WO2021022790A1 - 基于智能交互的主动风控方法和系统 - Google Patents

基于智能交互的主动风控方法和系统 Download PDF

Info

Publication number
WO2021022790A1
WO2021022790A1 PCT/CN2020/071590 CN2020071590W WO2021022790A1 WO 2021022790 A1 WO2021022790 A1 WO 2021022790A1 CN 2020071590 W CN2020071590 W CN 2020071590W WO 2021022790 A1 WO2021022790 A1 WO 2021022790A1
Authority
WO
WIPO (PCT)
Prior art keywords
risk
implementer
contact information
active
interaction
Prior art date
Application number
PCT/CN2020/071590
Other languages
English (en)
French (fr)
Inventor
姚磊
应亦丰
李娜
张哲�
Original Assignee
创新先进技术有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 创新先进技术有限公司 filed Critical 创新先进技术有限公司
Priority to US16/820,884 priority Critical patent/US11086991B2/en
Publication of WO2021022790A1 publication Critical patent/WO2021022790A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33—Querying
    • G06F16/3331—Query processing
    • G06F16/334—Query execution
    • G06F16/3344—Query execution using natural language analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0635—Risk analysis of enterprise or organisation activities
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G06Q40/04—Trading; Exchange, e.g. stocks, commodities, derivatives or currency exchange

Definitions

  • This disclosure mainly relates to risk control, and in particular to active risk control.
  • the traditional risk control plan is to conduct an in-depth analysis of the cases that have occurred, obtain the telephone numbers and modus operandi in them, and then carry out risk control based on these characteristics. From the perspective of user protection, the user has already been infringed and the loss has been caused, which greatly reduces the user experience.
  • third-party payment is a channel for fund transfer throughout the risk process, and the main places where risks occur are outside the payment process.
  • the beneficiary of the transfer of funds that is, the media for receiving payments, are mostly new emerging media, which cannot be identified before the case occurs through traditional risk control schemes.
  • the present disclosure provides an efficient active risk control solution based on intelligent interaction.
  • This program can take the initiative and identify beforehand.
  • This solution also has the ability of intelligent interaction.
  • intelligent interaction models instead of humans to actively communicate with risk implementers through various contact methods, all-weather high-efficiency work can be achieved.
  • the parallel working capacity of the machine can be expanded unlimitedly and the processing throughput can be improved.
  • This solution can also realize automated risk control. Based on the obtained risk medium and risk process, it can automatically enter the subsequent processing platform and automatically deploy related decision-making actions, thereby greatly improving the effect of risk defense.
  • an active risk control method based on intelligent interaction including: obtaining the contact information of the risk implementer; actively interacting with the risk implementer based on the contact information of the risk implementer and generating the active interaction Record; process active interaction records and extract risk characteristics; classify risks according to risk characteristics; and perform different risk controls on risks according to risk categories.
  • obtaining the contact information of the risk implementer further includes: collecting raw data related to the risk; performing semantic analysis on the raw data; marking the risk degree and determining the processing priority according to the semantic analysis result; and The processing priority extracts the contact information of the risk implementer.
  • the contact information of the risk implementer includes a phone number, an instant messaging account, an online publishing account, and an email address.
  • actively interacting with the risk implementer based on the risk implementer’s contact information further includes: actively contacting the risk implementer based on the risk implementer’s contact information; receiving information sent by the risk implementer; analyzing The information sent by the risk implementer to identify the interactive theme; confirm the risk implementer’s intention based on the interactive theme; and automatically generate a response based on the risk implementer’s intention.
  • the active interaction with the risk implementer is performed through an interactive agent based on the contact information of the risk implementer.
  • processing active interaction records and extracting risk characteristics further includes: converting the active interaction records into multimedia format; performing semantic analysis on the converted active interaction records; and performing risk process mining based on the semantic analysis results To extract risk characteristics.
  • the risk characteristics include risk medium characteristics, risk behavior characteristics, risk time characteristics, and risk geographic characteristics.
  • an active risk control system based on intelligent interaction
  • an acquisition module to obtain the contact information of the risk implementer
  • an active interaction module to communicate with the risk implementer based on the risk implementer’s contact information Active interaction and generate active interaction records
  • feature extraction module which processes active interaction records and extracts risk characteristics
  • classification control module which categorizes risks according to risk characteristics, and carries out different risk control on risks according to risk categories.
  • obtaining the contact information of the risk implementer by the obtaining module further includes: collecting raw data related to the risk; performing semantic analysis on the raw data; marking the risk degree and determining the processing priority according to the semantic analysis result; And extract the contact information of risk implementers according to the processing priority.
  • the contact information of the risk implementer includes a phone number, an instant messaging account, an online publishing account, and an email address.
  • the active interaction module that actively interacts with the risk implementer based on the risk implementer’s contact information further includes: actively contacting the risk implementer based on the risk implementer’s contact information; Information; analyze the information sent by the risk implementer to identify the interactive theme; confirm the risk implementer’s intention based on the interactive theme; and automatically generate a response based on the risk implementer’s intention.
  • the active interaction module performs active interaction with the risk enforcer based on the contact information of the risk enforcer through the interactive agent in the active interaction module.
  • the feature extraction module processing active interaction records and extracting risk features further includes: converting the active interaction records into multimedia format; performing semantic analysis on the converted active interaction records; and performing semantic analysis based on the results of the semantic analysis. Risk process mining to extract risk characteristics.
  • the risk characteristics include risk medium characteristics, risk behavior characteristics, risk time characteristics, and risk geographic characteristics.
  • a computer-readable storage medium storing instructions, and when these instructions are executed, the machine executes the aforementioned method.
  • Fig. 1 shows a flowchart of an active risk control method based on intelligent interaction according to an embodiment of the present disclosure
  • Fig. 2 shows a schematic diagram of a process of obtaining contact information of potential risk implementers according to an embodiment of the present disclosure
  • FIG. 3 shows a flowchart of a process of obtaining contact information of potential risk implementers according to an embodiment of the present disclosure
  • Fig. 4 shows a schematic diagram of a process of active interaction with a potential risk implementer according to an embodiment of the present disclosure
  • FIG. 5 shows a flowchart of a process of active interaction with a potential risk implementer according to an embodiment of the present disclosure
  • FIG. 6 shows a schematic diagram of a process of performing risk process mining and extracting risk features according to another embodiment of the present disclosure
  • Fig. 7 shows a flowchart of a process of performing risk process mining and extracting risk features according to another embodiment of the present disclosure
  • Fig. 8 shows a block diagram of an active risk control system based on intelligent interaction according to an embodiment of the present disclosure.
  • Telecommunications crimes refer to criminals fabricating false information, setting up scams, and carrying out remote, non-contact infringements on victims through telephone, text messages, and the Internet.
  • the criminal activity of network pyramid scheme is to use the network and other means to carry out the criminal activity of pyramid scheme.
  • it is more secretive. It uses ordinary people's love for money to develop offline. The development speed is very fast. The number of victims is large and widespread, which often has a serious impact on society.
  • Traditional risk prevention and control usually adopt passive risk control and human risk control methods.
  • Human risk control recognizes and summarizes the risks that occur manually, relies on expert rules, blacklist libraries, etc., requires accumulation of relevant experience, and cannot maintain long-term, high-efficiency work. In the process of confronting risk implementers, they are passively defensive and have low timeliness.
  • Passive risk control generally deploys risk control into a real-time prevention and control system after a comprehensive analysis based on cases that have occurred. At this time, the risk object has been successfully implemented, and the user has incurred a capital loss.
  • the successful medium for example, risk implementation account or account number, mobile phone number, etc.
  • the present disclosure proposes an active risk control scheme based on intelligent interaction. Since communication is the starting point of all risky behaviors, the current common communication channels are telecommunications interactions through phone numbers or text messages, and online interactions based on instant messaging accounts, online release accounts, and email addresses. Therefore, in the communication process, Need to know the contact information of the other party. Through the active identification of risk data, the contact information of risk implementers is discovered, and then active interaction is initiated. According to the records of the interaction process, the key behavioral characteristics can be identified, which can identify common risk techniques and discover new risk techniques in time. Finally, according to these common and new risk methods, carry out corresponding risk prevention and control and strategic layout.
  • Fig. 1 shows a flowchart of an active risk control method 100 based on intelligent interaction according to an embodiment of the present disclosure.
  • telecommunications crimes Take telecommunications crimes as an example.
  • telecommunications crimes that have emerged in an endless stream include: financial and wealth management related violations (for example, false winnings, rewards, refunds; securities related; credit card related; insurance related; margin related, etc.), gaming related violations , False recruitment/part-time related violations, identity impersonation-related violations, false shopping-related violations, online game transactions-related violations, and virtual goods-related violations, etc.
  • risk implementers Under normal circumstances, risk implementers will publish their contact information in public places, waiting for users to actively contact them. As far as telecommunications crime is concerned, the contact information of the risk perpetrator includes phone numbers, instant messaging accounts, online release accounts, and e-mail addresses. Those skilled in the art can understand that it is possible for risk implementers to use any other contact methods, and with the advancement of network and communication technology, risk implementers may of course choose new or advanced contact methods, which are also included within the technical solution of the present disclosure.
  • a crawler tool is used to collect the contact information of potential risk implementers.
  • Crawler tools are actively collecting and mining in open networks (such as forums, advertising alliances, classified information websites, etc.) to form a collection of potentially risky contact methods (for example, telephone number collection, instant messaging account collection, network publishing account collection, and Mailbox collection, etc.).
  • crawler tools according to certain risk identification rules and contact information collection rules, it is possible to collect suspected risky contact information in various public places on a regular basis.
  • the unprocessed raw data collected by the crawler tool can be output for further analysis.
  • a database of suspected risk implementers provided by a third party may be used.
  • a database of suspected risk implementers provided by a third party may be used.
  • the unprocessed raw data collected or collected can be processed to extract the contact information of potential risk implementers. This process will be described below with reference to the schematic diagram of FIG. 2 and the flowchart of FIG. 3.
  • an active interaction with the risk implementer is performed based on the contact information of the risk implementer and an active interaction record is generated.
  • the robot can carry out a large-scale proactive attack.
  • the robot is an interactive agent, and its back end is connected to a corresponding intelligent interaction model.
  • a robot is an entity that takes the initiative to attack, and it can be implemented in various ways.
  • the interactive agent and intelligent interaction model will be used as examples to expand the description.
  • the interaction agent establishes an information channel between the intelligent interaction model and the risk implementer, and is the transfer node of the message.
  • the intelligent interaction model has automatic interaction capabilities, can not be recognized as a machine by the other party, and supports biased guidance for specific scenarios, so as to obtain the desired risk information.
  • the interactive agent also supports parameter adjustments to play different character settings.
  • the intelligent interaction model includes three parts: understanding of the interactive theme, confirmation of the other party's intention, and automatic response generation. It mainly includes two models: an interactive intention understanding model that can understand the information sent by the risk implementer and obtain the other party's intention; The response generation model corresponding to the reply.
  • This active interaction with the risk implementer is recorded and saved as an active interaction record.
  • the active interaction record is processed and risk characteristics are extracted.
  • Active interaction records usually contain a complete risk process, from which valuable risk data can be discovered, and key points and obvious methods of risk can be summarized, which is the basis for the next step of analysis, prevention and control.
  • Processing active interaction records can include unified format conversion, semantic understanding of the interaction process, and risk process mining.
  • the mining of the risk process will obtain the risk characteristics related to the risk scenario.
  • Risk characteristics include risk medium characteristics, risk behavior characteristics, risk time characteristics, risk geographic characteristics, etc. Those skilled in the art can understand that for different risk scenarios, different risk medium characteristics, risk behavior characteristics, risk time characteristics, and risk geographic characteristics can be extracted, and different other risk characteristics can also be provided.
  • the risks are classified according to risk characteristics.
  • Risk time characteristics and risk geographical characteristics usually help to combine risk behavior characteristics and risk medium characteristics to classify risks.
  • the characteristics of risk media describe the tools and channels used by risk implementers and users for information communication and fund transfer during the risk occurrence process.
  • the characteristics of risk behavior describe the collection of behavior points where the behavior of the risk implementer is distinguished from normal behavior in the process of risk occurrence.
  • the external channel layer further monitors customer access and conversation suspicious behavior before the transaction; whether the counterparty in the transaction is on the suspicious list. Monitor business violations and suspicious operations at the internal channel level. Monitor infringement transactions within product services and cross-product infringement transactions at the product service layer. At the data integration layer, monitor cross-product and channel combination/complex infringement transactions.
  • the risk data is output to provide external risk services, for example, push suspicious phone numbers to telecom operators, push suspicious bank cards/accounts to relevant banks, etc. Wait.
  • Fig. 2 shows a schematic diagram of a process of obtaining contact information of potential risk implementers according to an embodiment of the present disclosure.
  • the process of obtaining the contact information of potential risk implementers is actually a process of active identification of risk implementers.
  • the risk implementer Before the user is violated, the risk implementer must communicate with the victim through certain means, so as to gradually achieve the purpose of illegally occupying the victim's money. Therefore, communication is the starting point for all risky behaviors.
  • the common communication channels are to make calls or send text messages through phone numbers, send text, pictures, videos, etc. through instant messaging accounts, publish account interaction through the Internet, and send links through e-mail. Therefore, in the communication process, you need to know the contact information of the other party.
  • the contact information of risk implementers can be mined to establish a data foundation for active interaction.
  • a crawler tool may be used to collect risk data.
  • Crawler tools actively collect and mine in open networks (such as forums, advertising alliances, classified information websites, etc.) to form a set of potentially risky contact information.
  • risk implementers will publish their contact information in public places, waiting for users to actively contact them.
  • the crawler tool according to the risk identification rules and the contact information collection rules, it is possible to collect suspected risky contact information in various public places on a regular basis.
  • the crawler collects unprocessed raw data and outputs it to the risk understanding platform for further analysis.
  • a database of suspected risk implementers provided by a third party may be used.
  • Those skilled in the art can understand that various methods can be used to collect the contact information of potential risk implementers, and the description of the above two embodiments does not constitute a limitation to the technical solution of the present disclosure.
  • Risk identification rules and contact information collection rules can be set in advance, or can be gradually learned and accumulated. Risk identification rules and contact information collection rules can be simple rules, rule sets, rule trees, rule flows, etc., or even a recognition engine incorporating natural language models and deep learning algorithms to identify different risk scenarios Risk and collect corresponding raw data containing risk information.
  • risk identification rules and contact information collection rules can be set related to financial and wealth management related violations (for example, false winnings, rewards, refunds; securities related; credit card related; insurance related; deposit Related keywords), gambling-related violations, false recruitment/part-time job-related violations, identity impersonation-related violations, false shopping-related violations, online game transaction-related violations, and virtual goods-related violations, etc., based on these key words Words are used to construct rules and dynamically update the rules adaptively as telecommunication crimes change.
  • financial and wealth management related violations for example, false winnings, rewards, refunds; securities related; credit card related; insurance related; deposit Related keywords
  • the crawler tool found that an online publishing account posted a potentially false reward message for college entrance examination candidates in a certain place, claiming that students who received the university admission notice will receive a one-time bursary of 2,800 yuan and leave a phone number For eligible students to contact by phone or SMS.
  • the crawler tool collects the original data including the online account number, phone number, acquisition conditions of the “bursary” and “bursary amount” based on the keyword “bursary”.
  • the risk understanding platform is used to process the raw data collected by the crawler tool. For the corresponding original data containing risk information, risk semantic understanding, risk level calibration and risk implementer contact information extraction can be carried out.
  • Risk semantic understanding performs semantic analysis on the original data collected by the crawler tool, confirms whether it is risk data according to the natural language understanding model and the parameters and thresholds designed in it, and preprocesses the data such as word segmentation and theme features.
  • the risk semantic understanding performs semantic analysis on the collected "students who receive the university admission notice will receive a one-time grant of 2,800 yuan", based on the natural language understanding model to understand the existence of "receive university admission notice” The condition of "student of books”. Further, it is confirmed that the data may be risk data, and the risk data is preprocessed to extract information such as the online release account number, phone number, "bursary” provider, “bursary” obtaining conditions, and "bursary amount”.
  • the risk level calibration is based on semantic understanding and possible infringement types to determine the degree of harm that the risk may cause.
  • the risk level gradually increases from level 1 to level 5 as the basis for subsequent processing priorities.
  • the risk implementer's contact information is extracted according to the priority, and the valuable information in the risk data is extracted, including the contact information of the other party and other information, thereby forming a set of potential risk implementers.
  • FIG. 3 shows a flowchart of a process 300 of obtaining contact information of potential risk implementers according to an embodiment of the present disclosure.
  • the crawler tool collects the original data including the online account number, phone number, “sponsorship” provider, “stipend” acquisition conditions, and “stipend amount” based on the keyword “stipend”.
  • semantic analysis is performed on the original data.
  • the risk semantic understanding performs semantic analysis on the collected "students who receive the university admission notice will receive a one-time grant of 2,800 yuan", based on the natural language understanding model to understand the existence of "receive university admission notice”
  • rough statistics or data extraction can be made on the number of eligible college entrance examination candidates in that place in that year, and the total amount can be roughly calculated.
  • the degree of risk is calibrated according to the semantic analysis result and the processing priority is determined.
  • the risk level is calibrated based on semantic understanding and possible infringement types, and the degree of harm that the risk may cause is judged.
  • the risk level gradually increases from level 1 to level 5 as the processing priority for subsequent processing. For example, in this embodiment of false rewards, the risk level is marked as 4.
  • the contact information of the risk implementer is extracted according to the processing priority.
  • the risk implementer's contact information is extracted according to the priority, and the valuable information in the risk data is extracted, including the original release method, the other party's contact information, the type of possible infringement and other information, thereby forming a data collection of potential risk implementers .
  • the potential risk implementer data collection can be stored as structured data.
  • Fig. 4 shows a schematic diagram of a process of active interaction with a potential risk implementer according to an embodiment of the present disclosure.
  • the process of active interaction with potential risk implementers is mainly carried out by the intelligent interaction platform.
  • the intelligent interaction platform is responsible for managing all interaction processes, and the platform can deploy multiple interaction agents.
  • the platform assigns risks to different interactive agents according to the different possible infringement types of risks, and at the same time uses the corresponding potential risk implementer data as the input of the interactive agent.
  • the platform automatically controls the number of agents working at the same time, the working hours of each agent, etc., according to the number of tasks to be handled with potential risks, and monitors the work of agents in real time and generates statistical data.
  • An interactive agent is an entity that takes the initiative, and its back end is connected to an intelligent interaction model. According to the different connection models, interaction agents are divided into voice agents, text agents, instant messaging agents, network publishing response agents, etc., respectively processing telephone voice interaction, short message interaction, instant message interaction, network publishing message interaction, and so on.
  • the interaction agent acts as a message transfer node, establishing an information channel between the intelligent interaction model and the risk implementer.
  • Each interactive agent can independently complete the active work.
  • multiple interactive agents can also jointly take the initiative to attack, in order to deal with potential gang risk implementers.
  • multiple interactive agents can be deployed to quickly increase the processing throughput of the system.
  • the intelligent interaction model has automatic interaction capabilities.
  • the specific capabilities include: not being recognized as a machine by the other party in the victimized scene; supporting inclined guidance to obtain the desired risk information; and supporting parameter adjustments to play different character settings set.
  • the intelligent interaction model can prepare relevant data and information for active interaction based on background data and operation.
  • the intelligent interaction model can support different character settings such as high school graduates who lack social experience, parents with relatively social experience, and class teachers with relatively social experience, and guide potential risk implementers Further explain the method of receiving the bursary, whether it involves the other party's risk operations such as providing the account number and requesting the student to transfer money in advance; and further guide the potential risk implementer to explain the situation of the support organization that provides the bursary.
  • the intelligent interaction model can also query the total number of students who meet the condition of "students who have received university admission notice", and multiply the total number of students with the stipend of 2,800 yuan per person To determine the total amount. Thus, to prepare information for subsequent active interaction. Based on the situation of the funding institution stated by the potential risk implementer, the intelligent interaction model can be used for query and subsequent interactions for verification.
  • the intelligent interaction model can understand the interaction theme, confirm the intention of the other party, and automatically generate responses.
  • the interactive theme After actively contacting, receive and analyze the voice, text, picture or video sent by the other party to understand the current interactive theme. Taking the embodiment of false rewards as an example, the interactive theme can be understood as "bursary award" based on the information sent by the other party.
  • the other party's expressions must be different.
  • the information of the other party can be converted into answers to some standard questions, such as, what is the process of obtaining grants? How many students have received grants (or, how many students can receive)? What kind of organization is the funding agency? What conditions or feedback do students need to meet in the future? and so on.
  • a natural language generation model is used, and a modifier tone or word with corresponding characteristics is added to generate a response to the intention.
  • a modifier tone or word with corresponding characteristics is added to generate a response to the intention.
  • For short message interaction just generate text; for voice interaction, the content needs to be converted into sound files; for instant messaging interaction, you need to maintain the corresponding personality or characteristics in terms of text, voice or emoticons, response speed, etc.; Online publishing or e-mail interaction, because the writing will be relatively free, it is necessary to maintain the corresponding writing style.
  • the intelligent interaction platform will record the interaction process, for example, the complete process of telephone communication between the voice interaction agent and the potential risk implementer.
  • the record contains a complete risk process, from which valuable risk data can be found, and the key points and obvious methods of risk can be summarized, which is the basis for the next analysis and prevention and control.
  • FIG. 5 shows a flowchart of a process 500 of active interaction with a potential risk implementer according to an embodiment of the present disclosure.
  • the risk implementer is actively contacted based on the contact information of the risk implementer.
  • the crawler tool found that an online publishing account posted a potentially false reward message for college entrance examination candidates in a certain place, claiming that students who received the university admission notice would receive a one-time bursary of 2,800 yuan, and left a phone number for eligible students to call Or SMS contact.
  • the intelligent interactive platform can actively call or send text messages to contact the risk implementer based on the contact information of the risk implementer.
  • the intelligent interaction platform can perform the active interaction according to different role settings, such as students, parents or teachers.
  • the information sent by the risk implementer is received.
  • the information sent by the risk implementer is analyzed to identify the subject of interaction.
  • the risk implementer's intention is confirmed based on the interactive theme.
  • the other party's intention can be confirmed based on the interactive theme.
  • the different expressions sent by the other party may have the same meaning behind it. Therefore, it is necessary to convert the other party's information into a standard question, and then identify the other party's intention and understand the other party's purpose.
  • a response is automatically generated based on the intent of the risk implementer.
  • a natural language generation model is used to generate a response to the intention by adding corresponding modifiers or words.
  • Fig. 6 shows a schematic diagram of a process of performing risk process mining and extracting risk features according to another embodiment of the present disclosure.
  • the process of mining the risk process and extracting risk features is executed by the risk data processing platform.
  • the risk data processing platform can perform multimedia format conversion, semantic analysis of interaction records, risk process mining, risk feature extraction and risk classification and marking.
  • the need for unified conversion of multimedia formats is due to the active interaction process through telephone voice, SMS/MMS, instant messaging (including text, voice, video, emoticons, etc.) and network publishing interaction.
  • unified conversion to text is required , Including voice content recognition and picture content recognition.
  • the active interaction of different character settings such as students, parents, and teachers supported by the intelligent interaction model is mainly carried out through phone calls and text messages. Therefore, phone voice and text messages will be uniformly converted into text , Including voice content recognition.
  • the semantic understanding of the interaction process is based on the natural language processing model to facilitate the understanding of the risk occurrence process.
  • the interactive theme is "bursary awards", so the focus will fall on the way to further receive bursaries, whether it involves risk operations such as the other party providing account numbers, requiring students to transfer funds in advance, and the support institutions that provide bursaries and many more.
  • the risk process can be characterized by a series of behavior points, and a collection of behavior points with a certain degree of distinction between risk behavior and normal behavior can be obtained.
  • the normal operation should be that the student provides the account number and transfers the funds to the account by the support agency providing the bursary; therefore, during the active interaction process, the other party provides the account number and requires the student to transfer the funds first
  • the point is the point of risk behavior.
  • the normal response should be the corresponding magnitude of the statistical data, and the risk behavior point when there is a response that does not meet the corresponding magnitude. Of course, it is understandable that there are other risk behavior points, which are not repeated here.
  • risk characteristics can be extracted, such as risk time characteristics, risk geographic characteristics, risk behavior characteristics, and risk medium characteristics.
  • the risk time feature is the time chain of the release date of the network release message and the subsequent post time.
  • the risk area feature is the area claimed in the network release message.
  • the phone number can be profiled during the interaction, and the current location of the phone number can be compared with the area claimed in the message.
  • the risk behavior feature is the set of risk behavior points formed by the described behavior points.
  • the characteristics of risk media are the tools and channels used by risk implementers and users for information communication and fund transfer during the occurrence of risk behaviors.
  • the bank account or payment APP account provided by the counterparty is a tool or channel for fund transfer.
  • risks can be classified and labeled. When each risk characteristic shows that the risk occurs inside the system, the risk is classified as an internal risk; when each risk characteristic shows that the risk occurs outside the system, the risk is classified as an external risk.
  • the risk medium characteristic shows that the fund transfer channel used by the risk offender is outside the system, thus classifying the risk as an external risk.
  • the processing of active interactive records allows valuable information points to be identified through semantic analysis, so that common risk techniques can be further obtained, new risk techniques can be discovered in time, and key behavior characteristics can be identified.
  • FIG. 7 shows a schematic diagram of a process 700 of performing risk process mining and extracting risk features according to another embodiment of the present disclosure.
  • multimedia format conversion is performed on the active interaction record.
  • Unified conversion of active interaction records into text including voice content recognition and image content recognition.
  • Semantic analysis is based on the natural language processing model for the converted active interaction records, in order to understand the risk occurrence process.
  • risk process mining is performed based on the semantic analysis result to extract risk features.
  • the risk process can be characterized by a series of behavior points, and a collection of behavior points with a certain degree of distinction between risk behavior and normal behavior can be obtained.
  • risk characteristics can be extracted, such as risk medium characteristics, risk behavior characteristics, risk time characteristics, and risk geographic characteristics.
  • FIG. 8 shows a block diagram of an active risk control system 800 based on intelligent interaction according to an embodiment of the present disclosure.
  • the system 800 includes an acquisition module 802, an active interaction module 804, a feature extraction module 806, and a classification control module 808.
  • the obtaining module 802 obtains the contact information of the risk implementer.
  • the contact information of the risk perpetrator includes phone numbers, instant messaging accounts, online release accounts, and e-mail addresses.
  • a crawler tool is used to collect the contact information of potential risk implementers.
  • a database of suspected risk implementers provided by a third party may be used.
  • the active interaction module 804 performs active interaction with the risk enforcer based on the contact information of the risk enforcer and generates an active interaction record.
  • the active interaction module 804 uses an interaction agent to establish an information channel between the intelligent interaction model and the risk implementer as a message transfer node.
  • the intelligent interaction model adopted by the active interaction module 804 has automatic interaction capabilities, can not be recognized as a machine by the counterparty, and supports biased guidance for specific scenarios, so as to obtain desired risk information.
  • the interactive agent also supports parameter adjustments to play different character settings.
  • the intelligent interaction model includes three parts: understanding of the interactive theme, confirmation of the other party's intention, and automatic response generation. It mainly includes two models: an interactive intention understanding model that can understand the information sent by the risk implementer and obtain the other party's intention; The response generation model corresponding to the reply. This active interaction with the risk implementer is recorded and saved as an active interaction record.
  • the feature extraction module 806 processes active interaction records and extracts risk features.
  • Active interaction records usually contain a complete risk process, from which valuable risk data can be discovered, and key points and obvious methods of risk can be summarized, which is the basis for the next step of analysis, prevention and control.
  • Processing active interaction records can include unified format conversion, semantic understanding of the interaction process, and risk process mining.
  • the mining of the risk process will obtain the risk characteristics related to the risk scenario.
  • Risk characteristics include risk medium characteristics, risk behavior characteristics, risk time characteristics, risk geographic characteristics, etc. Those skilled in the art can understand that for different risk scenarios, different risk medium characteristics, risk behavior characteristics, risk time characteristics, and risk geographic characteristics can be extracted, and different other risk characteristics can also be provided.
  • the classification control module 808 classifies the risks according to the risk characteristics.
  • the risks can be classified.
  • Risk time characteristics and risk geographical characteristics usually help to combine risk behavior characteristics and risk medium characteristics to classify risks.
  • the characteristics of risk media describe the tools and channels used by risk implementers and users for information communication and fund transfer during the risk occurrence process.
  • the characteristics of risk behavior describe the collection of behavior points where the behavior of the risk implementer is distinguished from normal behavior in the process of risk occurrence.
  • classification control module 808 performs different risk control on risks according to the types of risks.
  • the present disclosure proposes an active risk control scheme based on intelligent interaction, which can proactively attack and identify in advance.
  • risk information is obtained through user reports, that is, obtained after the risk occurs.
  • the life cycle is very short, and it is difficult to play a greater role in subsequent prevention and control.
  • the technical solution of the present disclosure can obtain relevant information before the risk occurs by taking the initiative to attack, that is, obtain before there is actual loss, so as to achieve early prevention and control.
  • This solution also has the ability of intelligent interaction.
  • intelligent interaction models instead of humans to actively communicate with risk implementers through various contact methods, all-weather high-efficiency work can be achieved.
  • the parallel working capacity of the machine can be expanded unlimitedly and the processing throughput can be improved.
  • This solution can also realize automated risk control. Based on the obtained risk medium and risk process, it can automatically enter the subsequent processing platform and automatically deploy related decision-making actions, thereby greatly improving the effect of risk defense.
  • the various steps and modules of the active risk control method and system based on intelligent interaction described above can be implemented by hardware, software, or a combination thereof. If implemented in hardware, the various illustrative steps, modules, and circuits described in the present invention can be combined with general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field programmable gate arrays (FPGA), Or other programmable logic components, hardware components, or any combination thereof to realize or execute.
  • a general-purpose processor may be a processor, microprocessor, controller, microcontroller, or state machine, etc. If implemented in software, various illustrative steps and modules described in conjunction with the present invention may be stored on a computer-readable medium or transmitted as one or more instructions or codes.
  • the software modules that implement various operations of the present invention can reside in storage media, such as RAM, flash memory, ROM, EPROM, EEPROM, registers, hard disks, removable disks, CD-ROMs, cloud storage, etc.
  • the storage medium may be coupled to the processor so that the processor can read and write information from/to the storage medium, and execute corresponding program modules to implement various steps of the present invention.
  • the software-based embodiments can be uploaded, downloaded or remotely accessed through appropriate communication means.
  • suitable communication means include, for example, the Internet, World Wide Web, Intranet, software applications, cables (including fiber optic cables), magnetic communication, electromagnetic communication (including RF, microwave, and infrared communication), electronic communication, or other such communication means.

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Human Resources & Organizations (AREA)
  • Databases & Information Systems (AREA)
  • General Physics & Mathematics (AREA)
  • Economics (AREA)
  • Strategic Management (AREA)
  • Finance (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Development Economics (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • Accounting & Taxation (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Game Theory and Decision Science (AREA)
  • Educational Administration (AREA)
  • Technology Law (AREA)
  • Tourism & Hospitality (AREA)
  • Quality & Reliability (AREA)
  • Operations Research (AREA)
  • Artificial Intelligence (AREA)
  • Computational Linguistics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Preparation Of Compounds By Using Micro-Organisms (AREA)

Abstract

一种基于智能交互的主动风控方法，包括：获取风险实施者的联系方式（102）；基于风险实施者的联系方式进行与风险实施者的主动交互并生成主动交互记录（104）；处理主动交互记录并提取风险特征（106）；根据风险特征将风险归类（108）；以及按照风险的类别对风险进行不同的风险控制（110）。

Description

基于智能交互的主动风控方法和系统 技术领域
本公开主要涉及风险控制，尤其涉及主动风险控制。
背景技术
随着互联网应用逐步深入到人们生活中的各个方面，用户面临的风险程度和种类都在不断增加，这些风险中以非法侵占用户资金最为突出。目前第三方支付蓬勃发展、交易过程逐步便捷，而相对地，人们对风险防范的意识还没有随之提升，最终表现为受侵害案件频繁发生。
传统上的风险控制方案，是对已发生的案件进行深入分析，获取其中出现的电话号码和作案手法，然后根据这些特征进行风险控制。从用户保护的角度上看，此时用户已经受到侵害，损失已经造成，极大降低了用户体验。通过分析当前风险可以发现，第三方支付在整个风险过程中是资金转移的渠道，风险发生的主要场所都在支付过程外。同时，资金转移的收益方，即收款的介质较多为新出现介质，通过传统的风险控制方案无法在案件发生前进行识别。
本领域需要一种高效的基于智能交互的主动风控方法和系统，能够在风险发生之前及时发现风险实施者，从而降低风险实施者作案的效率并降低用户受侵害的可能性。
发明内容
为解决上述技术问题，本公开提供了一种高效的基于智能交互的主动风控方案。此方案能够主动出击，事前识别。本方案还具备智能交互的能力，通过使用智能交互的模型来代替人工主动与风险实施者通过各种联系方式交流，可以实现全天候的高效率工作。同时，通过交互代理的设计，可以无限的扩展机器并行工作能力，提升处理吞吐量。本方案还能够实现自动化的风险控制。基于所获得的风险介质和风险过程，可以自动进入到后面的处理平台、自动部署相关决策动作，从而大大提升风险防御的效果。
在本公开一实施例中，提供了一种基于智能交互的主动风控方法，包括：获取风险实施者的联系方式；基于风险实施者的联系方式进行与风险实施者的主动交互并生成主动交互记录；处理主动交互记录并提取风险特征；根据风险特征将风险归类；以及按照 风险的类别对风险进行不同的风险控制。
在本公开的另一实施例中，获取风险实施者的联系方式进一步包括：采集与风险相关的原始数据；对原始数据进行语义分析；根据语义分析结果标定风险程度并确定处理优先级；以及按照处理优先级提取风险实施者的联系方式。
在本公开的又一实施例中，风险实施者的联系方式包括电话号码、即时通信账号、网络发布账号以及电子邮箱。
在本公开的另一实施例中，基于风险实施者的联系方式进行与风险实施者的主动交互进一步包括：基于风险实施者的联系方式主动联系风险实施者；接收风险实施者发出的信息；分析风险实施者发出的信息以识别交互主题；基于交互主题确认风险实施者的意图；以及根据风险实施者的意图自动生成回应。
在本公开的另一实施例中，基于风险实施者的联系方式进行与风险实施者的主动交互通过交互代理进行。
在本公开的又一实施例中，处理主动交互记录并提取风险特征进一步包括：将主动交互记录进行多媒体格式转换；对转换后的主动交互记录进行语义分析；以及基于语义分析结果进行风险过程挖掘以提取风险特征。
在本公开的另一实施例中，风险特征包括风险介质特征、风险行为特征、风险时间特征以及风险地域特征。
在本公开一实施例中，提供了一种基于智能交互的主动风控系统，包括：获取模块，获取风险实施者的联系方式；主动交互模块，基于风险实施者的联系方式进行与风险实施者的主动交互并生成主动交互记录；特征提取模块，处理主动交互记录并提取风险特征；以及归类控制模块，根据风险特征将风险归类，并按照风险的类别对风险进行不同的风险控制。
在本公开的另一实施例中，获取模块获取风险实施者的联系方式进一步包括：采集与风险相关的原始数据；对原始数据进行语义分析；根据语义分析结果标定风险程度并确定处理优先级；以及按照处理优先级提取风险实施者的联系方式。
在本公开的又一实施例中，风险实施者的联系方式包括电话号码、即时通信账号、网络发布账号以及电子邮箱。
在本公开的另一实施例中，主动交互模块基于风险实施者的联系方式进行与风险实 施者的主动交互进一步包括：基于风险实施者的联系方式主动联系风险实施者；接收风险实施者发出的信息；分析风险实施者发出的信息以识别交互主题；基于交互主题确认风险实施者的意图；以及根据风险实施者的意图自动生成回应。
在本公开的另一实施例中，主动交互模块基于风险实施者的联系方式进行与风险实施者的主动交互通过主动交互模块中的交互代理进行。
在本公开的又一实施例中，特征提取模块处理主动交互记录并提取风险特征进一步包括：将主动交互记录进行多媒体格式转换；对转换后的主动交互记录进行语义分析；以及基于语义分析结果进行风险过程挖掘以提取风险特征。
在本公开的另一实施例中，风险特征包括风险介质特征、风险行为特征、风险时间特征以及风险地域特征。
在本公开一实施例中，提供了一种存储有指令的计算机可读存储介质，当这些指令被执行时使得机器执行如前所述的方法。
提供本概述以便以简化的形式介绍以下在详细描述中进一步描述的一些概念。本概述并不旨在标识所要求保护主题的关键特征或必要特征，也不旨在用于限制所要求保护主题的范围。
附图说明
本公开的以上发明内容以及下面的具体实施方式在结合附图阅读时会得到更好的理解。需要说明的是，附图仅作为所请求保护的发明的示例。在附图中，相同的附图标记代表相同或类似的元素。
图1示出根据本公开一实施例的基于智能交互的主动风控方法的流程图；
图2示出根据本公开一实施例的获取潜在风险实施者的联系方式的过程的示意图；
图3示出根据本公开一实施例的获取潜在风险实施者的联系方式的过程的流程图；
图4示出根据本公开一实施例的与潜在风险实施者的主动交互的过程的示意图；
图5示出根据本公开一实施例的与潜在风险实施者的主动交互的过程的流程图；
图6示出根据本公开另一实施例的进行风险过程挖掘并提取风险特征的过程的示意图；
图7示出根据本公开另一实施例的进行风险过程挖掘并提取风险特征的过程的流程 图；
图8示出根据本公开一实施例的基于智能交互的主动风控系统的框图。
具体实施方式
为使得本公开的上述目的、特征和优点能更加明显易懂，以下结合附图对本公开的具体实施方式作详细说明。
在下面的描述中阐述了很多具体细节以便于充分理解本公开，但是本公开还可以采用其它不同于在此描述的其它方式来实施，因此本公开不受下文公开的具体实施例的限制。
近年来，伴随移动互联网、虚拟现实等技术的飞速发展，互联网金融的服务模式日趋多样化。通过虚拟网络实现的产品交易，越来越多的线上交易开始体现出用最少的交互、最具个性化的引导来促成交易。在客户享受灵活便捷服务的同时，受侵害风险呈现出更加隐蔽、专业的特点，发展出更多的作案手法和表现形式。受侵害风险多指利用第三方身份、虚假证件和资料，有团队、有组织地进行恶意骗贷。受侵害的对象有用户、诸如银行的金融机构、诸如应用APP的平台等等。
在受侵害对象为用户的情况下，从行为模式来说，网络犯罪行为可进一步分为电信犯罪行为和网络传销犯罪活动。电信犯罪行为是指犯罪分子通过电话、短信和网络方式，编造虚假信息、设置骗局，对受害者实施远程、非接触式侵害的行为。网络传销犯罪活动是利用网络等手段进行传销犯罪活动。它相比于电信犯罪行为更加隐秘，利用普通人爱财心理发展下线，发展速度非常快，受害者数量多且广泛，往往对社会造成严重影响。
传统风险防控通常采用被动风控和人力风控等方法。人力风控通过人工对发生的风险进行识别和总结，依赖专家规则、黑名单库等，需要累积相关经验，且无法保持长时间、高效率的工作。在与风险实施者对抗过程中，处于被动防守，时效性较低。而被动风控一般根据已经发生的案件，进行全面分析后将风险风控部署到实时防控体系中。此时该风险对象已经实施成功，用户已产生了资金损失，成功后的介质(例如，风险实施账户或账号、手机号码等)存活时间短，很容易被弃用。显然，这些方法已经不能适应新的风险挑战。
本公开提出了一种基于智能交互的主动风控方案。由于沟通是所有风险行为的起点，目前常见的沟通渠道是通过电话号码打电话或发短信进行电信交互，以及基于即时通信 账号、网络发布账号以及电子邮箱进行线上交互，因此在沟通过程中，需要知道对方的联系方式。通过风险数据的主动识别，挖掘出风险实施者的联系方式，继而发起主动交互。根据交互过程记录，能够识别出其中的关键行为特征，由此能够识别出常用风险手法、并能及时发现新的风险手法，。最后根据这些常用和新的风险手法，进行相应的风险防控和策略布局。
在下文中，将针对用户受到电信犯罪行为侵害的场景进行详细描述。然而，本领域技术人员可以理解，本公开的技术方案同样适用于涉及网络传销犯罪活动，诸如银行、保险机构等的金融机构受到侵害，以及诸如应用APP等的平台受到侵害以及其他场景下的侵害。
下文将基于附图具体描述根据本公开各个实施例的基于智能交互的主动风控方法和系统。
基于智能交互的主动风控方法
图1示出根据本公开一实施例的基于智能交互的主动风控方法100的流程图。
在102，获取风险实施者的联系方式。
风险实施者想要对用户实施侵害，其必定需要通过一定的手段与受害者进行沟通，达到非法占有受害者钱财的目的，因此沟通是所有风险行为的起点。
以电信犯罪行为为例，近年来层出不穷的电信犯罪行为有：金融理财相关侵害行为(例如，虚假中奖、奖励、退款；证券相关；信用卡相关；保险相关；保证金相关等)、博彩相关侵害行为、虚假招聘/兼职相关侵害行为、身份冒充相关侵害行为、虚假购物相关侵害行为、网游交易相关侵害行为、和虚拟商品相关侵害行为等等。
一般情况下，风险实施者会在公开场所发布自己的联系方式，等待用户主动与其进行联系。就电信犯罪行为而言，风险实施者的联系方式包括电话号码、即时通信账号、网络发布账号以及电子邮箱等等。本领域技术人员可以理解，风险实施者采用任何其他联系方式是可能的，并且随着网络和通信技术的进步，风险实施者当然有可能选择新的或先进的联系方式，这些联系方式也被纳入本公开的技术方案内。
在本公开一实施例中，采用爬虫工具来收集潜在风险实施者的联系方式。爬虫工具是在开放网络中(如论坛、广告联盟、分类信息网站等)进行主动采集和挖掘，形成有潜在风险的联系方式集合(例如，电话号码集合、即时通信账号集合、网络发布账号集合以及邮箱集合等)。通过爬虫工具，根据一定的风险识别规则和联系方式采集规则， 可以定期到各个公开场所采集疑似有风险的联系方式。爬虫工具所采集的未经处理的原始数据可被输出以供进一步分析。
在本公开另一实施例中，可采用第三方提供的疑似风险实施者数据库。本领域技术人员可以理解，可采用各种办法来收集潜在风险实施者的联系方式，以上两个实施例的描述并不构成对本公开技术方案的限定。
进一步地，风险实施者在公开场所发布自己的联系方式时，其发布内容往往含有其他有价值信息，例如具有‘中奖’、‘奖励’、‘退款’、‘招聘’、‘兼职’等敏感词，或者含有可疑链接等。
所采集或收集到的未经处理的原始数据可被处理，以提取出潜在风险实施者的联系方式。这一过程将在下文中参照图2的示意图和图3的流程图进行描述。
在104，基于风险实施者的联系方式进行与风险实施者的主动交互并生成主动交互记录。
基于风险实施者的联系方式，机器人能够进行大规模的主动出击。在本公开一实施例中，机器人是交互代理，其后端连接相应的智能交互模型。本领域技术人员可以理解，机器人是主动出击的实体，其可以有各种不同的实现方式。在下文中，将以交互代理和智能交互模型为例展开描述。
交互代理在智能交互模型和风险实施者之间建立信息通道，是消息的中转节点。智能交互模型则具备自动交互能力，能够不被对方识别为机器且支持针对特定场景的有倾向性的引导，从而获取希望得到的风险信息。交互代理还支持通过参数调整来扮演不同的人物设定。智能交互模型包含交互主题理解、对方意图确认、自动生成回应三个部分，主要包括两个模型：能够理解风险实施者发过来的信息并得到对方意图的交互意图理解模型；以及根据对方意图自动生成对应回复的回应生成模型。
通过交互代理和智能交互模型，与风险实施者的主动交互可基于风险实施者的联系方式来进行。这一过程将在下文中参照图4的示意图和图5的流程图进行描述。
该与风险实施者的主动交互被记录下来并保存为主动交互记录。
在106，处理主动交互记录并提取风险特征。
主动交互记录通常包含完整的风险过程，可以从中发现有价值的风险数据，总结出风险的关键点和显著手法，是下一步分析和防控的基础。
处理主动交互记录可包括格式统一转换、交互过程语义理解和风险过程挖掘。对风险过程的挖掘将获取相关于风险场景的风险特征。风险特征包括风险介质特征、风险行为特征、风险时间特征、风险地域特征等。本领域技术人员可以理解，针对不同风险场景，可提取不同的风险介质特征、风险行为特征、风险时间特征、风险地域特征，还可提供不同的其他风险特征。
处理主动交互记录并提取风险特征的具体过程将在下文中参照图6的示意图和图7的流程图进行描述。
在108，根据风险特征将风险归类。
就电信犯罪行为而言，风险分为内部风险和外部风险。外部风险主要包括当事人侵害、第三方侵害以及洗钱侵害，内部风险主要包括未经授权的行为与侵害。
根据以上所提取的风险特征，可将风险进行归类。风险时间特征和风险地域特征通常有助于结合风险行为特征和风险介质特征来将风险归类。风险介质特征描述在风险发生过程中风险实施者与用户发生信息沟通、资金转移等时所使用的工具和渠道。风险行为特征描述在风险发生过程中风险实施者的行为与正常行为有一定区分度的行为点集合。
当各个风险特征显示风险在系统内部发生，则将风险归类为内部风险；而当各个风险特征显示风险在系统外部发生，则将风险归类为外部风险。
根据风险特征将风险归类的过程将在下文中结合具体示例进行说明。
在110，按照风险的类别对风险进行不同的风险控制。
对于内部风险，基于不同的风险特征，建立有针对性的保护体系。
举例而言，可在系统内的不同层面进行进一步的风险控制。在外部渠道层进一步监控交易发生前的客户接入、会话可疑行为；交易发生中的交易对手是否在可疑名单中。在内部渠道层监控业务违规与可疑操作。在产品服务层监控产品服务内的侵害交易以及跨产品的侵害交易。在数据集成层监控跨产品、渠道的组合/复杂侵害交易。
而对于外部风险，由于存在本系统内无法独立处理的数据，因此将风险数据输出以对外提供风险服务，例如，将可疑电话号码推送给电信运营商、将可疑银行卡/账户推送给相关银行等等。
图2示出根据本公开一实施例的获取潜在风险实施者的联系方式的过程的示意图。
获取潜在风险实施者的联系方式的过程实际上是风险实施者主动识别的过程。
用户在受到侵害前，风险实施者必需通过一定手段与受害者进行沟通，从而逐步达到非法占有受害者钱财的目的，因此沟通是所有风险行为的起点。目前常见的沟通渠道是通过电话号码来打电话或发短信，通过即时通信账号发送文字、图片、视频等，通过网络发布账号互动，以及通过电子邮箱发送链接等等。因此在沟通过程中，需要知道对方的联系方式。通过风险数据的主动识别，可挖掘出风险实施者的联系方式，为主动交互建立数据基础。
在本公开一实施例中，可采用爬虫工具收集风险数据。爬虫工具在开放网络(例如论坛、广告联盟、分类信息网站等)中进行主动采集和挖掘，形成有潜在风险的联系方式集合。一般情况下，风险实施者会在公开场所发布自己的联系方式，等待用户主动与其进行联系。通过爬虫工具，根据风险识别规则和联系方式采集规则，可以定期到各个公开场所采集疑似有风险的联系方式。爬虫采集的是未经处理的原始数据，输出给风险理解平台进行进一步分析。
可以理解，可采用其他工具来收集风险数据。在本公开另一实施例中，可采用第三方提供的疑似风险实施者数据库。本领域技术人员可以理解，可采用各种办法来收集潜在风险实施者的联系方式，以上两个实施例的描述并不构成对本公开技术方案的限定。
风险识别规则和联系方式采集规则可事先设定，或者可逐步学习和积累。风险识别规则和联系方式采集规则可以是简单规则、规则集，也可以是规则树、规则流等，甚至还可以是纳入了自然语言模型和深度学习算法的识别引擎，以针对不同风险场景识别出风险并采集到含有风险信息的相应原始数据。
举例而言，与电信犯罪行为相关地，风险识别规则和联系方式采集规则可设定关联于金融理财相关侵害行为(例如，虚假中奖、奖励、退款；证券相关；信用卡相关；保险相关；保证金相关等)、博彩相关侵害行为、虚假招聘/兼职相关侵害行为、身份冒充相关侵害行为、虚假购物相关侵害行为、网游交易相关侵害行为、和虚拟商品相关侵害行为等等的关键词，依据这些关键词来构建规则并随着电信犯罪行为的变异而适应性地动态更新规则。
在本公开一实施例中，爬虫工具发现一网络发布账号发布针对某地高考生的潜在虚假奖励消息，声称收到大学录取通知书的学生将一次性获得助学金2800元，并留下了电话号码以供符合条件的学子们电话或短信联系。爬虫工具依据关键词“助学金”，采 集到包含网络发布账号、电话号码、“助学金”获取条件以及“助学金金额”等等的原始数据。
风险理解平台被用来处理爬虫工具采集的原始数据。针对含有风险信息的相应原始数据，可进行风险语义理解、风险程度标定和风险实施者联系方式提取。
风险语义理解对爬虫工具采集的原始数据进行语义分析，根据自然语言理解模型及其中设计的参数和阈值，确认是否为风险数据，并对数据做分词、主题特征等预处理。
在虚假奖励的实施例中，风险语义理解对所采集到的“收到大学录取通知书的学生将一次性获得助学金2800元”进行语义分析，基于自然语言理解模型理解存在“收到大学录取通知书的学生”这一条件。进一步地，确认该数据可能为风险数据，并预处理该风险数据，以提取出网络发布账号、电话号码、“助学金”提供机构、“助学金”获取条件以及“助学金金额”等信息。
接着基于语义理解，对不同联系方式做标记，例如通过电话语言沟通、通过短信联系、即时通信应用互动等类型，并对可能涉及的侵害类型作预判，例如虚假奖励、对象为无独立经济能力的学生。
由此，风险程度标定基于语义理解和可能侵害类型，判断该风险可能造成的危害程度，从1级到5级风险程度逐渐增大，作为后面处理优先级的依据。
根据可能危害程度，按优先级进行风险实施者联系方式提取，提取出风险数据中的有价值信息，包括对方的联系方式等信息，从而构成潜在风险实施者集合。
图3示出根据本公开一实施例的获取潜在风险实施者的联系方式的过程300的流程图。
在302，采集与风险相关的原始数据。
在虚假奖励的实施例中，爬虫工具依据关键词“助学金”，采集到包含网络发布账号、电话号码、“助学金”提供机构、“助学金”获取条件以及“助学金金额”等等的原始数据。
在304，对原始数据进行语义分析。
在虚假奖励的实施例中，风险语义理解对所采集到的“收到大学录取通知书的学生将一次性获得助学金2800元”进行语义分析，基于自然语言理解模型理解存在“收到大学录取通知书的学生”这一条件。进一步地，可基于“收到大学录取通知书的学生” 这一条件对该地该年符合条件的高考生人数进行粗略统计或数据提取，并粗略计算总金额。比对该网络发布账号所声称的资助机构的财力，并搜索该资助机构的历史数据。然后，确认该原始数据可能为风险数据，并预处理该风险数据，以提取出网络发布账号、电话号码、“助学金”提供机构、“助学金”获取条件以及“助学金金额”等信息。
在306，根据语义分析结果标定风险程度并确定处理优先级。
根据语义分析结果，对例如通过电话语言沟通、通过短信联系、即时通信应用互动等类型的不同联系方式做标记，并对可能涉及的侵害类型作预判，例如虚假奖励、对象为无独立经济能力的学生。
由此，基于语义理解和可能侵害类型进行风险程度标定，判断该风险可能造成的危害程度，从1级到5级风险程度逐渐增大，作为后续处理的处理优先级。例如，在虚假奖励的本实施例中，将其风险程度标记为4。
当然，本领域技术人员可以理解，风险程度的标定可按不同风控系统而不同，只要其可作为后续处理的优先级提示即可。
在308，按照处理优先级提取风险实施者的联系方式。
根据可能危害程度，按优先级进行风险实施者联系方式提取，提取出风险数据中的有价值信息，包括原始发布途径、对方的联系方式、可能侵害类型等信息，从而构成潜在风险实施者数据集合。该潜在风险实施者数据集合可被存储为结构化数据。
图4示出根据本公开一实施例的与潜在风险实施者的主动交互的过程的示意图。
与潜在风险实施者的主动交互的过程主要由智能交互平台来进行。智能交互平台负责管理全部交互过程，该平台可部署多个交互代理。平台根据风险的不同可能侵害类型将风险分配给不同的交互代理，同时将相应的潜在风险实施者数据作为该交互代理的输入。平台根据潜在风险待处置任务的多少，自动控制同时工作的代理数量、每个代理的工作时长等，并实时监控代理工作情况，生成统计数据。
交互代理是主动出击的实体，其后端连接智能交互模型。依据相连接模型的不同，交互代理分为语音代理、文本代理、即时通信代理、网络发布回应代理等，分别处理电话语音交互、短信交互、即时消息交互、网络发布消息交互等等。交互代理作为消息的中转节点，在智能交互模型和风险实施者之间建立信息通道。每一个交互代理均可以独立完成主动出击的工作。当然，多个交互代理亦可联合进行主动出击，以便于处置潜在的团伙风险实施者。在处理规模上，可以通过部署多个交互代理，来快速提升系统的处 理吞吐量。
智能交互模型具备自动交互能力，具体能力有：在受侵害场景下不被对方识别为机器；支持有倾向性的引导，从而获取希望得到的风险信息；以及支持通过参数调整来扮演不同的人物设定。在主动交互之前或期间，智能交互模型可基于后台数据和运行，为主动交互准备相关数据和信息。
以虚假奖励的实施例为例，智能交互模型可支持例如缺乏社会经验的高中毕业生、相对具备社会经验的家长、以及相对具备社会经验的班主任老师等不同的人物设定，引导潜在风险实施者进一步说明领取助学金的方式，是否涉及对方提供账号、要求学生先行转账等的风险操作；并进一步引导潜在风险实施者说明提供助学金的支助机构的情况。
基于该针对某地高考生的虚假奖励消息，智能交互模型还可查询该地符合“收到大学录取通知书的学生”这一条件的学生总人数，将每人助学金2800元与总人数相乘，确定总金额。由此，来为后续的主动交互做信息准备。基于潜在风险实施者说明的资助机构的情况，智能交互模型可进行查询并在后续进行相关交互，以供验证。
具体地，针对与潜在风险实施者的主动交互过程，智能交互模型可进行交互主题理解、对方意图确认、以及自动生成回应等。
首先进行交互主题理解：在主动联系后接收并分析对方发送的语音、文字、图片或视频，理解当前处于何种交互主题。以虚假奖励的实施例为例，可依据对方发送的信息理解交互主题为“助学金奖励”。
接着进行对方意图确认：对方发出的不同表述的语句，背后包含的意义可能一样，因此需要把对方的信息转换为标准问题，然后识别对方的意图，明白对方的目的。
在本实施例中，面对不同设定的角色，例如学生、家长或老师，对方的表述必然不同。由此可将对方的信息转换成对一些标准问题的回答，诸如，获取助学金的流程是什么？已领取助学金的学生有多少(或者，可领取的学生共有多少)？资助机构是什么性质的机构？学生后续需要满足什么条件或进行怎样的回馈？等等。
然后自动生成回应：基于对方意图，通过自然语言生成模型，加上有相应特点的修饰语气或词，生成对该意图的回应。对于短信交互，生成文本即可；对于语音交互，需要把内容转化为声音文件；对于即时通信交互，需要在所使用的文字、语音或表情符、回应速度等方面保持相应个性或特点；而对于网络发布或电子邮件交互，由于行文会相对比较自由，因此需要保持相应的行文风格。
对于整个主动交互，智能交互平台将进行交互过程记录，例如，语音交互代理与潜在风险实施者的电话沟通的完整过程。该记录包含完整的风险过程，可以从中发现有价值的风险数据，总结出风险的关键点和显著手法，是下一步分析和防控的基础。
图5示出根据本公开一实施例的与潜在风险实施者的主动交互的过程500的流程图。
在502，基于风险实施者的联系方式主动联系风险实施者。
爬虫工具发现一网络发布账号发布针对某地高考生的潜在虚假奖励消息，声称收到大学录取通知书的学生将一次性获得助学金2800元，并留下了电话号码以供符合条件的学子们电话或短信联系。
据此，智能交互平台可基于风险实施者的联系方式主动打电话或发短信联系风险实施者。当然，智能交互平台可按照不同的角色设定来进行该主动交互，例如学生、家长或老师。
在504，接收风险实施者发出的信息。
在506，分析风险实施者发出的信息以识别交互主题。
在接收到风险实施者发出的信息后，分析对方发送的语音、文字、图片或视频，理解当前处于何种交互主题。
在508，基于交互主题确认风险实施者的意图。
基于交互主题即可进行对方意图确认。对方发出的不同表述的语句，背后包含的意义可能一样，因此需要把对方的信息转换为标准问题，然后识别对方的意图，明白对方的目的。
在510，根据风险实施者的意图自动生成回应。
基于对方意图，通过自然语言生成模型，加上有相应特点的修饰语气或词，生成对该意图的回应。
对于短信交互，生成文本即可；对于语音交互，需要把内容转化为声音文件；对于即时通信交互，需要在所使用的文字、语音或表情符、回应速度等方面保持相应个性或特点；而对于网络发布或电子邮件交互，由于行文会相对比较自由，因此需要保持相应的行文风格。
图6示出根据本公开另一实施例的进行风险过程挖掘并提取风险特征的过程的 示意图。
进行风险过程挖掘并提取风险特征的过程由风险数据处理平台执行。针对主动交互过程记录，风险数据处理平台可进行多媒体格式转换、交互记录语义分析、风险过程挖掘、风险特征提取和风险分类标记。
需要多媒体格式统一转换是由于主动交互过程通过电话语音、短信/彩信、即时通信(包括文本、语音、视频、表情符等)以及网络发布交互进行，为后期处理方便起见，需进行统一转换为文本，包括语音内容识别和图片内容识别。
以虚假奖励的实施例为例，智能交互模型所支持的学生、家长、以及老师等不同的人物设定进行的主动交互主要通过电话和短信进行，因此，电话语音和短信文字将统一转换为文字，其中包括语音内容识别。
交互过程语义理解基于自然语言处理模型进行，以便于理解风险发生过程。在虚假奖励的实施例中，交互主题为“助学金奖励”，由此关注点将落到进一步领取助学金的方式，是否涉及对方提供账号、要求学生先行转账等风险操作，以及提供助学金的支助机构等等。
由此接着进行风险过程挖掘。风险过程可以通过一系列的行为点进行刻画，得到风险行为与正常行为有一定区分度的行为点集合。在虚假奖励的实施例中，正常操作应当是学生提供账号、由提供助学金的支助机构转账至该账号；由此，当在主动交互过程中，出现了对方提供账号、要求学生先行转账的行为点时，即为风险行为点。此外，在主动交互过程中，当问及奖励总规模时，正常回应应当是符合统计数据的相应量级，当出现不符合相应量级的回应时即为风险行为点。当然，可以理解，还存在其他风险行为点，在此不再赘述。
基于风险过程挖掘，即可提取风险特征，例如风险时间特征、风险地域特征、风险行为特征、以及风险介质特征。在虚假奖励的实施例中，风险时间特征为该网络发布消息发布日期以及后续跟帖时间的时间链。风险地域特征为网络发布消息中声称的地域，同时在交互过程中可进行电话号码画像，比对该电话号码现方位与消息中声称地域。风险行为特征则是所刻画的行为点形成的风险行为点集合。风险介质特征是在风险行为发生过程中，风险实施者与用户发生信息沟通、资金转移等时所使用的工具、渠道等。对方提供的银行账号或支付APP的账号为资金转移时的工具或渠道。
基于风险特征，可将风险归类并标记。当各个风险特征显示风险在系统内部发 生时，将风险归类为内部风险；而当各个风险特征显示风险在系统外部发生时，将风险归类为外部风险。
在虚假奖励的实施例中，风险介质特征显示风险侵害者使用的资金转移渠道在系统外部，由此将风险归类为外部风险。
主动交互记录的处理使得有价值信息点通过语义分析来识别，从而进一步地能够得到常用风险手法，并能及时发现新的风险手法，识别出其中的关键行为特征。
图7示出根据本公开另一实施例的进行风险过程挖掘并提取风险特征的过程700的示意图。
在702，将主动交互记录进行多媒体格式转换。
将主动交互记录统一转换为文本，其中包括语音内容识别和图片内容识别。
在704，对转换后的主动交互记录进行语义分析。
语义分析基于自然语言处理模型针对转换后的主动交互记录进行，以便于理解风险发生过程。
在706，基于语义分析结果进行风险过程挖掘以提取风险特征。
风险过程可以通过一系列的行为点进行刻画，得到风险行为与正常行为有一定区分度的行为点集合。基于风险过程挖掘，即可提取风险特征，例如风险介质特征、风险行为特征、风险时间特征以及风险地域特征。
基于智能交互的主动风控系统
图8示出根据本公开一实施例的基于智能交互的主动风控系统800的框图。
系统800包括获取模块802、主动交互模块804、特征提取模块806以及归类控制模块808。
获取模块802获取风险实施者的联系方式。
风险实施者想要对用户实施侵害，其必定需要通过一定的手段与受害者进行沟通，达到非法占有受害者钱财的目的，因此沟通是所有风险行为的起点。
一般情况下，风险实施者会在公开场所发布自己的联系方式，等待用户主动与其进行联系。就电信犯罪行为而言，风险实施者的联系方式包括电话号码、即时通信账号、网络发布账号以及电子邮箱等等。
在本公开一实施例中，采用爬虫工具来收集潜在风险实施者的联系方式。在本公开另一实施例中，可采用第三方提供的疑似风险实施者数据库。本领域技术人员可以理解，可采用各种办法来收集潜在风险实施者的联系方式，以上两个实施例的描述并不构成对本公开技术方案的限定。
主动交互模块804基于风险实施者的联系方式进行与风险实施者的主动交互并生成主动交互记录。
基于风险实施者的联系方式，能够进行大规模的主动出击。在本公开一实施例中，主动交互模块804采用交互代理在智能交互模型和风险实施者之间建立信息通道，作为消息的中转节点。主动交互模块804所采用的智能交互模型则具备自动交互能力，能够不被对方识别为机器且支持针对特定场景的有倾向性的引导，从而获取希望得到的风险信息。交互代理还支持通过参数调整来扮演不同的人物设定。智能交互模型包含交互主题理解、对方意图确认、自动生成回应三个部分，主要包括两个模型：能够理解风险实施者发过来的信息并得到对方意图的交互意图理解模型；以及根据对方意图自动生成对应回复的回应生成模型。该与风险实施者的主动交互被记录下来并保存为主动交互记录。
特征提取模块806处理主动交互记录并提取风险特征。
主动交互记录通常包含完整的风险过程，可以从中发现有价值的风险数据，总结出风险的关键点和显著手法，是下一步分析和防控的基础。处理主动交互记录可包括格式统一转换、交互过程语义理解和风险过程挖掘。对风险过程的挖掘将获取相关于风险场景的风险特征。风险特征包括风险介质特征、风险行为特征、风险时间特征、风险地域特征等。本领域技术人员可以理解，针对不同风险场景，可提取不同的风险介质特征、风险行为特征、风险时间特征、风险地域特征，还可提供不同的其他风险特征。
归类控制模块808根据风险特征将风险归类。
根据所提取的风险特征，可将风险进行归类。风险时间特征和风险地域特征通常有助于结合风险行为特征和风险介质特征来将风险归类。风险介质特征描述在风险发生过程中风险实施者与用户发生信息沟通、资金转移等时所使用的工具和渠道。风险行为特征描述在风险发生过程中风险实施者的行为与正常行为有一定区分度的行为点集合。
当各个风险特征显示风险在系统内部发生，则将风险归类为内部风险；而当各 个风险特征显示风险在系统外部发生，则将风险归类为外部风险。
进一步，归类控制模块808按照风险的类别对风险进行不同的风险控制。
对于内部风险，基于不同的风险特征，建立有针对性的保护体系。而对于外部风险，由于存在本系统内无法独立处理的数据，因此将风险数据输出以对外提供风险服务。
本公开提出了一种用于基于智能交互的主动风险控制方案，此方案能够主动出击，事前识别。传统上的风险信息是通过用户举报获得，即在风险发生后获取。同时，此时获取的介质由于已经作案成功，生命周期很短，很难在后面的防控中发挥较大作用。而本公开的技术方案通过主动出击，可以在风险发生前获得相关信息，即在没有实际损失前获取，做到提前防控。
本方案还具备智能交互的能力，通过使用智能交互的模型来代替人工主动与风险实施者通过各种联系方式交流，可以实现全天候的高效率工作。同时，通过交互代理的设计，可以无限的扩展机器并行工作能力，提升处理吞吐量。
本方案还能够实现自动化的风险控制。基于所获得的风险介质和风险过程，可以自动进入到后面的处理平台、自动部署相关决策动作，从而大大提升风险防御的效果。
以上描述的基于智能交互的主动风控方法和系统的各个步骤和模块可以用硬件、软件、或其组合来实现。如果在硬件中实现，结合本发明描述的各种说明性步骤、模块、以及电路可用通用处理器、数字信号处理器(DSP)、专用集成电路(ASIC)、现场可编程门阵列(FPGA)、或其他可编程逻辑组件、硬件组件、或其任何组合来实现或执行。通用处理器可以是处理器、微处理器、控制器、微控制器、或状态机等。如果在软件中实现，则结合本发明描述的各种说明性步骤、模块可以作为一条或多条指令或代码存储在计算机可读介质上或进行传送。实现本发明的各种操作的软件模块可驻留在存储介质中，如RAM、闪存、ROM、EPROM、EEPROM、寄存器、硬盘、可移动盘、CD-ROM、云存储等。存储介质可耦合到处理器以使得该处理器能从/向该存储介质读写信息，并执行相应的程序模块以实现本发明的各个步骤。而且，基于软件的实施例可以通过适当的通信手段被上载、下载或远程地访问。这种适当的通信手段包括例如互联网、万维网、内联网、软件应用、电缆(包括光纤电缆)、磁通信、电磁通信(包括RF、微波和红外通信)、电子通信或者其他这样的通信手段。
还应注意，这些实施例可能是作为被描绘为流程图、流图、结构图、或框图的 过程来描述的。尽管流程图可能会把诸操作描述为顺序过程，但是这些操作中有许多操作能够并行或并发地执行。另外，这些操作的次序可被重新安排。
所公开的方法、装置和系统不应以任何方式被限制。相反，本发明涵盖各种所公开的实施例(单独和彼此的各种组合和子组合)的所有新颖和非显而易见的特征和方面。所公开的方法、装置和系统不限于任何具体方面或特征或它们的组合，所公开的任何实施例也不要求存在任一个或多个具体优点或者解决特定或所有技术问题。
上面结合附图对本发明的实施例进行了描述，但是本发明并不局限于上述的具体实施方式，上述的具体实施方式仅仅是示意性的，而不是限制性的，本领域的普通技术人员在本发明的启示下，在不脱离本发明宗旨和权利要求所保护的范围情况下，还可做出很多更改，这些均落在本发明的保护范围之内。

Claims (15)

  1. 一种基于智能交互的主动风控方法，包括：
    获取风险实施者的联系方式；
    基于所述风险实施者的联系方式进行与所述风险实施者的主动交互并生成主动交互记录；
    处理所述主动交互记录并提取风险特征；
    根据所述风险特征将风险归类；以及
    按照所述风险的类别对所述风险进行不同的风险控制。
  2. 如权利要求1所述的方法，其特征在于，获取风险实施者的联系方式进一步包括：
    采集与所述风险相关的原始数据；
    对原始数据进行语义分析；
    根据语义分析结果标定风险程度并确定处理优先级；以及
    按照所述处理优先级提取所述风险实施者的联系方式。
  3. 如权利要求1所述的方法，其特征在于，所述风险实施者的联系方式包括电话号码、即时通信账号、网络发布账号以及电子邮箱。
  4. 如权利要求1所述的方法，其特征在于，基于所述风险实施者的联系方式进行与所述风险实施者的主动交互进一步包括：
    基于所述风险实施者的联系方式主动联系所述风险实施者；
    接收所述风险实施者发出的信息；
    分析所述风险实施者发出的信息以识别交互主题；
    基于所述交互主题确认所述风险实施者的意图；以及
    根据所述风险实施者的意图自动生成回应。
  5. 如权利要求1所述的方法，其特征在于，基于所述风险实施者的联系方式进行与所述风险实施者的主动交互通过交互代理进行。
  6. 如权利要求1所述的方法，其特征在于，处理主动交互记录并提取风险特征进一步包括：
    将所述主动交互记录进行多媒体格式转换；
    对转换后的所述主动交互记录进行语义分析；以及
    基于语义分析结果进行风险过程挖掘以提取风险特征。
  7. 如权利要求6所述的方法，其特征在于，所述风险特征包括风险介质特征、风险行为特征、风险时间特征以及风险地域特征。
  8. 一种基于智能交互的主动风控系统，包括：
    获取模块，获取风险实施者的联系方式；
    主动交互模块，基于所述风险实施者的联系方式进行与所述风险实施者的主动交互并生成主动交互记录；
    特征提取模块，处理所述主动交互记录并提取风险特征；以及
    归类控制模块，根据所述风险特征将风险归类，并按照所述风险的类别对所述风险进行不同的风险控制。
  9. 如权利要求8所述的系统，其特征在于，所述获取模块获取风险实施者的联系方式进一步包括：
    采集与所述风险相关的原始数据；
    对原始数据进行语义分析；
    根据语义分析结果标定风险程度并确定处理优先级；以及
    按照所述处理优先级提取所述风险实施者的联系方式。
  10. 如权利要求9所述的系统，其特征在于，所述风险实施者的联系方式包括电话号码、即时通信账号、网络发布账号以及电子邮箱。
  11. 如权利要求9所述的系统，其特征在于，所述主动交互模块基于所述风险实施者的联系方式进行与所述风险实施者的主动交互进一步包括：
    基于所述风险实施者的联系方式主动联系所述风险实施者；
    接收所述风险实施者发出的信息；
    分析所述风险实施者发出的信息以识别交互主题；
    基于所述交互主题确认所述风险实施者的意图；以及
    根据所述风险实施者的意图自动生成回应。
  12. 如权利要求9所述的系统，其特征在于，所述主动交互模块基于所述风险实施者的联系方式进行与所述风险实施者的主动交互通过所述主动交互模块中的交互代理进行。
  13. 如权利要求8所述的系统，其特征在于，所述特征提取模块处理所述主动交互记录并提取风险特征进一步包括：
    将所述主动交互记录进行多媒体格式转换；
    对转换后的所述主动交互记录进行语义分析；以及
    基于语义分析结果进行风险过程挖掘以提取风险特征。
  14. 如权利要求13所述的系统，其特征在于，所述风险特征包括风险介质特征、风险行为特征、风险时间特征以及风险地域特征。
  15. 一种存储有指令的计算机可读存储介质，当所述指令被执行时使得机器执行如权利要求1-8中任一项所述的方法。
PCT/CN2020/071590 2019-08-07 2020-01-11 基于智能交互的主动风控方法和系统 WO2021022790A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US16/820,884 US11086991B2 (en) 2019-08-07 2020-03-17 Method and system for active risk control based on intelligent interaction

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910726858.9A CN110598982B (zh) 2019-08-07 2019-08-07 基于智能交互的主动风控方法和系统
CN201910726858.9 2019-08-07

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US16/820,884 Continuation US11086991B2 (en) 2019-08-07 2020-03-17 Method and system for active risk control based on intelligent interaction

Publications (1)

Publication Number Publication Date
WO2021022790A1 true WO2021022790A1 (zh) 2021-02-11

Family

ID=68853746

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/071590 WO2021022790A1 (zh) 2019-08-07 2020-01-11 基于智能交互的主动风控方法和系统

Country Status (3)

Country Link
CN (1) CN110598982B (zh)
TW (1) TWI807172B (zh)
WO (1) WO2021022790A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115086046A (zh) * 2022-06-20 2022-09-20 支付宝(杭州)信息技术有限公司 智能交互的安全部署方法和系统

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110598982B (zh) * 2019-08-07 2022-02-22 创新先进技术有限公司 基于智能交互的主动风控方法和系统
US11086991B2 (en) 2019-08-07 2021-08-10 Advanced New Technologies Co., Ltd. Method and system for active risk control based on intelligent interaction
CN112862322A (zh) * 2021-02-10 2021-05-28 支付宝(杭州)信息技术有限公司 通用交互式风控方法和装置
TWI817106B (zh) * 2021-04-14 2023-10-01 台達電子工業股份有限公司 查詢回饋裝置以及方法

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106970911A (zh) * 2017-03-28 2017-07-21 广州中国科学院软件应用技术研究所 一种基于大数据和机器学习的防范电信诈骗系统及方法
CN107147621A (zh) * 2017-04-20 2017-09-08 微医集团（浙江）有限公司 互联网医疗黄牛风险控制的实现方法
US20170286653A1 (en) * 2012-06-29 2017-10-05 Microsoft Technology Licensing, Llc. Identity risk score generation and implementation
CN109146670A (zh) * 2018-08-27 2019-01-04 深圳前海微众银行股份有限公司 贷款反欺诈处理方法、装置及可读存储介质
CN109255697A (zh) * 2018-08-15 2019-01-22 普信恒业科技发展（北京）有限公司 一种基于人工智能的自动信用评估方法和系统
CN110598982A (zh) * 2019-08-07 2019-12-20 阿里巴巴集团控股有限公司 基于智能交互的主动风控方法和系统

Family Cites Families (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2005017802A2 (en) * 2003-08-15 2005-02-24 Providus Software Solutions, Inc. Risk mitigation and management
US9213827B2 (en) * 2012-09-27 2015-12-15 Intel Corporation Security data aggregation and business intelligence for web applications
US20140143010A1 (en) * 2012-11-16 2014-05-22 SPF, Inc. System and Method for Assessing Interaction Risks Potentially Associated with Transactions Between a Client and a Provider
CN106550155B (zh) * 2016-11-25 2019-05-17 上海欣方智能系统有限公司 对可疑号码进行诈骗样本甄别归类及拦截的方法及系统
CN107870994A (zh) * 2017-10-31 2018-04-03 北京光年无限科技有限公司 用于智能机器人的人机交互方法及系统
TWM557407U (zh) * 2017-11-23 2018-03-21 Ctbc Bank Co Ltd 信用審核系統
CN109376999A (zh) * 2018-09-20 2019-02-22 阿里巴巴集团控股有限公司 一种交易的管控方法、装置及设备
CN108989581B (zh) * 2018-09-21 2022-03-22 中国银行股份有限公司 一种用户风险识别方法、装置及系统
TWM577148U (zh) * 2019-01-03 2019-04-21 兆豐金融控股股份有限公司 評估金融風險的電子裝置
CN110046902A (zh) * 2019-01-15 2019-07-23 阿里巴巴集团控股有限公司 风险交易处理方法、装置及设备

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170286653A1 (en) * 2012-06-29 2017-10-05 Microsoft Technology Licensing, Llc. Identity risk score generation and implementation
CN106970911A (zh) * 2017-03-28 2017-07-21 广州中国科学院软件应用技术研究所 一种基于大数据和机器学习的防范电信诈骗系统及方法
CN107147621A (zh) * 2017-04-20 2017-09-08 微医集团（浙江）有限公司 互联网医疗黄牛风险控制的实现方法
CN109255697A (zh) * 2018-08-15 2019-01-22 普信恒业科技发展（北京）有限公司 一种基于人工智能的自动信用评估方法和系统
CN109146670A (zh) * 2018-08-27 2019-01-04 深圳前海微众银行股份有限公司 贷款反欺诈处理方法、装置及可读存储介质
CN110598982A (zh) * 2019-08-07 2019-12-20 阿里巴巴集团控股有限公司 基于智能交互的主动风控方法和系统

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115086046A (zh) * 2022-06-20 2022-09-20 支付宝(杭州)信息技术有限公司 智能交互的安全部署方法和系统
CN115086046B (zh) * 2022-06-20 2024-01-12 支付宝(杭州)信息技术有限公司 智能交互的安全部署方法和系统

Also Published As

Publication number Publication date
TW202107382A (zh) 2021-02-16
CN110598982B (zh) 2022-02-22
CN110598982A (zh) 2019-12-20
TWI807172B (zh) 2023-07-01

Similar Documents

Publication Publication Date Title
WO2021022790A1 (zh) 基于智能交互的主动风控方法和系统
Wang et al. Is this the era of misinformation yet: combining social bots and fake news to deceive the masses
US11086991B2 (en) Method and system for active risk control based on intelligent interaction
CN109214914A (zh) 一种基于通信开放平台的借贷信息审核方法及装置
CN106575327A (zh) 分析面部识别数据和社交网络数据以供用户鉴别
CN106453061A (zh) 一种识别网络诈骗行为的方法及系统
CN110880142B (zh) 一种风险实体获取方法及装置
US20220300977A1 (en) Real-time malicious activity detection using non-transaction data
Marzuoli et al. Uncovering the landscape of fraud and spam in the telephony channel
Bilz et al. Tainted love: A systematic literature review of online romance scam research
CN111062422B (zh) 一种套路贷体系化识别方法及装置
CN109711984B (zh) 一种基于催收的贷前风险监控方法及装置
Gangula et al. Enhanced Detection of Social Bots on Online Platforms using Semi-Supervised K-Means Clustering
CN116645134A (zh) 一种信用卡分期的推荐方法、装置、设备及介质
US20210357942A1 (en) Method and apparatus for identifying risky vertices
CN110766091B (zh) 一种套路贷团伙的识别方法及系统
Kim-Leffingwell et al. Money backfires: How Chinese investment fuels anti-China protests abroad
CN114363839A (zh) 一种诈骗数据的预警方法、装置、设备及存储介质
Idrees et al. Factors Affecting the Adoption of Branchless Banking in Pakistan
Ogala et al. Detecting Telecoms Fraud in a Cloud-Base Environment by Analyzing the Content of a Phone Conversation
CN117272201B (zh) 一种基于4w1h语言模型的金融行为异常检测方法及系统
Ghoshal Fake users and reviewers detection system
Desai et al. BIG DATA: TRANSFORMING EVIDENCE BASED POLICY MAKING
US20230139465A1 (en) Electronic service filter optimization
US11915313B2 (en) Using email history to estimate creditworthiness for applicants having insufficient credit history

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20849234

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20849234

Country of ref document: EP

Kind code of ref document: A1