction is performed on the multiple second texts to obtain keyword feature vectors; For the risk tags to be matched, the keyword feature vectors are classified using a tag classification model to obtain at least one keyword; The keywords are combined into the keyword sequence; Obtain a keyword sequence associated with the risk label to be matched, wherein the keyword sequence includes multiple keywords that are semantically related to the risk label to be matched; The first text and the keyword sequence are concatenated to obtain the target text; Predict the matching result between the account to be identified and the risk label to be matched based on the target text.

2. The method according to claim 1, characterized in that, The step of extracting features from the plurality of second texts based on the set of related words includes: The intermediate feature vector is determined based on the frequency of occurrence of related words in the multiple second texts in the related word set; Singular value decomposition is performed on the intermediate feature vector to obtain the keyword feature vector.

3. The method according to claim 1, characterized in that, The step of extracting features from the plurality of second texts based on the set of related words includes: For each of the plurality of second texts, determine the frequency of occurrence of all related words in the related word set in the second text; The text weights of all relevant words in the second text are determined based on the frequency of occurrence of all relevant words in the second text, the text length of the second text, and the average text length of the multiple second texts. The keyword feature vector is determined based on the text weights of all relevant words in the multiple second texts.

4. The method according to claim 1, characterized in that, The acquisition of the relevant word set includes: Multiple second texts from multiple accounts are segmented to obtain a word set, wherein the word set includes at least one word; The relevant word set is determined based on the frequency of occurrence of each word in the word set in the multiple second texts.

5. The method according to claim 1, characterized in that, The step of predicting the matching result between the account to be identified and the risk label to be matched based on the target text includes: The target text is transformed using the attention layer in the risk identification model to obtain a text vector; The text vector is transformed by the encoder in the risk identification model to obtain an enhanced text vector representation; Logistic regression is performed based on the text enhancement vector to obtain the matching result.

6. The method according to claim 5, characterized in that, The method further includes: Acquire training data and corresponding matching results, wherein the training data includes keyword sequence samples and second text samples from multiple accounts; The keyword sequence sample and the second text input sample are fed into the model to be trained for prediction to obtain the training prediction result; Based on the training prediction results and the matching results, the model to be trained is trained to obtain the risk identification model.

7. The method according to claim 1, characterized in that, The method further includes: Obtain multiple matching results with multiple risk tags to be matched, wherein the matching result is a risk identifier indicating whether the account to be identified has a risk; If any matching result indicates that the account to be identified is at risk, then transaction operations related to the account to be identified are prohibited.

8. The method according to claim 7, characterized in that, After obtaining multiple matching results with multiple risk tags to be matched, the method further includes: If multiple matching results indicate that the account to be identified is at risk, the risk level of the account to be identified is determined based on the number of matching results indicating that the account to be identified is at risk.

9. The method according to claim 1, characterized in that, The method further includes: Obtain multiple matching results with multiple risk labels to be matched, wherein the matching results are risk scores indicating that the account to be identified has risks; If the risk score of any matching result is greater than the scoring threshold, then the account to be identified is determined to be at risk. If the risk score of each matching result is less than the scoring threshold, the sum of the risk scores of the multiple matching results is calculated. If the sum of the risk scores is greater than the risk threshold, an alarm is issued for the account to be identified.

10. The method according to claim 1, characterized in that, After obtaining the keyword sequence associated with the risk label to be matched, the method further includes: The first text is input into the first text model to obtain the first text representation; The keyword sequence is input into the second text model to obtain the second text expression, wherein the first text model and the second text model share model parameters; Based on the first text expression and the second text expression, determine the similarity between the first text and the keyword sequence; Based on the similarity, the matching result between the account to be identified and the risk label to be matched is determined.

11. The method according to claim 1, characterized in that, The step of predicting the matching result between the account to be identified and the risk label to be matched based on the target text includes: The target text is mapped to a word vector space to generate a feature image; The feature image is convolved by a convolutional layer to obtain multiple feature maps; The pooling layer is used to perform pooling operations on the multiple feature maps to obtain the maximum eigenvalue of each feature map. Multiple maximum eigenvalues ​​are input into a fully connected layer to calculate the matching result.

12. An account identification device, characterized in that, include: The text acquisition module is used to acquire the first text corresponding to the account to be identified, wherein the first text is obtained by concatenating at least one text information associated with the account to be identified; The second text acquisition module is used to acquire multiple second texts from multiple accounts, wherein the second text of each account is obtained by concatenating at least one text information associated with each account; The related word set acquisition module is used to acquire a related word set, wherein the related word set includes at least one related word, and the related word is determined based on the frequency of occurrence of words in multiple second texts; The feature extraction module is used to extract features from multiple second texts based on a set of related words to obtain keyword feature vectors. The classification module is used to classify the keyword feature vectors based on the risk labels to be matched using a label classification model, thereby obtaining at least one keyword. The combination module is used to combine keywords into keyword sequences. A sequence acquisition module is used to acquire a keyword sequence associated with the risk label to be matched, wherein the keyword sequence includes multiple keywords that are semantically related to the risk label to be matched; The splicing module is used to splice the first text and the keyword sequence to obtain the target text; The prediction module is used to predict the matching result between the account to be identified and the risk label to be matched based on the target text.

13. An electronic device, characterized in that, include: processor; Memory for storing the executable instructions of the processor; The processor is configured to perform the account identification method according to any one of claims 1 to 11 by executing the executable instructions.

14. A computer-readable medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the account identification method as described in any one of claims 1 to 11.

15. A computer program product, characterized in that, The computer program product includes computer instructions stored in a computer-readable storage medium, a processor of a computer device reading the computer instructions from the computer-readable storage medium, and the processor executing the computer instructions to cause the computer device to perform the account identification method as described in any one of claims 1 to 11.

Images